A robust maximum correntropy forecasting model for time series with outliers

It is of great significance to develop a robust forecasting method for time series. The reliability and accuracy of the traditional model are reduced because the series is polluted by outliers. The present study proposes a robust maximum correntropy autoregressive (MCAR) forecasting model by examining the case of actual power series of Hanzhong City, Shaanxi province, China. In order to reduce the interference of the outlier, the local similarity between data is measured by the Gaussian kernel width of correlation entropy, and the semi-definite relaxation method is used to solve the parameters in MCAR model. The results show that the MCAR model in comparison with deep learning methods, in terms of the average value of the mean absolute percentage error (MAPE), performed better by 1.63%. It was found that maximum correntropy is helpful for reducing the interference of outliers.