Speed Control of BP Neural Networks Feed-Forward Compensation With Parameter Adaptation Based on Active Disturbance Rejection Control for PMSM

Permanent magnet synchronous motor(PMSM) control system is a time-varying nonlinear system. To address the issue of poor tracking performance in control systems caused by mismatch between motor flux linkage parameter and speed loop controller parameter, a feed-forward compensation method based on BP neural network(BPNN) with active disturbance rejection controller(ADRC) is proposed. The speed controller of the PMSM uses the composite control of the integral sliding mode control and ADRC algorithm to resolve the trade-off between overshoot and responsiveness in speed loop controllers through a linear tracking differentiator(LTD), to compensate the disturbance and improve the anti-disturbance capability by a linear extended state observer(LESO). In order to improve the control performance under the condition of parameter mismatch, the order and the number of the controller need to be increased, which increases the complexity of the design parameters. A BPNN load observer is trained in real time to learn and train the feed-forward compensation load torque, so as to eliminate the adaptability of load interference and controller parameter mismatch. The observer parameters can be trained and optimized in real time by the speed error value, has a good convergence without the need of the control system model and parameters tuning. Experimental results demonstrate that the BPNN load torque feed-forward compensation based on ADRC scheme exhibits superior dynamic performance, including reduced overshoot and faster response time. The control scheme demonstrates robustness, outstanding disturbance rejection capability, and stable parameter adaptability.