Abstract: The inevitable randomness, intermittence and uncertainty of wind power bring great challenges to grid connection, power dispatching and consumption. Effective assessment of wind power fluctuations through wind power forecasting is of great significance to reduce the risk of wind power uncertainty and promote the steady development of wind power. Aiming at the crucial problem of short-term wind power prediction accuracy in the current development of large-scale wind power, the sources and influences of short-term wind power prediction error are introduced, the principles, advantages and disadvantages, accuracy evaluation criteria and applicability of deterministic and probabilistic short-term wind power forecasting methods are elaborated. The latest research progress of key technologies to improve the accuracy of wind power prediction from the aspects of abnormal data detection and cleaning, missing data reconstruction, data feature selection or extraction, data clustering, data decomposition, optimization algorithm improvement and consideration of physical model are investigated and summarized, and finally the prospect of future development trend of wind power prediction technology is concluded, providing a reference for improving the short-term prediction accuracy of wind power, promoting the development of refined prediction technology, and ensuring the safe and stable operation of the system.

Key words: wind power forecasting, prediction accuracy, deep learning, interval prediction, combined forecasting model