Timesnet-based industrial park power demand forecasting system and method

The industrial park power demand forecasting system based on TimesNet solves the problems of insufficient identification of multi-period patterns and insufficient sensitivity to extreme value changes in existing demand forecasting technologies by using data processing, periodic feature analysis and two-dimensional tensor modeling, and achieves high-precision and stable demand forecasting.

CN122136807APending Publication Date: 2026-06-02JIANGXI THERMAL POWER CONSTR CORP
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
JIANGXI THERMAL POWER CONSTR CORP
Filing Date
2026-02-04
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing power forecasting technologies cannot effectively identify multi-period patterns in demand forecasting, lack sensitivity to extreme value changes, and have poor forecasting stability, making it difficult to meet the actual needs of power management and dispatching in modern industrial parks.

Method used

An industrial park power demand forecasting system based on TimesNet is adopted. The system uses a data processing module for cleaning and anomaly detection, a periodic feature analysis module for identifying the dominant period, a TimesNet modeling and feature extraction module for two-dimensional tensor structure modeling, and a feature fusion and prediction output module for multi-scale feature adaptive aggregation. The system also uses a multi-scale convolutional network to learn the features inside and outside the period.

Benefits of technology

It significantly improves the accuracy and stability of electricity demand forecasting, effectively identifies periodic patterns across multiple time scales, enhances sensitivity to extreme values, and achieves high-precision demand forecasting.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122136807A_ABST
    Figure CN122136807A_ABST
Patent Text Reader

Abstract

This invention provides a TimeNet-based system and method for predicting electricity demand in industrial parks, comprising a data processing module, a periodic feature analysis module, a TimeNet modeling and feature extraction module, and a feature fusion and prediction output module. The data processing module acquires high-frequency power data from an industrial park energy consumption monitoring system and performs cleaning, interpolation, and anomaly detection on the raw data. The periodic feature analysis module performs spectral analysis on the time series using Fast Fourier Transform to automatically identify the dominant period and its corresponding amplitude. The TimeNet modeling and feature extraction module inputs the periodically identified time series into a TimeNet model. The feature fusion and prediction output module performs Softmax weighted fusion of features extracted from different periods. This invention allows for the priority selection of a high-precision model in a cloud environment, while a lightweight model can be used in local systems with limited computing power, achieving an optimal balance between accuracy and efficiency.
Need to check novelty before this filing date? Find Prior Art