A Small Sample Load Forecasting Method Based on Hierarchical Data Ensemble Transfer Learning

Abstract

Accurate power load forecasting is essential for the safe and efficient operation of modern power systems. However, in practical scenarios, limited data availability caused by sensor failures or incomplete data collection poses significant challenges to conventional forecasting methods. To address this issue, this paper proposes a small-sample load forecasting method based on hierarchical ensemble transfer learning. The proposed approach employs a CNN–Informer parallel model with adaptive feature fusion to jointly capture local temporal patterns and long-range dependencies under data-scarce conditions. Furthermore, a goodness-of-fit-based hierarchical transfer learning strategy is introduced to rank and select source-domain datasets according to their relevance to the target domain, thereby reducing negative transfer. The final prediction is obtained by integrating relevance-weighted transfer learning with direct small-sample training through weighted averaging. Experimental results on a real-world small-sample load dataset demonstrate that the proposed method consistently outperforms representative baseline approaches in terms of forecasting accuracy and robustness.