Power quality analysis of an HID lamp network based on machine learning-assisted dynamic conductance model

Abstract

High-intensity discharge (HID) lamps employed in outdoor lighting are potential sources of harmonics in the distribution bus. These harmonics must be regulated as per IEEE 519-2022 and IEC 61000-3-2:2018 standards. A machine learning-assisted dynamic conductance model (ML-DCM) for HID lamps is developed for power quality analysis of lighting networks. ML-DCM offers superior accuracy and faster simulation speed and it is computationally lighter as compared to the previous lamp models owing to the data-driven approach. Extreme gradient boosting (XGBoost) regressor is the backbone of ML-DCM which is trained to predict the suitable model constants determining the physical processes of gas discharge given the inputs of lamp type, rated wattage and supply voltage. ML-DCM can reliably reproduce the electrical characteristics of an HID lamp and hence can be used as a virtual lamp load during computer simulation. A power quality analysis conducted on a 1.5-kW lighting network shows that harmonic current injections by individual lamps and distortions in the bus voltage waveform are below the maximum permissible limits. However, the harmonic distortion in the net load current violates the recommended limit of IEEE 519-2022 which has to be taken care of.