Abstract
The accuracy of wind power prediction is very important for the stable operation of a power system. Ultra-short-term wind speed forecasting is an effective way to ensure real-time and accurate wind power prediction. In this paper, a short-term wind speed forecasting method based on a support vector machine with a combined kernel function and similar data is proposed. Similar training data are selected based on the wind tendency, and a combination of two kinds of kernel functions is applied in forecasting using a support vector machine. The forecasting results for a wind farm in Ningxia Province indicate that a combination of kernel functions with complementary advantages outperforms each single function, and forecasting models based on grouped wind data with a similar tendency could reduce the forecasting error. Furthermore, more accurate wind forecasting results ensure better wind power prediction.
Highlights
As environmental issues have become more prominent, wind power has been rapidly developing as a clean renewable energy source [1,2,3]
Since the output power of a wind turbine is directly dependent on the actual wind speed, a research hotspot is to realize wind power prediction indirectly through wind speed prediction [3,4,5,6]
For the prediction of ultra-short-term wind speed, this paper proposes a support vector machine prediction model based on a combined kernel function and similar data
Summary
As environmental issues have become more prominent, wind power has been rapidly developing as a clean renewable energy source [1,2,3]. For the prediction of ultra-short-term wind speed, this paper proposes a support vector machine prediction model based on a combined kernel function and similar data. In this model, the training samples are extracted based on the trend of changes in wind speed, and a training model is established. The wind speed data prediction based on a wind field in Ningxia shows that the combination of a wavelet kernel function and polynomial kernel function has higher prediction accuracy than the single kernel functions. According to the trend of the wind speed, this paper proposes a classification modeling method that divides all training samples into three categories: rising, gradual, and decreasing. After normalization by formula (8), the sample component values of the training set and test set are between [0, 1]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have