同济大学计算机科学与技术学院 AI +大气海洋研究团队主要由两名正教授、一名助理教授及其指导下的15名博士生和12名硕士生所组成。主要从事人工智能与大气海洋科学的交叉科学研究。研究方向涵盖：人工智能及其可解释性、机器学习、神经网络、大气及海洋大数据分析、智能数据同化、AI 多圈层耦合、大模型误差溯源等。当前主持国家自然科学基金【原创探索】计划项目一项、国家自然科学基金联合基金【重点】项目一项、国家自然科学基金【面上】项目一项、国家重点研发计划课题两项、以及上海市科委重点项目一项。团队在该研究方向已发表高影响因子学术论文140余篇，获授国家发明专利16项。研发的“天行”气象大模型入选中国气象局人工智能天气预报大模型示范计划。研发的 SmaAt-UNet 海冰预测系统在国际海冰预测网络（SIPN）中预报精度及预报时效均排名全球第二，并为“雪龙2”号科考船2024航次提供了精准海冰预报。团队研究地点：同济大学嘉定校区济事楼316左实验室。

1. 国家自然科学基金联合基金【重点】项目“基于因果推断和物理引导的面向天气预报与气候预测的深度学习理论算法及可解释性研究”，项目编号：U2142211，2022.01-2025.12，主持。

2. 国家自然科学基金【面上】项目“多模态数据驱动的海气耦合台风概率预报模型”，项目编号：42075141，2021.01-2024.12，主持。

3. 国家重点研发计划“全球变化及应对”专项项目“大数据与深度学习方法创新地球系统模式发展及应用研究”之课题四“深度神经网络预测模型发展与动力模式结果订正方法研究”，课题编号2020YFA0608004，2020-11至2025-04，第二负责人。

4. 上海市2020年度“科技创新行动计划”社会发展科技攻关“公共安全/突发公共安全事件应急处理处置”专题项目“基于风云卫星智能精准观测针对极端天气事件的长三角航空运行安全应对研究”之课题二“针对CNOP的高效智能算法开发与应用“，课题编号 20dz1200702， 2020-09-01至2023-08-31，主持。

1. 国家自然科学基金联合基金【重点】项目“基于因果推断和物理引导的面向天气预报与气候预测的深度学习理论算法及可解释性研究”，项目编号：U2142211，2022.01-2025.12，主持。

2. 国家自然科学基金【面上】项目“多模态数据驱动的海气耦合台风概率预报模型”，项目编号：42075141，2021.01-2024.12，主持。

3. 国家重点研发计划“全球变化及应对”专项项目“大数据与深度学习方法创新地球系统模式发展及应用研究”之课题四“深度神经网络预测模型发展与动力模式结果订正方法研究”，课题编号2020YFA0608004，2020-11至2025-04，第二负责人。

4. 上海市2020年度“科技创新行动计划”社会发展科技攻关“公共安全/突发公共安全事件应急处理处置”专题项目“基于风云卫星智能精准观测针对极端天气事件的长三角航空运行安全应对研究”之课题二“针对CNOP的高效智能算法开发与应用“，课题编号 20dz1200702， 2020-09-01至2023-08-31，主持。

主持研发出的模型与系统：

【1】“天行”气象大模型，入选中国气象局人工智能天气预报大模型示范计划。

【2】 SmaAt-UNet 海冰预测系统，在国际海冰预测网络（SIPN）中预报精度及预报时效均排名全球第二，并为“雪龙2”号科考船2024航次提供了精准海冰预报。

发表的学术论文：

【1】TianXing: ALinear Complexity Transformer Model with Explicit Attention Decay for GlobalWeather Forecasting ，Yuan, Shijin; Wang, Guansong; Mu,Bin; Zhou, Feifan，Advances in Atmospheric Sciences |2025年

