Modern electrical power systems in developing sub-Saharan urban areas face critical structural vulnerabilities, characterized by severe distribution bottlenecks, rampant transformer overloads, and unmanaged demand-side dynamics. This research addresses these acute electrical power system challenges within the municipal grid of Offa, Kwara State, Nigeria, where rapid urbanization, localized commercial expansion, and a dense concentration of academic institutions have induced highly volatile domestic load profiles. To mitigate localized grid instability and optimize distribution asset utilization, this paper proposes a deep learning-enabled residential power consumption prediction system utilizing high-resolution advanced metering infrastructure (AMI) telemetry. The proposed framework integrates time-series electrical feature extraction, multi-layered deep artificial neural network (DNN) architectures, and Gaussian-form probabilistic modeling to deliver robust, multi-step ahead active power () forecasts at the individual household level. By identifying temporal loading signatures and mapping localized power system uncertainties, this interdisciplinary approach bridges power systems engineering, telemetry processing, and artificial intelligence. The resulting predictive model provides distribution network operators (DNOs) with actionable telemetry insights required to execute proactive peak load shedding, prevent distribution transformer failures, and formulate data-driven demand-side management (DSM) strategies tailored to the specific constraints of the Offa municipal grid.
JIMOH et al. (Sat,) studied this question.