Ultra-short-term Wind Power Prediction Research Based on WOA-BiLSTM-AM Model

Accurate and reliable probability prediction is a necessary tool to quantify the uncertainty of wind power output, and it is also an important technology to promote the development of new power systems. In this paper, a BiLSTM ultra-short-term wind power prediction method based on the Attention mechanism is proposed. This method takes the historical load data as input to model the internal dynamic change law of the learning feature, and introduces the Attention mechanism to give different weights to the implied state of BiLSTM by mapping weighting and learning parameter matrix, so as to reduce the loss of historical information and strengthen the influence of important information. At the same time, aiming at the difficulty of selecting the hyperparameters of the model, the WOA algorithm is proposed to realize the optimization selection of the hyperparameters of the model. The resulting RMSE is 160.6926 and R2 is 0.9771. By comparing the results with four typical predictive models, such as BP, LSTM, Bi-LSTM, and BiLSTM-Attention. It verifies that the Attention-BiLSTM-LSTM neural network method is more accurate and effective, and the WOA-BiLSTM-AM is a good method for ultra-short-term wind power prediction.