Research On Time Series Prediction Of Recurrent Neural Network Gas Data Based On Improved LSTM Model

Gas disaster has always been a major security issue in the field of coal mines.And the increase of coal mining depth makes it more difficult to ensure the safety of underground production.Therefore,how to accurately and quickly predict the gas data is very necessary,it can effectively reduce and prevent the occurrence of gas safety accidents,and can make timely response measures.In this paper,the time series of gas concentration is taken as the cut-in point,and the neural network is used as a tool to fit nonlinear complex problems.Then the recurrent neural network(RNN)and its evolution model,long-short memory model LSTM(Long Sh),which are very suitable for solving time series problems,are selected.Ort-term Memory)deeply studied the related problems of time series of gas concentration.The paper mainly completed three aspects:1.The cause of gas emission is very complicated,such as in-situ stress,gas pressure and coal seam burial depth.Therefore,gas emission is a complex nonlinear dynamic process.The paper finds out the main causes of coal and gas emission and extracts relevant data as the main basis for predicting gas concentration time series.2.This paper uses a variety of optimization algorithms,on the one hand,to solve the data in the absence or abnormal situation of the problem of non-smooth,on the other hand,the use of neural network optimization algorithm to improve the efficiency of neural network training.Next,RNN neural network model is used to simulate and predict,and LSTM neural network model is improved to solve the gradient dispersion problem caused by RNN model.3.Because the complexity of LSTM model is too high,this paper improves the LSTM model,and uses the principal component analysis method to put forward the PCA-LSTM-RNN neural network model.The experiment shows that this model can greatly improve the training efficiency of neural network under the premise of ensuring the accuracy.