Advancing Long‐Term Energy Demand Forecasting: A PRISMA‐Guided Systematic Review of AI and Econometric Models

Long‐term energy demand forecasts are used by governments, utilities, and planners when deciding future grid capacity, investment needs, and decarbonisation pathways. This review examines recent studies that used artificial intelligence, machine learning, hybrid, and statistical/econometric approaches for long‐term energy demand forecasting. The review followed PRISMA 2020. From 980 records identified in Scopus and IEEE Xplore for 2023–2025, 42 studies were retained. The papers were grouped into three themes: deep learning and hybrid models, ensemble machine learning models, and statistical/econometric models. The strongest deep learning and hybrid result reported was R 2 = 0.97, showing good ability to model nonlinear demand behaviour. Ensemble studies also reported improved robustness, including RMSE reductions of up to 37% and MAPE values below 2%. ARIMA, SARIMA, and LEAP remained useful where datasets were limited or policy assumptions had to be made explicit. The review also found repeated weaknesses in benchmarking, cross‐regional validation, uncertainty reporting, and labelled limitations sections. Overall, model selection for long‐term energy demand forecasting should consider not only accuracy but also data availability, interpretability, and policy use.