基于LSTM-TL的大坝变形监控模型研究OA

Aiming at the issue of low prediction accuracy caused by small sample sizes and irregular data in dam safety monito-ring models,this paper proposes a new deformation monitoring model that integrates transfer learning(TL)with a long short-term memory neural network(LSTM).The specific methodology is as follows:first,the source domain model and the target do-main model are constructed using finite element simulation data and measured data,respectively.Then,well-performing parame-ters are extracted from the source domain model and transferred to the target domain model as initial parameters.Finally,based on the measured data,the initial parameters of the target domain model are fine-tuned to obtain the optimal parameters,thereby gen-erating the final target domain monitoring model.Engineering application demonstrates that the LSTM-TL model can effectively suppress noise interference and overfitting induced by small sample of the data,significantly improving the model's accuracy.The regularity similarity(i.e.,non-numerical similarity)between the source domain and the target domain is the main factor influ-encing model accuracy.When the sample size is less than 150,the accuracy of the LSTM-TL model is significantly better than that of the traditional LSTM model;the performance of the two models becomes comparable when the sample size exceeds 150.The research findings can provide a reference for real-time dam safety monitoring and early warning.