A Method for Constructing Samples and Training Models for Predicting New Energy Power

By constructing a hybrid training sample set and using Bayesian data augmentation to generate virtual samples, the problem of model sensitivity to NWP error in new energy power prediction was solved, and high-precision and stable prediction was achieved under complex meteorological conditions.

CN122087450APending Publication Date: 2026-05-26HEFEI ZHONGKE LEINAO INTELLIGENCE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HEFEI ZHONGKE LEINAO INTELLIGENCE TECH CO LTD
Filing Date
2026-01-16
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing methods for predicting new energy power rely on a single data source and lack systematic modeling of numerical weather forecast errors, resulting in unstable prediction results and limited accuracy under complex meteorological conditions.

Method used

By simultaneously collecting measured meteorological data and NWP data, a hybrid training sample set is constructed. A Bayesian data augmentation method is used to generate virtual samples. Combined with gradient boosting trees and model weighted ensemble, a hierarchical model is established to alleviate the distribution mismatch problem.

Benefits of technology

It improves the stability and accuracy of new energy power forecasting, and can maintain the stability and reliability of forecast results under complex weather conditions, providing a more reliable basis for power grid dispatching decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122087450A_ABST
    Figure CN122087450A_ABST
Patent Text Reader

Abstract

This invention relates to the field of new energy power generation prediction technology, specifically, to a method for constructing new energy power prediction samples and training models. By simultaneously collecting historical actual power data, measured meteorological data, and historical numerical weather prediction data, measured meteorological sample sets and NWP sample sets are constructed respectively. The two sample sets are merged and source identification features are added to train a hybrid prediction model. A hierarchical model is established based on a Bayesian framework to analyze the systematic bias and random error distribution of NWP data, generating diverse virtual NWP samples and constructing a data augmentation training set. The augmented prediction model is trained using the augmented training set and then weighted and integrated with the baseline model, NWP model, and hybrid model to output the final prediction model. This invention, through innovative sample construction and training mechanisms, effectively solves the problem of inconsistent distribution between training and inference data, significantly improving the accuracy and robustness of power prediction.
Need to check novelty before this filing date? Find Prior Art