Wavelet kernel least square twin support vector regression for wind speed prediction.

Area of Science:

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

Background:

• Accurate wind speed (WS) prediction is essential for efficient wind power generation.
• Wind speed data is often non-stationary, posing challenges for traditional prediction models.
• Wavelet transforms offer a robust method for handling non-stationary time series data.

Purpose of the Study:

• To propose novel wavelet kernel-based least squares support vector regression (LSTSV R) models for wind speed prediction.
• To evaluate the performance of Morlet wavelet kernel LSTSV R and Mexican hat wavelet kernel LSTSV R models.
• To compare the proposed models against existing methods like twin support vector regression (TSVR), primal least squares support vector regression (PLSTSVR), and large-margin distribution machine-based regression (LDMR).

Main Methods:

• Development of Morlet wavelet kernel LSTSV R and Mexican hat wavelet kernel LSTSV R models.
• Utilizing hourly wind speed data from four meteorological stations in Tamil Nadu, India.
• Performance evaluation using metrics such as root mean square error (RMSE), mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), mean absolute scaled error (MASE), and R-squared (R²).

Main Results:

• The proposed wavelet kernel-based LSTSV R models demonstrated superior performance in wind speed prediction.
• Comparative analysis indicated that the novel models outperformed TSVR, PLSTSVR, and LDMR across all performance indicators.
• The wavelet-based approach effectively addressed the non-stationarity inherent in wind speed data.

Conclusions:

• Wavelet kernel-based LSTSV R models offer a significant advancement for accurate wind speed forecasting.
• These models provide a more reliable approach to predicting wind power generation.
• The findings highlight the potential of integrating wavelet transforms with LSTSV R for renewable energy applications.