A time series model based on hybrid-kernel least-squares support vector machine for short-term wind power forecasting

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Time Series Analysis

Background:

• Accurate short-term wind power forecasting is crucial for grid stability and energy management.
• Existing forecasting models face challenges in capturing complex wind power dynamics.
• Hybrid models offer potential for improved prediction accuracy.

Purpose of the Study:

• To propose a novel hybrid-kernel least-squares support vector machine (HKLSSVM) model for short-term wind power prediction.
• To enhance forecasting accuracy by decomposing, classifying, and reconstructing wind power time series.
• To evaluate the performance of the proposed HKLSSVM model against established benchmark models.

Main Methods:

• A three-stage process involving decomposition, classification, and reconstruction of wind power time series.
• Maximal Wavelet Decomposition (MWD) and Fuzzy C-Means algorithm for data decomposition and classification.
• Least-Squares Support Vector Machine (LSSVM) with optimized parameters using Non-dominated Sorting Genetic Algorithm II for forecasting.

Main Results:

• The proposed HKLSSVM model effectively decomposes and classifies wind power time series components.
• Optimized LSSVM models with different kernels are established for each classified component.
• Reconstructed forecasts from the HKLSSVM model show improved accuracy over benchmark models.

Conclusions:

• The HKLSSVM model offers a robust and accurate approach for short-term wind power forecasting.
• The decomposition-classification-reconstruction strategy significantly enhances prediction capabilities.
• HKLSSVM outperforms Empirical Mode Decomposition-LSSVM (EMD-LSSVM) and Wavelet Decomposition-LSSVM (WDLSSVM) models.