A hybrid load forecasting framework for future grid planning: NB Power case study

Abstract

Accurate long-term load forecasting (LTLF) is vital for strategic planning, infrastructure investment, and resilient grid operations, especially amid rising Behind-the-Meter (BTM) generation, Electric Vehicles (EVs), and Demand Side Management (DSM). This study develops a hybrid framework that integrates a Random Forest (RF) for trend extraction and an Artificial Neural Network (ANN) for residual modeling to forecast provincial electricity demand over a 20-year horizon. The approach includes a robust data preprocessing pipeline that harmonizes multi-resolution datasets and constructs scenario-based projections. Deterministic and probabilistic forecasts are generated, with the latter employing Monte Carlo simulations and time-varying noise injection to capture long-term uncertainty. Using NB Power as a case study, the model achieved an RMSE of 162.80 MW, an MAE of 112.70 MW, and a near-zero mean error of −5.36 MW. Results outperform traditional methods and enhance understanding of future load trajectories, supporting data-driven utility planning and risk management.