The Research Of Gas Concentration Prediction Method Based On Space-time Sequence

The underground gas accident is the most severe one among five major coal mine accident because of its destructive power and serious casualties.It is of great significance to analyze the characteristics of gas data and construct an effective gas concentration prediction model,which can early warn of gas anomalies and reduce the occurrence of gas accidents.This paper will analyze the space-time characteristics of gas migration in working face roadway and construct a space-time prediction model of gas concentration based on the combination of least squares support vector machine(LSSVM)and space-time modeling idea.The space-time model introduces then noise that may influence the prediction accuracy as long as it introduces the priori information.In order to reduce the influence of noise,a selective integration learning method is introduced to construct a spatial and temporal prediction model of gas concentration based on selective integration.The specific research content is as follows:(1)Some intelligent optimization algorithms is easy to fall into the local optimal solution problem so a dynamic ant colony algorithm(DACO)is used to optimize the penalty parameters and kernel function parameters of LSSVM,to establish a DACO-LSSVM model.The ant colony algorithm divides the solution space into sections.When initializing,the ant colonies,scattered and distributed in the solution space,are not easy to fall into the local optimal solution in the search process.At the same time,the dynamic global selection factor and the dynamic pheromone evaporation factor are adopted to improve the optimization accuracy and the solution speed.(2)The paper give the method to construct a space-time prediction model of gas concentration which combines the space-time modeling idea and DACO-LSSVM algorithm based on the study of gas concentration time series.That lead to a solution of the low prediction accuracy for the single point gas concentration prediction model.The time delay information and spatial position information are introduced into the gas concentration prediction model and determined the optimal spatial delay operator boundary value by K-means algorithm.Meanwhile the optimal time delay operator boundary value is obtained through experimental analysis,a sample similarity Gauss is proposed,by these the dynamic space-time weight matrix is constructed.Thesimulation experiment is carried out by using the gas data collected on the working face,which shows that the model has good prediction effect.(3)A space-time prediction model of gas concentration based on selective integration combining the selective integration learning idea with the space-time model is proposed to reduce the noise which is introduced by the space-time prediction model.Several basic time-space predictors are constructed whose the weights corresponding are combined to form a weight vector.And that vector will then be optimized by the artificial bee colony algorithm(ABC).Filtering out the basic time-space predictor with better performance according to the weight,the weights are normalized as the integrated weight of each basic time-space predictor,and the integrated model is obtained.Simulation experiments using test set data show that the model can improve the gas concentration prediction accuracy.