Open Access

Samer Chaaraoui, Stefanie Meilinger, Sebastian Houben, Thorsten Schneiders, Windmanagada Sawadogo

• In Ghana, unreliable public grid infrastructure greatly impacts rural healthcare, where diesel generators are commonly used despite their high financial and environmental costs. Photovoltaic (PV)-hybrid systems offer a sustainable alternative, but require robust, predictive control strategies to ensure reliability. This study proposes a sector-specific Model Predictive Control (MPC) approach, integrating advanced load and meteorological forecasting for optimal energy dispatch. The methodology includes a long-short-term memory (LSTM)-based load forecasting model with probabilistic Monte Carlo dropout, a customized Numerical Weather Prediction (NWP) model based on the Weather Research and Forecasting (WRF) framework, and deep learning-based All-Sky Imager (ASI) nowcasting to improve short-term solar predictions. By combining these forecasting methods into a seamless prediction framework, the proposed MPC optimizes system performance while reducing reliance on fossil fuels. This study benchmarks the MPC against a traditional rule-based dispatch system, using data collected from a rural health facility in Kologo, Ghana. Results demonstrate that predictive control greatly reduces both economic and ecological costs. Compared to rule-based dispatch, diesel generator operation and fuel consumption are reduced by up to 61.62% and 47.17%, leading to economical and ecological cost savings of up to 20.7% and 31.78%. Additionally, system reliability improves, with battery depletion events during blackouts decreasing by up to 99.42%, while wear and tear on the diesel generator and battery are reduced by up to 54.93% and 37.34%, respectively. Furthermore, hyperparameter tuning enhances MPC performance, introducing further optimization potential. These findings highlight the effectiveness of predictive control in improving energy resilience for critical healthcare applications in rural settings.