AI-driven forecast-integrated planning of hybrid renewable microgrids for climate-resilient rural electrification

Reliable renewable-resource characterization is critical for designing cost-effective and climate-resilient microgrids, particularly in rural regions where conventional planning often relies on outdated meteorological averages. This study proposes a forecast-integrated framework that couples deep-learning-based meteorological forecasting with techno-economic optimization of a hybrid renewable microgrid for a rural educational facility in Bangladesh. A 40-year dataset (1984–2024) of solar irradiance, wind speed, and ambient temperature was used to develop and validate six forecasting models: SVM, LSTM, BiLSTM, GRU, CNN–LSTM, and CNN–BiLSTM. The CNN–BiLSTM model achieved the best performance, with R 2 values of 0.980, 0.961, and 0.959 for solar irradiance, wind speed, and temperature, respectively. The resulting forecasts were integrated into HOMER Pro to evaluate seven microgrid configurations incorporating photovoltaic, wind, battery storage, hydrokinetic generation, and grid resources over a 25-year planning horizon. The optimal PV–wind–battery–grid configuration achieved a renewable fraction of approximately 90.9%, a cost of energy of $0.0471/kWh, and a net present cost of $17,225, while reducing annual CO 2 emissions by approximately 80.5% relative to the base case. A site-specific assessment further showed that hydrokinetic generation is technically feasible but economically non-optimal under the available river-flow conditions. Sensitivity analysis identified wind-turbine cost as the most influential component-level economic factor. The proposed framework provides a data-driven approach for robust renewable-energy planning, technology screening, and low-carbon rural electrification.