Harmonic classification of different lighting technologies using empirical mode decomposition and support vector machines

The impact of excessive harmonic distortion on electrical distribution networks highlights the critical need to understand harmonics in different lighting technologies to detect sources of issues and their effects. The increasing prevalence of nonlinear loads in power systems has contributed to significant harmonic pollution, deteriorating power line quality. This paper aims to address this issue by focusing on power line frequency detection and the classification of lighting technologies, combining empirical mode decomposition (EMD) with the one-versus-one support vector machine (OVO-SVM) approach. EMD is utilized to analyze signals through intrinsic mode functions, preserving the signal’s characteristics while extracting features related to harmonic distortion from various lighting types, including light emitting diodes, compact fluorescent lamps, and incandescent bulbs. The integration with MATLAB/Simulink and an Arduino Uno enables real-time monitoring, and the use of quantization and FFT implementations measures total harmonic distortion. In terms of classification, the OVO-SVM using polynomial kernel functions outperforms the RBF kernel functions across various training-to-testing splits (50/50, 60/40, 70/30, and 80/20). Additionally, the performance of the OVO-SVM with polynomial kernel functions aligns closely with the gradient boosting machine classifier, showing identical results in accuracy, precision, recall, and F1-score. This suggests that both models are highly effective in classifying the types of harmonic distortions associated with different lighting technologies, offering an efficient method for power quality disturbances detection and classification in power systems.