Hierarchical Net Load Forecasting in High Solar Penetration Grids

Abstract

The large-scale integration of solar photovoltaics (PV) poses a challenge to the accuracy of net-load forecasting which is essential to maintaining grid stability and operational reliability. While the degradation of forecast accuracy with increasing PV penetration is well-documented, existing research has exclusively examined this at a single, aggregated level, overlooking the inherently hierarchical nature of power grids. This study addresses this gap by providing the first comprehensive analysis of how increasing PV penetration impacts forecast accuracy across a power grid hierarchy and evaluating the efficacy of statistical reconciliation methods in this volatile environment. The results reveal that increasing PV penetration consistently degrades forecast accuracy across all hierarchical levels, with the most pronounced impact and introduction of bias at the disaggregated levels. A key finding is that PV reshapes the forecast error distributions, creating more peaked profiles with heavier tails, which increases the risk of large forecast errors. The study demonstrates that optimal reconciliation methods are essential for managing this uncertainty, outperforming bottom-up approach. Furthermore, the study identifies Ordinary Least Squares reconciliation to be more effective at aggregated levels while the MinT-Shrinkage method to be superior for the disaggregated series. This research provides the first evidence-based framework for selecting and applying hierarchical reconciliation strategies to ensure forecast coherence and reliability in grids with high solar penetration, offering critical guidance for grid operators and policymakers navigating the energy transition.

Keywords: Solar PV, Net Load, Forecasting, Hierarchical Time Series, Reconciliation