Deep learning-based multistep ahead wind speed and power generation forecasting using direct method

Abstract

Long-term effective and accurate wind power potential prediction, especially for wind farms, facilitates planning for the sustainable development of renewable energy. Accurate wind speed forecasting enhances wind power generation planning and reduces costs. Wind speed time series has nonlinearity, intermittence, and fluctuation, which makes the prediction difficult. Deep learning techniques can be beneficial when there is no specific structure to data. These techniques can predict wind speed with reasonable accuracy and reliability. In this study, four different algorithms, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and CNN-LSTM, for three different long-term horizons (6 months, 1 year, and 5 years) are successfully developed using the direct method. GRU method showed a higher degree of accuracy compared to other methods. In addition, it is confirmed that using a multivariate data set increases the model's accuracy compared to the univariate model. A computational cost analysis is also conducted to compare the proposed algorithms. Finally, the power production capacity of the wind farm at a given location, Zabol city, is calculated for the next five years, which is indispensable for planning, management, and economic analysis. The reasonable conformance between the real data and predicted ones is shown to confirm the capability of the proposed model to use in long-term wind speed forecasting.