Minute-Level Power-Quality, Machine-Learning Forecasts and Techno-Economic Assessment of a 100 kWp PV-Battery-EV Fast-Charging Microgrid in a Caribbean Small Island State

Small Island Developing States (SIDS) remain locked into diesel generation and fragile low-voltage grids. This study integrates a four-month, fifteen- minute SCADA record from a 100 kWp PV array, lithium-ion battery energy storage system ( BESS) and 50 kW DC fast-charger at a Trinidad and Tobago filling station to quantify power-quality losses, economics, and forecasting. Phase-resolved IEEE-519 analysis shows current total harmonic distortion (THD) breaching the 5% limit in 35%–45% of intervals, with 95th-percentile distortion peaking at 11%, triggering inverter curtailment. Consequently the dry-season specific yield is only 1.72 kWh kWp−1 day−1 (20.6 MWh total) and Net Present Value (NPV) stands at –US$0.72 M. K-means clustering isolates a “heavy-harmonic” mode affecting 15% of operation, while Light Gradient Boosting Machine (LightGBM) 15-min-ahead forecasts achieve Mean Absolute Error (MAE)= 6.1 kW—over 50% lower error than Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX). Modelling indicates that active harmonic filtering, bi-weekly cleaning and solar-aligned electric vehicle (EV) tariffs can lift annual yield by >20%, flip NPV to +US$0.18 M and avoid 2.9 kt CO2 over 30 years. These minute-level results provide the first Caribbean evidence that power-quality governance, machine-learning control, and tariff reform jointly unlock bankable PV-BESS-EV hubs, informing grid codes, and -mobility policy across SIDS.