Abstract: To address the problems that current operation and maintenance systems for mine belt conveyors have poor real-time perception of belt deviation faults and lack predictive and intelligent decision-making capabilities, a digital twin prediction and decision-making system for correction control of mine belt conveyors was designed based on the five-dimensional digital twin model. A multidimensional high-fidelity digital twin model integrating geometry, physics, behavior, and rules was constructed, and low-latency real-time synchronization between physical and virtual entities was realized based on the EMQX service platform. A BP-PID algorithm was designed. With deviation amount, deviation speed, and tension as inputs, it tuned PID parameters online and performed correction control, while predicting the deviation trend based on operating status data of the belt conveyor. An experimental platform was built with a DTL-800 mine belt conveyor as the control object, and comparative experiments were carried out at belt speeds of 0.5, 1.0, 1.5, and 2.0 m/s. The results showed that, compared with the PID algorithm, the BP-PID algorithm improved the response speed of correction control by an average of 33% and reduced the steady-state error by about 50%. The predictive correction experiment results showed that the system could warn of an increasing trend in deviation probability 20 min later, rapidly diagnose and locate the corresponding idler when deviation actually occurred, and accurately execute correction instructions, realizing an automated closed loop of "perception-prediction-decision-execution".