Short-term photovoltaic power prediction based on RF-SGMD-GWO-BiLSTM hybrid models

Accurate photovoltaic (PV) power prediction can support optimal scheduling and decision-making in energy systems. An innovative hybrid prediction model efficiently predicts complex time series data by integrating multiple advanced algorithms in a deep fusion approach. First, Random Forest (RF) is em...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Energy (Oxford) Ročník 316; s. 134545
Hlavní autoři: Yang, Shaomei, Luo, Yuman
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 01.02.2025
Témata:
algorithms
Bidirectional long and short-term memory networks
case studies
decision making
energy
Feature selection
geometry
Gray wolf optimization algorithm
memory
Power prediction
prediction
solar energy
time series analysis
Bidirectional long and short-term memory networks
Power prediction
Feature selection
Gray wolf optimization algorithm
ISSN:0360-5442
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Accurate photovoltaic (PV) power prediction can support optimal scheduling and decision-making in energy systems. An innovative hybrid prediction model efficiently predicts complex time series data by integrating multiple advanced algorithms in a deep fusion approach. First, Random Forest (RF) is employed for feature screening and optimization to eliminate redundant features, thereby enhancing model training efficiency and prediction accuracy when dealing with complex environmental data. Subsequently, the symplectic geometry model decomposition (SGMD) technique was utilized to break down historical power signals, extracting information from various frequency components and enhancing the input features for the Bidirectional Long and Short-Term Memory Network (BiLSTM) model. Thereafter, aiming at the problem of hyperparameter tuning of the BiLSTM model, the Gray Wolf Optimization Algorithm (GWO) was used for automatic optimization to improve prediction stability. In the case studies, the proposed model exhibited impressive performance metrics: in Case Study 1, it achieved an RMSE of 1.351, an MAE of 0.666, and a MAPE of 3.786, while in Case Study 2, it recorded an RMSE of 0.487, an MAE of 0.265, and a MAPE of 6.304. The model also shows scalability and robustness across diverse climatic conditions and PV technologies, confirming its applicability to real-world scenarios. •Highlighting the significance of selecting optimal meteorological features in PV power prediction models.•Introducing the SGMD algorithm to decompose PV power and enhance input feature quality for improved predictions.•Addressing the BiLSTM hyperparameter optimization challenge using the GWO algorithm to enhance model stability and accuracy.•Demonstrating BiLSTM's unique ability to capture both long- and short-term dependencies in time-series PV power data.•Validating the model's scalability and performance across diverse climatic conditions and PV technologies.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0360-5442
DOI:10.1016/j.energy.2025.134545