A long-term load forecasting method combined with deep learning
By utilizing date features and global pattern information in long-term load forecasting, combined with a short-term forecasting network, the problem of low model throughput in existing methods is solved, and accurate long-term load forecasting is achieved.
CN115829092BActive Publication Date: 2026-06-09SICHUAN UNIV
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2022-11-18
- Publication Date
- 2026-06-09
AI Technical Summary
Technical Problem
Existing long-term load forecasting methods cannot effectively utilize global pattern information from historical load sequences, resulting in low model throughput and an inability to accurately predict longer-term load sequences.
Method used
By acquiring and normalizing date features, a date representation is generated. Global pattern information is compressed into the date representation. Residual calculation is performed in conjunction with a short-term forecasting network, ultimately generating accurate long-term load forecast results.
Benefits of technology
It improves the robustness and accuracy of long-term load forecasting, reduces the modeling burden, and increases the model throughput.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115829092B_ABST
Abstract
The application provides a long-term load prediction method combined with deep learning, relates to the technical field of load prediction, and comprises the following steps: obtaining date characteristics of each date and performing normalization processing to obtain date representation; obtaining all historical load sequences to obtain global mode information and compress the global mode information into the date representation of the corresponding date to generate global prediction; obtaining historical observation values of a backtracking window to subtract the global prediction of the corresponding date to obtain residual errors of the backtracking window; inputting the residual errors of the backtracking window, the date representation of each date and the date representation of the next date into a short-term prediction network to obtain prediction residual errors of the next date; adding the prediction residual errors to the residual errors of the backtracking window, and repeating the previous step to obtain residual errors of all dates to be predicted; and combining the global prediction and the residual errors of all dates to be predicted to obtain a final prediction result; the application grasps the global mode of the sequence from the global information of the time representation, increases the model throughput, and realizes accurate long-term load prediction.
Need to check novelty before this filing date? Find Prior Art