【2】Incorporatingheat budget dynamics in a Transformer-based deep learning model for skillfulENSO prediction，Mu, Bin; 崔悦涵;Yuan, Shijin; Qin, Bo，NPJ CLIMATE AND ATMOSPHERICSCIENCE | 2024年 | 7卷 | 1期

【3】基于深度学习的全球热带气旋生成预测模型及其可解释性分析，穆斌; 王馨; 袁时金; 陈宇轩; 王冠淞等7名作者，中国科学:地球科学 | 2024年 | 54卷 | 12期 | 3708-3733页

【4】MultivariateUpstream Kuroshio Transport (UKT) Prediction and Targeted Observation SensitiveArea Identification of UKT Seasonal Reduction，穆斌;Yang-Hu, Yifan; Qin, Bo; 袁时金，OCEAN MODELLING | 2024年 | 189卷

【5】A generativeadversarial network-based unified model integrating bias correction anddownscaling for global SST，袁时金; 冯新; 穆斌; Qin, Bo; Wang, Xin等6名作者，Atmospheric and Oceanic Science Letters| 2024年 | 17卷 | 1期

【6】Toward aLearnable Climate Model in the Artificial Intelligence Era，Huang, Gang; Wang, Ya; Ham, Yoo-Geun; 穆斌;Tao, Weichen等6名作者，Advances inAtmospheric Sciences | 2024年 | 41卷 | 7期 | 1281-1288页

【7】A deeplearning-based bias correction model for Arctic sea ice concentration towardsMITgcm，袁时金; 朱师辰; Luo, Xiaodan; 穆斌，Ocean Modelling | 2024年 | 188卷

【8】Developingintelligent Earth System Models: An AI framework for replacing sub-modulesbased on incremental learning and its application，穆斌; 赵紫君; 袁时金; Qin, Bo; Dai, Guo-Kun等6名作者，Atmospheric Research | 2024年 | 302卷

【9】An extension toensemble forecast of conditional nonlinear optimal perturbation consideringnonlinear interaction between initial and model parametric uncertainties ，Mu, Bin; Zhao, Zi-Jun; Yuan, Shi-Jin; Chen, Xing-Rong; Qin, Bo等6名作者，Atmospheric Research | 2024年 | 311卷

【10】A deeplearning-based global tropical cyclogenesis prediction model and itsinterpretability analysis，Mu, Bin; Wang, Xin; Yuan,Shijin; Chen, Yuxuan; 王冠淞等7名作者，ScienceChina Earth Sciences | 2024年

【11】IceTFT v1.0.0:interpretable long-term prediction of Arctic sea ice extent with deep learning，穆斌; 罗晓丹; 袁时金; Liang,Xi，GEOSCIENTIFIC MODEL DEVELOPMENT | 2023年 | 16卷 | 16期 |4677-4697页

【12】A paralleledembedding high-dimensional Bayesian optimization with additive Gaussian kernelsfor solving CNOP，袁时金; 刘娅璇; Qin,Bo; 穆斌; Zhang, Kun，OceanModelling | 2023年 | 184卷

【13】A radiativetransfer deep learning model coupled into WRF with a generic fortran torchadaptor，穆斌; 陈璐; 袁时金; Qin, Bo，FRONTIERS IN EARTH SCIENCE | 2023年 | 11卷

【14】Dimensionshifting based intelligent algorithm framework to solve conditional nonlinearoptimal perturbation，袁时金; 刘娅璇;Zhang, Huazhen; 穆斌，Computers and Geosciences | 2023年 | 176卷

【15】NAO SeasonalForecast Using a Multivariate Air–Sea Coupled DeepLearning Model Combined with Causal Discovery，穆斌; 姜欣; 袁时金; 崔悦涵; Qin, Bo，Atmosphere | 2023年 | 14卷 | 5期

