Semi-Supervised Machine Learning Model for Sizing of Distributed Renewable Energy Sources

Abstract

This study presents a significant advancement in energy planning for grid-connected homes with plug-in electric vehicles (PEVs). A cutting-edge model has been developed to accurately size battery storage systems (BSS), small wind tur bines (SWT), and solar photovoltaic panels (SPV). The model considers real-world factors like grid limitations and component degradation, resulting in more realistic outcomes. To tackle the complex problem, a semi-supervised machine learning algorithm approach was employed, combining unsupervised and super vised methods. This innovative algorithm outperforms traditional machine learning techniques and metaheuristic methods. By analyzing a wide range of configurations using both labeled and unlabeled data, the optimal setup to minimize electricity costs is identified. In addition, a real-time, rule-based, and efficient home energy management system is presented. The study is based on real data from Australia, including temperature, wind speed, solar radiation, load, and economic and technical information on solar, wind, batteries, and plug-in electric vehicles. The results demonstrate that the proposed model significantly outperforms the conventional Group Method of Data Handling (GMDH), marking a significant advancement in energy planning technology.