Bilinear-DRTFT: Uncertainty prediction in electricity load considering multiple demand responses

Abstract

Accurately predicting electricity load is crucial for maintaining the stability of modern power systems. Most studies focus on deterministic predictions, while uncertainty prediction and demand responses (DR) need further research. This paper proposes Bilinear-DRTFT, a load forecasting model that supports uncertainty and considers DR. To comprehensively evaluate model performance, this study combines five deterministic evaluation metrics and three uncertainty evaluation metrics on the test set. The research findings include:1. Incorporating DR effectively enhances the model's performance in both deterministic and uncertainty prediction; 2. Model performance continually improves with refined DR, achieving better uncertainty reduction when incorporating dynamic DR signals considering critical peak electricity prices, reaching expected performance at 80 % and 90 % confidence levels with narrow PINAW; 3. After adding DR, high-quality feature engineering provides the basis for the simplification of feedforward network; 4. Analyzing monthly uncertainty predictions reveals specific time periods on which DR policies should focus. These conclusions are validated in two experimental regions.