1.School of Electrical Engineering, Shanghai University of Electric Power, Shanghai 200090, China;2.School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney 2052, Australia

Offshore wind power is faced with the deep coupling effect of complex and changeable meteorological and sea conditions, which leads to the accuracy of output forecasting to be improved. At the same time, the “black box” structure of the forecasting model leads to the low reliability of output forecasting results in the engineering applications. To address the above problems, an ultra-short-term offshore wind power output forecasting model is proposed based on the dual-stage attention long short-term memory (DALSTM) network. Based on the long short-term memory neural network, the dual-stage attention mechanism of feature space and time series is introduced to dynamically explore the potential correlations between the offshore wind power output and the input features. The importance evaluation is carried out from feature and time to improve the interpretability of the proposed DALSTM model. Finally, simulations are conducted on the supervisory control and data acquisition data collected from the Donghai Bridge offshore wind farm of China. The results show that the proposed DALSTM model can effectively forecast the ultra-short-term offshore wind power output, and has higher forecasting accuracy and stability than the traditional forecasting models, as well as reasonable interpretability.