When the time series algorithm is used to establish a thermal error compensation model for a Computer Numerical Controlled (CNC) Machine, it shows a shortcoming of forecasting robustness caused by the severe multiple collinearity. This paper proposes a method for improving the forecasting robustness of the time series algorithm. This algorithm combines the time series algorithm with the modeling algorithms which are able to suppress multiple collinearity. Thus, it not only provides more comprehensive temperature information by adding the temperature lag values in the thermal error model, but also deals with the severe multiple collinearity brought by the added temperature lag values. The Distribution Lag (DL) algorithm that belongs to time series algorithms and Principal Component Regression (PCR) algorithm that can suppress the multiple collinearity are selected as the examples, and a modeling method for establishing the thermal error compensation model of the machine tool is proposed by the Principal Component Distribution Lag (PCDL) algorithm. The forecasting accuracy and robustness of PCDL algorithm are compared with that of DL algorithm. The results show that the PCDL algorithm suppress the impact of multiple collinearity, so, its model′s forecasting accuracy and robustness are far better than that of DL model, and the forecasting accuracy is improved about 9 μm. The proposed method provides a good reference for the application of time series data modeling in different fields.