Forecasting and techno-economic optimization of residential PV-battery systems: A synergistic approach for system enhancement

This study develops an integrated framework to determine the optimal sizing of a residential solar PV and battery energy storage system (BESS) using machine learning and a metaheuristic optimization approach. Economic analysis is conducted to evaluate the financial feasibility of the optimal configuration. Four machine learning models including Bi-LSTM, CNN-LSTM, Transformer and XGBoost were compared to identify the most reliable model for forecasting a residential building’s PV power generation and electricity demand. Among these, XGBoost achieved the highest forecasting accuracy, with R² values of 0. 943 for PV generation and 0. 978 for load prediction, and was consequently selected to generate PV and load profiles. To enhance generalizability beyond the case-study region, a seven-class irradiance-based geographic model was introduced to evaluate optimal system sizing across varying solar resource potentials. The daily load and generation profiles were incorporated into the Particle Swarm Optimization (PSO) algorithm to determine the optimal sizing of approximately a 92 kW PV system and a 364 kWh BESS for the higher-irradiance region and a 108 kW PV system and 376 kWh BESS for the lower-irradiance region. For economic evaluation, the seven classes were divided into two categories. Category 1, which receives higher solar irradiance, achieved an LCOE of 0. 13/kWh, an IRR of 9. 1%, and an approximate ROI of 113%. Although Category 1 performed significantly better, Category 2 remains economically feasible. Monte Carlo-based sensitivity analysis confirmed the system’s robustness to variations in key financial parameters. The results highlight the effectiveness of the approach in determining optimal system sizing and assessing economic feasibility for residential solar PV-BESS.