The Research Of Gas Concentration Prediction Method Based On The Selective Integration Of Space-time Series

Gas disaster is one of the most common hidden dangers in coal mine safety production process. At present, the total amount of gas accidents which cause serious casualties and great economic losses in our country is large. Therefore, it is very important to improve the accuracy of gas concentration prediction to effectively prevent the occurrence of gas disasters.This paper proceeds from the view of time and space. It studies the characteristic of spatial-temporal data and the method to build a model of spatial-temporal sequences. Based on the thought of spatial-temporal model and traditional neural networks, a spatial-temporal neural network can be constructed by introducing the time and space delay operators. For the respect of learning algorithms, the learning algorithm for spatial-temporal model which is called STELM can be got by expanding extreme learning machine(ELM) algorithm. By taking spatial association as input weight of the spatial-temporal model, it simplifies the complexity of modeling. The learning algorithm can be confirmed only with the boundary value of time and space delay operators. For enhancing the performance of prediction further, selective ensemble learning methods are introduced. The modeling method named SERSTELM is proposed based on several STELM learners which are obtained by L1 norm sparse regularization. Combined with the gas density data in practice, the SERSTELM model can be applied and design a gas density forecast scheme. This gas density forecast scheme includes five steps: prediction of data collection, data preprocessing, model based generation, model selection and gas. The simulation results of STELM, BP, SVM and ELM show that STELM has a higher prediction accuracy; By analyzing the simulation results between the SERSTELM model and ELM, STELM, SERELM models respectively, it proves that the SERSTELM proposed in this paper has better prediction performance in gas density.