A Load Forecasting Model Based on Sequence-to-Image Preprocessing Framework

The smart grid is currently composed of many distributed energy resources. To manage these sources, Short-Term Load Forecasting (STLF) is regarded as a crucial role in the power dispatching and energy management domain. To improve the performance and accuracy of the STLF, this paper proposes a Sequence-to-Image preprocessing approach and Rolling Forecasting framework (S2I-RF) to help the model better catch the multi-level seasonality of the load series and overcome the cumulative error associated with the multi-step forecasting. To better extract the spatial pattern and the temporal relationship of the load images, a structure of the hybrid neural network model CNN-ConvBiLSTM-ECANet (CCBE-Net) is proposed. The technique can significantly improve the performance of the STLF in real-time or day-ahead forecasting.