Development of a Screening-Based Machine Learning Algorithm for PV Power Contribution Estimation

The growing penetration of distributed energy resources in distribution networks has highlighted the need to analyze power contribution characteristics under different operating and interconnection conditions. In particular, photovoltaic systems exhibit varying load-level power contribution patterns depending on their output levels and interconnection configurations, and machine-learning-based approaches have recently been investigated to efficiently estimate these contribution characteristics. However, conventional full-scale machine learning approaches require large datasets covering numerous PV operating and interconnection scenarios, resulting in substantial computational burden during dataset generation and model training. To address this issue, this paper proposes a screening-based machine learning algorithm for load-level PV power contribution estimation. A PV Impact-Range Index is developed by jointly considering active power contribution, voltage sensitivity, and electrical proximity to identify PV interconnection scenarios with meaningful contribution characteristics. Based on this index, non-informative scenarios are excluded, and a screening-scale dataset is constructed to train a computationally efficient learning model. Simulation results on a 19-bus distribution test system show that the proposed algorithm achieves prediction accuracy comparable to the full-scale learning approach while significantly reducing dataset generation and model training time. These results demonstrate that the proposed algorithm provides an efficient and practical solution for PV power contribution estimation in distribution networks.