This paper addresses the challenge of forecasting electricity consumption time series at a regional aggregation level. Due to limited measurement data and the lack of reliable validation benchmarks, generating accurate day-ahead forecasts for small geographic units remains difficult. The study focuses on overcoming these limitations by combining national-level forecasting with regional disaggregation approaches. In this way, it supports improved power system analysis and operation, particularly for grid calculation and stability assessment.

Among other applications, highly resolved consumption forecasts are needed to simulate electricity market prices. Additionally, they are essential to determinate and forecast vertical power flows (the combination of electrical energy production and consumption) at transmission grid buses.

The proposed methodology integrates two established procedures. First, a deep learning forecasting model, Neural Hierarchical Interpolation for Time Series (NHITS), is trained to produce a 24-hour day-ahead forecast at 15-minute resolution of Germany's total electricity consumption based on ENTSO-E time series data. This model decomposes predictions across multiple temporal scales using hierarchical interpolation, enabling efficient and accurate long-horizon time series forecasting, achieving an RMSE of 475 MW and a MAPE of 2.42 % on a 96-step ahead forecasting task. Second, regional consumption time series at the NUTS-3 level are generated using the open source DemandRegio model, which relies on statistical consumption data and standard load profiles. To ensure consistency between national and regional levels and to forecast on the regional level, the aggregated regional time series is compared to the NHITS national consumption forecasts. Scaling factors are then calculated at 15-minute intervals by aligning the aggregated regional data with the national wide forecasts. These factors are subsequently applied to adjust the regional time series accordingly.

The results demonstrate that this combined approach enables the generation of coherent and reliable day-ahead electricity demand forecasts at NUTS-3 level, disaggregated into households, commercial/trade/services (CTS), and industry sectors. The method ensures that regional forecasts remain consistent with national totals while preserving spatial granularity.