Research on the application of artificial intelligence in coal mine gas monitoring and prediction

This study investigates the application of artificial intelligence techniques in coal mine gas monitoring and prediction, aiming to construct more efficient and accurate gas concentration prediction models to reduce the risk of gas explosion in coal mine production.Due to the limited performance of traditional prediction methods in dealing with high-dimensional and dynamic three-dimensional mining environments, this study employs a fusion model based on temporal convolutional network (TCN) and temporal generative adversarial network (TimeGAN), TCN-TimeGAN, to predict the gas concentration.The model combines the interval sampling advantage of TCN and the time series characteristics of TimeGAN, and through four processes of embedding, recovering, by generating and discriminating gas concentration time sequences, the time-dependent features of gas concentration data can be effectively captured, thus improving the prediction accuracy and timeliness.In this study, gas concentration data from September 2020 through December 2021 were used as the basis, through data cleaning and outlier processing, it is found that the gas concentration data has obvious timedependence, which is suitable for using time series modeling.Embedding and Recovery Networks via TCN-TimeGAN Modeling, the gas concentration data are mapped to a low-dimensional feature space, a generative network is then used to generate new time series data from random noise, and the model parameters are optimized by combining the discriminative network in order to improve the quality and consistency of the generated data.In particular, to cope with the problem of gradient instability of generative adversarial networks during training, In this paper, Wasserstein distance is introduced to optimize the loss function and a gradient penalty term is added during the training process to improve the stability of model training and the realism in the samples generated.In addition, this study also explores the prediction performance of combining LSTM networks for gas concentration.The standard recurrent neural network (RNN) faces the problem of gradient vanishing in the processing of long time-dependent data, whereas the improved LSTM overcomes this problem through memory cells and gating mechanism for real-time prediction task of gas concentration.In this study, the LSTM is further extended to three-dimensional spatial input data, experiments demonstrate the prediction accuracy