Deformation prediction of gravity dam based on ISADE-OSELM

Deformation prediction of gravity dams is important measures to ensure dam safety, and is difficult areas of study. In order to better carry out gravity dam safety monitoring, this paper introduced Online Sequential Extreme Learning Machine (OSELM) algorithm to establish gravity dam deformation prediction model. The parameters of OSELM model can be updated by incremental learning, that is different from the traditional batch learning of establish monitor model. The advantage of this learning method is that it helps the model to incorporate the information contained in the latest monitoring data. It optimizes the traditional modeling approach. In addition, the selection method of initial parameters in Self Adaptive Differential Evolution (SADE) is improved, and an Improved Self Adaptive Differential Evolution (ISADE) algorithm is proposed with better global optimization ability. Then it is used to optimize the OSELM model, thus improving the accuracy of OSELM model. In this paper, an example is analyzed by stepwise, ELM, OSELM and ISADE-OSELM models respectively. Generally speaking, the predicted results of ELM and OSELM basically meet the engineering needs. However, the MSE and MAE values of ISADE-OSELM are the smallest, indicating that the accuracy of the model is relatively high, the R2 value of ISADE-OSELM is the largest, indicating that the overall fitting degree of the model is relatively best. That is the comprehensive performance of ISADE-OSELM is relatively optimal. The example analysis shows that the prediction performance of ISADE-OSELM model is improved, and the model can predicte the gravity dam deformation with high accuracy.