A stepwise time series regression procedure for water demand model identification
- Univ. of Tennessee, Knoxville (USA)
Annual time series water demand has traditionally been studied through multiple linear regression analysis. Four associated model specification problems have long been recognized: (1) the length of the available time series data is relatively short, (2) a large set of candidate explanatory or input variables needs to be considered, (3) input variables can be higholy correlated with each other (multicollinearity problem), and (4) model error series are often highly autocorrelated or even nonstationary. A stepwise time series regression identification procedure is proposed to alleviate these problems. The proposed procedure adopts the sequential input variable selection concept of stepwise regression and the three-step time series model building strategy of Bos and Jenkins. Autocorrelated model error is assumed to follow an autoregressive integrated moving average (ARIMA) process. The stepwise selection procedure begins with a univariate time series demand model with no input variables. Subsequently, input variables are selected and inserted into the equation one at a time until the last entered variable is found to be statistically insignficant. The order of insertion is determined by a statistical measure called between-variable partial correlation. This correlation measure is free from the contamination of serial autocorrelation. Three data sets from previous studies are employed to illustrate the proposed procedure. The results are then comparewd with those from their orginal studies.
- OSTI ID:
- 5339258
- Journal Information:
- Water Resources Research; (United States), Vol. 26:9; ISSN 0043-1397
- Country of Publication:
- United States
- Language:
- English
Similar Records
Short-term forecasts gain in accuracy. [Regression technique using ''Box-Jenkins'' analysis]
Physics-Informed Gaussian Process Regression for States Estimation and Forecasting in Power Grids