diff --git a/MindFlow/NOTICE b/MindFlow/NOTICE index 9547ae1e81c207ceecc74e1af31f5841d77ab4f6..ac3f62273b55a828c5f64b26048c4267b459c0aa 100644 --- a/MindFlow/NOTICE +++ b/MindFlow/NOTICE @@ -1,3 +1,3 @@ MindSpore MindFlow -Copyright 2019-2022 Huawei Technologies Co., Ltd +Copyright 2019-2025 Huawei Technologies Co., Ltd diff --git a/MindFlow/README.md b/MindFlow/README.md index 4e8c6ccdbb519b58ef9b836068f43c39ffa09c50..b63ea25179723913db15d37bd50aa3d8fde182c4 100644 --- a/MindFlow/README.md +++ b/MindFlow/README.md @@ -8,22 +8,27 @@ [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg?style=flat)](https://gitee.com/mindspore/mindscience/pulls) [![LICENSE](https://img.shields.io/github/license/mindspore-ai/mindspore.svg?style=flat)](https://github.com/mindspore-ai/mindspore/blob/master/LICENSE) -# **MindFlow** +# MindFlow -## **MindFlow介绍** +## MindFlow介绍 流体仿真是指通过数值计算对给定边界条件下的流体控制方程进行求解，从而实现流动的分析、预测和控制，其在航空航天、船舶制造以及能源电力等行业领域的工程设计中应用广泛。传统流体仿真的数值方法如有限体积、有限差分等，主要依赖商业软件实现，需要进行物理建模、网格划分、数值离散、迭代求解等步骤，仿真过程较为复杂，计算周期长。AI具备强大的学习拟合和天然的并行推理能力，可以有效地提升流体仿真效率。 -MindFlow是基于[昇思MindSpore](https://www.mindspore.cn/)开发的流体仿真领域套件，支持航空航天、船舶制造以及能源电力等行业领域的AI流场模拟，旨在于为广大的工业界科研工程人员、高校老师及学生提供高效易用的AI计算流体仿真软件。 +MindFlow是基于[昇思MindSpore](https://www.mindspore.cn/)开发的流体仿真领域套件，支持航空航天、船舶制造以及能源电力等行业领域的AI流场仿真、AI气动外形设计、AI流动控制，旨在于为广大的工业界科研工程人员、高校老师及学生提供高效易用的AI计算流体仿真软件。
MindFlow Architecture
-## **最新消息** +## 最新消息 -- 🔥`2024.03.22` 以“为智而昇，思创之源”为主题的昇思人工智能框架峰会2024在北京国家会议中心召开，北京国际数学研究中心教授、国际机器学习研究中心副主任董彬介绍，基于MindSpore和MindFLow套件，团队打造了AI解偏微分方程领域的基础模型PDEformer-1，能够直接接受任意形式PDE作为输入，通过在包含300万条一维PDE样本的庞大数据集上进行训练，PDEformer-1展现出了对广泛类型的一维PDE正问题的迅速且精准求解能力。 -- 🔥`2024.03.22`以“为智而昇，思创之源”为主题的昇思人工智能框架峰会2024在北京国家会议中心召开，中国科学院院士、中国空气动力学会理事长唐志共介绍，基于昇思MindSpore和MindFlow套件，团队首创了生成式气动设计大模型平台，面向多种应用场景，打破传统设计范式，将设计时长由月级缩短到分钟级，满足概念设计要求[相关新闻](https://tech.cnr.cn/techph/20240323/t20240323_526636454.shtml)。 +- 🔥`2025.03.30` MindFlow 0.3.0版本发布，详见[MindFlow 0.3.0](RELEASE_CN.md)。 +- 🔥`2024.11.04` 2024科学智能峰会由北京大学计算机学院、北京科学智能研究院主办。会上，北京大学博雅特聘教授、北京国际数学研究中心教授、国际机器学习研究中心副主任董彬介绍，基于MindSpore和MindFlow套件，发布可以直接处理任意 PDE 形式的端到端解预测模型PDEformer-2，同时适用于含时与不含时的方程。模型使用约 40TB 的数据集进行预训练，能够对具有不同边界条件、求解区域、变量个数的二维方程直接推理，快速获得任意时空位置的预测解。此外，PDEformer-2 作为正问题解算子的可微分代理模型，也可以用于求解各类反问题，包括基于有噪声的时空散点观测进行全波反演以恢复方程中的波速场等。这为模型支持包括流体、电磁等领域的众多物理现象及工程应用的研究打下良好基础。[相关新闻](https://www.mindspore.cn/news/newschildren?id=3481&type=news) +- 🔥`2024.10.13` 智能流体力学产业联合体第三次全体会议在陕西西安索菲特人民大厦成功举办，产业联合体代表及关心联合体的学术界、产业界专家共计超过两百位嘉宾现场参会。会上专家分享了《AI流体仿真及MindSpore实践》报告，介绍了昇思MindSpore AI框架使能大模型全流程开发的能力及MindSpore Flow流体仿真套件相关进展，同时，也介绍了与产业联合体的主要伙伴们的联合创新成果，展示了AI+流体力学在大飞机、气动外形设计等国计民生场景的应用实践。[相关新闻](https://www.mindspore.cn/news/newschildren?id=3424&type=news) +- 🔥`2024.09.23` “风雷”气动外形设计大模型平台在四川绵阳举办的“智能赋能空天科技创新”博士后学术交流活动上发布。“风雷”大模型平台由中国空气动力研究与发展中心基于昇思MindSpore AI框架及MindSpore Flow流体力学套件研制，用于辅助设计人员在飞行器概念设计阶段开展气动外形生成。“风雷”可实现满足性能指标的气动外形端到端生成，适配多场景、多类型气动外形设计，且设计方案覆盖多样性需求。学术交流活动上，中国科学院院士唐志共向全体参会专家介绍了风雷大模型的技术框架和应用案例，他表示：“AI给空气动力学提供了新的研究范式，为学科发展和空天科技创新发展注入了新的活力，生成式气动外形设计平台加速了气动外形概念设计，可助力设计范式智能化转型”。[相关新闻](https://www.mindspore.cn/news/newschildren?id=3405&type=news) +- 🔥`2024.07.04` 以“以共商促共享 以善治促善智”为主题的2024世界人工智能大会在上海召开。会上，基于昇思MindSpore框架打造的南方电网公司研发成果“驭电”智能仿真大模型获得最高奖SAIL大奖。“驭电大模型既能精准刻画新型电力系统的安全边界，又能精细安排各类电源的发电计划，确保大电网安全的前提下，动态优化电网运行方式，解决新能源变化无常、难以计划带来的难题，最大限度提高新能源的利用率。”南方电网公司战略规划部总经理郑外生介绍。[相关新闻](https://business.cctv.com/2024/07/04/ARTICo0MOGKfEyWdRf3QTyGo240704.shtml) +- 🔥`2024.03.22` 以“为智而昇，思创之源”为主题的昇思人工智能框架峰会2024在北京国家会议中心召开，北京国际数学研究中心教授、国际机器学习研究中心副主任董彬介绍，基于MindSpore和MindFlow套件，团队打造了AI解偏微分方程领域的基础模型PDEformer-1，能够直接接受任意形式PDE作为输入，通过在包含300万条一维PDE样本的庞大数据集上进行训练，PDEformer-1展现出了对广泛类型的一维PDE正问题的迅速且精准求解能力。 +- 🔥`2024.03.22` 以“为智而昇，思创之源”为主题的昇思人工智能框架峰会2024在北京国家会议中心召开，中国科学院院士、中国空气动力学会理事长唐志共介绍，基于昇思MindSpore和MindFlow套件，团队首创了生成式气动设计大模型平台，面向多种应用场景，打破传统设计范式，将设计时长由月级缩短到分钟级，满足概念设计要求[相关新闻](https://tech.cnr.cn/techph/20240323/t20240323_526636454.shtml)。 - 🔥`2024.03.20` MindFlow 0.2.0版本发布，详见[MindFlow 0.2.0](RELEASE_CN.md)。 -- 🔥`2023.11.04`中国(西安)人工智能高峰论坛在西安市雁塔区高新国际会议中心召开，由西北工业大学与华为联合研发的首个面向飞行器的流体力学大模型“秦岭·翱翔”正式发布。该模型是西工大流体力学智能化国际联合研究所携手华为AI4Sci Lab在国产开源流体计算软件风雷的基础上，依托昇腾AI澎湃算力及昇思MindSpore AI框架共同研发的面向飞行器流体仿真的智能化模型，[相关新闻](https://mp.weixin.qq.com/s/Rhpiyf3VJYm_lMBWTRDtGA)。 +- 🔥`2023.11.04` 中国(西安)人工智能高峰论坛在西安市雁塔区高新国际会议中心召开，由西北工业大学与华为联合研发的首个面向飞行器的流体力学大模型“秦岭·翱翔”正式发布。该模型是西工大流体力学智能化国际联合研究所携手华为AI4Sci Lab在国产开源流体计算软件风雷的基础上，依托昇腾AI澎湃算力及昇思MindSpore AI框架共同研发的面向飞行器流体仿真的智能化模型，[相关新闻](https://mp.weixin.qq.com/s/Rhpiyf3VJYm_lMBWTRDtGA)。 - 🔥`2023.08.02` MindFlow 0.1.0版本发布，详见[MindFlow 0.1.0](https://mindspore.cn/mindflow/docs/zh-CN/r0.1/index.html)。 - 🔥`2023.07.06` 以“智联世界 生成未来”为主题的2023世界人工智能大会在上海世博中心开幕，来自中国商用飞机有限责任公司上海飞机设计研究院的三维超临界机翼流体仿真重器“东方.翼风”获得世界人工智能大会最高奖项——SAIL奖，该模型是由中国商用飞机有限责任公司上海飞机设计研究院与华为基于国产昇腾AI基础软硬件平台及昇思MindSpore AI框架研发的面向机翼复杂流动仿真场景的智能化模型，[相关新闻](https://www.thepaper.cn/newsDetail_forward_23769936)。 - 🔥`2023.05.21` 智能流体力学产业联合体第二次全体会议在杭州西湖大学成功举办，昇思MindSpore协办本次会议，三位中国科学院院士、产业联合体代表及关心联合体的学术界、产业界专家共计百位嘉宾现场参会。面向飞行器的首个流体力学大模型————“秦岭·翱翔”大模型预发布，该模型是由西北工业大学流体力学智能化国际联合研究所与华为基于国产昇腾AI基础软硬件平台及昇思MindSpore AI框架，共同研发的面向飞行器流体仿真的智能化模型，[相关新闻](http://science.china.com.cn/2023-05/23/content_42378458.htm)。 @@ -33,27 +38,37 @@ MindFlow是基于[昇思MindSpore](https://www.mindspore.cn/)开发的流体仿 ## 论文 -Ye Z, Huang X, Liu H, et al. Meta-Auto-Decoder: A Meta-Learning Based Reduced Order Model for Solving Parametric Partial Differential Equations[J]. Communications on Applied Mathematics and Computation. [[Paper]](https://link.springer.com/article/10.1007/s42967-023-00293-7) +[2024] Li X, Deng Z, Feng R, et al. Deep learning-based reduced order model for three-dimensional unsteady flow using mesh transformation and stitching[J]. Computers & Fluids. [[Paper]](https://arxiv.org/pdf/2307.07323) -Deng Z, Wang J, Liu H, et al. Prediction of transactional flow over supercritical airfoils using geometric-encoding and deep-learning strategies. Physics of Fluids 35, 075146 (2023). [[Paper]](https://pubs.aip.org/aip/pof/article-abstract/35/7/075146/2903765/Prediction-of-transonic-flow-over-supercritical?redirectedFrom=fulltext) +[2024] Wang Q, Ren P, Zhou H, et al. P $^ 2$ C $^ 2$ Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics[J]. arXiv preprint.[[Paper]](https://arxiv.org/pdf/2411.00040) + +[2024] Zeng B, Wang Q, Yan M, et al. PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems[J]. arXiv preprint. [[Paper]](https://arxiv.org/pdf/2410.01337) + +[2024] Ye Z, Huang X, Chen L, et al. Pdeformer-1: A foundation model for one-dimensional partial differential equations[J]. arXiv preprint. [[Paper]](https://arxiv.org/pdf/2407.06664) + +[2024] Li Z, Wang Y, Liu H, et al. Solving the boltzmann equation with a neural sparse representation[J]. SIAM Journal on Scientific Computing. [[Paper]](https://arxiv.org/pdf/2302.09233) + +[2024] Ye Z, Huang X, Chen L, et al. Pdeformer: Towards a foundation model for one-dimensional partial differential equations[J]. arXiv preprint.[[Paper](https://arxiv.org/abs/2402.12652)] + +[2024] Ye Z, Huang X, Liu H, et al. Meta-Auto-Decoder: A Meta-Learning Based Reduced Order Model for Solving Parametric Partial Differential Equations[J]. Communications on Applied Mathematics and Computation. [[Paper]](https://link.springer.com/article/10.1007/s42967-023-00293-7) + +[2024] Li Z, Wang Y, Liu H, et al. Solving Boltzmann equation with neural sparse representation[J]. SIAM Journal on Scientific Computing. [[Paper]](https://epubs.siam.org/doi/abs/10.1137/23M1558227?journalCode=sjoce3) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) + +[2023] Deng Z, Wang J, Liu H, et al. Prediction of transactional flow over supercritical airfoils using geometric-encoding and deep-learning strategies[J]. Physics of Fluids. [[Paper]](https://pubs.aip.org/aip/pof/article-abstract/35/7/075146/2903765/Prediction-of-transonic-flow-over-supercritical?redirectedFrom=fulltext) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady) -Rao C, Ren P, Wang Q, et al. Encoding physics to learn reaction–diffusion processes[J]. Nature Machine Intelligence, 2023: 1-15. [[Paper]](https://arxiv.org/abs/2106.04781) +[2023] Rao C, Ren P, Wang Q, et al. Encoding physics to learn reaction–diffusion processes[J]. Nature Machine Intelligence. [[Paper]](https://arxiv.org/abs/2106.04781) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn) -Li Z, Wang Y, Liu H, et al. Solving Boltzmann equation with neural sparse representation[J]. SIAM Journal on Scientific Computing, Vol. 46, Iss. 2 (2024). -[[Paper]](https://epubs.siam.org/doi/abs/10.1137/23M1558227?journalCode=sjoce3) -[[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) - -Deng Z, Liu H, Shi B, et al. Temporal predictions of periodic flows using a mesh transformation and deep learning-based strategy[J]. Aerospace Science and Technology, 2023, 134: 108081. [[Paper]](https://www.sciencedirect.com/science/article/pii/S1270963822007556) +[2023] Deng Z, Liu H, Shi B, et al. Temporal predictions of periodic flows using a mesh transformation and deep learning-based strategy[J]. Aerospace Science and Technology. [[Paper]](https://www.sciencedirect.com/science/article/pii/S1270963822007556) -Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Differential Equations with a Point Source[C]//IJCAI. 2022: 3839-3846. [[Paper]](https://gitee.com/link?target=https%3A%2F%2Fwww.ijcai.org%2Fproceedings%2F2022%2F0533.pdf) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) +[2022] Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Differential Equations with a Point Source[C]. IJCAI. [[Paper]](https://gitee.com/link?target=https%3A%2F%2Fwww.ijcai.org%2Fproceedings%2F2022%2F0533.pdf) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) ## 特性 -- [MindSpore自动微分详解](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/features/mindspore_grad_cookbook.ipynb) +[MindSpore自动微分详解](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/features/mindspore_grad_cookbook.ipynb) -- [基于MindFlow求解PINNs问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/features/solve_pinns_by_mindflow) +[基于MindFlow求解PINNs问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/features/solve_pinns_by_mindflow) ## 应用案例 @@ -64,9 +79,12 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe |[东方.