A Transfer Learning Method for Forecasting Masked-Load With Behind-the-Meter Distributed Energy Resources

With increasing installation of behind-the-meter distributed energy resources (DERs) at the household level, the power load profile has been significantly masked. As a result, original load forecasting models are becoming not suitable for the continuously masked-load. Besides, present masked-load may not have sufficient samples to train an accurate forecasting model. This letter proposed a transfer learning-based method to solve this problem by capturing different but related relationships between the unmasked-load and masked-load datasets. The common feature vectors from the two datasets are extracted by adversarial training, then the feature vectors from masked-load could be compatible with a predictor which is based on feature vectors from unmasked-load, thus masked-load could be accurately predicted. Simulation results show that the proposed method has higher accuracy compared to the benchmark models.