An ultra‐short‐term wind speed prediction model using LSTM based on modified tuna swarm optimization and successive variational mode decomposition

• Published in: Energy science & engineering 2022-08, Vol.10 (8), p.3001-3022
• Main Authors: Tuerxun, Wumaier, Xu, Chang, Guo, Hongyu, Guo, Lei, Zeng, Namei, Cheng, Zhiming
• Format: Article
• Language: English
• Subjects: Algorithms; Alternative energy sources; Artificial intelligence; Belief networks; Data acquisition; Decomposition; Kalman filters; long short‐term memory; Machine learning; Methods; modified tuna swarm optimization algorithm; Neural networks; Optimization; parameter optimization; Particle swarm optimization; Power control; Prediction models; Renewable resources; successive variational mode decomposition; Supervisory control and data acquisition; Support vector machines; Turbines; Wind farms; Wind power; Wind speed; wind speed prediction; wind turbine; Wind turbines
• ISSN: 2050-0505; 2050-0505
• DOI: 10.1002/ese3.1183

Accurate ultra‐short‐term wind speed prediction is extremely important for the power control of wind farms, the safe dispatch of power systems, and the stable operation of power grids. At present, most wind farms mainly rely on supervisory control and data acquisition systems to obtain operation and maintenance data which includes operating characteristics of wind turbines. In the ultra‐short‐term wind speed prediction, a long short‐term memory network is one of the commonly used deep learning methods. To address the problem that improper selection of long short‐term memory network's hyperparameters may affect the prediction results, In the present study, a hybrid prediction model based on the long short‐term memory and the modified tuna swarm optimization algorithm was established, and was used to predict after the wind speed sample data had been decomposed by successive variational mode decomposition method. The experimental results reveal that the proposed model effectively improved the accuracy of wind speed prediction for wind farms compared with the support vector regression, deep belief networks, and long short‐term memory models optimized by particle swarm optimization algorithm. The successive variational mode decomposition method is used to decompose preprocessed wind speed SCADA data. The original tuna swarm optimization (TSO) algorithm was improved and tested. TSO algorithm was used to optimize the parameters of the long short‐term memory.