御风](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady) | [二维翼型流场数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/airfoil/2D_steady/) | ViT | ✔️ | ✔️ | |[FNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/fno1d) | [一维Burgers方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | FNO1D | ✔️ | ✔️ | |[KNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/kno1d) | [一维Burgers方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | KNO1D | ✔️ | ✔️ | +|[SNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d) | [一维Burgers方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | SNO1D | ✔️ | ✔️ | |[FNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno2d) | [二维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | FNO2D | ✔️ | ✔️ | +|[SNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d) | [二维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | SNO2D | ✔️ | ✔️ | +|[KNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) | [二维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | KNO2D | ✔️ | ✔️ | |[FNO3D方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) | [二维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | FNO3D | ✔️ | ✔️ | -|[KNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) | [二维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | KNO2D | ✔️ | ✔️ | +|[SNO3D方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d) | [三维NS方程数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes_3d/) | SNO3D | ✔️ | ✔️ | |[CAE-LSTM方法求解二维黎曼问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [二维黎曼问题数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/riemann/) | CAE-LSTM | ✔️ | ✔️ | |[CAE-LSTM方法求解Shu-Osher问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [一维Shu-Osher波数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/shu_osher/) | CAE-LSTM | ✔️ | ✔️ | |[CAE-LSTM方法求解Sod激波管问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [一维Sod激波管数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/sod/) | CAE-LSTM | ✔️ | ✔️ | @@ -77,6 +95,9 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe |[CAE-Transformer方法求解二维圆柱绕流问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_transformer) | [低雷诺数圆柱绕流数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-transformer/) | CAE-Transformer | ✔️ | ✔️ | |[FNO2D和UNET2D方法预测多时间步跨声速翼型复杂流场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady) | [二维跨声速翼型复杂流场数据集](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_driven/airfoil/2D_unsteady/) | FNO2D/UNET2D | ✔️ | ✔️ | |[HDNN方法预测流固耦合系统流场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/fluid_structure_interaction) | [流固耦合系统数据集](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_driven/fluid_structure_interaction/) | HDNN | ✔️ | ✔️ | +|[CascadeNet预测圆柱尾迹脉动速度时空场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net) | [CascadeNet圆柱尾迹脉动数据集](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/research/Cascade_Net/) | CascadeNet | ✔️ | ✔️ | +|[MultiScaleGNN求解压力泊松方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn) | [MultiScaleGNN压力泊松方程数据集](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/research/MultiScaleGNN/) | MultiScaleGNN | ✔️ | ✔️ | +|[基于神经算子网络的涡轮级流场预测与不确定性优化设计](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq) | [涡轮级子午面流场数据集](https://gitee.com/link?target=https%3A%2F%2Fdownload-mindspore.osinfra.cn%2Fmindscience%2Fmindflow%2Fdataset%2Fapplications%2Fresearch%2Fturbine_uq%2F) | UNet/FNO | ✔️ | ✔️ | ### 数据-机理融合驱动 @@ -86,6 +107,9 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | [PeRCNN方法求解二维Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/burgers_2d) | [PeRCNN数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PeRCNN/) | PeRCNN | ✔️ | ✔️ | | [PeRCNN方法求解三维反应扩散方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d) | [PeRCNN数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PeRCNN/) | PeRCNN | ✔️ | ✔️ | | [AI湍流模型](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/ai_turbulence_modeling) | - | MLP | ✔️ | ✔️ | +| [物理编码消息传递图神经网络PhyMPGN求解时空PDE](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn) | [PhyMPGN数据集](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PhyMPGN/) | PhyMPGN | | ✔️ | +| [数据与物理混合驱动下的物理场预测模型](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction) | [Allen-Cahn数据集](https://download.mindspore.cn/mindscience/mindflow/dataset/periodic_hill_2d/) | UNet2D | ✔️ | ✔️ | +| [融合物理机理的复杂流动温度场预测](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition) | - | SDNO | ✔️ | ✔️ | ### 物理驱动 @@ -106,17 +130,19 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe |[META-PINNs算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/meta_pinns) | - | PINNs | ✔️ | ✔️ | |[MOE-PINNs算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns) | - | PINNs | ✔️ | ✔️ | |[R-DLGA算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/r_dlga) | - | PINNs | ✔️ | ✔️ | +|[NSFNets方法求解不可压缩 Navier-Stokes 方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets) | - | NSFNets | ✔️ | ✔️ | ### CFD | 案例 | 格式 | GPU | NPU | -|:------------:|:-------------:|:---------:|:-------| +|:------------:|:-------------:|:---------:|:------:| |[Sod激波管](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/sod) | Rusanov | ✔️ | - | |[Lax激波管](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/lax) | Rusanov | ✔️ | - | |[二维黎曼问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/riemann2d) | - | ✔️ | - | |[库埃特流动](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/couette) | - | ✔️ | - | +|[二维声波方程求解](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic) | CBS | - | ✔️ | -## **安装教程** +## 安装教程 ### 版本依赖关系 @@ -124,7 +150,8 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | MindFlow | 分支 | MindSpore | Python | |:--------:|:----------------------------------------------------------------------:|:-----------:|:------:| -| master | [master](https://gitee.com/mindspore/mindscience/tree/master/MindFlow) | \ | \>=3.7 | +| master | [master](https://gitee.com/mindspore/mindscience/tree/master/MindFlow) | \ | \>=3.9 | +| 0.3.0 | [r0.7]() | \ | \>=3.7 | | 0.2.0 | [r0.6](https://gitee.com/mindspore/mindscience/tree/r0.6/MindFlow) | \>=2.2.12 | \>=3.7 | | 0.1.0 | [r0.3](https://gitee.com/mindspore/mindscience/tree/r0.3/MindFlow) | \>=2.0.0 | \>=3.7 | | 0.1.0rc1 | [r0.2.0](https://gitee.com/mindspore/mindscience/tree/r0.2.0/MindFlow) | \>=2.0.0rc1 | \>=3.7 | @@ -137,17 +164,16 @@ pip install -r requirements.txt ### 硬件支持情况 -| 硬件平台 | 操作系统 | 状态 | -| :------------ | :-------------- | :--- | -| Ascend | Linux | ✔️ | -| GPU | Linux | ✔️ | +| MindSpore Flow版本 | 硬件平台 | 操作系统 | 状态 | +| :------------ | :-------------- | :--- | ------------- | +| 0.1.0rc1/0.1.0/0.2.0/0.3.0 | Ascend | Linux | ✔️ | +| 0.1.0rc1/0.1.0/0.2.0 | GPU | Linux | ✔️ | ### pip安装 ```bash -# gpu and ascend are supported -export DEVICE_NAME=gpu -pip install mindflow_${DEVICE_NAME} +# ascend is supported +pip install mindflow_ascend ``` ### 源码安装 @@ -165,13 +191,6 @@ cd {PATH}/mindscience/MindFlow bash build.sh -e ascend -j8 ``` -- 编译GPU后端源码。 - -```bash -export CUDA_PATH={your_cuda_path} -bash build.sh -e gpu -j8 -``` - - 安装编译所得whl包。 ```bash @@ -179,7 +198,7 @@ cd {PATH}/mindscience/MindFLow/output pip install mindflow_*.whl ``` -## **社区** +## 社区 ### 加入MindFlow SIG @@ -229,7 +248,7 @@ MindSpore AI+科学计算专题，北京大学董彬老师[Learning and Learning 感谢以下开发者做出的贡献 🧑‍🤝‍🧑： -yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboqiang, chengzeruizhi, libokai, yangge, longzichao, qiuyisheng, haojiwei, leiyixiang, huangxiang, huxin,xingzhongfan, mengqinghe, lizhengyi, lixin, liuziyang, dujiaoxi, xiaoruoye, liangjiaming +yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboqiang, chengzeruizhi, libokai, yangge, longzichao, qiuyisheng, haojiwei, leiyixiang, huangxiang, huxin,xingzhongfan, mengqinghe, lizhengyi, lixin, liuziyang, dujiaoxi, xiaoruoye, liangjiaming, zoujingyuan, wanghaining, juliagurieva, guoqicheng, chenruilin, chenchao, wangqineng, wubingyang, zhaoyifan ### 合作伙伴 @@ -282,11 +301,11 @@ yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboq -## **贡献指南** +## 贡献指南 - 如何贡献您的代码，请点击此处查看：[贡献指南](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/CONTRIBUTION_CN.md) - 需要算力的用户，请参考[启智社区云脑使用指南](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/%E5%90%AF%E6%99%BA%E6%8C%87%E5%8D%97.pdf), [NPU使用录屏](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/npu%E4%BD%BF%E7%94%A8.MP4), [GPU使用录屏](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/gpu%E4%BD%BF%E7%94%A8.MP4) -## **许可证** +## 许可证 [Apache License 2.0](http://www.apache.org/licenses/LICENSE-2.0) diff --git a/MindFlow/README_EN.md b/MindFlow/README_EN.md index 7c901eb4ed36effa5b6622563399a37d3676d7aa..5edb4ffe57cf04813646420362f7d4ed1e4a072f 100644 --- a/MindFlow/README_EN.md +++ b/MindFlow/README_EN.md @@ -8,19 +8,24 @@ ENGLISH | [简体中文](README.md) [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg?style=flat)](https://gitee.com/mindspore/mindscience/pulls) [![LICENSE](https://img.shields.io/github/license/mindspore-ai/mindspore.svg?style=flat)](https://github.com/mindspore-ai/mindspore/blob/master/LICENSE) -# **MindFlow** +# MindFlow -## **Introduction** +## Introduction -Flow simulation aims to solve the fluid governing equation under a given boundary condition by numerical methods, so as to realize the flow analysis, prediction and control. It is widely used in engineering design in aerospace, ship manufacturing, energy and power industries. The numerical methods of traditional flow simulation, such as finite volume method and finite difference method, are mainly implemented by commercial software, requiring physical modeling, mesh generation, numerical dispersion, iterative solution and other steps. The simulation process is complex and the calculation cycle is long. AI has powerful learning fitting and natural parallel inference capabilities, which can improve the efficiency of the flow simulation. +Fluid simulation refers to solving the governing equations of fluid dynamics under given boundary conditions through numerical computation, enabling analysis, prediction, and control of flow behaviors. It has been widely applied in engineering design across industries such as aerospace, shipbuilding, and energy/power. Traditional numerical methods for fluid simulation (e.g., finite volume and finite difference methods), mainly implemented through commercial software, involve multiple complex steps including physical modeling, mesh generation, numerical discretization, and iterative solving, resulting in lengthy computational cycles. AI demonstrates powerful learning capabilities and natural parallel inference capabilities, which can effectively enhance the efficiency of fluid simulations. -MindSpore Flow is a flow simulation suite developed based on [MindSpore](https://www.mindspore.cn/). It supports AI flow simulation in industries such as aerospace, ship manufacturing, and energy and power. It aims to provide efficient and easy-to-use AI computing flow simulation software for industrial research engineers, university professors, and students. +MindFlow, developed based on [MindSpore](https://www.mindspore.cn/), is a fluid simulation suite supporting AI-enabled flow field simulation, aerodynamic shape design, and flow control for industries including aerospace, shipbuilding, and energy/power. It aims to provide industrial researchers, engineers, academic faculty, and students with an efficient and user-friendly AI-powered computational fluid dynamics (CFD) simulation software.
MindFlow Architecture
