Abstract: Low-concentration gas direct combustion systems often face the problem of abnormal fluctuations in gas concentration, especially during the gas mixing process. If the abnormal concentration is not discovered and corrected in time, it may cause serious safety accidents. Improving the accuracy and adaptability of the prediction of abnormal gas concentration after mixing has become an urgent problem to be solved. To address this problem, a strategy for the prediction and regulation of abnormal gas concentrations based on low-concentration gas direct combustion system is proposed. Firstly, a normalization flow normalized dataset is constructed based on the data of main gas concentration obtained from sensors such as methane and temperature, and multi-source data fusion is performed on the dataset; further, an adaptive prediction and regulation model for abnormal gas concentration is designed, and the coding feature extraction module is used to extract features from the experimental data, the abnormal gas concentration prediction module is used to predict the abnormal gas concentration in real time, and the decoding regulation module is used to regulate the abnormal gas concentration value using adaptive upsampling, the performance of the designed model is comprehensively evaluated based on ablation experiments, comparative experiments, and adaptive experiments; finally, the adaptive prediction and regulation model for abnormal gas concentration is integrated into a low-concentration gas direct-combustion system mixed gas concentration abnormality prediction and regulation platform jointly developed with a graphical user interface (GUI). The results show that the model’s accuracy (Acc) is 95.70%, precision (Pre) is 95.83%, mean square error (MSE) is 0.068, and frame rate per second (FPS) is 98.3 fps/s, which is better than Transformer, Mamba-2, and other models. The maximum error value of the mixed gas concentration anomaly prediction and regulation platform integrated into the low-concentration gas direct combustion system does not exceed 3%, which can timely and effectively regulate abnormal fluctuations in gas concentration and ensure the stable operation of the low-concentration gas direct combustion system.