High-Precision Identification of Power Quality Disturbances Under Strong Noise Environment Based on FastICA and Random Forest

The continuous integration of fluctuating distributed generators and nonlinear power electronic equipment have produced severe signal contamination and induced various power quality (PQ) problems to modern power systems. PQ disturbances (PQD) greatly ruin user experience and also bring significant power losses. Therefore, a high-precision machine learning-based PQD identification model is proposed in this article, which combines the advantages of the modified fast independent component analysis method and the improved random forest classifier. First, ten types of PQD models are established, and fast independent component analysis is adopted to denoise the PQD sample signals mixed with Gaussian noises. Second, the discrete wavelet transform is utilized to extract the statistical and wavelet-related features from the denoised PQD samples, so as to form the desired feature set. Finally, a random forest-based PQD identification model is proposed. Compared with several existing models, the proposed model has higher identification accuracy and stronger feasibility under strong noise environment, which could provide valuable information for future PQ management.