Comparative Analysis of Machine Learning Algorithms (Linear Regression, Random Forest, and XGBoost) with Explainable AI for Optimal BESS Sizing and Techno-Economic Assessment in Industrial PV Self-Consumption: A PRISMA 2020 Systematic Review and Empirical Study in Andalusia

The recent volatility in global energy markets has accelerated the adoption of photovoltaic (PV) self-consumption and Battery Energy Storage Systems (BESS) in the industrial sector. However, optimal sizing remains complex. This pre-registered study uses a two-fold approach: (1) an AI-assisted Systematic Literature Review (SLR) following the PRISMA 2020 statement to map the state-of-the-art in ML applications for PV-BESS sizing, and (2) an empirical comparative study evaluating three supervised ML algorithms — Linear Regression (baseline interpretable), Random Forest, and XGBoost — enhanced with Explainable AI (XAI) techniques (SHAP values and feature importance). The models predict optimal battery capacity (kWh) and payback period (years) using real industrial load profiles, solar irradiance, and electricity tariffs from Andalusia (Spain). Expected outcomes include identification of the best-performing algorithm, quantification of XAI benefits for industrial decision-making, and clear research gaps to guide future work. The protocol is fully pre-registered to ensure transparency and reproducibility.