Sensorless control of a PMSM based on an RBF neural network-optimized ADRC and SGHCKF-STF algorithm

For the problem of the rotor position estimation and control accuracy of permanent magnet synchronous motor (PMSM), this paper proposes a PMSM sensorless based on radial basis function (RBF) neural network optimized Automatic disturbance rejection control (RBF-ADRC) and strong tracking filter (STF) improved square root generalized fifth-order cubature Kalman filter (SGHCKF-STF). The Automatic disturbance rejection control (ADRC) has strong robustness, but there are many parameters and difficult to adjust. Now we use RBF neural network to adjust the parameters in ADRC online so as to improve the robustness and anti-disturbance ability. In order to improve the estimation accuracy of rotor position and speed, the orthogonal triangle (QR) decomposition and STF are introduced on the basis of the generalized fifth-order cubature Kalman filter (GHCKF) to design the SGHCKF-STF algorithm that not only ensure the non-positive nature of the covariance matrix but also improve the ability to cope with sudden changes in state during the filtering process. Experimental results show that the combination of RBF-ADRC and SGHCKF-STF improve the sensorless control effect of the PMSM to some extent.