Research Of Dam Deformation Prediction Based On ICS-MKELM And Bootstrap Method

Dam deformation prediction has important theoretical and practical significance for analyzing the service status of dams and ensuring the safe operation of dams.However,existing deformation prediction models such as statistical models and neural network models use linear regression and gradient descent modeling methods,which is difficult to express high-dimensional nonlinear features of deformation sequences and is easy to fall into local optimum.Therefore,the prediction accuracy of the model needs further improvement.At the same time,when modeling and predicting deformation observation data,the random noise of the monitoring data,the subjective selection of the model structure,and the random setting of the parameters have a great influence on the uncertainty of the prediction results.However,the current research can only analyze dam deformation by point prediction method,which lacks effective quantification of uncertainty in deformation prediction,making it difficult to judge the credibility of the results.In view of the above problems,this paper proposes a dam deformation prediction model based on the improved cuckoo search-multiple kernel extreme learning machine(ICS-MKELM)algorithm and the Bootstrap method to achieve accurate prediction of deformation values.At the same time,it realizes the characterization of the uncertainty of the prediction results through the interval forms.The main research contents are as follows:(1)To solve the problem that the accuracy of dam deformation prediction model needs to be improved,a prediction algorithm for dam deformation point of improved cuckoo optimization multiple kernel extreme learning machine is proposed.A high-precision multiple kernel extreme learning machine(MKELM)is constructed by combining the kernel extreme learning machine with good applicability for nonlinear problems and a hybrid kernel with generalization and learning ability.At the same time,the improved cuckoo search(ICS)algorithm based on inertia weight and chaos theory is used to optimize the kernel parameters and regular coefficients in the multiple kernel extreme learning machine to further compensate for the problem that the model is easy to fall into local optimum.(2)To make up for the lack of considering the influence of uncertainty factors in the model and data in the existing prediction models,a dam deformation prediction model based on improved cuckoo optimization multiple kernel extreme learning machine and interval prediction method is proposed.Based on the improved cuckoo optimization multiple kernel extreme learning machine point prediction algorithm,the Bootstrap interval prediction method is introduced to estimate the uncertainty influence caused by the subjectivity of model and sample structure,the random setting of model parameters in the model and repeated sampling.Then,the prediction intervals are constructed so as to achieve quantitative characterization of the uncertainty of the results.(3)Taking a project in southwest China as a case study,the application of dam deformation prediction was carried out.The deformation prediction of the dam is carried out by using the proposed prediction model based on improved cuckoo optimization MKELM and Bootstrap.The point prediction and interval prediction results corresponding to the prediction methods such as the standard cuckoo optimized multiple kernel extreme learning machine,single kernel extreme learning machine and extreme learning machine are compared and analyzed.The results show that the model has superiority and consistency.The improvement rate of the two indicators of mean square error and width range can reach36.45% and 93.48%,respectively,which indicates that the model has good point and interval prediction accuracy.