Abstract: The prediction results of mining subsidence development of Knothe time function model is not consistent to the actual situation.In order to solve the problem and improve the prediction accuracy of mining subsidence based on Knothe time function model,the defects of the model is discussed in detail,and the improved Knothe time function mining subsidence prediction model is proposed.Based on the actual mining subsidence monitoring data of 6300 fully mechanized working face of Hongqi Iron Mine in Wu'an County,Hebei Province,the improved Knothe time function mining subsidence prediction model is established,and its prediction accuracy is analyzed.The study results show that during the monitoring process of mining subsidence for 387 d,the error between prediction values of the improved model and actual measured values is 0.2~73.6 mm,the average error is 35.2 mm,which is superior to the ones of BP neural network model (the error between predictions values and actual measured values is 8.1~143 mm,the average error is 49.9 mm),SVM model (the error between prediction values and actual measured values is 0.7~105.1 mm,the average error is 35.8 mm) and probability integral method model (the error between prediction values and actual measured values is 18.2~180.5 mm,the average error is 102.6 mm),which further indicated that the improved model proposed in this paper is help for improving the mining subsidence prediction accuracy of the mine.

Key words: Mining subsidence, Knothe time function, BP neural network, SVM, Probability integral method