Short-term wind speed forecasting based on Gaussian process regression and particle filter

Improving the wind speed prediction accuracy of wind farm will help enhance the power grid stability and economy. Noise or data loss often appears in historical wind speed sequences. These abnormal values will lead to inaccurate estimation of model parameters. Therefore, the affecting the prediction accuracy. Therefore, the detection and correction of abnormal value is the prerequisite and necessary measure to effectively analyze the law of wind speed. In this paper, a short-term wind speed forecasting model combining Gaussian process regression and particle filter is established, which realizes online dynamic detection and correction of outliers. Firstly, the state space equation is established by the Gaussian process regression in the training sample set. The particle filter algorithm is then used to estimate the current measurement value. The residuals between the estimated and measured values are analyzed and the anomalous values are detected according to the principle of 3σ. Secondly, after the anomaly being corrected, the Gaussian process regression forecasting model is reconstructed. The particle filter algorithm is repeatedly used to estimate the latest measurement value during the process of 15 mins ahead wind speed forecasting, realizing the online dynamic detection and correction of outliers. The case study show that the particle filter algorithm can effectively detect the abnormal values and reduce the wind speed prediction error, the average absolute percentage error and root mean square error are reduced to 8.92% and 0.5826 m/s respectively when the wind speed is predicted 15 minutes ahead. © 2020, Editorial Board of Acta Energiae Solaris Sinica. All right reserved.