Gas Gushing Forecasting Based On Difference Grey Radial Basis Function Neural Network

Gas explosion, coal and gas outburst and gas asphyxiation are the major disasters against safe production of coal mine in China and gas emission quantity over the safety threshold in coal mine is the direct factor which results in gas disasters. Therefore, the precision of the prediction of gas emission quantity is very important to safe production of coal mine.As we know, gas emission quantity is influenced by many complicated factors such as geology, the state of coal layer, gas quantity in coal and wall rock, exploiting scale, exploiting technology, mine structure, and these factors are highly nonlinear. In recent years, the neural network which develops rapidly has high nonlinear mapping and parallel processing capabilities, such as BP neural network, radial basis function neural network, etc. These methods are no doubt very suitable for modeling of gas emission. However, the precision of neural network is directly affected by the random effects of sample data. Because of the presence of noise, the historical data of gas emission quantity show a general state of chaos, which accords with the grey theory which can be used to solve the problem of the random effects of sample data in neural network. Meanwhile, radial basis function neural network works better than BP neural network on the prediction of gas emission quantity and has its own features which are neat training, fast constringency, good approximation to any function, so grey theory and radial basis function neural network are combined together.There are two main factors which affect the precision of grey radial basis function neural network model. One factor is how to combine two models and the other is the precision accuracy of both models. So in the paper some work is done as follows: (1) On the combination of grey theory and radial basis function neural network, a valid method is put forward in the paper according to the advantage of grey theory weakening the random effects of sample data and the features of radial basis function neural network which are neat training, fast constringency, good approximation to any function. This method improves the prediction accuracy to a certain extent.(2) Considering in recent years, in the field of gas prediction, although many reseachers have used different methods to combine grey theory and neural network together and improve the gas prediction model, most reseachers just combine the traditional models together and only a few reseachers use one of two improved models and the reseachers who use both improved models are less, which is the bottleneck of the grey neural network model. Meanwhile, some reseachers have done some good work on the improvement of both models. So the improved GM(1,1) model and the improved radial basis function neural network model are introduced with the method mentioned in (1) to construct the difference grey radial basis function neural network model to improve the forecasting accuracy. Therefore, on the forecasting performance of both models, two improved models which can work better are used in the thesis.(3) Considering the powerful data processing capability of MATLAB, the convenience of functions of neural network toolbox and the good visual environment, some program is done to do the simulation experiment on radial basis function neural network model, grey radial basis function neural network model and difference grey radial basis function neural network model. The simulation results show that the relative error of radial basis function neural network model is higher than grey radial basis function neural network model and the relative error of grey radial basis function neural network model is higher than the difference grey radial basis function neural network model. Moreover, the relative error of the difference grey radial basis function neural network model varies within a reasonable range, which shows that the accuracy of the difference grey radial basis function neural network model is improved and the performance of the model is good and acceptable.