The Research Of Gas Concentration Prediction Based On Space-time Neural Network Model

Currently, the gas disaster is one of the disasters threating our country’s coal mine safety production process, so extracting the factors influencing gas concentration change from coal mine actual monitoring data, fully tapping the rule of gas concentration change, establishing accurate and reliable gas concentration prediction model have great significance to effectively prevent the occurrence of gas disaster.Based on space-time sequence data analysis and modeling, this paper first discusses the characteristics of space-time sequence: space-time auto correlation, space-time partial correlation, space-time stationarity. Then this paper studies the modeling steps of traditional space-time model STARMA: spatial weight matrix establish, model identification, parameter estimation and diagnosis. Based on theory of traditional neural network, bringing in time delay operator and space delay operator, a new space-time neural network based on time and space analysis is constructed by space extension using a single hidden layer feedforward neural network. And a new kind of STELM learning algorithm is proposed by training space-time neural network using extreme learning machine.For the gas concentration sequence of each monitoring points in the coal mine gas monitoring system is not only related to its historical data information in time, gas concentration sequence of different monitoring points also exists certain relevance in space, this paper puts forward the proposed space-time neural network model in gas concentration prediction. Combining with the actual coal mine gas concentration sequence data, a gas concentration prediction scheme based on space-time neural network model is designed; it determines the space and time delay parameters of the space-time neural network model respectively using K-means clustering algorithm and GP algorithm. This paper explains that the proposed STELM algorithm performance is better than the traditional neural network learning algorithm through simulation and comparison analysis of the space-time neural network STELM learning algorithm and the traditional neural network learning algorithm; and proves that the accuracy of the proposed space-time neural network model is higher than traditional STARMA model through simulation and comparison analysis of the presented space-time neural network prediction model of gas concentration prediction and the traditional STARMA model.