Abstract: Aiming at the problems of the poor controllability and maneuverability and low accuracy of the conventional mining subsidence method,using the BP neural network model to conduct the mining subsidence by fitting of the heights of mining area is a ideal mining subsidence prediction method.But the traditional BP neural network model is the back propagation algorithm,the connection weights and thresholds of the BP neural network system can be obtained by experimental calculation in many times,the deficiencies of the BP neural network are easily falling into local minimum values,slow convergence,and so on.The parameters of the BP neural network model area optimized by genetic algorithm (GA) to improve the generalization ability of the BP neural network model to establish the genetic BP neural network model (GA-BP).25 sets of monitoring points values that are conducted the third level measurement of the first working face of the mining area are used as training samples (15 sets of monitoring points values) and prediction samples (the other 10 sets of monitoring points values) of the genetic BP neural network (GA-BP) respectively,the experimental results show that the internal precision and external precision of the genetic BP neural network model (GA-BP) are higher than the BP neural network model and quadric surface fitting method,besides that,the residuals of the BP neural network model (GA-BP) are lower than the BP neural network model and quadric surface fitting method.The above research results show that the genetic BP neural network model (GA-BP) is good to realize the high-precision mining subsidence prediction.

Key words: Mining subsidence, Genetic algorithm, BP neural network, Genetic BP neural network, Evaluation fitting, Quadric surface fitting