Ultra-short-term wind power forecasting based on data-driven approach is one of the key foundations when large scale wind power integrated into the power grid. According to the forecasting process, the basic thoughts and limitations of existing methods are analyzed from the point of view of data mining approach, machine learning algorithms and wind power curve. Furthermore, the new prediction idea of offline data-driven/deep learning algorithms and online application is concluded, the evaluation methods of information screening are given, the latest research progress of deep learning algorithms in data-driven forecasting is summarized. Finally, the current position of ultra-short term wind power forecasting is summarized, that is transition from model driven to data-driven and transfer from machine learning to deep learning, and it is pointed out that the alternation of algorithms and data fusion will be the research trends in the future. This work is supported by National Key Research and Development Program of China (No. 2018YFB0904200).