Key points are not available for this paper at this time.
Accurate prediction of air conditioning (AC) loads at the feeder level is imperative for the effective implementation of targeted demand response and effective peak load management. However, it is challenged by data heterogeneity and scarcity, especially for new feeders. Traditional methods often fail under such conditions. This paper proposes a meta-learning-based Dynamic Hierarchical Forecasting Framework (DHFF) explicitly designed for efficient few-shot load forecasting. A core gating network dynamically fuses predictions from a global unified model and a feeder specific model, adapting based on context. The framework’s effectiveness was validated through dual testing scenarios. Outstanding performance was achieved in data-rich environments, with an RMSE of 0.0135, improving upon a strong LSTM baseline by nearly 80%. Furthermore, rigorous few-shot experiments confirmed its primary design goal: under extreme data scarcity (e.g., 5% data), DHFF demonstrated superior accuracy, improving up to 7.01% over standard approaches by intelligently leveraging generalized knowledge. These results validate DHFF as an adaptive, high-performing solution across both data-scarce and data-rich feeder forecasting scenarios.
Long et al. (Wed,) studied this question.