Short-term wind power forecasting using the hybrid model of improved variational mode decomposition and Correntropy Long Short -term memory neural network. (1st January 2021)

• Record Type: Journal Article
• Authors: Duan, Jiandong; Wang, Peng; Ma, Wentao; Tian, Xuan; Fang, Shuai; Cheng, Yulin; Chang, Ying; Liu, Haofan
• Is Part Of: Energy. Volume 214(2021)
• Publication Date: 2021-01-01
• Subjects: Short-term wind power forecasting -- Long short-term memory neural network -- Maximum Correntropy criterion -- Variational mode decomposition -- Sample entropy; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.118980
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Nowadays, various wind power forecasting methods have been developed to improve wind power utilization. Most of these techniques are designed based on the mean square error (MSE) loss, which are very suitable for the assumption that the error distribution obeys the Gaussian distribution. However, there are many outliers in real wind power data due to many uncertain factors such as weather, temperature, and other random factors. Meanwhile, the highly nonlinear process of converting wind energy into wind power may changes the statistical characteristics of errors. Therefore, the prediction model established based on the traditional MSE loss may lead to unsatisfactory results. As a result, a robust short-term wind power hybrid forecasting model based on Long Short-term Memory (LSTM) neural network with Correntropy combining an improved variational mode decomposition (IVMD) and Sample Entropy (SE) is proposed in this work. The IVMD in which the parameter K in the IVMD is determined by the Maximum Correntropy Criterion (MCC) is used to decompose the original wind power data and the decomposed subseries is reconstructed by SE to improve the prediction efficiency. Then the MCC is also utilized to replace the MSE in the classic LSTM network to develop a novel robust hybrid model to forecast the wind power. Finally, four experiments were conducted using real data from two wind farms in China at different sampling intervals to evaluate the effectiveness of the proposed method, and the results show that proposed method is more effective than other traditional methods. Highlights: An adaptive VMD decomposition algorithm with MCC is proposed. A new data preprocessing model is proposed by combining with the sample entropy. The LSTM network with MCC loss is proposed in the forecasting process. A novel IVMD-SE-MCC-LSTM model is proposed for short-term wind power forecasting. Experiments are performed to evaluate the efficacy of the proposed method.