Study on Prediction of Surface Subsidence Coefficient Based on SSA-XGBoost Model
In order to solve the problems of limited accuracy,inconsistent model,complicated calculation and poor generalization of the current prediction model of surface subsidence coefficient,a prediction model based on Extreme Gradient Boosting Tree(XGBoost)optimized by Sparrow Search Algorithm(SSA)was established by analyzing the influencing factors of surface subsidence coefficient in 35 existing sample data sets.The SSA-XGBoost model was used to iteratively learn the nonlinear mapping relationship between surface subsidence coefficient with coal seam mining height,coal seam dip angle,overburden lithology,ratio of buried depth and mining height,ratio of bedrock thickness and overburden thickness,as well as ratio of bedrock thickness and mining height.The predicted surface subsidence coefficient based on the SSA-XGBoost model was obtained,and its accuracy was verified by using goodness of fit,root mean square error,and average absolute percentage error.The results show that the goodness of fit of SSA-XGBoost model is 0.9516,the root mean square error of prediction is only 0.0206,and the average absolute percentage error is only 2.47％.Compared with other models(XGBoost model,BP neural network model and random forest algorithm model),the goodness of fit of SSA-XGBoost model is improved by 15.79％,111％and 62.81％respectively,the root mean square error of prediction is reduced by 43.25％,76.61％and 73.72％respectively,and the average absolute percentage error is reduced by 46.99％,73.52％and 75.99％respectively.It is found that the prediction results of subsidence coefficient based on SSA-XGBoost have a better fitting effect and a higher accuracy.
万方数据
维普
