Predicting Of Gas Emission Quantity In The Heading Face Based On The GA-RBF Neural Network

The coal production of China rank first in the world. However, coal mine accidents do occur frequently recent years, and the number of casualties of coal mine accidents only falls behind that in the traffic accident. The safety of coal mine is confronted with enormous challenges as the increasing of the depth of coal seam, the improving of production capacity, and the complicated geological condition. In consideration of that the gas disaster is always one of the main disasters threaten the safety of coal mine, predicting the gas emission accurately and quickly is necessary for the formulation of gas prevention measures.The dissertation takes mining and nature as two aspects of factors, analyzing the effect of coal seam’s gas content, depth, gas pressure, atmospheric pressure, wind quantity, capacity, etc. to gas emission. It pointed out the limitation of traditional gas emission prediction method that it can’t clearly express the complex and nonlinear relationship between gas emission and its influence factor. The RBF neural network can overcame those disadvantages because of its fault tolerance, adaptability and its capability approaching the nonlinear function.The RBF neural network has global optimization search and optimal appreciation. The topology structure, number of hidden units, centers, widths and weights are key factors in the determination of the network’s performance. As a global optimization algorithm, genetic algorithm has strong robustness. It prevents the result of the network from being trapped at local minima and solved the low speed of training. GA changes the crossover and mutation probability with adaptability, which avoiding repetition search effectively and increasing the search efficiency. The dissertation proposed a model that use genetic algorithm optimize RBF neural network’s number of hidden units, center locations, widths and weights. Structuring GA-RBF neural network based on Matlab and predicting the gas emission of two coal faces, the performance of the model is good and acceptable.