【16】ErrorEvolutions and Analyses on Joint Effects of SST and SL via Intermediate CoupledModels and Conditional Nonlinear Optimal Perturbation Method，穆斌; 秦小云; 袁时金; Qin, Bo，JOURNAL OF MARINE SCIENCE AND ENGINEERING | 2023年 | 11卷 | 5期

【17】Estimating thetropical cyclone wind structure using physics-incorporated networks，袁时金; 尤钱湖; 穆斌; 秦博; Xu Jing，FRONTIERS IN EARTH SCIENCE | 2023年 | 10卷

【18】PIRT: APhysics-Informed Red Tide Deep Learning Forecast Model ConsideringCausal-Inferred Predictors Selection，穆斌; 秦博; 袁时金; Wang, Xin; Chen, Yuxuan，IEEE Geoscience and Remote Sensing Letters | 2023年 | 20卷

【19】CAU: ACausality Attention Unit for Spatial-temporal Sequence Forecast，Qin, Bo; Meng, Fanqing; Fang, Xianghui; Dai, Guokun; 袁时金等6名作者，IEEE Transactions on Multimedia | 2023年 | 1-15页

【20】ENSO-GTC: ENSODeep Learning Forecast Model With a Global Spatial-Temporal TeleconnectionCoupler，穆斌; 秦博; 袁时金，Journal of Advances in Modeling Earth Systems | 2022年 | 14卷 | 12期

【21】Featureextraction-based intelligent algorithm framework with neural network forsolving conditional nonlinear optimal perturbation，袁时金;张华桢; 刘娅璇; 穆斌，Soft Computing | 2022年 | 26卷 | 14期 | 6907-6924页

【22】A deep learningurban traffic congestion forecast model blending the temporal continuity andperiodicity，穆斌; Huang, Yuxi，ACMInternational Conference Proceeding Series | 2022年 |602-607页

【23】EnsembleForecast for Tropical Cyclone Based on CNOP-P Method: A Case Study of WRF Modeland Two Typhoons，袁时金; Shi Bo; 赵紫君; 穆斌; Zhou Fei-fan等6名作者，JOURNAL OF TROPICAL METEOROLOGY | 2022年 | 28卷 | 2期 | 121-138页

【24】Simulation,precursor analysis and targeted observation sensitive area identification fortwo types of ENSO using ENSO-MC v1.0 ，穆斌; 崔悦涵; 袁时金; 秦博，GEOSCIENTIFICMODEL DEVELOPMENT | 2022年 | 15卷| 10期 | 4105-4127页

【25】OptimalPrecursors Identification for North Atlantic Oscillation Using the ParallelIntelligence Algorithm，穆斌; 李婧; 袁时金; 罗晓丹; Dai, Guokun，ScientificProgramming | 2022年 | 2022卷

【26】The NAOVariability Prediction and Forecasting with Multiple Time Scales Driven by ENSOUsing Machine Learning Approaches，穆斌; 李婧; 袁时金; Luo, Xiaodan，ComputationalIntelligence and Neuroscience | 2022年 | 2022卷

【27】GCN Modelcombined with Bi-GRU for traffic prediction，穆斌; Zhen,Lin，Proceedings of SPIE - The International Society forOptical Engineering | 2022年 | 12259

【28】ENSO-ASC 1.0.0:ENSO deep learning forecast model with a multivariate air-sea coupler，穆斌; 秦博; 袁时金，GEOSCIENTIFICMODEL DEVELOPMENT | 2021年 | 14卷| 11期 | 6977-6999页

【29】The ELM Modelwith Residual Compensation Based on ARIMA for North Atlantic Oscillation IndexPrediction，Luo, Xiaodan; 袁时金; 穆斌; Li, Jing，ACM International ConferenceProceeding Series | 2021年 | 122-126页

【30】An improvedcontinuous tabu search algorithm with adaptive neighborhood radius andincreasing search iteration times strategies，袁时金; 徐运佳; 穆斌; Zhang, Linlin; Ren, Juhui等7名作者，International Journal on ArtificialIntelligence Tools | 2021年 | 30卷 | 2期

