Imputation of missing wind speed data based on low-rank matrix approximation

Abstract

Wind speed data usually contain missing values due to data observing device failure, incorrect data recording, and so on. Using an effective method to impute the missing and error wind speed data before predicting the future wind speed is necessary. The traditional methods are based on simple statistical models or complex machine learning models. In this paper, the method based on low-rank matrix approximation is introduced for imputing missing wind speed data. Firstly, wind speed data are converted into a low-rank matrix form. Then a low-rank matrix approximation named Grouse algorithm is used on the wind speed matrix which contains missing values. Finally, the missing values in wind speed data are imputed by the approximate matrix. This study is carried on the real wind speed data obtained in Saihanba. Compared with the three traditional methods: mean, multiple imputation, and k nearest neighbors, experiment results show that our method imputes more accurately than other methods in dealing with missing and incorrect wind speed data.