A multi-level transfer learning method for multi-level load forecasting of urban building groups

By employing multi-level transfer learning and cross-level dynamic attention mechanisms, a multi-level load forecasting model for urban building clusters is constructed. This solves the problems of data sparsity and single feature fusion, achieving high-precision load forecasting and adapting to the intelligent operation of urban energy systems.

CN122198240APending Publication Date: 2026-06-12CHINA ACAD OF BUILDING RES
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA ACAD OF BUILDING RES
Filing Date
2026-03-18
Publication Date
2026-06-12

AI Technical Summary

Technical Problem

Existing technologies for load forecasting in urban building clusters suffer from problems such as strong data dependence, insufficient model generalization ability, neglect of knowledge association between levels, and limited feature fusion, resulting in decreased forecast accuracy and poor adaptability, making it difficult to meet energy dispatching needs.

Method used

A multi-level transfer learning approach is adopted, which constructs a three-level progressive model of individual buildings, functional clusters and city-level regions. It combines spatiotemporal graph convolutional networks and cross-level dynamic attention mechanisms to integrate multi-dimensional features, dynamically capture spatiotemporal correlations, and optimize the load prediction model.

Benefits of technology

It improves the accuracy and robustness of load forecasting, adapts to the load forecasting needs of different cities, and provides reliable support for energy dispatching and grid optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122198240A_ABST
    Figure CN122198240A_ABST
Patent Text Reader

Abstract

The application discloses a kind of multi-level transfer learning's city building group multi-level load prediction method, including the load monitoring data and building data of the collection preset city, the load monitoring data and the building data are preprocessed;Based on the load monitoring data and the building data, construct monomer building basic model, according to space-time graph convolution network, space-time correlation modeling is carried out to obtain functional group load prediction model;Multi-dimensional feature fusion is used to construct city-level regional load prediction model, according to the monomer building basic model, the functional group load prediction model and the city-level regional load prediction model, construct city building group multi-level load prediction model;Cross-level dynamic attention mechanism is used to optimize city building group multi-level load prediction model, input the data to be predicted into the city building group multi-level load prediction model optimized, and output prediction result.
Need to check novelty before this filing date? Find Prior Art