-## **Latest News** +## Latest News -- 🔥`2024.03.22` MindSpore Artificial Intelligence Framework Summit 2024 was held in Beijing National Convention Center. Professor Dong Bin, affiliated with both the Beijing International Center for Mathematical Research and the Center for Machine Learning Research at Peking University, revealed that the team has developed a groundbreaking model in the realm of AI-driven PDEs, named PDEformer-1. Leveraging the MindSpore and MindFlow suites, this model is uniquely capable of directly ingesting any PDE format as input. Through extensive training on a comprehensive dataset encompassing 3 million 1D PDE samples, it has demonstrated impressive speed and precision in resolving a broad spectrum of 1D PDE forward problems. +- 🔥`2025.03.30` MindFlow 0.3.0 released. Details: [MindFlow 0.3.0](RELEASE.md/). +- 🔥`2024.11.04` The 2024 AI for Science Forum, hosted by Peking University’s School of Computer Science and Beijing AI for Science Institute, featured a keynote by Professor Dong Bin (Boya Distinguished Professor at PKU, Deputy Director of the International Machine Learning Research Center). He announced PDEformer-2, an end-to-end solution prediction model based on MindSpore and MindFlow, capable of directly processing arbitrary PDE forms (both time-dependent and time-independent). Pre-trained on a ~40TB dataset, PDEformer-2 can infer solutions for 2D equations with varying boundary conditions, domains, and variables, rapidly predicting solutions at arbitrary spatiotemporal points. Additionally, as a differentiable surrogate model for forward problem solution operators, PDEformer-2 supports solving inverse problems, including full-wave inversion to recover wave velocity fields from noisy spatiotemporal scatter observations. This lays a foundation for modeling diverse physical phenomena and engineering applications in fluid dynamics, electromagnetics, and beyond. [Related News](https://www.mindspore.cn/news/newschildren?id=3481&type=news) +- 🔥`2024.10.13` The 3rd General Assembly of the Intelligent Fluid Mechanics Industry Consortium was successfully held in Xi’an, Shaanxi, with over 200 attendees from academia and industry. An expert presented *AI Fluid Simulation and MindSpore Practices*, highlighting MindSpore’s capabilities in whole process of AI model development and MindFlow’s advancements. He also showcased collaborative innovations with consortium partners, demonstrating AI+fluid mechanics applications in national priority scenarios like large aircraft development and aerodynamic design. [Related News](https://www.mindspore.cn/news/newschildren?id=3424&type=news) +- 🔥`Sep. 23, 2024` The **"PHengLEI"** Aerodynamic Shape Design Platform was launched at the "AI Empowers Aerospace Innovation" Postdoctoral Forum in Mianyang, Sichuan. Developed by the China Aerodynamics Research & Development Center using MindSpore and MindFlow, this generative AI platform assists designers in conceptual aerodynamic shape design. "Fenglei" enables end-to-end shape design that meets performance metrics, supports multi-scenario/multi-type design, and ensures solution diversity. Academician Tang Zhigong introduced its technical framework and applications, stating: *"AI offers a new paradigm for aerodynamics, injecting vitality into aerospace innovation. Generative aerodynamic design accelerates conceptual design and drives intelligent transformation of aero-design methodologies."* [Related News](https://www.mindspore.cn/news/newschildren?id=3405&type=news) +- 🔥`Jul. 04, 2024` At the 2024 World Artificial Intelligence Conference (WAIC) in Shanghai themed *"Governing AI for good and for all"*, China Southern Power Grid’s **"Yudian"** Intelligent Simulation Model, built on MindSpore, won the prestigious SAIL (Superior AI Leader) Award. Zheng Waisheng, General Manager of CSG’s Strategic Planning Department, explained: *"Yudian precisely delineates safety boundaries of new-type power systems and optimizes generation schedules. It dynamically adjusts grid operations to address renewable energy volatility, maximizing utilization rates while ensuring grid stability."* [Related News](https://business.cctv.com/2024/07/04/ARTICo0MOGKfEyWdRf3QTyGo240704.shtml) +- 🔥`2024.03.22` MindSpore Artificial Intelligence Framework Summit 2024 was held in Beijing National Convention Center. Professor Dong Bin, affiliated with both the Beijing International Center for Mathematical Research and the Center for Machine Learning Research at Peking University, revealed that the team has developed a foundation model in the realm of AI-driven PDEs, named PDEformer-1. Leveraging the MindSpore and MindFlow suites, this model is uniquely capable of directly ingesting any PDE format as input. Through extensive training on a comprehensive dataset encompassing 3 million 1D PDE samples, it has demonstrated impressive speed and precision in resolving a broad spectrum of 1D PDE forward problems. - 🔥`2024.03.22` MindSpore Artificial Intelligence Framework Summit 2024 was held in Beijing National Convention Center. Tang Zhigong, academician of Chinese Academy of Sciences and chairman of the Chinese Aerodynamic Society, introduced that the team created the generative aerodynamic design model platform based on MindSpore and MindFlow. Platform is oriented to a variety of application scenarios and breaks the traditional design paradigm. It shortens the design periods from the month level to the minute level, and meet the conceptual design requirements. [News](https://tech.cnr.cn/techph/20240323/t20240323_526636454.shtml). - 🔥`2024.03.20` MindFlow 0.2.0 is released, [Page](RELEASE.md). - 🔥`2023.11.07`The China (Xi'an) Artificial Intelligence Summit Forum was held at the High-tech International Conference Center in Yanta District, Xi'an, and the first large-scale fluid dynamics model for aircraft, "Qinling·AoXiang", jointly developed by Northwestern Polytechnical University and Huawei, was officially released. The model is an intelligent model for aircraft fluid simulation jointly developed by the International Joint Institute of Fluid Mechanics and Intelligence of Northwestern Polytechnical University and Huawei AI4Sci Lab on the basis of the domestic open-source fluid computing software Fenglei, relying on the surging computing power of Ascend AI and the MindSpore AI framework, [page](https://mp.weixin.qq.com/s/Rhpiyf3VJYm_lMBWTRDtGA). @@ -33,27 +38,37 @@ MindSpore Flow is a flow simulation suite developed based on [MindSpore](https:/ ## Publications -Ye Z, Huang X, Liu H, et al. Meta-Auto-Decoder: A Meta-Learning Based Reduced Order Model for Solving Parametric Partial Differential Equations[J]. Communications on Applied Mathematics and Computation. [[Paper]](https://link.springer.com/article/10.1007/s42967-023-00293-7) +[2024] Li X, Deng Z, Feng R, et al. Deep learning-based reduced order model for three-dimensional unsteady flow using mesh transformation and stitching[J]. Computers & Fluids. [[Paper]](https://arxiv.org/pdf/2307.07323) -Deng Z, Wang J, Liu H, et al. Prediction of transactional flow over supercritical airfoils using geometric-encoding and deep-learning strategies. Physics of Fluids 35, 075146 (2023). [[Paper]](https://pubs.aip.org/aip/pof/article-abstract/35/7/075146/2903765/Prediction-of-transonic-flow-over-supercritical?redirectedFrom=fulltext) +[2024] Wang Q, Ren P, Zhou H, et al. P $^ 2$ C $^ 2$ Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics[J]. arXiv preprint.[[Paper]](https://arxiv.org/pdf/2411.00040) + +[2024] Zeng B, Wang Q, Yan M, et al. PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems[J]. arXiv preprint. [[Paper]](https://arxiv.org/pdf/2410.01337) + +[2024] Ye Z, Huang X, Chen L, et al. Pdeformer-1: A foundation model for one-dimensional partial differential equations[J]. arXiv preprint. [[Paper]](https://arxiv.org/pdf/2407.06664) + +[2024] Li Z, Wang Y, Liu H, et al. Solving the boltzmann equation with a neural sparse representation[J]. SIAM Journal on Scientific Computing. [[Paper]](https://arxiv.org/pdf/2302.09233) + +[2024] Ye Z, Huang X, Chen L, et al. Pdeformer: Towards a foundation model for one-dimensional partial differential equations[J]. arXiv preprint.[[Paper](https://arxiv.org/abs/2402.12652)] + +[2024] Ye Z, Huang X, Liu H, et al. Meta-Auto-Decoder: A Meta-Learning Based Reduced Order Model for Solving Parametric Partial Differential Equations[J]. Communications on Applied Mathematics and Computation. [[Paper]](https://link.springer.com/article/10.1007/s42967-023-00293-7) + +[2024] Li Z, Wang Y, Liu H, et al. Solving Boltzmann equation with neural sparse representation[J]. SIAM Journal on Scientific Computing. [[Paper]](https://epubs.siam.org/doi/abs/10.1137/23M1558227?journalCode=sjoce3) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) + +[2023] Deng Z, Wang J, Liu H, et al. Prediction of transactional flow over supercritical airfoils using geometric-encoding and deep-learning strategies[J]. Physics of Fluids. [[Paper]](https://pubs.aip.org/aip/pof/article-abstract/35/7/075146/2903765/Prediction-of-transonic-flow-over-supercritical?redirectedFrom=fulltext) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady) -Rao C, Ren P, Wang Q, et al. Encoding physics to learn reaction–diffusion processes[J]. Nature Machine Intelligence, 2023: 1-15. [[Paper]](https://arxiv.org/abs/2106.04781) +[2023] Rao C, Ren P, Wang Q, et al. Encoding physics to learn reaction–diffusion processes[J]. Nature Machine Intelligence. [[Paper]](https://arxiv.org/abs/2106.04781) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn) -Li Z, Wang Y, Liu H, et al. Solving Boltzmann equation with neural sparse representation[J]. SIAM Journal on Scientific Computing, Vol. 46, Iss. 2 (2024). -[[Paper]](https://epubs.siam.org/doi/abs/10.1137/23M1558227?journalCode=sjoce3) -[[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) - -Deng Z, Liu H, Shi B, et al. Temporal predictions of periodic flows using a mesh transformation and deep learning-based strategy[J]. Aerospace Science and Technology, 2023, 134: 108081. [[Paper]](https://www.sciencedirect.com/science/article/pii/S1270963822007556) +[2023] Deng Z, Liu H, Shi B, et al. Temporal predictions of periodic flows using a mesh transformation and deep learning-based strategy[J]. Aerospace Science and Technology. [[Paper]](https://www.sciencedirect.com/science/article/pii/S1270963822007556) -Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Differential Equations with a Point Source[C]//IJCAI. 2022: 3839-3846. [[Paper]](https://gitee.com/link?target=https%3A%2F%2Fwww.ijcai.org%2Fproceedings%2F2022%2F0533.pdf) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) +[2022] Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Differential Equations with a Point Source[C]. IJCAI. [[Paper]](https://gitee.com/link?target=https%3A%2F%2Fwww.ijcai.org%2Fproceedings%2F2022%2F0533.pdf) [[Code]](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) ## Features -- [MindSpore Grad](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/features/mindspore_grad_cookbook.ipynb) +[MindSpore Grad](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/features/mindspore_grad_cookbook.ipynb) -- [Solve Pinns by MindFlow](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/features/solve_pinns_by_mindflow) +[Solve Pinns by MindFlow](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/features/solve_pinns_by_mindflow) ## Applications @@ -64,9 +79,12 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | [DongFang.YuFeng](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady) | [2D Airfoil Flow Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/airfoil/2D_steady/) | ViT | ✔️ | ✔️ | | [Solve Burgers Equation by FNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/fno1d) | [1D Burgers Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | FNO1D | ✔️ | ✔️ | | [Solve Burgers Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/kno1d) | [1D Burgers Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | KNO1D | ✔️ | ✔️ | +| [Solve Burgers Equation by SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d) | [1D Burgers Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/burgers/) | SNO1D | ✔️ | ✔️ | | [Solve Navier-Stokes Equation by FNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno2d) | [2D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | FNO2D | ✔️ | ✔️ | -| [Solve Navier-Stokes Equation by FNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) | [2D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | FNO3D | ✔️ | ✔️ | -| [Solve Navier-Stokes Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) | [2D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | KNO2D | ✔️ | ✔️ | +| [Solve Navier-Stokes Equation by SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d) | [2D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | SNO2D | ✔️ | ✔️ | +| [Solve Navier-Stokes Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) | [2D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes/) | KNO2D | ✔️ | ✔️ | +| [Solve Navier-Stokes Equation by FNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) | [3D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes_3d/) | FNO3D | ✔️ | ✔️ | +| [Solve Navier-Stokes Equation by SNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d) | [3D Navier-Stokes Equation Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/navier_stokes_3d/) | SNO3D | ✔️ | ✔️ | | [Solve 2D Riemann Problem by CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [2D Riemann Problem Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/riemann/) | CAE-LSTM | ✔️ | ✔️ | | [Solve Shu-Osher Problem by CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [1D Shu-Osher Problem Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/shu_osher/) | CAE-LSTM | ✔️ | ✔️ | | [Solve 1D Sod Shock Tube Problem by CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) | [1D Sod Problem Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-lstm/sod/) | CAE-LSTM | ✔️ | ✔️ | @@ -77,6 +95,9 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | [Solve 2D Cylinder Flow by CAE-Transformer](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_transformer) | [Low Reynolds Cylinder Flow Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_driven/cae-transformer/) | CAE-Transformer | ✔️ | ✔️ | |[Predict Multi-timestep Complicated Transonic Airfoil by FNO2D and UNET2D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady) | [2D Transonic Airfoil Dataset](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_driven/airfoil/2D_unsteady/) | FNO2D/UNET2D | ✔️ | ✔️ | |[Predict Fluid-structure Interaction System by HDNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/fluid_structure_interaction) | [Fluid-structure Interaction System Dataset](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_driven/fluid_structure_interaction/) | HDNN | ✔️ | ✔️ | +|[Prediction of spatiotemporal field of pulsation velocity in cylindrical wake by Cascade Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net) | [CascadeNet Cylinder Wake Pulsating Dataset](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/research/Cascade_Net/) | CascadeNet | ✔️ | ✔️ | +|[MultiScaleGNN for Solving Pressure Poisson Equation](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn) | [MultiScaleGNN Pressure Poisson Equation](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/research/MultiScaleGNN/) | MultiScaleGNN | ✔️ | ✔️ | +|[Turbine Stage Flow Field Prediction and Uncertainty Optimization Design Based on Neural Operator Networks](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq) | [Turbine Cascade Meridional Flow Field Dataset](https://gitee.com/link?target=https%3A%2F%2Fdownload-mindspore.osinfra.cn%2Fmindscience%2Fmindflow%2Fdataset%2Fapplications%2Fresearch%2Fturbine_uq%2F) | UNet/FNO | ✔️ | ✔️ | ### Data-Mechanism Fusion @@ -86,6 +107,9 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | [Solve 2D Burgers Equation by PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/burgers_2d) | [PeRCNN Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PeRCNN/) | PeRCNN | ✔️ | ✔️ | | [Solve 3D Reaction-Diffusion Equation by PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d) | [PeRCNN Dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PeRCNN/) | PeRCNN | ✔️ | ✔️ | | [AI Turb Model](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/ai_turbulence_modeling) | - | MLP | ✔️ | ✔️ | +| [Physics-encoded Message Passing Graph Network PhyMPGN solving spatiotemporal PDE systems](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn) | [PhyMPGN dataset](https://download-mindspore.osinfra.cn/mindscience/mindflow/dataset/applications/data_mechanism_fusion/PhyMPGN/) | PhyMPGN | | ✔️ | +| [Physical Field Prediction Model Driven by Data and Physics Hybridization](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction) | [Allen-Cahn dataset](https://download.mindspore.cn/mindscience/mindflow/dataset/periodic_hill_2d/) | UNet2D | ✔️ | ✔️ | +| [Fusion of Physical Mechanism for Predicting Complex Flow Temperature Fields](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition) | _ | SDNO | ✔️ | ✔️ | ### Physics Driven @@ -106,6 +130,7 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe |[META-PINNs Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/meta_pinns) | - | PINNs | ✔️ | ✔️ | |[MOE-PINNs Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns) | - | PINNs | ✔️ | ✔️ | |[R-DLGA Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/r_dlga) | - | PINNs | ✔️ | ✔️ | +|[NSFNets: Physics-informed neural networks for the incompressible Navier-Stokes equations](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets) | - | NSFNets | ✔️ | ✔️ | ### CFD @@ -115,8 +140,9 @@ Huang X, Liu H, Shi B, et al. A Universal PINNs Method for Solving Partial Diffe | [Lax Shock Tube](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/lax) | Rusanov | ✔️ | - | | [2D Riemann Problem](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/riemann2d) | - | ✔️ | - | | [Couette Flow](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/couette) | - | ✔️ | - | +| [2D Acoustic Wave Equation CBS Solver](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic) | CBS | - | ✔️ | -## **Installation** +## Installation ### Version Dependency @@ -124,7 +150,8 @@ Because MindFlow is dependent on MindSpore, please click [MindSpore Download Pag | MindFlow | Branch | MindSpore | Python | | :------: | :--------------------------------------------------------------------: | :---------: | :----: | -| master | [master](https://gitee.com/mindspore/mindscience/tree/master/MindFlow) | \ | \>=3.7 | +| master | [master](https://gitee.com/mindspore/mindscience/tree/master/MindFlow) | \ | \>=3.9 | +| 0.3.0 | [r0.7]() | \ | \>=3.7 | | 0.2.0 | [r0.6](https://gitee.com/mindspore/mindscience/tree/r0.6/MindFlow) | \>=2.2.12 | \>=3.7 | | 0.1.0 | [r0.3](https://gitee.com/mindspore/mindscience/tree/r0.3/MindFlow) | \>=2.0.0 | \>=3.7 | | 0.1.0rc1 | [r0.2.0](https://gitee.com/mindspore/mindscience/tree/r0.2.0/MindFlow) | \>=2.0.0rc1 | \>=3.7 | @@ -137,20 +164,19 @@ pip install -r requirements.txt ### Hardware -| Hardware | OS | Status | -| :------------ | :-------------- | :----- | -| Ascend | Linux | ✔️ | -| GPU | Linux | ✔️ | +| MindSpore Flow version | Hardware | OS | Status | +| :------------ | :-------------- | :----- | :----- | +| 0.1.0rc1/0.1.0/0.2.0/0.3.0 | Ascend | Linux | ✔️ | +| 0.1.0rc1/0.1.0/0.2.0 | GPU | Linux | ✔️ | -### **pip install** +### pip install ```bash -# gpu and ascend are supported -export DEVICE_NAME=gpu -pip install mindflow_${DEVICE_NAME} +# ascend is supported +pip install mindflow_ascend ``` -### **source code install** +### source code install - Download source code from Gitee. @@ -165,13 +191,6 @@ cd {PATH}/mindscience/MindFlow bash build.sh -e ascend -j8 ``` -- Compile in GPU backend. - -```bash -export CUDA_PATH={your_cuda_path} -bash build.sh -e gpu -j8 -``` - - Install the compiled .whl file. ```bash @@ -179,7 +198,7 @@ cd {PATH}/mindscience/MindFLow/output pip install mindflow_*.whl ``` -## **Community** +## Community ### Join MindFlow SIG @@ -229,7 +248,7 @@ We will continue to release [open source internship tasks](https://gitee.com/min Thanks goes to these wonderful people 🧑‍🤝‍🧑: -yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboqiang, chengzeruizhi, libokai, yangge, longzichao, qiuyisheng, haojiwei, leiyixiang, huangxiang, huxin,xingzhongfan, mengqinghe, lizhengyi, lixin, liuziyang, dujiaoxi, xiaoruoye, liangjiaming +yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboqiang, chengzeruizhi, libokai, yangge, longzichao, qiuyisheng, haojiwei, leiyixiang, huangxiang, huxin,xingzhongfan, mengqinghe, lizhengyi, lixin, liuziyang, dujiaoxi, xiaoruoye, liangjiaming, zoujingyuan, wanghaining, juliagurieva, guoqicheng, chenruilin, chenchao, wangqineng, wubingyang, zhaoyifan ### Community Partners @@ -282,11 +301,11 @@ yufan, wangzidong, liuhongsheng, zhouhongye, zhangyi, dengzhiwen, liulei, guoboq -## **Contribution Guide** +## Contribution Guide - Please click here to see how to contribute your code:[Contribution Guide](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/CONTRIBUTION_CN.md) - For users who are in need of AI chips, please refer to [the document of Open Intelligence](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/%E5%90%AF%E6%99%BA%E6%8C%87%E5%8D%97.pdf), [NPU tutorials](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/npu%E4%BD%BF%E7%94%A8.MP4), [GPU tutorials](https://download-mindspore.osinfra.cn/mindscience/mindflow/tutorials/gpu%E4%BD%BF%E7%94%A8.MP4) -## **License** +## License [Apache License 2.0](http://www.apache.org/licenses/LICENSE-2.0) diff --git a/MindFlow/RELEASE.md b/MindFlow/RELEASE.md index d400e52f89dde9bbd9db059d0f2bd951beb7e698..e8fb5d27d7d59752c369e13e5a0f03dbced812c1 100644 --- a/MindFlow/RELEASE.md +++ b/MindFlow/RELEASE.md @@ -2,7 +2,61 @@ [查看中文](./RELEASE_CN.md) -MindSpore Flow is a flow simulation suite developed based on MindSpore. It supports AI flow simulation in industries such as aerospace, ship manufacturing, and energy and power. It aims to provide efficient and easy-to-use AI computing flow simulation software for industrial research engineers, university professors, and students. +MindSpore Flow is a flow simulation suite developed based on MindSpore. It supports AI flow simulation in industries such as aerospace, ship manufacturing, and energy and power under AI fluid simulation, AI aerodynamc design and AI flow control applications. It aims to provide efficient and easy-to-use AI computing flow simulation software for industrial research engineers, university professors, and students. + +## MindSpore Flow 0.3.0 Release Notes + +### Major Feature and Improvements + +#### Data Driven + +-[STABLE] [Burgers_SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d)/[Navier_Stokes_SNO2D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d)/[Navier_Stokes_SNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d): Applications sovling one-dimension Burgers Equation, two/three-dimension Navier Stokes Equation by Spectral Neural Operator under data driven method are added. + +-[STABLE] [API-SNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/sno.py): Spectral Neural Operator (including SNO and U-SNO) APIs are added, utilizing polynomial transformations to transform computations into a spectral space similar to FNO