【31】TyphoonIntensity Forecasting Based on LSTM Using the Rolling Forecast Method，袁时金; Wang, Cheng; 穆斌; Zhou, Feifan; Duan,Wansuo，ALGORITHMS | 2021年 | 14卷 | 3期

【32】Efficientexecutions of community earth system model onto accelerators using GPUs，袁时金; Wang, Cheng; 穆斌; 罗晓丹，ACM International Conference Proceeding Series | 2020年 | 192-199页

【33】CNOP-P-BasedParameter Sensitivity Analysis for North Atlantic Oscillation in CommunityEarth System Model Using Intelligence Algorithms，穆斌; 李婧; 袁时金; 罗晓丹; Dai,Guokun，ADVANCES IN METEOROLOGY | 2020年 | 2020卷

【34】ApplyingConvolutional LSTM Network to Predict El Ni?o Events: Transfer Learning fromthe Data of Dynamical Model and Observation，穆斌; Ma,Shaoyang; 袁时金; Xu, Hui，ICEIEC2020 - Proceedings of 2020 IEEE 10th International Conference on ElectronicsInformation and Emergency Communication，2020年 | 215-219页

【35】DataAssimilation by Artificial Neural Network using Conventional Observation forWRF Model，袁时金; Shi, Bo; 穆斌，ACMInternational Conference Proceeding Series | 2020年 |62-67页

【36】Multi-scaledownscaling with bayesian convolution network for ENSO SST pattern，穆斌; 秦博; 袁时金，Proceedings- 2020 5th International Conference on Electromechanical Control Technology andTransportation, ICECTT 2020 | 2020年 | 359-362页

【37】A ClimateDownscaling Deep Learning Model considering the Multiscale Spatial Correlationsand Chaos of Meteorological Events，穆斌; 秦博; 袁时金; 秦小云，MathematicalProblems in Engineering | 2020年 | 2020卷

【38】Prediction ofnorth atlantic oscillation index associated with the sea level pressure usingDWT-LSTM and DWT-ConvLSTM networks，穆斌; 李婧; 袁时金; 罗晓丹，MathematicalProblems in Engineering | 2020年 | 2020卷

【39】ApplyingConvolutional LSTM Network to Predict El Nino Events: Transfer Learning fromThe Data of Dynamical Model and Observation，穆斌; 马少阳; 袁时金; Xu, Hui，PROCEEDINGSOF 2020 IEEE 10TH INTERNATIONAL CONFERENCE ON ELECTRONICS INFORMATION ANDEMERGENCY COMMUNICATION (ICEIEC 2020) | 2020年 | 215-219页

【40】NAO IndexPrediction using LSTM and ConvLSTM Networks Coupled with Discrete WaveletTransform，穆斌; 李婧; 袁时金; 罗晓丹; Dai, Guokun，Proceedingsof the International Joint Conference on Neural Networks | 2019年 | 2019-July卷，匈牙利布达佩斯

【41】ENSOForecasting over Multiple Time Horizons Using ConvLSTM Network and RollingMechanism，穆斌; Peng, Cheng; 袁时金;Chen, Lei，Proceedings of the International JointConference on Neural Networks | 2019年 | 2019-July卷，匈牙利布达佩斯

【42】IdentifyingTyphoon Targeted Observations Sensitive Areas Using the Gradient DefinitionBased Method，穆斌; Ren, Juhui; 袁时金; Zhou, Feifan，ASIA-PACIFIC JOURNAL OFATMOSPHERIC SCIENCES | 2019年 | 55卷 | 2期 | 195-207页

【43】Prediction ofnorth atlantic oscillation index with convolutional LSTM based on ensembleempirical mode decomposition，袁时金; 罗晓丹; 穆斌; Li, Jing; Dai, Guokun，Atmosphere | 2019年 | 10卷 | 5期

