Ultra-short-term forecasting of wind power based on multi-task learning and LSTM. (July 2023)

• Record Type: Journal Article
• Authors: Wei, Junqiang; Wu, Xuejie; Yang, Tianming; Jiao, Runhai
• Is Part Of: International journal of electrical power & energy systems. Volume 149(2023)
• Publication Date: 2023-07
• Subjects: Maximal information coefficient (MIC) -- Multi-task learning (MTL) -- Long short-term memory (LSTM) neural network -- Wind power prediction; Electrical engineering -- Periodicals; Electric power systems -- Periodicals; Électrotechnique -- Périodiques; Réseaux électriques (Énergie) -- Périodiques; Electric power systems; Electrical engineering; Periodicals; 621.3
• DOI: 10.1016/j.ijepes.2023.109073
• Languages: English
• ISSNs: 0142-0615

Abstract

Highlights: This paper first conducts MIC correlation analysis on wind power series and wind speed series, and then conducts hysteresis correlation analysis on wind power series and wind speed series based on MIC. Based on MIC correlation and considering the parameter characteristics of LSTM, this paper constructed the input sequence of neural network. This paper takes wind speed prediction as an auxiliary task and wind power prediction as the main task, adopts multi-task learning framework to build long short-term memory (LSTM) network, and adopts grid search to optimize network parameters. Abstract: In order to achieve high precision ultra-short-term prediction of wind power, a new ultra-short-term prediction method for wind power is proposed by combining the maximal information coefficient (MIC) with multi-task learning (MTL) and long short-term memory (LSTM) network. First, the correlation analysis method is used to analyze the MIC correlation of wind power sequence and wind speed sequence, the MIC correlation between the alternative sequence, the wind power sequence and the wind speed sequence, respectively. The feature input sequence of the neural network is constructed base on the correlation analysis results. Second, taking wind speed prediction as the auxiliary task and wind power prediction as the main task, LSTM based prediction network was constructed using MTL framework, and the network parameters were optimized by grid search. Finally, based on the historical data of a wind farm in the United States, the case study verifies that the proposed method gains higher prediction accuracy than other existing methods modeling wind speed as a feature, such as single-task LSTM neural network, BP neural network and traditional ARIMA model.