architecture. Its advantage lies in effectively reducing system bias caused by aliasing errors. + +-[STABLE] [API-Attention](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/attention.py): Refactoring most commonly used Transformer class networks such as Attention, MultiHeadAttention, AttentionBlock, and ViT network interfaces. + +-[STABLE] [API-Diffusion](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/diffusion.py): A complete set of training and inference interfaces for diffusion models are added with support of two mainstream diffusion methods of DDPM and DDIM. Meanwhile the entire process of diffusion model training and inference can be completed through the simple and easy-to-use interfaces of Diffusion Scheduler, Diffusion Trainer, Diffusion Pipeline, and Diffusion Transformer. + +-[STABLE] [API-Refactor_Core](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/mindflow/core): Refactor of mindflow.core by fusion of mindflow.common, mindflow.loss and mindflow.operators. + +-[RESEARCH] [CascadeNet](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net): CascadeNet case is added, It uses surface pressure, Reynolds number, and a small number of wake velocity measurement points as inputs to predict the spatiotemporal field of cylinder wake pulsation velocity through a generative adversarial network with scale transfer topology structure. + +-[RESEARCH] [MultiScaleGNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn): A multi-scale graph neural network case to solve the large-scale pressure Poisson equation is added, which supports the use of projection method (or fractional step method) to solve incompressible Navier Stokes equations. + +-[RESEARCH] [TurbineUQ](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq): A case study of turbine stage flow field prediction and uncertainty optimization design is added with a combination of Monte Carlo method with deep learning methods to quantitative evaluation of uncertainty. + +#### Data-Mechanism Fusion + +-[STABLE] [PhyMPGN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn): An application of PhyMPGN, a physical equation solving model based on graph neural networks for the problem of flow around a cylinder is added. PhyMPGN can solve Burgers, FitzHugh-Nagumo, Gray-Scott and other equations in unstructured grids. Related [paper](http://arxiv.org/abs/2410.01337) has been received as ICLR 2025 Spotlight. + +-[RESEARCH] [Heat_Conduction](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction): A case study of steady-state heat conduction physics field prediction driven by data and physics is added. + +-[RESEARCH] [SuperPosition](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition): SDNO, an operator neural network based on the superposition principle, is added for predicting the temperature field of complex flow patterns in aircraft engine internal flow cascades. + +#### Physics Driven + +-[RESEARCH] [NSFNets](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets): Navier Stokes Flow Networks (NSFNets) are added. It is a [highly cited paper](https://www.sciencedirect.com/science/article/pii/S0021999120307257) for solving ill posed problems (such as partially missing boundary conditions or inversion problems. + +#### Solver + +-[STABLE] [CBS solver](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic): Application of CBS acoustic equation solver for solving two-dimensional acoustic equations in complex parameter fields is added. The CBS solver solves the acoustic equation in the frequency domain and has spectral accuracy in all spatial directions, with higher accuracy than the finite difference method. Reference: [Osnabrugge et al. 2016](https://linkinghub.elsevier.com/retrieve/pii/S0021999116302595) + +#### Optimizer + +-[STABLE] [API-AdaHessian second-order optimizer](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/core/optimizers.py): AdaHessian second-order optimizer based on the second-order information provided by the diagonal elements of the Hessian matrix for optimization calculations is added. Tests achieved a loss reduction over 20% compared with Adam under the same number of steps. + +#### Foundation Model + +-[RESEARCH] [PDEformer](https://github.com/functoreality/pdeformer-2): PDEformer supports to solve [one dimensional](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/pdeformer1d)/[two dimensional](https://github.com/functoreality/pdeformer-2) general partial differential equations with time with a superior of accuracy to domain model by foundation model under Zero-Shot occasions. + +### Contributors + +Thanks to the following developers for their contributions: + +hsliu_ustc, gracezou, mengqinghe0909, Yi_zhang95, b_rookie, WhFanatic, xingzhongfan, juliagurieva, GQEm, chenruilin2024, ZYF00000, chenchao2024, wangqineng2024, BingyangWu-pkusms21, Bochengz, functoreality, huangxiang360729, ChenLeheng, juste_une_photo. + +Contributions to the project in any form are welcome! ## MindSpore Flow 0.2.0 Release Notes @@ -10,23 +64,31 @@ MindSpore Flow is a flow simulation suite developed based on MindSpore. It suppo #### Data Driven -- [STABLE] [Airfoil2D_Unsteady](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady): Transonic airfoil flow is simulated by data-driven methods (using FNO2D and Unet2D). -- [STABLE] [API-FNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/fno.py): FNO1D, FNO2D and FNO3D APIs are refactored to improve the commonality. "Channels_last" and "channels_first" input formats are supported. Activation functions can be set respectively. Users can set compute data type of SpectralConvDft and FNO-skip. Hyper parameters of projection and lifting layers, residual connection and positional embedding are supported. -- [STABLE] [API-UNet2D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/unet2d.py): UNet2D API are refactored. Users can define the improving and reducing of UpConv and DownCov by 'base_channels'. Data formats of 'NCHW' and 'NHWC' are supported. +\- [STABLE] [Airfoil2D_Unsteady](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady): Transonic airfoil flow is simulated by data-driven methods (using FNO2D and Unet2D). + +\- [STABLE] [API-FNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/fno.py): FNO1D, FNO2D and FNO3D APIs are refactored to improve the commonality. "Channels_last" and "channels_first" input formats are supported. Activation functions can be set respectively. Users can set compute data type of SpectralConvDft and FNO-skip. Hyper parameters of projection and lifting layers, residual connection and positional embedding are supported. + +\- [STABLE] [API-UNet2D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/unet2d.py): UNet2D API are refactored. Users can define the improving and reducing of UpConv and DownCov by 'base_channels'. Data formats of 'NCHW' and 'NHWC' are supported. #### Data-Mechanism Fusion -- [STABLE] [API-Percnn](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/percnn.py): The percnn API is added to learn the spatiotemporal evolution rules of physical fields on coarse grids through the recursive convolutional neural network. By default, the input of two physical components is supported. The number of conv layers and kernel size can be customized to implement applications on different physical phenomena. -- [STABLE] [PeRCNN-gsrd3d](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d): Add case of solving 3d GS reaction-diffusion equation by PeRCNN. +\- [STABLE] [API-Percnn](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/percnn.py): The percnn API is added to learn the spatiotemporal evolution rules of physical fields on coarse grids through the recursive convolutional neural network. By default, the input of two physical components is supported. The number of conv layers and kernel size can be customized to implement applications on different physical phenomena. + +\- [STABLE] [PeRCNN-gsrd3d](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d): Add case of solving 3d GS reaction-diffusion equation by PeRCNN. #### Physics Driven -- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): Boltzmann equation with D1V3-BGK and secondary collision term is solved. The relevant papers are published in [SIAM Journal on Scientific Computing](https://www.siam.org/publications/journals/siam-journal-on-scientific-computing-sisc). -- [STABLE] [Periodic Hill](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/periodic_hill): Periodic hill flow are solved by PINNs. -- [STABLE] [Possion](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/continuous): Poisson equations with periodic and robin boundary conditions are solved by PINNs. -- [RESEARCH] [Cma_Es_Mgda](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cma_es_mgda): Add CMA-ES and Multi-objective Gradient Optimization Algorithm(mgda) to solve PINNs. -- [RESEARCH] [Moe_Pinns](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns): Support MOE-PINNs. -- [RESEARCH] [Allen-Cahn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/allen_cahn): Allen-Cahn equation is solved by PINNs. +\- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): Boltzmann equation with D1V3-BGK and secondary collision term is solved. The relevant papers are published in [SIAM Journal on Scientific Computing](https://www.siam.org/publications/journals/siam-journal-on-scientific-computing-sisc). + +\- [STABLE] [Periodic Hill](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/periodic_hill): Periodic hill flow are solved by PINNs. + +\- [STABLE] [Possion](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/continuous): Poisson equations with periodic and robin boundary conditions are solved by PINNs. + +\- [RESEARCH] [Cma_Es_Mgda](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cma_es_mgda): Add CMA-ES and Multi-objective Gradient Optimization Algorithm(mgda) to solve PINNs. + +\- [RESEARCH] [Moe_Pinns](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns): Support MOE-PINNs. + +\- [RESEARCH] [Allen-Cahn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/allen_cahn): Allen-Cahn equation is solved by PINNs. ### Contributors @@ -42,17 +104,19 @@ Contributions to the project in any form are welcome! #### Data Driven -- [STABLE] [CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) : Support data-driven implementation of convolutional autoencoder-long short memory neural network for processing unsteady compressible flow. -- [STABLE] [Move Boundary Hdnn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/move_boundary_hdnn) : Support data-driven implementation of HDNN network for solving unsteady flow field problems with moving boundaries. +\- [STABLE] [CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) : Support data-driven implementation of convolutional autoencoder-long short memory neural network for processing unsteady compressible flow. + +\- [STABLE] [Move Boundary Hdnn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/move_boundary_hdnn) : Support data-driven implementation of HDNN network for solving unsteady flow field problems with moving boundaries. #### Data-Mechanism Fusion -- [STABLE] [PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn) : Support physical encoded recursive Convolutional neural network (PeRCNN). +\- [STABLE] [PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn) : Support physical encoded recursive Convolutional neural network (PeRCNN). #### Physics Driven -- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) : Support PINNs method for solving Boltzmann equations. -- [STABLE] [Poisson with Point Source](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) : Support PINNs method to solve Poisson's equation with point source. +\- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann) : Support PINNs method for solving Boltzmann equations. + +\- [STABLE] [Poisson with Point Source](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source) : Support PINNs method to solve Poisson's equation with point source. ## MindSpore Flow 0.1.0.rc1 Release Notes @@ -60,8 +124,9 @@ Contributions to the project in any form are welcome! #### Data Driven -- [STABLE] [KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d): Provide the Kupmann KNO neural operator to improve the simulation accuracy of NS equations. -- [STABLE] [DongFang·YuFeng](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady): Provide a large model of Dongfang Yufeng, supporting end-to-end rapid simulation of airfoils. +\- [STABLE] [KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d): Provide the Kupmann KNO neural operator to improve the simulation accuracy of NS equations. + +\- [STABLE] [DongFang·YuFeng](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady): Provide a large model of Dongfang Yufeng, supporting end-to-end rapid simulation of airfoils. ## MindSpore Flow 0.1.0-alpha Release Notes @@ -69,19 +134,19 @@ Contributions to the project in any form are welcome! #### Data Driven -- [STABLE] Various neural networks are supported, including fully connected networks, residual networks, Fourier neural operators and Vision Transformer. Dataset merging and multiple data formats are supported. High level API is provided for training and evaluation. Multiple learning rates and losses are supported. +\- [STABLE] Various neural networks are supported, including fully connected networks, residual networks, Fourier neural operators and Vision Transformer. Dataset merging and multiple data formats are supported. High level API is provided for training and evaluation. Multiple learning rates and losses are supported. #### Data-Mechanism Fusion -- [STABLE] [PDE-Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net): A physics plus data driven deep learning method, PDE-Net, is provided for unsteady flow field prediction and regression of PDEs. +\- [STABLE] [PDE-Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net): A physics plus data driven deep learning method, PDE-Net, is provided for unsteady flow field prediction and regression of PDEs. #### Physics Driven -- [STABLE] Solve partial differential equations (PDEs) based on physics informed neural network. PDEs and basic equations can be defined by sympy. Users can calculate the Hessian and Jacobian matrix of network output to input. Basic geometrics, time domains and their operations are supported, which can be used for sampling within the geometric region and on the boundary. +\- [STABLE] Solve partial differential equations (PDEs) based on physics informed neural network. PDEs and basic equations can be defined by sympy. Users can calculate the Hessian and Jacobian matrix of network output to input. Basic geometrics, time domains and their operations are supported, which can be used for sampling within the geometric region and on the boundary. #### Differentiable CFD Solver -- [STABLE] An end-to-end differentiable compressible CFD solver, MindFlow-CFD, is introduced. WENO5 reconstruction, Rusanov flux, Runge-Kutta integrator are supported. Symmetry, periodic, solid wall and Neumann boundary conditions are supported. +\- [STABLE] An end-to-end differentiable compressible CFD solver, MindFlow-CFD, is introduced. WENO5 reconstruction, Rusanov flux, Runge-Kutta integrator are supported. Symmetry, periodic, solid wall and Neumann boundary conditions are supported. ### Contributors diff --git a/MindFlow/RELEASE_CN.md b/MindFlow/RELEASE_CN.md index 31327703cb3775193d1848cfae684338629e4d56..982d9706fc21f073d1f3fcd79bd5b82c969c685f 100644 --- a/MindFlow/RELEASE_CN.md +++ b/MindFlow/RELEASE_CN.md @@ -2,7 +2,52 @@ [View English](./RELEASE.md) -MindSpore Flow是基于昇思MindSpore开发的流体仿真领域套件，支持航空航天、船舶制造以及能源电力等行业领域的AI流场模拟，旨在于为广大的工业界科研工程人员、高校老师及学生提供高效易用的AI计算流体仿真软件。 +MindSpore Flow是基于昇思MindSpore开发的流体仿真领域套件，支持航空航天、船舶制造以及能源电力等行业领域的AI流场仿真、AI气动外形设计、AI流动控制，旨在于为广大的工业界科研工程人员、高校老师及学生提供高效易用的AI计算流体仿真软件。 + +## MindSpore Flow 0.3.0 Release Notes + +### 主要特性和增强 + +#### 数据驱动 + +* [STABLE] [Burgers_SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d)/[Navier_Stokes_SNO2D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d)/[Navier_Stokes_SNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d): 新增数据驱动下的谱神经算子求解一维Burgers方程、二维/三维纳维·斯托克斯方程案例。 +* [STABLE] [API-SNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/sno.py): 新增谱神经算子（包括SNO和U-SNO）API，利用多项式变换将计算转换到频谱空间的类FNO架构，其优势在于有效减小由混淆误差引起的系统偏差。 +* [STABLE] [API-Attention](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/attention.py): 重构transformer类网络常用的Attention、MultiHeadAttention、AttentionBlock、ViT网络接口。 +* [STABLE] [API-Diffusion](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/diffusion.py): 新增扩散模型全套训练、推理接口，支持DDPM、DDIM业界两种主流扩散方式，通过简洁易用的DiffusionScheduler、DiffusionTrainer、DiffusionPipeline、DiffusionTransformer接口即可完成扩散模型训练推理全流程。 +* [STABLE] [API-Core_Refactor](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/mindflow/core): 将原mindflow.common，mindflow.loss，mindflow.operators优化重构为mindflow.core。 +* [RESEARCH] [CascadeNet](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net): 新增CascadeNet案例，以表面压强、雷诺数和少量尾迹速度测点作为输入，通过具有尺度传递拓扑结构的生成对抗网络，预测圆柱尾迹脉动速度时空场。 +* [RESEARCH] [MultiScaleGNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn): 新增多级图神经网络案例，求解大规模压力泊松方程，可以支撑使用投影法（或分步法）求解不可压缩纳维 · 斯托克斯方程。 +* [RESEARCH] [TurbineUQ](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq): 新增涡轮级流场预测与不确定性优化设计案例，通过蒙特卡洛结合深度学习的方法来完成不确定性的量化评估。 + +#### 数据-机理融合驱动 + +* [STABLE] [PhyMPGN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn): 新增基于图神经网络的物理方程求解模型 PhyMPGN 针对圆柱绕流问题的应用案例。PhyMPGN 能够在非结构网格下实现 Burgers、FitzHugh-Nagumo、Gray-Scott 等方程的求解，相关[论文](http://arxiv.org/abs/2410.01337)被接收为 ICLR 2025 Spotlight。 +* [RESEARCH] [Heat_Conduction](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction): 新增数据与物理混合驱动下的稳态热传导物理场预测案例。 +* [RESEARCH] [SuperPosition](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition): 新增叠加原理的算子神经网络SDNO，用于航空发动机内流叶栅复杂流态下温度场的预测。 + +#### 物理驱动 + +* [RESEARCH] [NSFNets](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets): 新增纳维 · 斯托克斯流动网络（NSFNets），该网络对应[论文](https://www.sciencedirect.com/science/article/pii/S0021999120307257)为该领域高被引论文，用以求解不适定问题（例如部分边界条件缺失）或反演问题。 + +#### 求解器 + +* [STABLE] [CBS求解器](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic): 新增声波方程求解器 CBS 在复杂参数场下求解二维声波方程的应用案例。CBS 求解器在频域求解声波方程，在空间各方向均具有谱精度，精度高于有限差分法。相关论文参考 [Osnabrugge et al. 2016](https://linkinghub.elsevier.com/retrieve/pii/S0021999116302595) + +#### 优化器 + +* [STABLE] [API-AdaHessian二阶优化器](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/core/optimizers.py): 新增AdaHessian二阶优化器，基于Hessian矩阵对角元提供的二阶信息进行优化计算，实测相同step数下相比Adam实现20%以上的loss下降。 + +#### 基础模型 + +* [RESEARCH] [PDEformer](https://github.com/functoreality/pdeformer-2): PDEformer支持[一维](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/pdeformer1d)/[二维](https://github.com/functoreality/pdeformer-2)含时偏微分方程通用求解，典型场景下Zero-Shot求解精度超过专有模型。 + +### 贡献者 + +感谢以下开发者做出的贡献： + +hsliu_ustc, gracezou, mengqinghe0909, Yi_zhang95, b_rookie, WhFanatic, xingzhongfan, juliagurieva, GQEm, chenruilin2024, ZYF00000, chenchao2024, wangqineng2024, BingyangWu-pkusms21, Bochengz, functoreality, huangxiang360729, ChenLeheng, juste_une_photo. + +欢迎以任何形式对项目提供贡献！ ## MindSpore Flow 0.2.0 Release Notes @@ -10,23 +55,23 @@ MindSpore Flow是基于昇思MindSpore开发的流体仿真领域套件，支持 #### 数据驱动 -- [STABLE] [Airfoil2D_Unsteady](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady): 支持数据驱动(FNO2D和Unet2D两种backbone)下跨声速翼型复杂流场的多时间步预测。 -- [STABLE] [API-FNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/fno.py): 重构FNO1D、FNO2D、FNO3D API，提升接口的通用性，支持"channels_last"和"channels_first"两种输入数据格式，支持mlp层和FNOBlock层分别设置激活函数，支持SpectralConvDft和FNO skip分别设置计算精度，支持设置projection和lifting中间层参数，支持选择残差增强和嵌入位置信息。 -- [STABLE] [API-UNet2D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/unet2d.py): 重构UNet2D API，新增base_channels作为基准通道数，以控制上/下采样的通道数增/减，支持"NCHW"和"NHWC"两种输入数据格式。 +* [STABLE] [Airfoil2D_Unsteady](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady): 支持数据驱动(FNO2D和Unet2D两种backbone)下跨声速翼型复杂流场的多时间步预测。 +* [STABLE] [API-FNO1D/2D/3D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/fno.py): 重构FNO1D、FNO2D、FNO3D API，提升接口的通用性，支持"channels_last"和"channels_first"两种输入数据格式，支持mlp层和FNOBlock层分别设置激活函数，支持SpectralConvDft和FNO skip分别设置计算精度，支持设置projection和lifting中间层参数，支持选择残差增强和嵌入位置信息。 +* [STABLE] [API-UNet2D](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/unet2d.py): 重构UNet2D API，新增base_channels作为基准通道数，以控制上/下采样的通道数增/减，支持"NCHW"和"NHWC"两种输入数据格式。 #### 数据-机理融合驱动 -- [STABLE] [API-Percnn](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/percnn.py): 新增percnn API，通过递归卷积神经网络，在粗网格上学习物理场时空演化规律，默认支持两个物理分量的输入可自定义调节conv layer数量及kernel size，实现在不同物理现象上的应用。 -- [STABLE] [PeRCNN-gsrd3d](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d): 新增PeRCNN求解三维GS反应扩散方程的案例。 +* [STABLE] [API-Percnn](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/mindflow/cell/neural_operators/percnn.py): 新增percnn API，通过递归卷积神经网络，在粗网格上学习物理场时空演化规律，默认支持两个物理分量的输入可自定义调节conv layer数量及kernel size，实现在不同物理现象上的应用。 +* [STABLE] [PeRCNN-gsrd3d](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d): 新增PeRCNN求解三维GS反应扩散方程的案例。 #### 物理驱动 -- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): 支持D1V3的BGK以及二次碰撞项的玻尔兹曼方程求解，相关论文发表在《[SIAM Journal on Scientific Computing](https://www.siam.org/publications/journals/siam-journal-on-scientific-computing-sisc)》。 -- [STABLE] [Periodic Hill](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/periodic_hill): 支持PINNs方法求解周期山流通问题。 -- [STABLE] [Possion](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/continuous): 添加PINNs求解poisson方程时对periodic以及robin边界条件的支持。 -- [RESEARCH] [Cma_Es_Mgda](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cma_es_mgda): 支持CMA-ES和多目标梯度优化算法（mgda）结合求解PINNs问题。 -- [RESEARCH] [Moe_Pinns](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns): 支持多专家模型求解PINNs问题。 -- [RESEARCH] [Allen-Cahn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/allen_cahn): 反应扩散的Allen-Cahn和NS特解的Kovasznay流是常见的物理过程，通过PINNs方式以无监督方式完成对特定I/BC的求解。 +* [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): 支持D1V3的BGK以及二次碰撞项的玻尔兹曼方程求解，相关论文发表在《[SIAM Journal on Scientific Computing](https://www.siam.org/publications/journals/siam-journal-on-scientific-computing-sisc)》。 +* [STABLE] [Periodic Hill](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/periodic_hill): 支持PINNs方法求解周期山流通问题。 +* [STABLE] [Possion](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/continuous): 添加PINNs求解poisson方程时对periodic以及robin边界条件的支持。 +* [RESEARCH] [Cma_Es_Mgda](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cma_es_mgda): 支持CMA-ES和多目标梯度优化算法（mgda）结合求解PINNs问题。 +* [RESEARCH] [Moe_Pinns](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns): 支持多专家模型求解PINNs问题。 +* [RESEARCH] [Allen-Cahn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/allen_cahn): 反应扩散的Allen-Cahn和NS特解的Kovasznay流是常见的物理过程，通过PINNs方式以无监督方式完成对特定I/BC的求解。 ### 贡献者 @@ -42,17 +87,17 @@ hsliu_ustc, Yi_zhang95, zwdeng, liulei277, chengzrz, mengqinghe0909, xingzhongfa #### 数据驱动 -- [STABLE] [CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm): 支持数据驱动的卷积自编码器–长短时记忆神经网络求解非定常可压缩流动。 -- [STABLE] [Move Boundary Hdnn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/move_boundary_hdnn): 支持数据驱动的HDNN网络求解动边界的非定常流场。 +* [STABLE] [CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm): 支持数据驱动的卷积自编码器–长短时记忆神经网络求解非定常可压缩流动。 +* [STABLE] [Move Boundary Hdnn](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/move_boundary_hdnn): 支持数据驱动的HDNN网络求解动边界的非定常流场。 #### 数据-机理融合驱动 -- [STABLE] [PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn): 支持物理编码递归卷积神经网络(Physics-encoded Recurrent Convolutional Neural Network，PeRCNN)。 +* [STABLE] [PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn): 支持物理编码递归卷积神经网络(Physics-encoded Recurrent Convolutional Neural Network，PeRCNN)。 #### 物理驱动 -- [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): 支持PINNs方法求解玻尔兹曼方程。 -- [STABLE] [Poisson with Point Source](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source): 支持PINNs方法求解带点源的泊松方程。 +* [STABLE] [Boltzmann](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/boltzmann): 支持PINNs方法求解玻尔兹曼方程。 +* [STABLE] [Poisson with Point Source](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/poisson/point_source): 支持PINNs方法求解带点源的泊松方程。 ## MindSpore Flow 0.1.0.rc1 Release Notes @@ -60,8 +105,8 @@ hsliu_ustc, Yi_zhang95, zwdeng, liulei277, chengzrz, mengqinghe0909, xingzhongfa #### 数据驱动 -- [STABLE] [KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d): 支持KNO神经算子，提升NS方程仿真精度 -- [STABLE] [东方.御风](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady): 东方御风大模型，支持翼型端到端快速仿真。 +* [STABLE] [KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d): 支持KNO神经算子，提升NS方程仿真精度 +* [STABLE] [东方.御风](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady): 东方御风大模型，支持翼型端到端快速仿真。 ## MindSpore Flow 0.1.0-alpha Release Notes @@ -69,19 +114,19 @@ hsliu_ustc, Yi_zhang95, zwdeng, liulei277, chengzrz, mengqinghe0909, xingzhongfa #### 数据驱动 -- [STABLE] 提供了多种神经网络，包括全连接网络、残差网络、傅里叶神经算子、Vision Transformer，支持多种数据格式的读取和数据集的合并，MindFlow提供了模型训练和推理的高阶API，支持多种学习率和损失函数的使用。 +* [STABLE] 提供了多种神经网络，包括全连接网络、残差网络、傅里叶神经算子、Vision Transformer，支持多种数据格式的读取和数据集的合并，MindFlow提供了模型训练和推理的高阶API，支持多种学习率和损失函数的使用。 #### 数据-物理融合驱动 -- [STABLE] [PDE-Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net): 提供数据-物理融合驱动的深度学习方法PDE-Net，用于流场的时序预测和偏微分方程的回归。 +* [STABLE] [PDE-Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net): 提供数据-物理融合驱动的深度学习方法PDE-Net，用于流场的时序预测和偏微分方程的回归。 #### 物理驱动 -- [STABLE] 支持物理信息神经网络求解偏微分方程，使用sympy定义微分方程及基本方程的求解，支持计算神经网络输出对输入的一阶和二阶导数矩阵，提供了基本几何形状、时域的定义及其操作，便于几何区域内和边界上的采样。 +* [STABLE] 支持物理信息神经网络求解偏微分方程，使用sympy定义微分方程及基本方程的求解，支持计算神经网络输出对输入的一阶和二阶导数矩阵，提供了基本几何形状、时域的定义及其操作，便于几何区域内和边界上的采样。 #### 可微分CFD求解器 -- [STABLE] 我们推出了端到端可微分的可压缩CFD求解器MindFlow-CFD，支持WENO5重构、Rusanov通量以及龙格-库塔积分，支持对称、周期性、固壁及诺依曼边界条件。 +* [STABLE] 我们推出了端到端可微分的可压缩CFD求解器MindFlow-CFD，支持WENO5重构、Rusanov通量以及龙格-库塔积分，支持对称、周期性、固壁及诺依曼边界条件。 ### 贡献者 @@ -89,4 +134,4 @@ hsliu_ustc, Yi_zhang95, zwdeng, liulei277, chengzrz, mengqinghe0909, xingzhongfa hsliu_ustc, Yi_zhang95, zwdeng, liulei277, chengzrz, liangjiaming2023, yanglin2023 -欢迎以任何形式对项目提供贡献！ \ No newline at end of file +欢迎以任何形式对项目提供贡献！ diff --git a/MindFlow/applications/README.md b/MindFlow/applications/README.md index 1a9943a059399feda705dae2e96919050ff8ca64..39ef85593fb9d1f08433eba87b190df29b8524bb 100644 --- a/MindFlow/applications/README.md +++ b/MindFlow/applications/README.md @@ -23,9 +23,12 @@ MindFlow覆盖了物理驱动、数据驱动、数据机理融合的AI流体仿 - [东方.御风](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_steady) - [FNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/fno1d) - [KNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/kno1d) + - [SNO方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d) - [FNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno2d) - - [FNO3D方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) + - [SNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d) - [KNO方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) + - [FNO3D方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) + - [SNO3D方法求解NS方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d) - [CAE-LSTM方法求解二维黎曼问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) - [CAE-LSTM方法求解Shu-Osher问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) - [CAE-LSTM方法求解Sod激波管问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) @@ -36,11 +39,17 @@ MindFlow覆盖了物理驱动、数据驱动、数据机理融合的AI流体仿 - [CAE-Transformer方法求解二维圆柱绕流问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_transformer) - [FNO2D和UNET2D方法预测多时间步跨声速翼型复杂流场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady) - [HDNN方法预测流固耦合系统流场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/fluid_structure_interaction) + - [CascadeNet预测圆柱尾迹脉动速度时空场](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net) + - [MultiScaleGNN求解压力泊松方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn) + - [基于神经算子网络的涡轮级流场预测与不确定性优化设计](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq) - 数据-机理融合驱动 - [PDE-NET方法求解对流扩散方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net) - [PeRCNN方法求解二维Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/burgers_2d) - [PeRCNN方法求解三维反应扩散方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d) - [AI湍流模型](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/ai_turbulence_modeling) + - [物理编码消息传递图神经网络PhyMPGN求解时空PDE](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn) + - [数据与物理混合驱动下的物理场预测模型](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction) + - [融合物理机理的复杂流动温度场预测](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition) - 物理驱动 - [PINNs方法求解Burgers方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/burgers) - [PINNs方法求解圆柱绕流](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/cylinder_flow_forward) @@ -57,10 +66,12 @@ MindFlow覆盖了物理驱动、数据驱动、数据机理融合的AI流体仿 - [META-PINNs算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/meta_pinns) - [MOE-PINNs算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns) - [R-DLGA算法](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/r_dlga) + - [NSFNets方法求解不可压缩 Navier-Stokes 方程](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets) - CFD - [Sod激波管](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/sod) - [Lax激波管](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/lax) - [二维黎曼问题](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/riemann2d) - [库埃特流动](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/couette) + - [声波方程求解](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic) 代码贡献指导：请参考[教程](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/CONTRIBUTION_CN.md) \ No newline at end of file diff --git a/MindFlow/applications/README_EN.md b/MindFlow/applications/README_EN.md index 0840154d71dc52ddb0d3375a7a1168bed62c076a..dbbc5bbae9d203b3ffc72e454aa24e21289d6bde 100644 --- a/MindFlow/applications/README_EN.md +++ b/MindFlow/applications/README_EN.md @@ -25,9 +25,9 @@ The differentiable CFD solver mainly solves the control equation of fluid dynami - [Solve Burgers Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/kno1d) - [Solve Burgers Equation by SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/burgers/sno1d) - [Solve Navier-Stokes Equation by FNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno2d) + - [Solve Navier-Stokes Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) - [Solve Navier-Stokes Equation by SNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno2d) - [Solve Navier-Stokes Equation by FNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/fno3d) - - [Solve Navier-Stokes Equation by KNO](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/kno2d) - [Solve Navier-Stokes Equation by SNO3D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/navier_stokes/sno3d) - [Solve 2D Riemann Problem by CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) - [Solve Shu-Osher Problem by CAE-LSTM](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_lstm) @@ -39,11 +39,17 @@ The differentiable CFD solver mainly solves the control equation of fluid dynami - [Solve 2D Cylinder Flow by CAE-Transformer](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cae_transformer) - [Predict Multi-timestep Complicated Transonic Airfoil by FNO2D and UNET2D](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_driven/airfoil/2D_unsteady) - [Predict Fluid-structure Interaction System by HDNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/fluid_structure_interaction) + - [Prediction of spatiotemporal field of pulsation velocity in cylindrical wake by Cascade Net](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/cascade_net) + - [MultiScaleGNN for Solving Pressure Poisson Equation](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/multiscale_gnn) + - [Turbine Stage Flow Field Prediction and Uncertainty Optimization Design Based on Neural Operator Networks](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/turbine_uq) - Data-Mechanism Fusion - [Solve Convection-Diffusion Equation by PDE-NET](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/pde_net) - [Solve 2D Burgers Equation by PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/burgers_2d) - [Solve 3D Reaction-Diffusion Equation by PeRCNN](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/percnn/gsrd_3d) - [AI Turb Model](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/ai_turbulence_modeling) + - [Physics-encoded Message Passing Graph Network PhyMPGN solving spatiotemporal PDE systems](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/data_mechanism_fusion/phympgn) + - [Physical Field Prediction Model Driven by Data and Physics Hybridization](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/heat_conduction) + - [Fusion of Physical Mechanism for Predicting Complex Flow Temperature Fields](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/superposition) - Physics Driven - [Solve Burgers Equation by PINNs](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/burgers) - [Solve 2D Cylinder Flow by PINNs](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/physics_driven/navier_stokes/cylinder_flow_forward) @@ -60,10 +66,12 @@ The differentiable CFD solver mainly solves the control equation of fluid dynami - [META-PINNs Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/meta_pinns) - [MOE-PINNs Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/moe_pinns) - [R-DLGA Algorithm](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/r_dlga) + - [NSFNets: Physics-informed neural networks for the incompressible Navier-Stokes equations](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/research/nsf_nets) - CFD - [Sod Shock Tube](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/sod) - [Lax Shock Tube](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/lax) - [2D Riemann Problem](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/riemann2d) - [Couette Flow](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/couette) + - [2D