Residual-corrected TSMixer with a RIME-enhanced decomposition strategy for photovoltaic power prediction

•Proposes a RIME-enhanced two-stage decomposition for deep feature extraction.•Bayesian optimization and residual correction enhance the model’s performance.•The model’s excellent balance of accuracy and efficiency is verified by experiments. To address the challenges of volatility and uncertainty i...

Full description

Saved in:
Bibliographic Details
Published in:Electric power systems research Vol. 253
Main Authors: Cao, Lipeng, Shao, Xing, Wang, Cuixiang, Gao, Jun
Format: Journal Article
Language:English
Published: Elsevier B.V 01.04.2026
Subjects:
Bayesian optimization algorithm
Photovoltaic power prediction
Residual correction
TSMixer
Two-stage decomposition
Two-stage decomposition
Photovoltaic power prediction
Bayesian optimization algorithm
TSMixer
Residual correction
ISSN:0378-7796
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Proposes a RIME-enhanced two-stage decomposition for deep feature extraction.•Bayesian optimization and residual correction enhance the model’s performance.•The model’s excellent balance of accuracy and efficiency is verified by experiments. To address the challenges of volatility and uncertainty in photovoltaic (PV) power prediction, this paper proposes a hybrid prediction model based on two-stage decomposition and residual correction (RC). The model first employs Variational Mode Decomposition (VMD), optimized by the RIME algorithm, for primary decomposition, followed by secondary decomposition of the resulting high-frequency residuals using Complete Ensemble EMD with Adaptive Noise (CEEMDAN) to achieve deep feature extraction. Subsequently, a TSMixer model, fine-tuned by the Bayesian Optimization Algorithm (BOA), is used to predict each component. Finally, the prediction results are rectified through an error correction mechanism. Comprehensive experiments based on datasets from multiple sites with diverse climatic conditions–including Alice Springs and Yulara in Australia, and Xinjiang in China–show that the proposed model demonstrates significant performance advantages over several comparative models, validating its effectiveness, robustness, and generalizability in complex PV power prediction tasks.
ISSN:0378-7796
DOI:10.1016/j.epsr.2025.112554