Abstract：Wind power prediction is an important basis for monitoring and information management of wind farms. It requires high prediction accuracy when using ultra-short-term wind power prediction to balance load and optimize spinning reserve. Due to the complex environment of the wind farm and many uncertainties of wind speed, the wind power timing signals are non-stationary and random. Recurrent Neural Network (RNN) is suitable for time series tasks, but the non-periodic and non-stationary timing signal will increase the difficulty of network learning. To overcome the interference of non-stationary signal in the prediction task and improve the prediction accuracy of wind power, an ultra-short-term wind power prediction method combining signal decomposition and multi-branch neural network was proposed. Firstly, the original wind power timing sequence signal was decomposed by Empirical Mode Decomposition (EMD) to reconstruct the data tensor, then the convolution layer and Gated Recurrent Unit (GRU) layer were used to extract the local features and trend features, and finally the prediction results were obtained through feature fusion and full connection layer. The experimental results of a wind field data set in Inner Mongolia show that compared with the Autoregressive Integrated Moving Average model (ARIMA), the proposed method improves the prediction accuracy by nearly 30 percentage points, which verifies the effectiveness of the proposed method.
