Wind Power Prediction Based on Ransac Data Screening and Seq2seq-Bigru Model

Abstract

Accurate wind power prediction is the primary way to overcome the variability and fluctuations of wind power generation and alleviate the power grid's peak regulation and voltage regulation. In order to improve the wind power prediction accuracy, a wind power prediction method based on RANSAC noise screening and the Seq2Seq-Attention-BiGRU model is proposed in this paper. By introducing the noise screening technology of RANSAC anti-noise regression, an effective and stable noise screening is realized in the large sample noise datasets by carrying out robust regression, which follows the relationship between wind speed and power. Then, the Seq2Seq-Attention short-term wind power prediction model is established and verified by comparison with the measured data of wind farms in different seasons. Besides, the BiGRU error correction model is established to further improve wind power prediction accuracy. The comparison results of existing prediction methods and correction methods show that the proposed prediction method is effective in improving prediction accuracy. Comparison with the existing forecasting method (LSTM) showed that the RANSAC-Seq2Seq-BiGRU forecasting method obtained a decrease in the RMSE by 26.0%, 43.6%, 46.8%, and 48.4% in January, March, June, and September, respectively, and the MSE decreased by 19.7%, 43.4%, 41.0%, and 46.2%, respectively.

Keywords: Wind power prediction、RANSAC、Seq2Seq、Attention、BiGRU