【44】INTELLIGENTALGORITHMS FOR SOLVING CNOP AND THEIR APPLICATIONS IN ENSO PREDICTABILITY ANDTROPICAL CYCLONE ADAPTIVE OBSERVATIONS，穆斌; ZhangLin-lin; 袁时金; 钱一闻; 温仕成等7名作者，JOURNAL OF TROPICAL METEOROLOGY | 2019年 | 25卷 | 1期 | 63-81页

【45】The OptimalPrecursors for ENSO Events Depicted Using the Gradientdefinition-based Methodin an Intermediate Coupled Model ，穆斌; Ren, Juhui; 袁时金; Zhang, Rong-Hua; Chen, Lei等6名作者，Advances in Atmospheric Sciences | 2019年 |36卷 | 12期 | 1381-1392页

【46】Optimalprecursors of double-gyre regime transitions with an adjoint-free method，袁时金; 李糜; Wang, Qiang; Zhang, Kun; 张华桢等6名作者，Journal of Oceanology and Limnology |2019年 | 37卷 | 4期 | 1137-1153页

【47】CNOP-P-basedparameter sensitivity for double-gyre variation in ROMS with simulatedannealing algorithm，袁时金; 张华桢; 李糜; 穆斌，Journal of Oceanology and Limnology |2019年 | 37卷 | 3期 | 957-967页

【48】A modifieddirect search algorithm based on kernel density estimator with three mappingstrategies for solving nonlinear optimization，Zhang,Lin-Lin; 穆斌; 袁时金，Journal ofComputers (Taiwan) | 2019年 | 30卷 | 4期 | 17-30页

【49】ParallelPCA-Based Bacterial Foraging Optimization Algorithm for Identifying OptimalPrecursors of North Atlantic Oscillation，穆斌; Jing Li; 袁时金; 罗晓丹; Guokun Dai，2019IEEE 21st International Conference on High Performance Computing andCommunications; IEEE 17th International Conference on Smart City; IEEE 5thInternational Conference on Data Science and Systems (HPCC/SmartCity/DSS).Proceedings | 2019年 | 1171-7页

【50】A novelapproach for solving CNOPs and its application in identifying sensitive regionsof tropical cyclone adaptive observations，Zhang,Linlin; 穆斌; 袁时金; Zhou, Feifan，NONLINEAR PROCESSES IN GEOPHYSICS | 2018年 |25卷 | 3期 | 693-712页

【51】Paralleldynamic search fireworks algorithm with linearly decreased dimension numberstrategy for solving conditional nonlinear optimal perturbation，穆斌; 赵珺晖; 袁时金; 颜景豪，Proceedings of the International Joint Conference on Neural Networks| 2017年 | 2017-May卷 | 2314-2321页，美国阿拉斯加

【52】CNOP-BasedSensitive Areas Identification for Tropical Cyclone Adaptive Observations withPCAGA Method ，Zhang, Lin-Lin; 袁时金; 穆斌; Zhou, Fei-Fan，ASIA-PACIFICJOURNAL OF ATMOSPHERIC SCIENCES | 2017年 | 53卷 | 1期 | 63-73页

【53】An efficientapproach based on the gradient definition for solving conditional nonlinearoptimal perturbation ，穆斌; Ren, Juhui; 袁时金，Mathematical Problems in Engineering | 2017年| 2017卷

【54】CACO-LD:Parallel Continuous Ant Colony Optimization with Linear Decrease Strategy forSolving CNOP，袁时金; 陈韵怡; 穆斌，Lecture Notes in Computer Science (including subseries Lecture Notesin Artificial Intelligence and Lecture Notes in Bioinformatics) | 2017年 | 10637 LNCS卷 | 494-503页

