Modeling and simulation of current control for permanent magnet synchronous motors based on multilayer perceptron neural networks

In the traditional finite control set model predictive current control(FCS-MPCC) for a permanent magnetic synchronous motor (PMSM), periodic delay caused by computational latency and hardware register update mechanisms leads to control commands lagging behind actual motor states, thereby impairing dynamic response and control stability of the PMSM. To address the issue, this paper introduces a two-step finite control set model predictive current control(FCS-MPCC) method. By predicting two-step current states simultaneously and generating control commands for the current time step in the previous cycle, the introduced method effectively reduces the impact of control delay and improves prediction accuracy. However, while the two-step FCS-MPCC method enhances control performance, the more complex computational logic increases the computational burden, limiting its real-time applicability. To overcome the limitation, the paper proposes a method based on the multilayer perceptron(MLP) neural network, which replaces traditional model predictive control strategies with a data-driven method. By learning the optimization rules of the two-step FCS-MPCC, the MLP neural network can replicate its control performance without requiring online computational efforts. The simulation results demonstrate that the proposed method exhibits strong robustness under secondary load disturbances, further validating its potentials for application in motor control.