Elman neural network considering dynamic time delay estimation for short-term forecasting of offshore wind power

Accurately forecasting the output power of offshore wind turbines is a key way to improve power quality and ensure stable operation of the power grid. The existing works focus on utilizing historical data or designing effective models while ignoring the fundamental issue of whether there is a time delay effect between each monitoring variable and wind power output. Therefore, a short-term offshore wind power prediction method considering dynamic delay effects is proposed to intuitively capture power prediction information. Firstly, based on the nonlinear coupling relationship, dynamic sliding windows matching different average mean periods are introduced. Then, the dynamic delay time is calculated based on coupled Granger causality analysis, and the multiple delay relationships between the variables are defined. Finally, the Elman network is used to achieve short-term offshore wind power forecasting. The feasibility and compatibility of the proposed method are verified by the actual operation data of offshore wind turbines for 10 consecutive days. The results show that the dynamic sliding window technology can accurately extract the dynamic time delay relationship between the process monitoring variables. The proposed monitoring strategy has the best accuracy on all mainstream metrics compared to other methods. The average MAE of the 10-day wind power prediction results reached 0.0025, while the average operating time was 4.0869 s. The proposed method has good stability and potential for application in the field of accurate forecasting of offshore wind turbine output power.