Project TERESA: A GIS-based multifactorial framework utilizing supervised machine learning for nation-scale electrification planning

This paper presents a large-scale electrification strategy developed within the project "Technology for Rural Electrification in Sub-Saharan Africa" (TERESA). The approach proposed enhances the existing electrification framework GISEle by improving the population aggregation, introducing a novel survey-based methodology for energy demand estimation in spatially clustered communities, and incorporating a rule-based meta-heuristic algorithm to solve the optimization problem. Initially, iterative DBSCAN clusters population data. Electric grid and nighttime lighting open-source datasets then determine the electrification stage of communities. Subsequent steps involve load profile estimation: Multi-Tier Framework surveys gather household, educational, healthcare, and commercial activities data in selected communities. This represents a significant innovation in modeling cluster-level demand by systematically integrating survey insights into scalable energy planning. Considering the non-linearity between energy consumption, socioeconomic and resource consuming on-field campaigns, a supervised machine learning model extrapolates the energy demand of all the communities recognized by the clustering procedure. Lastly, a rule-based approach is utilized to determine each community's means of electrification, and a genetic algorithm is employed for expanding the national grid. The approach was applied to the Zambezia region-the second most populous and least electrified province in Mozambique. In this context, the method enabled demand estimation for 1,292 communities, leveraging and transferring insights derived from 726 on-field surveys to support broader regional planning.