Seismic response prediction model of asphalt concrete core sand-gravel dams and its application
Although deep learning has been widely used to predict the nonlinear seismic response of structures,how to construct its network framework and how to select its hyperparameters are still controversial issues,because either of them may lead to problems such as low computational efficiency and low-accuracy predictions.The seismic response of asphalt concrete core sand-gravel(ACCSG)dams is usually depicted by a data series,which can actually be mined and predicted by a time series prediction model.This paper presents a long short-term memory(LSTM)neural network model that is based on the genetic algorithm(GA)and the particle swarm optimization(PSO)algorithm.This GAPSO-LSTM model overcomes the drawback of low prediction accuracy caused by difficulty in determining the hyperparameters of the traditional network structure,and achieves the accurate prediction goal of the nonlinear dynamic response of an ACCSG dam.Its prediction accuracy is compared with the convolutional neural network(CNN)model,LSTM single neural network model,and PSO-LSTM neural network model without GA optimization.The results show that compared with the other network models,the GAPSO-LSTM network model has higher prediction accuracy for the seismic response of an ACCSG dam.It overcomes the blindness of subjective selection of hyperparameters,and relieves the local convergence problem of the PSO algorithm,thus providing a new idea for seismic performance evaluation of ACCSG dams.
国家科技期刊平台
NSTL
万方数据
