复杂系统不确定性量化和模型验证的最新进展.pptx

《复杂系统不确定性量化和模型验证的最新进展.pptx》由会员分享，可在线阅读，更多相关《复杂系统不确定性量化和模型验证的最新进展.pptx（46页珍藏版）》请在三个皮匠报告上搜索。

1、Recent developments in uncertainty quantification and model validation for complex systems,Roger Ghanem,USC,New York Scientific Data Summit 2024:Addressing Data Challengesin Digital TwinsBrookhaven National Laboratory September 1617,2024,MOTIVATION,RESOLVED MODELS OF PHYSICAL PHENOMENA CANNOT BE EXE

2、CUTED IN REALTIME.WE TYPICALLY INTRODUCE ERRORS TRYING TO RUN FASTER:MUST REALIZE THAT ERRORS ARE INTRODUCEDIDENTIFY THE ERRORSANALYZE THE ERRORS(math or heuristics)ASSESS CONSEQUENCES OF ERRORS,THRUST AREAS,STOCHASTIC MODELINGFASTMATHAUTONOMY,EXAMPLE THAT RUNS ACROSS THRUSTS:MODEL VALIDATION,VERIFI

3、CATION:Ascertain that we are solving the equations correctly.TEST PREDICTIONS AGAINST EXACT SOLUTION OF EQUATIONSVALIDATION:Ascertain that we are solving the correct equations.TEST PREDICTIONS AGAINST DATAMore reasonably:Determine to what extent the data does not invalidate the model.,TRADITIONAL TH

4、INKING ABOUT VALIDATION,COMPARES MODEL PREDICTION TO EVIDENCEUSING UNCERTAINTY QUANTIFICATION TO PROVIDE A CLOUD OF PREDICTIONS THAT IS REASONABLE,TRADITIONAL THINKING ABOUT VALIDATION,COMPARES MODEL PREDICTION TO EVIDENCEUSING UNCERTAINTY QUANTIFICATION TO PROVIDE A CLOUD OF PREDICTIONS THAT IS REA

5、SONABLE,WE OFTEN DO NOT HAVE MODELS(COMPLETELY DATA DRIVEN)OROPERATING CONDITIONS BETWEEN MODELS AND EVIDENCE ARE DIFFERENT AND CANNOT BE RECONCILED,TRADITIONAL THINKING ABOUT VALIDATION,COMPARES MODEL PREDICTION TO EVIDENCEUSING UNCERTAINTY QUANTIFICATION TO PROVIDE A CLOUD OF PREDICTIONS THAT IS R

6、EASONABLE,WE OFTEN DO NOT HAVE MODELS(COMPLETELY DATA DRIVEN)OROPERATING CONDITIONS BETWEEN MODELS AND EVIDENCE ARE DIFFERENT AND CANNOT BE RECONCILED,WE SHOULD STILL BE ABLE TO VALIDATE A MODEL,APPROACH,INSIGHT FROM THE HUMAN BODY:WE CAN RELATE BODY TEMPERATURE AND BLOOD PRESSURE TAKEN AT THE WRIST