Dual-meta pool method for wind farm power forecasting with small sample data

The power prediction is important for the safe and efficient operation of new wind farms with small data. However, there is a lack of research on improving the prediction accuracy. Here we propose a novel dual -meta pool model to realize multi-step prediction based on learning the data knowledge contained in relevant wind farms. By analysing the essence of neural network, we defined meta-data to express and extract the data knowledge. For meta-data classification, we designed a new unsupervised classification method based on the accuracy of data tail part. A convolutional neural network is used to learn the mapping relationship between meta-data and their labels. To make multi-step prediction, we proposed a novel meta-method which contains Hilbert spatial transformation, data extension and neural network. To verify the performance, eight comparison models are used to predict the last 30 data of two wind farms which only contain 96 data. The results show that the proposed dual-meta pool method has smaller prediction errors than other models, which average mean absolute errors are 13.33 and 2.42, respectively.

Automated summary

This article proposes a novel dual-meta pool model to improve the prediction accuracy of power in new wind farms with small data. The model learns the data knowledge contained in relevant wind farms to achieve multi-step prediction.