Photovoltaic power prediction method based on combination of TCN-BiGRU optimized by KLA and CatBoost model

By optimizing the combined prediction method of TCN-BiGRU and CatBoost using KLA, the problem of long-term time dependence and difficulty in capturing feature interactions in photovoltaic power prediction is solved, and higher accuracy and robustness of photovoltaic power prediction are achieved.

CN122243234APending Publication Date: 2026-06-19CHINA THREE GORGES UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA THREE GORGES UNIV
Filing Date
2026-03-10
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing photovoltaic power prediction methods struggle to simultaneously and efficiently capture long-term time-series dependencies and complex feature interactions, and their hyperparameter optimization is inefficient, resulting in limited prediction accuracy.

Method used

A combined prediction method using Kirchhoff's Law (KLA) algorithm to optimize temporal convolutional networks, bidirectional gated recurrent units (TCN-BiGRU), and a classification boosting library (CatBoost) is adopted. The high-level features of time-series data are deeply mined by TCN-BiGRU and then fused with the original features before being input into the CatBoost model for final prediction.

Benefits of technology

It achieves higher accuracy and stronger robustness in photovoltaic power prediction under various weather conditions, significantly improving prediction accuracy and stability, and performing particularly well in complex scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122243234A_ABST
    Figure CN122243234A_ABST
Patent Text Reader

Abstract

A photovoltaic power generation prediction method based on KLA-optimized TCN-BiGRU and CatBoost model combination includes: acquiring historical power data and related meteorological data of the target photovoltaic power plant, and preprocessing the data to form a dataset; using KLA to efficiently optimize the key hyperparameters of the TCN-BiGRU and CatBoost models respectively; inputting the preprocessed data, divided by weather type, into the optimized TCN-BiGRU hybrid network for training and preliminary prediction; concatenating the extracted deep temporal features with the features of the preprocessed original meteorological and power data to construct a fused feature matrix, which is then used as input to the optimized CatBoost model for training and final prediction; and outputting the final photovoltaic power generation sequence predicted by the CatBoost model. This method aims to solve the problems of existing technologies where models struggle to simultaneously handle long-term temporal dependencies and complex feature interactions, as well as the low efficiency of hybrid model parameter optimization, achieving higher accuracy and stronger robustness in photovoltaic power prediction.
Need to check novelty before this filing date? Find Prior Art