基于蜣螂优化算法与长短期记忆网络的大坝变形预测组合模型研究OA

To address the problems that traditional dam safety monitoring models struggle to accurately characterize the complex nonlinear relationship between influencing factors and effect quantities,and that the single LSTM model relies on manual hyperparameter tuning and is prone to local optima,a combined prediction model based on the Dung Beetle Optimizer(DBO)and Long Short-Term Memory(LSTM)network is proposed.The DBO is used to globally optimize key hyperparameters of LSTM,such as the initial learning rate,the number of hidden units,and the number of iterations,to construct the DBO-LSTM model,which is verified using monitoring data of the horizontal displacement of the wire extensometer at the crest of a concrete gravity dam.The results show that the prediction accuracy of the DBO-LSTM model is significantly superior to that of the traditional multiple regression model and the single LSTM model.Its accuracy is comparable to that of the GA-LSTM model,but it has faster convergence speed,simpler parameter setting,and more stable optimization results.The DBO-LSTM combined model can effectively improve the accuracy and stability of dam deformation prediction in complex time-series prediction scenarios,but its advantages are not obvious in simple time-series data or strong noise environments,and it has a relatively high computational cost.