Research on an Ultra-Short-Term Wind Speed Prediction Method Based on Improved Combined Neural Networks

Abstract

Ultra-short-term wind speed prediction is the key to ensure the implementation effect of wind turbine pitch angle feedforward control, and has an important impact on improving the environmental adaptability of wind turbines. In order to improve the prediction accuracy, an ultra-short-term wind speed prediction method based on an improved combined neural networks was proposed. In this method, BP neural network and long short-term memory (LSTM) neural network, which are suitable for time series prediction and have strong nonlinear learning ability, are selected for weighted combination to eliminate the large error that may exist in a single neural network. At the same time, to improve the combination effect, the differential evolution (DE) algorithm was used to optimize the combination weight. The method was applied to the ultra-short-term wind speed prediction of a wind farm. Compared with the results of single neural network prediction and equal weight combined neural networks prediction, the effectiveness of the proposed method in improving the prediction accuracy was verified.

Keywords

• wind power
• ultra-short-term wind speed prediction
• bp neural network
• long short-term memory (lstm) neural network
• differential evolution (de) algorithm