Multi-step interval prediction of ultra-short-term wind power based on CEEMDAN-FIG and CNN-BiLSTM

Aiming at the uncertainty of wind power and the low accuracy of multi-step interval prediction, an ultra-short-term wind power multi-step interval prediction method based on complete ensemble empirical mode decomposition with adaptive noise-fuzzy information granulation (CEEMDAN-FIG) and convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) is proposed. Firstly, the CEEMDAN is used to decompose the wind power time series into several sub-components to reduce the non-stationary characteristics of the wind power time series. Then, different components are selected for FIG, and the maximum value sequence, average value sequence, minimum value sequence gotten from FIG, and the remaining components without FIG are combined with the wind speed data, wind direction data, and the temperature data. They all are input into the CNN-BiLSTM combined prediction model to obtain the initial wind power prediction interval. The prediction results of the maximum value sequence, the average value sequence, and the minimum value sequence are respectively superimposed on the prediction results of the remaining components to obtain the upper limit, point prediction, and lower limit of the initial prediction interval. Finally, the improved coverage width criterion is used as the objective function to optimize the interval, and the forecast interval of wind power under a given confidence level is generated. Taking the actual operating data of a certain unit of a wind farm as an example, the validity of the proposed model is verified.