The authors first describe a class of discrete linear time series models capable of representing nonstationary as well as stationary behavior. In control problems these models are used to describe disturbances to the system. Dynamic models which represent relationships between variables which control and are controlled are later introduced. The identification, fitting, checking and practical use of such models in forecasting and control are discussed. The models employed are empirico-mechanistic in that while they can be interpreted as descriptions of physical phenomena having the right general character they do not claim to represent exact physical reality and are fitted to data empirically. An important principle in the choice of such models is that, they should, while adequately representing the data, contain the fewest possible number of parameters. This is called the principle of parsimony or of parsimonious parametrization. Author
Prepared in cooperation with Lancaster Univ. (England). Dept. of Systems Engineers.