Gas concentration prediction model based on SSA-LSTM

Abstract

In order to better capture the time-varying patterns and effective information of gas concentration, and achieve precise prediction of gas concentration in coal working faces, a gas concentration prediction model based on SSA-LSTM is proposed by optimizing the long short term memory (LSTM) network using sparrow search algorithm (SSA). The model uses the mean replacement method to process missing and abnormal data in the original gas concentration time series data, followed by normalization and wavelet threshold denoising. The performance differences between SSA and grey wolf optimization (GWO) and particle swarm optimization (PSO) algorithms are compared and tested. The result verifies the advantages of SSA in terms of optimization precision, convergence speed, and adaptability. By utilizing the adaptability of SSA, the hyperparameters of LSTM, such as learning rate, number of hidden layer nodes, and regularization parameters, are sequentially optimized to improve the global optimization capability and avoid the prediction model falling into local optimum. The obtained optimal hyperparameter combination is substituted into the LSTM network model and the prediction results are output. Comparing SSA-LSTM with LSTM, GWO-LSTM, and PSO-LSTM gas concentration prediction models, the experimental results show that the root mean square error (RMSE) of the gas concentration prediction model based on SSA-LSTM is reduced by 77.8%, 58.9%, and 69.7% compared to LSTM, PSO-LSTM, and GWO-LSTM, respectively. The mean absolute error (MAE) decreases by 83.9%, 37.8%, and 70%, respectively. The LSTM prediction model optimized by SSA has higher prediction precision and robustness compared to traditional LSTM models.

Keywords

• gas concentration prediction
• time series prediction
• deep learning
• long short-term memory network
• sparrow search algorithm
• hyperparameter optimization