Research on gas concentration prediction based on VMD-CNN-BiLSTM artificial intelligence algorithm

To address the difficulty in simultaneously improving the prediction accuracy and robustness of existing coal mine gas concentration prediction models, which limits the reliability of gas safety monitoring and early warning, a hybrid prediction model integrating variational mode decomposition (VMD), a convolutional neural network (CNN), and a bidirectional long short-term memory network (BiLSTM), termed VMD-CNN-BiLSTM, was proposed. First, VMD was employed to decompose and reconstruct the original gas concentration time-series data, thereby reducing the adverse effects of data nonstationarity, noise interference, and mode mixing on prediction modeling. Subsequently, a coupled CNN-BiLSTM network was constructed. The CNN was used to extract local fluctuation characteristics and multiscale features from each modal component, whereas the BiLSTM was employed to capture bidirectional long-term temporal dependencies in the gas concentration time-series data. The extracted features were then fused to generate the final prediction results. Experimental validation was conducted using gas concentration monitoring data collected from a working face in the Pingdingshan mining area, and the proposed model was compared with the CNN, long short-term memory network (LSTM), BiLSTM, CNN-LSTM, and CNN-BiLSTM models. The results demonstrated that the VMD-CNN-BiLSTM model achieved the best prediction performance, with a coefficient of determination (R²) of 98.1597%, a mean squared error (MSE) of 0.00069499, and a mean absolute error (MAE) of 0.019251. Compared with the benchmark models, the proposed model increased R² by 3.7540 to 32.4854 percentage points, while reducing MSE by 0.0014 to 0.0128 and MAE by 0.0121 to 0.0667. These results indicate that integrating VMD with CNN-BiLSTM can effectively enhance the feature representation and temporal modeling capabilities for nonstationary gas concentration time-series data. The proposed model can therefore provide technical support for advance gas concentration prediction, safety monitoring, and early warning in coal mines.