This initiative is spearheaded by a PhD student at the Responsible Artificial Intelligence Lab under the Energy theme The mini-grid system is vital for solving the energy deficit in rural Sub-Saharan Africa, especially communities far from the national grid or island communities isolated by water. It is proven in scientific literature to be one of the most efficient solutions for achieving universal electricity access by 2030, as indicated by the Sustainable Development Goals (SDGs).
Accurate electricity demand estimation is crucial for mini-grid site selection, system sizing, revenue projection, and ensuring seamless and robust energy solutions. However, predicting electricity demand in unelectrified communities lacking historical data poses significant challenges, with inaccuracies impacting capital costs and system efficiency.
This study addresses this gap by establishing a novel relationship between population growth, household connections, and appliance adoption, factors often overlooked in rural electrification models. The first phase involves a Population-Household-Appliance-Growth (PHAG) framework to estimate electricity demand. The output is fed into a hybrid deep learning model in the second phase, achieving superior performance (MSE: 0.0004, MAE: 0.0147, RMSE: 0.0199, R²: 0.9890, NRMSE: 0.2279). Validation against actual consumption revealed under-predictions of 8.50% and 1.37% for maximum and minimum peak loads, respectively, and an aggregated demand error of 0.40%.
The results demonstrate that combining appliance inventory surveys and time of use data with historical records, augmented by deep learning, effectively predicts initial electricity demand for unelectrified communities. This methodology provides a robust foundation for scalable, cost-effective mini-grid implementation in rural areas, ensuring sustainable energy access and development.
