Ultra-short-term wind speed prediction based on TCN-MCM-EKF

Abstract

Ultra-short-term wind speed prediction plays a key role in renewable energy management in ensuring the efficient operation of wind power systems. In recent years, temporal convolutional network algorithms have attracted much attention in this field. However, their prediction effectiveness is challenged when facing nonlinear, nonsmooth, and Gaussian noise problems. In this research, an ultra-short-term wind speed prediction method based on a temporal convolutional network, a Monte Carlo method, and an extended Kalman filter is proposed. The combined model fully utilizes the superior time-dependent capture capability of the temporal convolutional network, takes into account the uncertainty of wind speed data by simulating random particles through the Monte Carlo method, and improves the prediction accuracy by updating the temporal convolutional network model parameters based on the observed wind speed measurements through the extended Kalman filter. By training and predicting this data from Los Angeles wind farms in the United States, this research found that the combined temporal convolutional network-Monte Carlo method-extended Kalman filter model significantly outperforms other wind speed prediction methods in terms of accuracy. Applying the model to predict data from wind farms in New Jersey and Wyoming, USA, similarly yielded excellent results. This research provides technical support to improve the operational efficiency of wind power systems, optimize energy utilization, and promote the integration of renewable energy into the grid, emphasizing the overall goal of the research, which is to contribute to the sustainable development of renewable energy through innovative approaches.