The accelerating energy transition, marked by the rapid adoption of distributed energy resources and electrification technologies such as electric vehicles and heat pumps, is reshaping electricity demand patterns and challenging traditional forecasting approaches. Aggregate models, commonly used for grid planning, are increasingly inadequate to capture the spatial and temporal heterogeneity required for effective infrastructure reinforcement. This research addresses these limitations by proposing a hierarchical demand forecasting framework tailored for medium- to long-term planning, enabling a more accurate representation of localized consumption dynamics while preserving system-wide consistency.

The study is grounded in an extensive analysis of real-world data from the E-REDES Open Data portal, covering the entire Portuguese territory. A key requirement for Distribution System Operators (DSOs) is forecast coherence. Thus, projections at national or district levels must be mathematically consistent with those at finer spatial resolutions, such as municipalities and parishes. By focusing on monthly data, the approach emphasizes structural consumption trends and supports capacity planning decisions in a high-renewable, increasingly electrified context.

Methodologically, the model adopts a hierarchical structure aligned with Portugal’s administrative divisions, national, district, municipality, and parish. This multi-level framework captures the diversity of consumption patterns driven by land use, socio-economic factors, and uneven technological adoption. The diffusion of emerging loads, such as EVs and heat pumps, is explicitly treated as a spatially heterogeneous process. To ensure consistency across scales, the framework integrates both bottom-up and top-down reconciliation techniques, producing forecasts that are coherent throughout the hierarchy.

The implementation is delivered as a customized Python package, leveraging the Nixtla ecosystem (StatsForecast and HierarchicalForecast) for time series modelling, alongside ArcGIS for geospatial data management. Preliminary results demonstrate the model’s ability to preserve data integrity while delivering high-resolution insights into localized demand growth, offering a valuable decision-support tool for DSOs in optimizing grid investments and managing emerging load impacts.

By integrating spatial and temporal dimensions within a unified hierarchical framework, this work significantly enhances the ability to capture demand dynamics in modern power systems. The proposed open-source solution, validated with national-scale data, provides a robust foundation for future research on cross-vector energy interactions and supports more informed, localized energy system planning.