Hybrid Robust Regression for Harmonic Contribution Assessment in Power Distribution Networks

As renewable energy and power electronic devices penetrate power distribution networks, harmonic pollution has become increasingly severe. Because of random user behavior and time-varying operating conditions, harmonic voltages and currents vary widely, making conventional deterministic regression methods inadequate. To address this problem, this paper proposes a hybrid robust estimation method combining LTS, ridge regression, and IRLS (HRE-LRI) for harmonic contribution assessment. The three-stage estimator synergistically handles outliers, multicollinearity, and non-normal errors using only standard robust-statistics constants, while cross-validation adaptively selects the key ridge parameter, as validated by ablation experiments. A projection-based method is then established to quantify the harmonic contribution of each known source and the background. Comprehensive simulations on the IEEE 33-bus system are conducted, with partial least squares (PLS) and recursive least squares (RLS) as benchmarks. Additional comparisons are also provided, with recently proposed harmonic impedance estimation methods including Huber/M-estimation and differential recursive estimation. Results demonstrate that the proposed method maintains high accuracy and robustness under harsh conditions, including frequent source switching, strong current correlation, and background noise. Simulation results show that HRE-LRI maintains high accuracy and robustness in harmonic contribution estimation under different operating conditions.