Abstract: Aiming at the problems of low transmission efficiency and high failure rate in traditional asynchronous motor drive of belt conveyors, high manufacturing cost and demagnetization risk in permanent magnet motor drive, and large low-speed torque ripple in switched reluctance motor drive, a direct instantaneous torque control (DITC) strategy based on BP neural network nonlinear model is innovatively proposed to improve the control accuracy of high-power switched reluctance motor and reduce torque ripple in a certain mine of Pingdingshan coal industry group. Firstly, a 2×400kW switched reluctance semi-direct drive(SRSD) system for belt conveyors was innovatively designed, and a high-precision prediction model for the torque and magnetic flux of the switched reluctance motor was established using BP neural network. Then, combining the torque variation law of the switched reluctance motor in the commutation zone with the PWM control concept, an improved DITC control strategy is proposed, which takes torque error as the input and uses PWM to control phase current within the torque error threshold to improve the smoothness of motor operation. Finally, simulation tests were conducted on the underground transportation conditions of the belt conveyor under no-load and variable load conditions, and the following results were obtained: the improved DITC control strategy could improve the dynamic response of the switched reluctance semi direct drive system, effectively reducing torque ripple by 18.7% to 39.1% under various loads, improving the stability of belt conveyor operation, and providing reference for the application of the SRSD system on belt conveyors.