Optimization on GA-BP neural network of coal and gas outburst hazard prediction

Abstract

This paper presents a genetic algorithm and back propagation neural network (GA-BP-NN) outburst prediction model with a structure of 6 × 13 × 1 according to basic theory of coal and gas outburst hazard classification prediction of coal mine and genetic algorithm, back propagation and neural network. Particularly, we also construct an application of outburst prediction of coal mine. From the learning of living examples of an area in Shanxi province in China, we could safely draw the conclusions as followed: a proper number of learning samples is 12∼18 when there are 6 input neurons of outburst prediction; In addition, the network generalization capability could be enhanced by increasing number of classes which belong to the training samples and decreasing distances of sample intervals; When the Logsig delivery function is taken in output layer, the pattern classification of network is best and the critical value of outburst prediction criterion is 0.5; When the pattern classification of network is best, other parameters have little influence on the network capability. The application and conclusions could be taken in Prediction of Coal and Gas Outburst of coal mining and contribute greatly to production safety of coal mine.