Integrating AI-Based Electricity Demand Forecasting With Solar Grid Planning to Enhance Sustainability and Reliability
Abstract
Proper electricity-demand forecasting is essential for reliable power-system planning, particularly in urban networks facing rapid demand growth and transformer overloading. However, many previous studies have treated load forecasting and renewable-energy integration as separate tasks, which limits their usefulness for practical planning. This study develops an integrated forecasting–planning framework that links AI-based electricity-demand forecasting with photovoltaic (PV) system design and transformer-loading assessment. The framework is applied to real daily data from the Al-Intisar 132/33 kV substation in Mosul, Iraq, covering electrical load, temperature, population, and date-related variables for the period 2022–2024. Fourteen forecasting models from four methodological categories were evaluated: machine-learning models, deep-learning models, hybrid models, and statistical time-series models. Model performance was assessed using mean absolute error (MAE), root-mean-square error (RMSE), coefficient of determination (R2), symmetric mean absolute percentage error (SMAPE), and computational time. Among the evaluated models, SARIMAX achieved the best performance on the 2024 hold-out test set, with R2 = 0.999998 and SMAPE = 1.28%. The selected SARIMAX model was then used to forecast future electricity demand up to 2030, and the resulting peak-demand estimates were used in PVsyst to design a 160 MWp grid-connected PV system. The integration impact was assessed using ETAP, focusing on load-flow-based transformer-loading evaluation and comparison of operating conditions with and without PV integration. The results show that the proposed PV system substantially reduces the grid-supplied load and relieves transformer overloading, supporting improved operational reliability of the Al-Intisar network. The main contribution of this work is the application-oriented integration of load forecasting, PV design, and transformer-loading assessment for a real 132/33 kV substation in Mosul.