Research On Gas Concentration Prediction Method Based On Deep Learning

Gas accident is the main reason that affects the safety work of underground coal mine.Real-time and accurate monitoring of gas concentration is the main measure to prevent and control gas accident.However,the conventional prediction method of gas concentration has great limitations.Therefore,the establishment of accurate,efficient and real-time prediction method of gas concentration has important theoretical value and practical significance for strengthening national coal mine production capacity,improving industrial management level,improving enterprise economic benefits and protecting workers’ life safety.In order to improve the prediction accuracy of gas concentration,this paper proposes a prediction method of gas concentration based on deep learning.A large number of experiments show that the proposed method can further improve the overall prediction accuracy of the model compared with the traditional prediction algorithm.The main work of this study is as follows:(1)In terms of data selection,this study firstly takes the spatio-temporal characteristics of gas data as the starting point,and uses the deep learning algorithm to build the prediction model of gas concentration.The input of traditional gas concentration prediction method only considers the time characteristic of gas concentration series,and lacks the consideration of spatial information.In this paper,by combining the time series factor and the spatial topological factor,the attribute requirements of the input data are clarified,which not only improves the prediction accuracy of the model but also greatly enhances the stability of the model.(2)In terms of gas concentration prediction model construction,this study enhanced the correlation between each gate unit by improving the gated unit of LSTM,so as to improve the ability to extract long-term sequence features.In order to avoid the traditional long and short term memory neural network to extract excessive redundant feature information of gas concentration and other monitoring data in limited storage space,this paper uses the improved long and short term memory network to extract relevant feature,and builds Encoder-Decoder model based on this to predict the long term sequence of gas concentration.The improved snake optimization algorithm(SO)is used to simplify the parameter adjustment process of the neural network to avoid the local optimal solution in the training process.(3)In terms of residual sequence correction,this paper uses the adaptive noise-complete ensemble empirical mode decomposition(CEEMDAN)algorithm to decompose the residual sequence,then takes each eigenmode function as a new round of input data,and then combines the predicted values of multiple eigenmode functions through the variable weight combination method as the residual correction sequence of the initial predicted values.Further improve the prediction accuracy of the model.In order to verify the effectiveness of the proposed method,the experiment was carried out on the mine real data set.The experimental results show that the improved long and short-term memory neural network prediction model under the EncoderDecoder framework proposed in this paper is superior to other comparison models and can achieve accurate long-term gas concentration prediction.Figure [26] Table [4] Reference [80]...