Comprehensive energy wind power prediction method based on modal decomposition noise reduction and GAT-BiLSTM

By employing signal decomposition and noise reduction using ICEEMDAN, sample entropy, and Savitzky-Golay filters, combined with feature fusion models of GAT and BiLSTM, the problems of noise interference and correlation with meteorological factors in wind power prediction were solved, achieving high-precision and robust prediction results and providing reliable support for the optimized operation of integrated energy systems.

CN122246682APending Publication Date: 2026-06-19CHINA THREE GORGES UNIV
1 Cites 0 Cited by

Patent Information

Application Number
CN202610241939.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-02-28
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing technologies for wind power forecasting are subject to high-frequency noise interference and have difficulty in mining the spatiotemporal correlation of multidimensional meteorological factors, resulting in low forecast accuracy and poor robustness, which cannot meet the needs of efficient and economical operation of integrated energy systems.

Method used

An improved adaptive noise-complete ensemble empirical mode decomposition (ICEEMDAN) combined with a sample entropy algorithm is used to identify and filter out high-frequency noise. Useful signals are retained through a Savitzky-Golay filter. A graph attention network (GAT) is constructed to mine the spatial correlation of meteorological features. A bidirectional long short-term memory network (BiLSTM) is combined to capture temporal dependence, forming a GAT-BiLSTM combined prediction model.

Benefits of technology

It has achieved higher accuracy and improved robustness in wind power forecasting, providing reliable decision support, offering a scientific basis for the economic dispatch and safe operation of integrated energy systems, increasing the renewable energy absorption rate and reducing system operating costs.

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

Abstract

The invention discloses an integrated energy wind power prediction method based on modal decomposition noise reduction and GAT-BiLSTM, belongs to the technical field of renewable energy prediction, and aims to solve the problem that wind power prediction is interfered by high-frequency noise, and precision is low due to insufficient spatio-temporal correlation mining of multi-dimensional meteorological factors. The method comprises the following steps: firstly, collecting meteorological data such as wind power, wind speed and wind direction, preprocessing the meteorological data through an isolated forest algorithm, decomposing a wind power sequence by using ICEEMDAN, positioning a high-frequency noise mode in combination with sample entropy, and reconstructing a signal after filtering through a Savitzky-Golay filter; and a GAT-BiLSTM model is constructed, feature spatial correlation is mined through GAT, bidirectional time dependence is extracted through BiLSTM, and wind power prediction is realized. According to the method, accurate noise reduction of the wind power signal is realized, the wind power-meteorological space-time correlation is deeply fused, the prediction precision and robustness are remarkably improved, and reliable decision support is provided for economic dispatching of a comprehensive energy system.
Need to check novelty before this filing date? Find Prior Art