An Evaluation of Direct and Indirect Strategies for Hierarchical Net-Load Forecasting

Abstract

Power systems require accurate and coherent net-load forecasts across all levels of the grid hierarchy. A key methodological choice is between a direct strategy (forecasting net-load holistically) and an indirect strategy (forecasting gross load and renewable generation separately). This paper provides the first systematic evaluation of these two approaches within a hierarchical net-load forecasting framework, including the subsequent impact of statistical forecast reconciliation. Using load data from a three-level electricity network and simulated solar PV generation, this study compares the performance of both strategies using neural network auto-regression forecasting model and five reconciliation techniques. The findings show that the direct forecasting strategy yields statistically significant and more accurate forecasts across the hierarchy. However, the direct approach higher accuracy is coupled with over-forecasting bias. The analysis of reconciliation techniques revealed that while optimal methods improved the direct strategy's accuracy, they failed to correct its bias. Conversely, the simple Bottom-Up method, when applied to the indirect strategy, provided the least biased, though less accurate, overall forecast. The study concludes that while the direct approach is more robust for maximizing forecast accuracy, the final choice of strategy and reconciliation method must be aligned with specific operational objectives.