Techno-Economic Sizing and Optimization of Hybrid Solar Photovoltaic-Battery Energy Storage System (PV-BESS) For Time of Use (TOU) Tariffs Using Machine Learning

Malaysia's National Energy Transition Roadmap and Thirteenth Malaysia Plan set ambitious renewable energy targets, yet the new Regulatory Period 4 (RP4) tariff structure has reshaped the cost pressures facing non-domestic medium-voltage (MV) consumers, whose bills are dominated by a demand charge on peak power. This creates an opportunity to cut cost by shaving peak demand, and it raises a practical concern that this study addresses the extent to which an MV consumer can save by switching tariff alone, and the additional saving achievable by installing a hybrid solar photovoltaic and battery energy storage system (PV-BESS). Existing work leaves a gap here. Storage-sizing studies typically size the battery against a load assumed to be already known, while short-term load-forecasting studies show that tree-based models predict demand accurately but are rarely coupled to the sizing decision. Moreover, earlier Malaysian studies used the pre-2025 tariff and did not separate the effect of the tariff choice from that of the PV-BESS investment. This study addresses that gap by evaluating three cases for a single non-domestic MV facility: a General-tariff baseline (A), the same facility on the Time-of-Use (TOU) tariff without equipment (B), and a forecast-driven PV-BESS under the TOU tariff (C). Using a full year of hourly load data (CLEMD) and matched NASA POWER weather data, three machine learning models (ANN, LightGBM, XGBoost) were trained and compared. The best forecast was used to size the PV array in PVsyst, size the battery by a demand-threshold method, and schedule peak-shaving dispatch, after which each case was billed under RP4 rates and assessed by payback, NPV, IRR and ROI. XGBoost was the most accurate (MAPE = 3.57%, R² = 0.973). Notably, switching tariff alone was not beneficial, it raised the bill by RM87,950, because the facility's peak sits inside the charged TOU window. The entire saving came from the PV-BESS (941 kWp, 620 kW/2,100 kWh), which held the billed demand at 1,200 kW, cut annual billed demand by 33%, and lowered the bill by about RM948,000 per year (23.3%), paying back in 4.1 years (NPV RM4.41 million, IRR 21.4%), contingent on the GITA incentive. By isolating the tariff effect from the investment effect, the study gives Malaysian MV consumers a defensible basis for deciding which tariff to adopt and whether to invest.