Research On Prediction Of Gas Concentration Based On Multi-factor Fusion Based On Deep Learning

China is rich in natural coal resources,and the coal mining industry has become one of the important social and economic pillars of our national economy,maintaining the sustainable development of the national economy.However,the occurrence of gas accidents has brought great damage to the people and economic development of the mining area.Accurately predicting gas concentration and identifying gas abnormal signs in advance are important means to avoid gas accidents.In order to improve the accuracy of gas concentration prediction,the multi-factor fusion gas concentration prediction based on deep learning is studied in this paper.First of all,the coal mine monitoring big data is pretreated by Raida criterion and LOCF method.Secondly,the data are derived from the characteristics and supervised.After the low-order features are derived from the features,more high-order features and cross-term features are generated.In the process of supervision,the temporal and spatial factors of gas concentration in each time series are combined into complete multi-factor fusion characteristics,and the spatio-temporal characteristic system and supervised learning samples of the gas concentration prediction model are constructed.Then,the optimal algorithm is selected as the core algorithm of the multi-factor fusion gas concentration prediction model.Under the condition of 12 different lag steps,the LSTM algorithm is selected from single-factor LSTM algorithm,RNN algorithm,SVR algorithm and BPNN algorithm.Then,using Bi-LSTM algorithm to capture the historical data and future data information of coal mine monitoring data,under several lag steps,the multi-factor fusion gas concentration prediction model based on Bi-LSTM and the multi-factor gas concentration prediction model based on LSTM are evaluated,and the gas concentration prediction is compared.The experimental results show that the RMSE and MAE indexes of the multi-factor fusion gas concentration prediction model based on Bi-LSTM are the smallest,and the model prediction accuracy is the highest.Finally,a multi-factor fusion gas concentration prediction model based on CNN is constructed,and the gas concentration samples of multi-factor fusion are identified as one-dimensional images,and more deep features are mined through convolution kernel to improve the accuracy of gas concentration prediction.Under six different lag steps,when the lag step is greater than 30,the prediction effect of the multi-factor fusion gas concentration prediction model based on CNN is higher than that of the multi-factor fusion gas concentration prediction model based on Bi-LSTM.In this paper,by combining the time series data and spatial factor data affecting gas concentration,in the process of feature derivation and sample supervision,a spatio-temporal feature system of multi-factor fusion affecting the change of gas concentration is formed.combined with the deep learning algorithm,a multi-factor fusion gas concentration prediction model based on deep learning is constructed,which improves the robustness and accuracy of the gas concentration prediction model.Figure 25 Table 25 Reference 80...