Digital Harmonic Identification of Power Systems Based on Multi-Agent Collaboration

Harmonic distortion analysis plays a crucial role in power quality monitoring. This paper proposes a harmonic classification framework based on multi-agent Support Vector Machines (SVM), combining Bayesian optimization and the Alternating Direction Method of Multipliers (ADMM) for distributed hyperparameter tuning. First, the power system signals are preprocessed, and harmonic features are extracted using Orthogonal Matching Pursuit (OMP). Then, a Bayesian optimization strategy is applied to tune the hyperparameters of the SVM classifier, incorporating a loss function to prevent overfitting. To ensure hyperparameter consistency among distributed agents, ADMM is used to collaboratively adjust the SVM parameters across multiple nodes. Experimental results show that ADMM further ensures the global convergence of hyperparameters across the entire network. This framework significantly enhances harmonic classification performance and is suitable for multi-agent power quality monitoring systems.