The Research Of Coal Mine Gas Concentration Prediction Based On Quotient Space

Safety in coal mines is a nonlinear and dynamical system, which integrated human conditions, management facilities, geological conditions and other conditions. Although the change of this system has certain regularities, it can also be affected by many factors, such as economic and environment. Different factors and ways will result in incidents of the different levels, which makes the prediction of underground accurately more difficult. Nevertheless, our countries and many scholars still propose possible method about prediction from different theories, different ways and different practices. The traditional gas prediction faces many challenges in the practical application. However, neural network has the characteristics of self-learning, adaptive and so on, which could extract the knowledge characteristics dynamically on gas from historical data. So, it is more suitable for solving some problems about gas warning. But the traditional BP neural network usually considers the impact of the single factor as the gas concentration in the gas prediction and this method is a low forecasting accuracy. If adding the other input vectors artificially in the input vector of the BP network, we make the data too large, and result in a large burden on model training; the speed of training is slow; the effect is also not good.Therefore, this paper proposes a constructive network based on quotient space—CA-BP model, which is a prediction method combined with BP network. Firstly, based on the quotient space, we select many factors of the gas prediction with the association rules of the data mining, in order to get the main factors describing the trends of gas concentration. Then, through the self-organizing classification capacity of CA-BP, the gas concentration factors are classified, and on the basis of qualitative classification, we can do classification prediction by BP network.The work includes mainly the following sections:1. We get the knowledge describing the gas concentration based on quotient space with association rules algorithm, and obtain the effective input variables from many factors. And then we can not only elect a major variable, but also remove the unimportant variables.2. Introduct the traditional modeling method of BP prediction model, establish network prediction model, and determine the number of hidden layer nodes by empirical formula of experts. Then according to the defect of the above model, we propose the CA-BP model, and consider the problem with the quotient space of granular computing theory at different levels.3. The comparative analysis of results show this method applied to the coal mine gas warning is feasible and effective.