<p>As shallow coal resources become increasingly depleted, coal mining has gradually shifted to deep deposits. During deep mining, coal-and-gas outbursts represent one of the most severe hazards. At present, real-time early warning of such disasters primarily relies on the time frequency characteristics of signals such as gas concentration, acoustic emission, and electromagnetic radiation. However, due to the complex underground environment, real-time monitoring data obtained from sensors are often contaminated by substantial interference signals. Consequently, early warning methods based on a single signal cannot simultaneously achieve both accuracy and comprehensiveness. To address this issue, this paper proposes a comprehensive early warning method based on deep learning and multi-source signal fusion. The method analyzes and identifies precursor features of gas concentration, acoustic emission, and electromagnetic radiation that are associated with outburst risks. A predictive model is constructed by integrating convolutional neural networks and long short-term memory networks to dynamically forecast multi-source signals. On this basis, risk probabilities are derived from the predicted values of each signal. Finally, the coefficient of variation method is employed to fuse the risk probabilities of different signals, thereby yielding a comprehensive risk coefficient for coal-and-gas outbursts and enabling intelligent, integrated early warning. The results demonstrate that the proposed deep learning-based multi-source fusion method can effectively identify outburst precursors. Moreover, it fully exploits information from multiple monitoring signals, achieves robust signal fusion, and is capable of issuing an advanced warning up to 2 h before the occurrence of outburst precursors. This paper thus provides an effective technical reference for multi-source signal fusion-based early warning of coal-and-gas outburst disasters.</p>