Method, device and equipment for coordinated prediction of ac and dc loads, and storage medium

By extracting time-series features bidirectionally from the AC/DC load forecasting model and calculating attention weights, a weighted composite loss function is constructed, which solves the problem of neglecting the correlation between AC and DC load forecasting and achieves high-precision collaborative forecasting results.

CN122199194APending Publication Date: 2026-06-12GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
Filing Date
2026-02-04
Publication Date
2026-06-12

AI Technical Summary

Technical Problem

Existing load forecasting methods neglect the inherent correlation and mutual influence between AC and DC loads, resulting in forecast results that deviate from reality and have limited accuracy.

Method used

By acquiring multidimensional feature sets at multiple sampling times, using a shared feature extraction layer to extract time-dependent features bidirectionally, combining an attention layer to calculate weights, inputting them into AC and DC load prediction layers for collaborative prediction, and constructing a weighted composite loss function for model training.

Benefits of technology

It enables accurate prediction of AC and DC load power, improves prediction accuracy and reliability, adapts to the real-time requirements of hybrid power grids, and provides accurate data support for power grid dispatch and power allocation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122199194A_ABST
    Figure CN122199194A_ABST
Patent Text Reader

Abstract

The application relates to a coordinated prediction method, device and equipment for AC / DC loads, and a storage medium. The method comprises the following steps: first, obtaining a multi-dimensional feature set related to AC / DC loads in a preset area at multiple sampling time points; then inputting the feature set into a trained prediction model shared feature extraction layer to bidirectionally extract time sequence dependent features and splice hidden state vectors at the multiple sampling time points; subsequently, inputting the hidden state vectors into a model attention layer to calculate and normalize attention scores between the hidden state vectors to determine corresponding weights, and obtaining a context vector through weighted summation; finally, inputting the context vector into AC and DC load prediction layers of the model respectively to output predicted values of AC / DC load powers, and realizing the coordinated prediction of AC / DC loads through the process of "feature collection-bidirectional time sequence extraction-attention weighted fusion-double load prediction". The method can effectively improve the accuracy and reliability of AC / DC load power prediction.
Need to check novelty before this filing date? Find Prior Art