ANN-APFFT Harmonic Detection Method Considering Power System Security

Abstract

This study proposes a harmonic detection method based on artificial neural network and all-phase Fast Fourier Transform to solve three problems in power systems: inaccurate nonlinear feature extraction, noise sensitivity, and slow dynamic response. It uses adaptive particle swarm optimization algorithm to improve neural network parameter initialization and sparse decomposition to enhance spectrum anti_interference ability. This builds a tandem architecture for rapid feature extraction and precise parameter estimation. Experimental results revealed that the phase measurement accuracy reached 2.05°, and under standard simulation conditions the frequency tracking error did not exceed 40mHz. The average signal-to-noise ratio improvement reached 12.3dB across the tested noisy environments. Meanwhile, the algorithm convergence time of this method in transient events such as motor starting, capacitor switching, and short-circuit faults was reduced to within 105ms, 50ms, and 150ms, respectively. The research integrates intelligent feature learning with precision spectrum analysis. It effectively balances accuracy and robustness in harmonic detection for complex power grids. This provides technical support for power quality monitoring and safety assessment.