Behind-the-Meter PV Power Disaggregation via Ensemble Machine Learning Methods

In recent years, the rise in popularity of distributed photovoltaic (PV) systems has presented new challenges to the operation of the grid. To enable effective grid control and optimization, accurate estimation of PV generation is crucial. Yet, many distributed rooftop PV installations are installed behind the meter, limiting grid operators' access to their actual generation data. This paper presents a novel machine learning model to address this problem. Firstly, the peak generation power features of the users are extracted using the net load distribution characteristics, after which a robust PV disaggregation framework is developed by utilizing known user data within the same geographical area and combining it with meteorological information. The approach is applied to the targeted users, thus allowing the accurate disaggregation of PV generated power for individual consumers. A case study on publicly available dataset from the Australian Grid demonstrates the effectiveness of the model in achieving accurate and reliable PV power disaggregation, providing a valuable reference for managing the grid scheduling operations.