Distributionally Robust Bayesian Optimisation for Adaptive Voltage–Frequency Control via CVaR‐Guided Harmonic‐Aware ADRC Autotuning in Inverter‐Based Microgrids

ABSTRACT

Inverter‐based microgrids must regulate voltage and frequency under islanding, load swings, parameter uncertainty, and harmonic stress while maintaining power‐quality performance. Conventional proportional‐integral (PI) and proportional‐resonant controllers, as well as nominally tuned active disturbance rejection control (ADRC), face a speed–robustness trade‐off; meanwhile, Bayesian optimisation (BO), particle swarm optimisation (PSO), and genetic algorithm (GA) autotuning can improve mean performance while overlooking transient and harmonic tail events. This paper proposes a conditional value‐at‐risk–guided distributionally robust Bayesian optimisation (DR‐BO) framework for ADRC autotuning. Two extended state observer (ESO)‐based ADRC channels regulate the P–f and Q – V paths through compensated PI laws with saturation and anti‐windup protection. DR‐BO selects gains across grid‐connected/islanded operation, loading levels, step/ramp disturbances, and harmonic stress. Supervisory event‐window data are used only for scenario construction, identification, and DR‐BO evaluation; the deployed ADRC/ESO/PI loop runs at the fast digital controller period. Compared with fixed‐gain PI and nominal BO/PSO/GA baselines, the method reduces overshoot by 60%–65%, shortens 2% settling time by 67% (170 to 56 ms), and maintains simulated total harmonic distortion at 4.4% within a 5% screening envelope, with stronger stability margins. The workflow is offline/commissioning‐oriented, harmonic‐aware, and does not claim hardware certification or field readiness.