中文摘要：

通过对近52年的极移数据进行二阶差分,发现极移不规则变化的活跃程度存在明显的分段特征。其中活跃部分大约持续6 100d,与地震的近6 500d的长活动周期以及章动的18.6a周期较接近;然后以ARMA模型为例,选取状况不同的4个不规则变化时段的极移数据进行模型的拟合与预报,并进行对比。结果表明：1若拟合数据处于活跃期,建模及预报的精度相对于处于平静期时较差;2拟合数据同时包含活跃期与平静期,拟合的残差出现病态分布,模型建立不成功;3拟合精度相当的不同模型,预测数据处于平静期要比处于活跃期时精度要好。

英文摘要：

Firstly,through second order difference of polar motion data that has lasted for about 52 years we find out that the active degree of irregular variation of polar motion exists obvious sectional characteristics,in which the active part lasts about 6 100 days,close to the nearly 6 500 days of the long cycle of activity of earthquake and the 18.6year cycle of the change of day.Then the ARMA model is taken as an example and four sessions of polar motion data that has different irregular variation for model fitting and predicting are chosen,the results of which are compared.The conclusions are as follows：1The precision of model fitting and predicting in active part are relatively lower when compared to those in quiet period;2The distribution of model fitting residuals is pathological and the model fitting failed when the data is used from both active part or quiet part;3Even if two models have the same precision of fitting,the precision of predicting can be different,the precision is higher when the data is predicted in the quiet part.