Dam Deformation Prediction Model Based On ELM Optimized Improved Bat Algorithm

As an important part of dam safety monitoring system,the dam deformation forecast method research has become a hot topic of scholars.However,dam deformation prediction has strong nonlinear relationship and complex influencing factors.Traditional prediction models have been difficult to meet the accuracy requirements of DAMS and other water conservancy projects due to complex parameter setting and unstable prediction performance.Therefore,this paper applies the extreme learning machine(ELM)in machine learning to the dam deformation prediction field,and proposes a dam deformation prediction model with improved bat algorithm(IBA)to optimize the parameters of ELM.At the same time,a dam deformation prediction model based on bat algorithm(BA)and genetic algorithm(GA)to optimize ELM parameters was established to compare and verify the optimization effect of IBA algorithm and The BP neural network model of dam deformation prediction based on gradient descent method is established to verify the superiority of the prediction performance of extreme learning machine.In each iteration of Ba algorithm,bat individuals will gradually approach the current optimal individual position,and finally the whole bat population is limited in the same interval.This problem is not obvious when solving low dimensional problems,but in reality,it is often faced with the problem of solving high dimensional space.Therefore,in practical application,it is easy to lead to poor diversity of bat population,and bat individuals are easy to fall into the local optimal value,reducing the final convergence accuracy.Aiming at the existing problems of bat algorithm,this paper integrates the crossover and mutation mechanism of GA algorithm into the optimization process of bat algorithm to improve the diversity of bat population and enhance the ability of bat individuals to break through the local optimal value in the iterative process.Based on IBA algorithm,the initial network connection weights and thresholds of elm are optimized to construct IBA＿Elm dam deformation prediction model.The IBA＿The minimum error between the predicted value and the measured value of elm model is taken as the solution problem of IBA algorithm,so as to construct the IBA algorithm with the network connection weights and thresholds of extreme learning machine as independent variables＿Elm model.The minimum value of IBA-ELM model prediction error as the final target of IBA algorithm,to construct the extreme learning machine network connection weights and threshold for IBA-ELM model of the independent variables,the mean square error of prediction results and the measured values as fitness function formula of IBA algorithm,the optimal parameter combination by IBA algorithm,plug in ELM to construct IBA-ELM dam deformation forecasting model.The model was applied to an engineering example,and the prediction results of IBAELM model BA-ELM model GA-ELM model and IBA-BP model were verified through th e prediction analysis of the dam settlement data of Lishan Reservoir.The accuracy was ev aluated by four precision evaluation indexes of mean absolute error mean absolute percenta ge error root mean square error.The experimental results show that the average absolute er ror MAE of the predicted and measured values of the three models is 0.286 mm,0.332 mm,0.515 mm,0.396 mm,the average absolute percentage error MAPE is 0.361%,0.418%,0.649%,0.499%,and the root mean square error RMSE is 0.360 mm,0.406 mm,0.617 mm,0.484 mm.Acc ording to the indicators,the prediction accuracy of each model is IBA-ELM>BA-ELM>IB A-BP>GA-ELM,and the prediction accuracy of IBA-ELM model is higher than that of the other three models,which indicates that IBA-ELM model can effectively improve the pred iction ability of dam deformation and provide a decision-making basis for dam deformation monitoring.The IBA ELM model is also used to predict the settlement data of the dam m onitoring points SZ2 and SZ6,and the experimental results show that the model has a goo d prediction effect on the other two points,which proves that the model has a strong gene ralization ability.