Hardware in the Loop-Driven Parameter Identification for Wind Turbine Controllers Using a Hybrid WaOA-GMM Method

With the large-scale integration of renewable energy sources into power systems, the dynamic characteristics of renewable generators have a critical impact on grid security and stability. However, wind turbine manufacturers usually do not disclose key internal controller parameters, making it difficult for grid operators to establish accurate white-box models for assessing the electromagnetic transient (EMT) behavior of wind farms. To address this “black-box” challenge, this paper proposes a parameter identification framework driven by hardware-in-the-loop (HIL) simulation waveforms, combining the Walrus Optimization Algorithm (WaOA) and the Gaussian Mixture Model (GMM) to accurately identify the controller parameters of white-box wind turbine models. First, an HIL simulation platform was built, where the actual controller (or the manufacturer’s black-box model) is connected to a real-time simulator. By designing typical fault scenarios (e.g., voltage sags), high-fidelity response waveforms were recorded. Second, a generic white-box model covering both doubly fed induction generator (DFIG) and permanent magnet synchronous generator (PMSG) wind turbines was developed on the CloudPSS platform, and Sobol global sensitivity analysis was used to screen the dominant parameters that significantly influence transient responses. Third, a GMM was constructed to model the probabilistic relationship between the dominant parameters and the simulation errors, and the posterior parameter distribution was derived using conditional probability invariance. Finally, an improved WaOA was adopted to directly optimize the controller parameters; each walrus position represents a candidate parameter set, and in each iteration, the posterior distribution from the GMM guides the generation of “newborn walruses” (high-quality candidate parameter sets), greatly accelerating convergence and improving identification accuracy. An HIL case study on a PMSG-based direct-drive wind turbine demonstrates that the proposed method can accurately identify the white-box model parameters, achieving an active-power error below 5%, without relying on proprietary manufacturer information. The experimental validation in this work focuses on the PMSG configuration; the extension to DFIG-based wind turbines is identified as future work.