Abstract: To address the problems of control lag, inability to perform online dynamic adjustment, and poor anti-interference ability in existing motor power control methods for scraper conveyors under different working conditions, this study combined the Long Short-Term Memory (LSTM) network with PI control and proposed an adaptive control method for multi-motor power balance of scraper conveyors based on LSTM-PI. On the basis of traditional PI control, this method introduced the LSTM network to predict motor output power based on motor current and the difference between the rated chain speed and the actual chain speed. The difference between the predicted motor power value and the actual value was used to realize online dynamic adjustment of PI control parameters through nonlinear mapping. The output electromagnetic torque of the motor was then adjusted through the PI controller, thereby achieving multi-motor power balance of the scraper conveyor. The simulation results showed that the LSTM network could effectively track changes in actual power, and the prediction results showed strong temporal continuity and trend consistency, avoiding large fluctuations in predicted values. Under typical working conditions such as start-up, sudden load change, and load imbalance, LSTM-PI control could effectively suppress chain speed fluctuation and current oscillation. The fluctuation amplitudes of chain speed and current were reduced by 3% and 4%, respectively, compared with traditional PI control, thereby improving operating stability. Under boundary working conditions such as high load, low-speed heavy load, and unilateral eccentric load, LSTM-PI could complete power regulation more quickly and smoothly, suppress power fluctuation, and improve power regulation efficiency and anti-interference ability.