Abstract: In order to predict surface subsidence deformation more accurately,a multi network collaborative calculation strategy was adopted based on the Adaboost algorithm to improve the BP neural network. The improved neural network was trained on actual subsidence data to predict the maximum subsidence,influence angle tangent,and inflection point offset. The three predicted parameters were introduced into the probability integration method,and a surface subsidence formula was established. The improvement effect and surface subsidence formula were verified separately. The study results show that:① By comparing the calculation accuracy of the BP neural network before and after improvement,the results showed that the error of the BP neural network without Adaboost algorithm improvement it is obviously greater than the improved BP neural network,indicating that the calculation accuracy of the BP neural network based on Adaboost correction has been effectively improved; ② Based on the BP neural network,the maximum subsidence,influence angle tangent,and inflection point offset are predicted. Combined with probability analysis method,the description of surface subsidence above the main section of the goaf after stable subsidence can be achieved. Taking the surface of the 3301 goaf in a certain mine in Southwestern Shandong Province as the study example,the improved BP neural network is used to predict the maximum subsidence,influence angle tangent,and inflection point offset,and then the surface subsidence curve is given. Compared with the on-site measurement results,the comparison results show that the maximum error of the improved BP neural network is less than 0. 105 m,and the maximum relative error is 4. 3%,which proves the reliability of the calculation method in this paper.

Key words: subsidence,BP neural network,goaf,Adaboost algorithm,error analysis