Machine Learning Approaches for Demand Forecasting and Intelligent Dispatch in Nigerian Decentralised Energy Systems

Nigeria’s persistent national grid instability has necessitated a paradigm shift from centralised generation expansion towards predictive and decentralised resilience strategies. This paper presents a PRISMA-inspired systematic review examining the efficacy of machine learning-based demand forecasting and intelligent dispatch for enhancing the reliability of hybrid microgrid systems within the Nigerian context. Through a systematic search across Google Scholar, ScienceDirect, Nature/Scientific Reports, PubMed/PMC, and MDPI, 47 records were identified, with 10 peer-reviewed studies meeting eligibility criteria for narrative synthesis after PRISMA screening. The synthesis reveals that ensemble and hybrid deep learning architectures, particularly Random Forest, XGBoost, and Prophet-LSTM models, consistently outperform single-algorithm baselines, achieving high predictive accuracy with coefficients of determination up to 0.9992 and error reductions of 12-18%. Integration of these forecasting models with mixed-integer linear programming and model predictive control-based energy management systems demonstrates substantial operational benefits, including reduction of grid dependency from 100% to 28.6%, diesel consumption reduction, and carbon emissions reduction of up to 27.5%. The review proposes a context-specific three-layer forecasting-to-dispatch framework comprising data, modelling, and dispatch layers tailored to Nigeria’s data constraints and climatic dynamics. The paper concludes that scalable deployment hinges less on novel algorithmic innovation and more on open data infrastructure, local model recalibration, and integrated forecasting-dispatch implementation in pursuit of Sustainable Development Goal 7.