Abstract: To address the lack of comprehensive reviews focusing on the suitability of coal spontaneous combustion monitoring and early warning technologies for various underground scenarios, this paper reviews monitoring technologies applicable under real underground conditions. It summarizes coal spontaneous combustion monitoring technologies suitable typical coal mine scenarios along with their current research status (goaf areas, working face-roadway). It analyzes the principles and current research on coal spontaneous combustion prediction technologies based on indicator gases and machine learning. It introduces existing coal spontaneous combustion stage division methods and hierarchical early warning strategies. An intelligent development path is proposed: constructing a “one-stop, visualized, and intelligent” early warning platform for coal spontaneous combustion in mines to realize real-time continuous visualization monitoring of key information. With the deep integration of large models and the coal industry, coal spontaneous combustion early warning technology will advance toward an intelligent model of “multimodal analysis-accurate prediction-proactive prevention and control.” The paper points out the development directions of coal spontaneous combustion monitoring and early warning technology: ① continue studying the multi-physical field coupling mechanisms and disaster-causing mechanisms of coal spontaneous combustion to provide theoretical support for proactive early warning and control; ② focus on overcoming bottlenecks in applying new monitoring technologies underground based on coal spontaneous combustion characteristics and actual working conditions, providing technical support for constructing multiparameter dynamic monitoring networks; ③ deeply integrate machine learning, digital twins, and other technologies to build a 3D visualized intelligent early warning platform based on multimodal prediction models; ④ cultivate large models specialized for the coal mining industry, distilling lightweight vertical domain models based on general large models, thereby supporting the intelligent development of coal mines.