Online Load Forecasting Model Considering Concept Drift
Ying Fan, Ling Luo, Pengfei Zhang, Na Wang, Yuxin Chen, Lili Wang · 2025
The traditional online learning models update model parameters based on sample data within a fixed-length time window, overlooking the impact of concept drift. To address this, this paper proposes an online learning model based on concept drift, which dynamically adjusts the size of the time window to ensure the model's adaptability to the most recent data. Firstly, a concept drift model is established. Then, the load data is updated to detect the occurrence of concept drift, followed by dynamically adjusting the size of the time window. Finally, using the sample data within the time window, load forecasting is performed based on an Attention-LSTM model. The proposed method is applied to forecast load data from a region in China. The experimental results demonstrate that the proposed method effectively enhances the accuracy of load forecasting.