Real-time power load prediction method and system based on hybrid model

By optimizing the hyperparameters of linear and nonlinear prediction models through hybrid models and combinatorial optimization algorithms, and combining trend compensation and online calibration, the problem of unstable load prediction in existing technologies is solved, and high-precision and stable prediction of power load is achieved.

CN122338733APending Publication Date: 2026-07-03YUNNAN POWER GRID CO LTD
0 Cites 0 Cited by

Patent Information

Application Number
CN202610437917.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-03
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing load forecasting technologies struggle to accurately capture both linear trends and nonlinear fluctuations in loads in real-time or near-real-time scenarios. Furthermore, deep learning models are sensitive to hyperparameters, leading to unstable forecasting performance and insufficient generalization ability.

Method used

A real-time power load forecasting method based on a hybrid model is adopted. The hyperparameters of the linear and nonlinear forecasting models are optimized by combining optimization algorithms, trend compensation is performed by combining linear and nonlinear forecasting sequences, and online calibration is performed using residual sequences to ensure the stability and accuracy of the forecasting results.

Benefits of technology

It achieves collaborative modeling of linear trends and nonlinear fluctuations in load, improves the stability and accuracy of forecasts, reduces reliance on manual parameter tuning, and enhances the robustness and adaptability of real-time rolling forecasts.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to a real-time power load prediction method and system based on a hybrid model. The method comprises the steps of obtaining current period load data, forming a load sequence in combination with historical loads, and obtaining corresponding weather data; performing timestamp alignment, exception handling and normalization on the load sequence and the weather data to obtain a standardized input sequence; determining a to-be-optimized hyper-parameter set formed by a linear prediction model order parameter, a nonlinear prediction model network structure and a training parameter, and completing optimization through a combinatorial optimization algorithm to update training configuration of the two models; respectively outputting a first prediction sequence and a second prediction sequence for the next period in the current period, and compensating the first prediction sequence by taking the second prediction sequence as a trend term to obtain a fused prediction sequence; and obtaining a fusion prediction sequence of the previous period to the current period, constructing a residual sequence with the load data of the current period, determining a deviation item, carrying out online calibration on the fusion prediction sequence of the current period, and outputting a calibrated load prediction result.
Need to check novelty before this filing date? Find Prior Art