The Design of Position Estimator for PMSM by Using Diagonal Recurrent Neural Network

In order to implement the sensorless electric drive for permanent magnet synchronous motor (PMSM), the rotor position observers were designed based on diagonal recurrent neural network (DRNN). The neural rotor position estimation process is separated into two neural observers, an stator current estimator, and angular velocity estimator. The former is used to map the estimated angular velocity and the applied terminal voltage to the estimated stator current. The neural velocity estimators used to map input voltage and current to the estimated PMSM angular velocity. The angle estimation block generates rotor angle by integrating the estimated angular velocity, with adjustment derived from the current estimation error. Through experimentation, 12 neurons in the hidden recurrent layer were found to produce good results for the neural current observer. The output layer contains two neurons: one for each the direct and quadrature axis current estimates. The simulation results show the estimated rotor angle error is small, and the robust property is good when step load changes.