A PSO-BPNN-Based PI Parameter Tuning Strategy for the Inner Current Loop of a Vienna Rectifier with FPGA-Based Hardware-in-the-Loop Verification
TL;DR
A PSO-enhanced BPNN-based PI parameter-tuning strategy (PSO-BPNN-PI), which achieves the lowest input-current THD and a balanced overall performance, while remaining competitive in tracking accuracy and maintaining its advantage over a wide inductance-mismatch range.
Abstract
To address the difficulty of conventional fixed-parameter PI controllers in the inner current loop of Vienna rectifiers in simultaneously achieving a fast dynamic response and overshoot suppression, this paper proposes a PSO-enhanced BPNN-based PI parameter-tuning strategy (PSO-BPNN-PI). The global search capability of particle swarm optimization (PSO) is exploited to optimize the initial weights of the back-propagation neural network (BPNN), mitigating the local-minima problem of random initialization and yielding a better-performing set of inner-loop PI parameters. On the equivalent current-loop model, the proposed strategy reduces the step-response overshoot to only 3.35%. On a full switching model of the Vienna rectifier, the proposed strategy is evaluated under steady-state operation, load disturbance, and filter-inductance variation and is further benchmarked against conventional PI, randomly initialized BPNN-PI, and LADRC controllers in dynamic disturbance tests. The proposed method achieves the lowest input-current THD and a balanced overall performance, while remaining competitive in tracking accuracy and maintaining its advantage over a wide inductance-mismatch range. Finally, a Hardware-in-the-Loop (HIL) platform with a heterogeneous CPU/FPGA architecture is established, on which the offline-optimized PI parameters are deployed; the results show stable tracking of the 750 V DC-bus voltage, rapid recovery from sudden load steps, and a well-balanced neutral-point voltage, confirming the engineering feasibility of the offline-optimized parameters.