An enhanced model day-ahead electricity price forecasting method based on dynamic feature selection

By combining dynamic feature selection and heterogeneous parallel prediction with XGBoost and Transformer models, the problems of static feature selection and convenient model deployment in electricity price prediction are solved, achieving high accuracy and rapid response in the complex environment of the electricity market.

CN122155769APending Publication Date: 2026-06-05XIDIAN UNIV HANGZHOU RES INST +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
XIDIAN UNIV HANGZHOU RES INST
Filing Date
2026-02-02
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing electricity price forecasting technologies, when faced with the complex environment of the electricity market, have static feature selections that cannot be dynamically adjusted, resulting in limited forecasting accuracy and applicability in emergency scenarios. Furthermore, traditional models struggle to balance the structured logic and long-term trend evolution in electricity data, neglecting the practical deployment convenience and operational efficiency of the models.

Method used

An enhanced model based on dynamic feature selection is adopted, which combines XGBoost and Transformer models. Through multi-objective Bayesian optimization algorithm and multi-head attention mechanism, feature weights are adjusted in real time and heterogeneous parallel prediction is achieved, and a hybrid model is constructed to improve prediction accuracy and response speed.

Benefits of technology

It significantly improves the model's response speed and prediction accuracy when power market conditions change abruptly, ensuring robustness and efficient deployment under different volatile environments, and facilitating its application in actual trading systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122155769A_ABST
    Figure CN122155769A_ABST
Patent Text Reader

Abstract

The application discloses an enhanced model day-ahead electricity price prediction method based on dynamic feature selection, which extracts features from the original data set through a multi-head attention mechanism for multidimensional correlation analysis, generates a feature weight vector in real time to realize dynamic screening of the optimal feature subset, thereby suppressing redundant information and nonlinear noise features at the source; then, an enhanced XGBoost model is used to deeply mine the structured relationship of power market data, simultaneously combined with a Transformer model to capture the long-range time series dependence characteristics of the electricity price sequence, and through a gating mechanism to perform dynamic weighted fusion, to realize the complementary integration of spatial features and time series features; finally, a multi-objective Bayesian optimization algorithm is introduced to obtain a trained enhanced model through multi-objective collaborative optimization, and then prediction is performed. The application significantly improves the prediction accuracy and robustness, effectively solves the deployment problem of the model in a resource-limited environment, and has high engineering application value and power market prediction practicality.
Need to check novelty before this filing date? Find Prior Art