Multi-modal electricity load forecasting method and system based on news and time series embedding fusion technology, and medium

CN122371104APending Publication Date: 2026-07-10STATE GRID SHANGHAI MUNICIPAL ELECTRIC POWER CO
View PDF 0 Cites -1 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
STATE GRID SHANGHAI MUNICIPAL ELECTRIC POWER CO
Filing Date
2026-06-10
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing electricity load forecasting methods struggle to fully utilize various dynamic external factors when facing complex power environments, leading to a decline in generalization ability and robustness. Furthermore, deep learning models have shortcomings in multimodal data fusion and interpretability, making it difficult to accurately capture rapid load fluctuations and sudden peaks.

Method used

We employ a generative agent-based RRFU framework to filter news texts, generate summaries, and use a knowledge-enhanced bimodal coding architecture for encoding and cross-modal alignment. This integrates structured payload sequences with unstructured news texts and utilizes a large language model to generate interpretable prediction results.

Benefits of technology

It enables rapid response and high-precision prediction of emergencies, improves the robustness and interpretability of the model, enhances the ability to capture complex load fluctuations, and improves the reliability and security of power grid dispatch.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122371104A_ABST
    Figure CN122371104A_ABST
Patent Text Reader

Abstract

This invention relates to a multimodal electricity load forecasting method, system, and medium based on news and temporal embedding fusion technology. The method includes: acquiring multi-source news text and a multivariate load sequence for the current period; using a generative agent-based RRFU framework to perform relevance filtering on the multi-source news text and generate a news summary; encoding the multivariate load sequence and news summary using a knowledge-enhanced bimodal coding architecture to obtain temporal embeddings and text semantic embeddings, respectively; performing cross-modal alignment and fusion based on the temporal embeddings and text semantic embeddings to obtain fused features; and predicting the fused features to output the predicted electricity load value for the future step. Compared with existing technologies, this invention has the advantages of overcoming the prediction bottleneck of traditional methods in complex scenarios and improving the accuracy, robustness, and interpretability of electricity load forecasting.
Need to check novelty before this filing date? Find Prior Art