【55】ParallelModified Artificial Bee Colony Algorithm for Solving Conditional NonlinearOptimal Perturbation，Ren, Juhui; 袁时金; 穆斌，Proceedings - 18th IEEE InternationalConference on High Performance Computing and Communications, 14th IEEEInternational Conference on Smart City and 2nd IEEE International Conference onData Science and Systems, HPCC/SmartCity/DSS 2016 | 2016年 | 333-340页，澳大利亚悉尼

【56】PCAFP forSolving CNOP in Double-Gyre Variation and Its Parallelization on Clusters，袁时金; 李糜; 穆斌; Wang,Jingpeng，Proceedings - 18th IEEE InternationalConference on High Performance Computing and Communications, 14th IEEEInternational Conference on Smart City and 2nd IEEE International Conference onData Science and Systems, HPCC/SmartCity/DSS 2016 | 2016年 | 284-291页，澳大利亚悉尼

【57】PCGD: Principalcomponents-based great deluge method for solving CNOP，温仕成; 袁时金; 穆斌; Li,Hongyu; Ren, Juhui，2015 IEEE CONGRESS ON EVOLUTIONARYCOMPUTATION (CEC) | 2015年 | 1513-1520页

【58】PCAGA:Principal Component Analysis Based Genetic Algorithm for Solving ConditionalNonlinear Optimal Perturbation，穆斌; Zhang, Linlin; 袁时金; Li, Hongyu，2015 INTERNATIONAL JOINTCONFERENCE ON NEURAL NETWORKS (IJCNN) | 2015年

【59】Paralleldynamic step size sphere-gap transferring algorithm for solving conditionalnonlinear optimal perturbation，袁时金; 颜景豪; 穆斌; Li, Hongyu，Proceedings- 2015 IEEE 17th International Conference on High Performance Computing andCommunications, 2015 IEEE 7th International Symposium on Cyberspace Safety andSecurity and 2015 IEEE 12th International Conference on Embedded Software andSystems, HPCC-CSS-ICESS 2015 | 2015年 | 559-565页

【60】PPSO: PCA basedparticle swarm optimization for solving conditional nonlinear optimalperturbation，穆斌; 温仕成; 袁时金; Li, Hongyu，Computers and Geosciences |2015年 | 83卷 | 65-71页

【61】A ParallelSensitive Area Selection-Based Particle Swarm Optimization Algorithm for FastSolving CNOP，Yuan Shijin, Ji Feng, Yan Jinghao, Mu Bin，22nd International Conference on Neural Information Processing(ICONIP)，土耳其伊斯坦布尔

【62】ParallelCooperative Co-evolution Based Particle Swarm Optimization Algorithm forSolving Conditional Nonlinear Optimal Perturbation，YuanShijin, Zhao Li, Mu Bin，22nd International Conferenceon Neural Information Processing (ICONIP)，土耳其伊斯坦布尔

【63】Paralleldynamic step size sphere-gap transferring algorithm for solving conditionalnonlinear optimal perturbation，Yuan Shijin, YanJinghao, Mu Bin, Li Hongyu，17th IEEE InternationalConference on High Performance Computing and Communications, IEEE 7thInternational Symposium on Cyberspace Safety and Security and IEEE 12thInternational Conference on Embedded Software and Systems, 美国纽约

【64】PCAGA:Principal Component Analysis Based Genetic Algorithm for Solving ConditionalNonlinear Optimal Perturbation，Bin Mu，Linlin Zhang，Shijin Yuan，Hongyu Li，2015 International JointConference on Neural Networks (IJCNN)，爱尔兰基拉尼

【65】User-QoS-basedWeb Service Clustering for QoS Prediction，Fuxin Chen,Shijin Yuan, Bin Mu，the 22nd IEEE InternationalConference on Web Services， CCF-B，美国纽约

【66】PCGD: Principalcomponents-based great deluge method for solving CNOP，Wen,Shicheng，Yuan, Shijin，Mu, Bin，Li, Hongyu，Ren, Juhui，IEEE Congress on Evolutionary Computation, CEC 2015，日本仙台