基于时间序列分解的海面变化预测

Sea Level Prediction Based on Time Series Model

对吴淞站1955-2001年月平均潮位序列采用时间序列分解方法进行分析,采用奇异谱分析方法提取长期趋势,采用比率平均法和剩余法分别提取季节因素和循环变化,应用乘法模型拟合试验的效果良好.以1955-1996年数据为基础,建立长期趋势的自回归(AR)模型,对1997-2001年间进行了预测试验,并以该时段实测数据作为验证.试验结果表明月均序列预测值的最大相对误差不超过±10%,年均序列预测值的相对误差最大不超过±4%.这一系列方法的综合运用可适用于较长期的海面变化预测.

Analyzing the monthly average tidal records of Wusong tidal gauge station from 1955 to 2001, SSA （singular spectrum analysis） ,ratio average method and residual method are respectively used to extract the components of trends, cycles, and seasons. Using multiplication model to fit tidal records shows that the fitting series can be perfectly compared with origin series. Using AR model to predict the trends of tidal records of 1955 - 1996, then using multiplication model to predict the sea-level changes of 1997 -2001, and comparing prediction results with the original data, we find that those two series are fairly comparable. The maximum error of prediction of monthly average tidal records and yearly average tidal records do not exceed ＋ 10% and ＋4%. It shows that time-series model provide an efficient way to predict sealevel changes.