中文摘要：

为更好地实现天极偏差序列的预报，采用了最小二乘（least—squares，LS）拟合外推与自回归模型（autoregressive，AR）联合预报方法（LS＋AR），并对预报精度进行评估。结果表明，LS＋AR模型对CPO序列1—5d的短期预报有较高的精度，而对5-90d的预报，dr方向大约为110μas，dy方向大约为130μas。而国际上常用的Lambert模型的预报精度比较平稳，其中出方向在120μas左右，曲方向在220μas左右。结果表明，LS＋AR模型的预报精度总体上优子Lambert模型。

英文摘要：

A joint model of least squares （LS） and autoregression （AN） is proposect to prealct （forecast） the celestial pole offsets （CPO）. The forecasting accuracy is evaluated by using observations which shows that the CPO prediction accuracy of LS＋AR model is better for 1~,5 d than 5~90 d. For 5~90 d prediction, the error is about 110 μas in dx direction and 130 gas in dy direction respectively. In contrast, the accuracy of Lambert model for 1~90 d prediction is about 120 μas in dx direction and 220 μas in dy direction. It shows that the joint model can get more accurate results than Lambert model.