A hybrid GAN based autoencoder approach with attention mechanism for wind speed prediction

Accurate forecasting of wind speed is essential for the effective utilization of wind power. For forecasting algorithms to produce accurate results, high-dimensional input is necessary. The method of obtaining wind speed data, however, runs into a number of issues since data measurement equipment fails. Accurate imputation and effective feature extraction are required for precise wind speed forecasting (WSF). Thus, this paper proposed a hybrid wind speed prediction model consisting of a generative adversarial network (GAN), and an attention mechanism-based convolutional long short-term memory network autoencoder (AM-CLSTMAE). The GAN is used for the effective missing data imputation (MDI) of wind speed values based on the data distribution, and AM-CLSTMAE extracts the spatio-temporal characteristics to accurately predict the wind speed. The proposed model is evaluated using two test cases comprehensively for the MDI, and WSF. The 5-minute wind speed data for the two test cases is collected from the wind farms located in Leicester, and Portland. Different comparison models are used to evaluate the proposed model using various evaluation indices. The two test cases indicated that the proposed model achieved an improvement of 60%, and 63% in the MDI, 42%, and 40% in the WSF for two test cases, respectively.