抄録
This paper deals with the nonlinear statistical modeling of steady state river quality. The methodology used for modeling is a revised GMDH (Group Method of Data Handling) of generating optimal intermediate polynomials. By using the measured data of river quality such as BOD and DO concentrations in the Bormida river, Italy, we intend to construct two kinds of nonlinear steady state models, and they are compared with the Rinaldi's linear physical model. In the steady state model I, we intend to discover a suitable structure of the Bormida river by using no a priori information of the system structure. It is shown that the revised GMDH model gives much better prediction for DO concentration compared with the physical model. In the steady state model II, we intend to approximate the river quality in the Bormida river as a polynomial of the input variables. It is shown that it is difficult to approximate the DO part of the model as a polynomial of the input variables, because the system structure for the DO concentration is very complex.
