Implementation of Various Neural-Network-Based Adaptive Speed PI Controllers for Dual-Three-Phase PMSM
Zhenxiao Yin, Hang Zhao
Abstract
This paper provides a preliminary study on applying Neural Network (NN) based proportional and integral (PI) controllers with the positional PI principle. This method is set into the speed loop of a dual-three-phase permanent magnet synchronous motor (PMSM), where the vector space decomposition method (VSD) is utilized. The proposed methods are single-layer neural network (SNN), backpropagation neural network (BPNN), and radial basis function neural network (RBFNN). These methods aim to reduce the overshoot of the speed tracking in control problems. By optimizing the current reference output, the copper loss can also be reduced at the same time. Finally, the control performances using traditional PI, SNN-based PI, BPNN-based PI, and RBFNN-based PI are compared by adopting a self-defined scorecard with different evaluation indices.