A Validation-Guided Framework for Target-Specific Pathway Configuration in Daily Electricity-Supply Forecasting Under Data Constraints

Abstract

Accurate daily forecasts of thermal generation, renewable generation, hydropower, and imported electricity are essential for spot-market trading and day-ahead scheduling. However, short training histories, incomplete operational information, and distinct target drivers constrain accuracy and generalizability. To address these challenges, this study proposes a Delta-enhanced, validation-guided framework for configuring target-specific forecasting pathways. The Delta representation captures inter-day changes in thermal and hydropower generation. For the internal renewable-generation target, a regional target-day weather-forecast interface is applied retrospectively. It aggregates 13 hierarchical regional forecast columns for each of 24 weather variables and incorporates six calendar variables. For each target, chronological validation selects an admissible pathway defined by forecast-origin information, target representation, predictor family, optional temporal representation, and output rule. Once selected, it is refitted and fixed for subsequent forecasting. The evaluation covers four internal targets from a regional power system in China and six national-scale targets from France. The 97-day evaluation yielded R2 values of 0.8084, 0.7337, 0.7782, and 0.5212 for thermal generation, renewable generation, hydropower, and imported electricity, respectively. The frozen thermal and renewable pathways reduced RMSE by 23.2% and 50.9%, respectively, relative to Persistence, a baseline that carries the previous-day value forward. These findings support target-specific forecasting under data constraints.