Abstract:To eliminate the disadvantages of the BP neural network model in predicting the surface subsidence of the mining area, which is limited in accuracy stability, we take a mining area as an example. We select nine factors and the maximum subsidence value affecting the surface subsidence of the mining area, including the elastic modulus, Poisson’s ratio, cohesion, et al, as initial sample data. The BP neural network is optimized by Kalman filtering(KF), and then the constructed KF-BP model is regarded as a weak predictor in adaptive boosting(AdaBoost) algorithm, and each weak predictor is weighted and combined into a strong predictor through the final weight distribution. This study uses MATLAB to establish BP neural network model, KF-BP model, AdaBoost-BP model and AdaBoost-KF-BP model to train and predict the actual settlement monitoring data of the mining area. The results show that AdaBoost-KF-BP model has the highest stability, and its accuracy is significantly improved compared with other models.
YUAN Xitun,WEN Yongxiao,CHEN Xinyu. Prediction Method and Applicability of Mining Area Surface Subsidence Based on Multi-Model Fusion[J]. jgg, 2023, 43(3): 232-238.