A new method for short-term load forecasting based on weighted fusion of base models

•Expanding economic and climate features as inputs to enhance the model’s ability to capture load changes.•Introducing singular value decomposition technology to remove noise from load data, while preserving key features and trends, to improve data quality.•Select multiple typical models as sub mode...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Applied thermal engineering Ročník 280; s. 128313
Hlavní autori: Bian, Kai, Xu, Jing, Wang, Haohao, Ding, Junfeng
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier Ltd 01.12.2025
Predmet:
Combined model
Data denoising
Feature selection
Parameter optimization
Short-term load forecasting
Combined model
Feature selection
Data denoising
Short-term load forecasting
Parameter optimization
ISSN:1359-4311
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:•Expanding economic and climate features as inputs to enhance the model’s ability to capture load changes.•Introducing singular value decomposition technology to remove noise from load data, while preserving key features and trends, to improve data quality.•Select multiple typical models as sub model samples through the system, and choose the three with the best performance as the research focus.•Using the SFOA optimization algorithm to optimize the weights of the combined model and achieve higher prediction accuracy. In recent years, integrated energy systems have developed rapidly, integrating electricity, heat, cooling, and other energy sources into cross-energy coupling systems for efficient energy optimization. The setting and dynamic adjustment of their operating parameters depend heavily on power load forecasting, so improving forecasting accuracy is key to optimizing system operation. However, the load data is affected by interference from non-stationary signals and the introduction of a significant amount of noise. The traditional single-model approach struggles to handle complex and variable forecasting tasks, failing to meet the high-precision load forecasting demands. Given such problems, this research proposes a combined forecasting model utilizing SVD denoising preprocessing and multi-base model fusion to address the limitations of single-model approaches and enhance load forecasting precision. We used one year of load data from an Indian power plant to verify our proposed model. Firstly, SVD preprocesses the original data to remove noise and abnormal fluctuations. Unlike common modal decomposition methods that are prone to causing modal aliasing, SVD has stronger universality for nonlinear and non-stationary power load data and avoids such defects. Secondly, a weighted combination model integrating three base models is constructed, with SFOA searching for the minimum error weight to determine the optimal combination weight. The results showed that the MSE of the single BP model is 2.3709 MW, while the combined model proposed is just 1.2764 MW, which is 46 % lower. This intuitively proves the advantages of the combined model, providing data support for integrated energy system development and optimal resource allocation.
ISSN:1359-4311
DOI:10.1016/j.applthermaleng.2025.128313