Research On Prediction Method Of Mining Pressure In Coal Face Based On LSTM

China is the largest coal producer in the world,but the safety production of the coal industry has always faced severe challenges.In the era of big data and artificial intelligence,it is currently an important research direction to utilize these technologies to solve practical problems in coal production.Since 2017,Xi'an University of Science and Technology,relying on key projects of the National Natural Science Foundation of China,has begun to explore the utilization of big data and deep learning to improve the safety in coal mining,including researching on the mining pressure prediction of mining face.The research on pressure prediction in mining face mainly includes single sequence pressure prediction based on deep learning LSTM network,and the effect of time window setting on the performance of LSTM single sequence prediction,and spatio-temporal prediction method of mining pressure based on time series correlation analysis and LSTM.The specific work is as follows.(1)The LSTM network is able to capture long-term dependence,and can be used for sequence modeling to capture the temporal dependence of the sequence.Stacked LSTMs stacks multiple hidden layers,and each layer of network can further abstract information,which is conducive to improve the network's abstract expression ability.Therefore,A three-layer stacked LSTM network is designed to achieve single sequence mining pressure prediction.(2)Researches show that,there is a close relationship between the prediction accuracy of the model and the length of historical data which is used to construct the sample set by the coordinate delay method.Given the length of historical data,within a certain range of accuracy,the predictable data length of the model is limited.Therefore,The time window parameters that influence the performance of the model is studied,which is the best historical data length and the longest predicted data length.The design proposes a search method for the best historical data length and the longest predicted data length.(3)For single sequence mining pressure prediction,only the temporal dependence of the sequence is modeled.When mining of the working face,the spatial structure of the rock layer changes dynamically,and the sequence of the mining pressure measured by the support monitoring system is spatially related.It is beneficial to improve the prediction accuracy of mining pressure by applying the correlation to spatio-temporal prediction of mining pressure.The correlation coefficient is used to study the correlation between sequences,and the three-layer Stacked LSTMs is designed as the spatio-temporal prediction model to forecast mining pressure.Finally,the thesis takes the mining pressure data of 14160 working face in Pingmei No.8 Coal Mine for an instance to complete single sequence mining pressure prediction and the analysis of the factors influencing the model performance.The results show that the time window parameters influence the performance of the model,and reasonable time window parameters improves the prediction accuracy of the model.For the spatio-temporal prediction of mining pressure,the thesis takes the mining pressure data of Yujialiang 52306 working face for an instance,and makes a verification experiment of spatio-temporal prediction method of LSTM network based on time series correlation analysis.The results show that,compared with single-sequence mining pressure prediction,the spatio-temporal prediction model effectively improves the prediction accuracy of mining pressure.