铁道科学与工程学报2024，Vol.21Issue(4)：1602-1612,11.DOI:10.19713/j.cnki.43-1423/u.T20231072
基于LSTM神经网络的牵引站电气设备耦联体系地震响应预测
Seismic response prediction of electrical equipment interconnected system of traction station based on LSTM neural network
摘要
Abstract
In the railway traction substation,the flexible conductor-electrical equipment interconnected system has strong geometric nonlinearity.To improve the analysis efficiency of the system,an improved recursive prediction method for the flexible conductor-electrical equipment interconnected system was proposed.The prediction model was established based on recursive long short-term memory(LSTM)neural network and the Dropout regularization.The theoretical analysis model of the interconnected system between the flexible conductors and electrical equipment was established,fully considering the coupling effects of flexible conductors on adjacent equipment.Besides,to fully reflect the generalization ability of the model,41 seismic time histories with large differences in peak ground acceleration(PGA),frequency spectrum,and duration were selected.According to the recursive scheme,the selected seismic time histories,along with the displacement responses obtained by the theoretical analysis model of the flexible conductor-electrical equipment interconnected system,were subjected to sliding-window slicing treatment and established the mapping relationship between model input features and output labels.In addition,the model was used to predict the seismic displacement responses of the interconnected system,and several evaluation indices were used to evaluate the model performance comprehensively.The results indicated that the LSTM recursive prediction model exhibits excellent performance in seismic response prediction.When combined with the Dropout regularization,it effectively prevents model overfitting and improves the adaptability of the model.For seismic time-history data with significant variations,the model can rapidly predict earthquake responses with lower errors and higher correlations,demonstrating high accuracy,efficiency,and generalization capability.This method can quickly and accurately predict the seismic responses of the flexible conductor-electrical equipment interconnected system at any time point,providing a new research idea for the seismic design of railway traction substations.关键词
长短期记忆神经网络/电气设备/软导线/耦联体系/地震响应预测Key words
long short-term memory neural network/electrical equipment/flexible conductor/interconnected system/seismic response prediction分类
动力与电气工程引用本文复制引用
郭彦颜,陈雅芳,何畅,余玉洁,何紫薇,蒋丽忠..基于LSTM神经网络的牵引站电气设备耦联体系地震响应预测[J].铁道科学与工程学报,2024,21(4):1602-1612,11.基金项目
国家自然科学基金资助项目(52008406) （52008406）
湖南省自然科学基金资助项目(2021JJ40737) （2021JJ40737）
