To address the uncertainty caused by integrating wind power into the electricity grid, accurate wind speed forecasting is highly desired. However, historical wind speed data of new wind farms may be insufficient for training a well-performed forecasting model. To address this issue, short-term wind speed forecasting with convolutional neural network (CNN) based on information of neighboring wind farms is studied in this paper. In the proposed approach, the CNN is employed to migrate the intrinsic features of wind speed changes to newly built wind farms. To evaluate the performance of the proposed approach, wind speed data collected from three wind farms in China is utilized and multi-step-ahead forecasting is considered. The computational results prove the proposed approach outperforms benchmarking methods Support Vector Regression, Kernel Ridge Regression, and CNN by only considering data of the target wind farm. INDEX TERMS Wind speed forecasting, transfer learning, neural networks, wind energy.

Abstract-Wind speed forecasting is an important work in the design of wind farm planning. Because of the wind speed itself has a timing sequence and autocorrelation, the paper propose wind speed forecast model for wind farms based on time series analysis. In order to test the validity of time series analysis model, using ARMA (P, q) function. In the example, compared the distribution characteristics of wind speed prediction with the actual characteristics of wind speed distribution, verify the time series model proposed in this paper use for the feasibility of wind speed forecast.