观测分析中的回归-时序列模型

Regression time-series model for analyzing observational data

大坝观测数据经常规回归分析后的残差序列一般并非为白噪声 .考虑将回归拟合与随机型时间序列方法结合 ,先对大坝位移数据按水位、温度、时效等物理因素作回归分析 ,再对回归残差作时序列建模处理 .实例采用Box- Jenkins法和由自相关、偏自相关函数及 AIC准则进行模型识别 ,建立时序列模型 .应用示例的计算表明 ,这样获得的回归 -时序列模型能很好拟合实测数据 ,提高精度 ,误差序列也符合白噪声要求 .

Analysis of observational data on a dam is very important for monitoring the dam safety. The regression method is usually used. However, the residual series obtained through the model often does not satisfactorily dealt with the problem of white noise. This paper presents a wethod for analysing observational date by using a random time series model in combination with regression analysis. First, the observational data were used to form a regression equation using the water level, temperature and time effect as regression factors; then the time series model was identified by the Box Jenkins method. This modeling and method were successfully applied to practical displacement data on an arch dam in Zhejiang Province. Results showed that the fitting curve coincided well with the measured curve; that the accuracy of forecast was improved; and that the final error series by passed the problem of white noise. It means that the regression time series modeling and method are feasible for analysis of observational data.