AI-Driven Optimization of a Hybrid PV–Wind–BESS Microgrid for a Rural Educational Institution in Developing Countries

Abstract

Hybrid renewable microgrid planning in HOMER Pro often relies on outdated meteorological data sets, which can lead to inaccurate component sizing and misestimation of economic performance under future climatic conditions. This study develops a forecasting-integrated optimization framework for a grid-connected PV–wind–battery microgrid serving a rural school in Bangladesh. A hybrid CNN–LSTM model trained on 40 years of NASA POWER data generates long-term (2025–2050) projections, achieving an R² of 0.9803 and the lowest prediction errors among eight artificial intelligence models. The forecasted data are integrated into HOMER Pro to enable climate-informed system design. The optimal configuration (36 kW photovoltaic [PV], 15 kW wind turbine, 22 kWh battery energy storage system, 18.4 kW converter) delivers 62.84 MWh/year of electricity, including 47.30 MWh from PV and 10.90 MWh from wind, with only 4.64 MWh purchased from the grid and 31.11 MWh exported annually. The system achieves a net present cost of USD 29,744, a levelized cost of energy of 0.0397 USD/kWh, an operating cost of 19.55 USD/year, a renewable fraction of 91.7%, and negligible unmet load (0.0031%). A discounted payback period of 7.73 years and a 12.4% internal rate of return confirm its financial viability. Annual CO₂ emissions are reduced by 81.5% compared with grid-only operation. Sensitivity analysis identifies solar irradiance and discount rate as dominant performance drivers. These findings demonstrate that integrating accurate meteorological forecasting with hybrid microgrid optimization enhances renewable penetration, reduces fossil-fuel dependence, and improves long-term system reliability.