Abstract: The operating environment of coal mine electromechanical systems is highly complex, and traditional fixed threshold methods and manual control approaches struggle to promptly respond to operational changes such as load fluctuations, abnormal temperature rises, and vibration disturbances. Addressing the needs for AI-driven adaptive control in coal mine electromechanical systems, this paper establishes a four-tier technical framework encompassing perception, data acquisition, modeling, and application. It proposes a control strategy that integrates reinforcement learning with model-based predictive control to support parameter self-tuning, load forecasting, and risk assessment, supported by an application analysis using a scraper conveyor system. Results demonstrate that AI-based adaptive control reduces unit transportation energy consumption by 15.2%, improves fault early-warning accuracy to 93.5%, enhances equipment operational efficiency by 12.8%, shortens average control response time from 180 seconds to 35 seconds, and decreases unplanned shutdowns from 7 to 3 instances, demonstrating its effectiveness in improving the operational safety, economic efficiency, and continuity of coal mine electromechanical systems.