Acoustic Wave Equation CBS Solver](https://gitee.com/mindspore/mindscience/tree/master/MindFlow/applications/cfd/acoustic) How to contribute: Please refer to [Tutorial](https://gitee.com/mindspore/mindscience/blob/master/MindFlow/CONTRIBUTION_CN.md) \ No newline at end of file diff --git a/MindFlow/docs/mindflow_archi_old.png b/MindFlow/docs/mindflow_archi_0.1.0.png similarity index 100% rename from MindFlow/docs/mindflow_archi_old.png rename to MindFlow/docs/mindflow_archi_0.1.0.png diff --git a/MindFlow/docs/mindflow_archi_0.2.0.png b/MindFlow/docs/mindflow_archi_0.2.0.png new file mode 100644 index 0000000000000000000000000000000000000000..6198c3df566b6785e38d01096a136a3e01475d58 Binary files /dev/null and b/MindFlow/docs/mindflow_archi_0.2.0.png differ diff --git a/MindFlow/docs/mindflow_archi_cn.png b/MindFlow/docs/mindflow_archi_cn.png index 6198c3df566b6785e38d01096a136a3e01475d58..871ed98e2e4b3b7dbb0e7491e2b658651b161998 100644 Binary files a/MindFlow/docs/mindflow_archi_cn.png and b/MindFlow/docs/mindflow_archi_cn.png differ diff --git a/MindFlow/docs/mindflow_archi_en.png b/MindFlow/docs/mindflow_archi_en.png index 64e4e52950702b35708cabd6e8ec0d08771555d2..e297a7fce307520b451a054864de0f4cbea608c1 100644 Binary files a/MindFlow/docs/mindflow_archi_en.png and b/MindFlow/docs/mindflow_archi_en.png differ diff --git a/MindFlow/mindflow/cell/neural_operators/sno.py b/MindFlow/mindflow/cell/neural_operators/sno.py index ef88d346c7c7bbc179ac76d7f3994925b8989467..9d5f7b71f73a4e02bcf8427598e1fcd8f8f9fb3d 100644 --- a/MindFlow/mindflow/cell/neural_operators/sno.py +++ b/MindFlow/mindflow/cell/neural_operators/sno.py @@ -41,9 +41,9 @@ class SNOKernelCell(nn.Cell): compute_dtype (dtype.Number): The computation type. Default: ``mstype.float32``. Inputs: - - **x** (Tensor) - Tensor of shape :math:`(batch\_size, in\_channels, height, width)`. + - **x** (Tensor) - Tensor with shape :math:`(batch\_size, in\_channels, height, width)`. Outputs: - - **output** (Tensor) -Tensor of shape :math:`(batch\_size, out\_channels, height, width)`. + - **output** (Tensor) -Tensor with shape :math:`(batch\_size, out\_channels, height, width)`. Raises: TypeError: If `in_channels` is not an int. TypeError: If `out_channels` is not an int. @@ -153,29 +153,27 @@ class SNO(nn.Cell): Args: in_channels (int): The number of channels in the input space. out_channels (int): The number of channels in the output space. - hidden_channels (int): The number of channels of the SNO layers input and output. Default: 64. - num_sno_layers (int): The number of spectral layers. Default: 3. + hidden_channels (int): The number of channels of the SNO layers input and output. Default: ``64``. + num_sno_layers (int): The number of spectral layers. Default: ``3``. data_format (str): The input data channel sequence. Default: ``channels_first``. - transforms (list(list(mindspore.Tensor))): The list of direct and inverse polynomial transforms on x, y and z axis, respectively. The list has the following structure: [[transform_x, inv_transform_x], - ... [transform_z, inv_transform_z]]. The shape of transformation matrix should be (n_modes, resolution), + [transform_z, inv_transform_z]]. The shape of transformation matrix should be (n_modes, resolution), where n_modes is the number of polynomial transform modes, resolution is spatial resolution of input - in the corresponding direction. The shape of inverse transformation is (resolution, n_modes). Default: None. - - kernel_size (int): Specifies the height and width of the convolution kernel in SNO layers. Default: 5. - num_usno_layers (int): The number of spectral layers with UNet skip blocks. Default: 0. - num_unet_strides (int): The number of convolutional downsample blocks in UNet skip blocks. Default: 1. + in the corresponding direction. The shape of inverse transformation is (resolution, n_modes). Default: ``None``. + kernel_size (int): Specifies the height and width of the convolution kernel in SNO layers. Default: ``5``. + num_usno_layers (int): The number of spectral layers with UNet skip blocks. Default: ``0``. + num_unet_strides (int): The number of convolutional downsample blocks in UNet skip blocks. Default: ``1``. activation (Union[str, class]): The activation function, could be either str or class. Default: ``gelu``. - compute_dtype (dtype.Number): The computation type. Default: mstype.float32. + compute_dtype (dtype.Number): The computation type. Default: ``mstype.float32``. Should be ``mstype.float32`` or ``mstype.float16``. mstype.float32 is recommended for the GPU backend, mstype.float16 is recommended for the Ascend backend. Inputs: - - **x** (Tensor) - Tensor of shape :math:`(batch\_size, in_channels, resolution)` + - **x** (Tensor) - Tensor with shape :math:`(batch\_size, in_channels, resolution)`. Outputs: - Tensor of shape :math:`(batch\_size, out_channels, resolution)`. + Tensor with shape :math:`(batch\_size, out_channels, resolution)`. Raises: TypeError: If `in_channels` is not an int. @@ -295,7 +293,7 @@ class SNO1D(SNO): r""" The 1D SNO, which contains a lifting layer (encoder), multiple spectral transform layers and a projection layer (decoder). - See documentation for base class, `SNO`. + See documentation for base class, :class:`mindflow.cell.SNO`. Example: >>> import numpy as np @@ -339,7 +337,7 @@ class SNO2D(SNO): r""" The 2D SNO, which contains a lifting layer (encoder), multiple spectral transform layers and a projection layer (decoder). - See documentation for base class, `SNO`. + See documentation for base class, :class:`mindflow.cell.SNO`. Example: >>> import numpy as np @@ -389,7 +387,7 @@ class SNO3D(SNO): r""" The 3D SNO, which contains a lifting layer (encoder), multiple spectral transform layers and a projection layer (decoder). - See documentation for base class, `SNO`. + See documentation for base class, :class:`mindflow.cell.SNO`. Example: >>> import numpy as np diff --git a/docs/api_python/mindflow/cell/mindflow.cell.SNO.rst b/docs/api_python/mindflow/cell/mindflow.cell.SNO.rst new file mode 100644 index 0000000000000000000000000000000000000000..c9e6227abcede9f6b5a8e1f0bf79e8f7ba700fed --- /dev/null +++ b/docs/api_python/mindflow/cell/mindflow.cell.SNO.rst @@ -0,0 +1,37 @@ +mindflow.cell.SNO +========================= + +.. py:class:: mindflow.cell.SNO(in_channels, out_channels, hidden_channels=64, num_sno_layers=3, data_format="channels_first", transforms=None, kernel_size=5, num_usno_layers=0, num_unet_strides=1, activation="gelu", compute_dtype=mstype.float32) + + 谱神经算子（Spectral Neural Operator, SNO）基类，包含一个提升层（编码器）、多个谱变换层（谱空间的线性变换）和一个投影层（解码器）。 + 这是一种类似FNO的架构，但使用多项式变换（Chebyshev、Legendre等）替代傅里叶变换。 + 详细信息请参考谱神经算子论文 `Spectral Neural Operators `_ 。 + + 参数： + - **in_channels** (int) - 输入中的通道数。 + - **out_channels** (int) - 输出中的通道数。 + - **hidden_channels** (int) - SNO层输入和输出的通道数。默认值： ``64``。 + - **num_sno_layers** (int) - 谱层数量。默认值： ``3``。 + - **data_format** (str) - 输入数据的通道顺序。默认值： ``channels_first``。 + - **transforms** (list(list(mindspore.Tensor))) - 沿x、y、z轴的正变换和逆多项式变换列表。结构形式为：[[transform_x, inv_transform_x], [transform_z, inv_transform_z]]。变换矩阵形状应为(n_modes, resolution)，其中n_modes为多项式变换模式数，resolution为对应方向输入的空间分辨率。逆变换矩阵形状为(resolution, n_modes)。默认值： ``None``。 + - **kernel_size** (int) - 指定SNO层中卷积核的高度和宽度。默认值： ``5``。 + - **num_usno_layers** (int) - 带UNet跳跃连接的谱层数量。默认值： ``0``。 + - **num_unet_strides** (int) - UNet跳跃连接中卷积下采样块的数量。默认值： ``1``。 + - **activation** (Union[str, class]) - 激活函数，支持字符串或类形式。默认值： ``gelu``。 + - **compute_dtype** (dtype.Number) - 计算数据类型。默认值： ``mstype.float32``。 + 可选``mstype.float32``或``mstype.float16``。GPU后端推荐使用float32，Ascend后端推荐使用float16。 + + 输入： + - **x** (Tensor) - shape为 :math:`(batch\_size, in_channels, resolution)` 的张量。 + + 输出： + - shape为 :math:`(batch\_size, out_channels, resolution)` 的张量。 + + 异常： + - **TypeError** - 如果 `in_channels` 不是int。 + - **TypeError** - 如果 `out_channels` 不是int。 + - **TypeError** - 如果 `hidden_channels` 不是int。 + - **TypeError** - 如果 `num_sno_layers` 不是int。 + - **TypeError** - 如果 `transforms` 不是list。 + - **ValueError** - 如果 `transforms` 长度不在(1, 2, 3)范围内。 + - **TypeError** - 如果 `num_usno_layers` 不是int。 diff --git a/docs/api_python/mindflow/cell/mindflow.cell.SNO1D.rst b/docs/api_python/mindflow/cell/mindflow.cell.SNO1D.rst new file mode 100644 index 0000000000000000000000000000000000000000..6da9f35a844fb33550c6138e4844b765ba1f509b --- /dev/null +++ b/docs/api_python/mindflow/cell/mindflow.cell.SNO1D.rst @@ -0,0 +1,6 @@ +mindflow.cell.SNO1D +========================= + +.. py:class:: mindflow.cell.SNO1D(in_channels, out_channels, hidden_channels=64, num_sno_layers=3, data_format="channels_first", transforms=None, kernel_size=5, activation="gelu", compute_dtype=mstype.float32) + + 一维谱神经算子，包含一个提升层（编码器）、多个谱变换层（谱空间的线性变换）和一个投影层（解码器）。参见基类文档 :class:`mindflow.cell.SNO`。 diff --git a/docs/api_python/mindflow/cell/mindflow.cell.SNO2D.rst b/docs/api_python/mindflow/cell/mindflow.cell.SNO2D.rst new file mode 100644 index 0000000000000000000000000000000000000000..cee9ea99319cb91a74169cfaf898bc15d4ae1db6 --- /dev/null +++ b/docs/api_python/mindflow/cell/mindflow.cell.SNO2D.rst @@ -0,0 +1,6 @@ +mindflow.cell.SNO2D +========================= + +.. py:class:: mindflow.cell.SNO2D(in_channels, out_channels, hidden_channels=64, num_sno_layers=3, data_format="channels_first", transforms=None, kernel_size=5, activation="gelu", compute_dtype=mstype.float32) + + 二维谱神经算子，包含一个提升层（编码器）、多个谱变换层（谱空间的线性变换）和一个投影层（解码器）。参见基类文档 :class:`mindflow.cell.SNO`。 diff --git a/docs/api_python/mindflow/cell/mindflow.cell.SNO3D.rst b/docs/api_python/mindflow/cell/mindflow.cell.SNO3D.rst new file mode 100644 index 0000000000000000000000000000000000000000..66bc7cc3a51cab772134a09a55810e66051e80e6 --- /dev/null +++ b/docs/api_python/mindflow/cell/mindflow.cell.SNO3D.rst @@ -0,0 +1,6 @@ +mindflow.cell.SNO3D +========================= + +.. py:class:: mindflow.cell.SNO3D(in_channels, out_channels, hidden_channels=64, num_sno_layers=3, data_format="channels_first", transforms=None, kernel_size=5, activation="gelu", compute_dtype=mstype.float32) + + 三维谱神经算子，包含一个提升层（编码器）、多个谱变换层（谱空间的线性变换）和一个投影层（解码器）。参见基类文档 :class:`mindflow.cell.SNO`。 diff --git a/docs/api_python/mindflow/mindflow.cell.rst b/docs/api_python/mindflow/mindflow.cell.rst index 2f2d19ef87491591e8a7c521d5916a4edc799c0e..eb677b002596ae513859ac4b407019f0ffb9f6f4 100644 --- a/docs/api_python/mindflow/mindflow.cell.rst +++ b/docs/api_python/mindflow/mindflow.cell.rst @@ -25,6 +25,10 @@ mindflow.cell mindflow.cell.PDENet mindflow.cell.PeRCNN mindflow.cell.ResBlock + mindflow.cell.SNO + mindflow.cell.SNO1D + mindflow.cell.SNO2D + mindflow.cell.SNO3D mindflow.cell.UNet2D mindflow.cell.ViT mindflow.cell.get_activation diff --git a/docs/api_python_en/mindflow/mindflow.cell.rst b/docs/api_python_en/mindflow/mindflow.cell.rst index a4c79c080e024f0c31a3cb8e4568390023e51caf..f4136a8e7353ea29c256a106fedc037babbb7467 100644 --- a/docs/api_python_en/mindflow/mindflow.cell.rst +++ b/docs/api_python_en/mindflow/mindflow.cell.rst @@ -25,6 +25,10 @@ mindflow.cell mindflow.cell.PDENet mindflow.cell.PeRCNN mindflow.cell.ResBlock + mindflow.cell.SNO + mindflow.cell.SNO1D + mindflow.cell.SNO2D + mindflow.cell.SNO3D mindflow.cell.UNet2D mindflow.cell.ViT mindflow.cell.get_activation