Short-term wind speed prediction based on time series trends and periodic characteristics

Abstract In order to improve the accuracy of short-term wind speed prediction and ensure the stable operation of wind farms, a hybrid model combining contrast learning, convolutional neural network, and Transformer network is proposed to decouple and predict the period and trend characteristics of wind speed series to improve the accuracy of wind speed prediction. Based on the mutual information theory of contrast learning, the model uses the contrast loss function to train the adaptive convolutional module to extract the general information between the data of different periods and obtain the periodic features in the data. The trend characteristics of data are extracted through the Transformer network, and the two types of characteristics are combined for prediction. The model is verified by using an open data set, and the results show that the prediction error is effectively reduced by using this model to predict short-term wind speed.