Gas Emission Forecasting and Fore-warning Research of A Working Face Based on Improved BP Network

摘要

According to the training samples set is composed of the mean and the maximum value of the absolute outflow of gas in the 1483 Working Face of the mine of Chongqing, China, in continuous 117days, the gas emission predicting model based on the time-series neural network was put forward through the combination of the artificial neural network with nonlinear dynamic processing capabilities and the MATLAB software. The speed and accuracy of neural network training of the predicting model was improved by using L-M algorithms (Levenberg-Marquardt algorithms).The model is (6-8-1) structure, that is to say, the number of joints of the input layer is 6, the number of the hidden layer is 8, and the number of the output layer is 1.The transmitting function of hidden layer is chosen S-(logsig) function, the transmitting function of output layer is chosen the Purelin-function. The maximal recycled times is 1000 and the average root of goal error is 0.01 can be concluded by Taylor Series and Newton Formula.By comparing with the predicting result and the monitoring results in the mine, the relative error between the mean predicting result and the monitoring result of gas emission is 0.37%-6.07%, the relative error of the maximum is 2.15%-11.05%. It's considered that the predicting results can be used to satisfy the need of present security management. In order to make fore-warning before gas concentration coming to dangerous limits and adopt measures in time, the system of gas emission forecasting and fore-warning at a working face was set up through the combination of BP network nonlinear time-series predicting model and the monitoring system of the gas sensors in a mine, which can renew the training sample continually and make predicting process rolls continually. This research provides a theoretical foundation for the gas emission forecasting and fore-warning at coal working faces.