Model Structure Identification from Experimental Data

Methods for identifying the structure of dynamic mathematical models for water quality by reference to experimental field data are discussed . The context of the problem of model structure identification is described by briefly reviewing the steps involved in the overall process of system identification. These steps include experimental design; choice of model type; model structure identification; parameter estimation; and verification/validation. Two examples of approaches to solving the problem of model structure identification are presented. The first example is concerned with identifying the structure of a black box (input/output) model for the variations of gas production in the anaerobic digestion process of wastewater treatment. Correlation analysis is used as the principal method of solution, although it is found to have significant limitations for certain kinds of data. The second example addresses the more difficult problem of identifying the structure of an internally descriptive ("mechanistic") model form. The application of an extended Kalman-filtering algorithm to this problem is discussed in detail. The approach is illustrated with a model for phytoplanktonbiochemical-oxygen-demand (BOD) interaction in a freshwater river system. Chapter 11 MODEL STRUCTURE IDENTIFICATION FROM EXPERIMENTAL DATA