基于PCA-SSA-XGBoost算法的拱坝应力预测模型研究OACSTPCD

Because the dam stress is affected by many factors such as water level and temperature,the interrelationship between these influencing factors will cause multiple collinear problems,which may easily lead to the pseudo-regression phenomenon of the prediction model using this as the input.In addition,the existing stress prediction models based on machine learning algorithms are prone to overfitting due to too many training features and overtraining,and their prediction accuracy needs to be improved.In response to above problems,an improved limit gradient lifting algorithm(PCA-SSA-XGBoost)based on principal component analysis(PCA)and sparrow search algorithm(SSA)is proposed to construct an arch dam stress prediction model.The model first uses principal component analysis to reduce the dimensions of the parameters to reduce the impact of multicollinearity of the influencing factors,and then,the hyperparameters of XGBoost are optimized through the SSA algorithm to avoid the overfitting of traditional algorithm and further improve the prediction performance of model.The model is applied to a concrete arch dam project in southwest China to process,analyze and predict stress and stress-related monitoring data,and the prediction result is combined with that of multiple linear regression models(MVLR),neural network models(RBFNN)and limit gradient lifting regression prediction models(XGBR).The comparisons show that the stress prediction model based on the PCA-SSA-XGBoost algorithm can overcome the multicollinearity and overfitting problems of input variables,and has superiority in prediction accuracy.