Abstract
The study to predict effluent quality indicated by TRS (transparency), SS, TOD and TOC from sewage treatment plant carried out using the improved Group Method of Data Handling (GMDH) reported by IHARA. Generally, absolute mean value of error was used as criterion function for selection in the GMDH.
In this paper, relative error is used as a criterion function for selection, because the data of effluent quality from sewage treatment plant does not show normal distribution in some cases, and also, correlation coefficient is used as the function in order to take similality of the variation patterns of the predicted values and of the measured values more important than closeness of the two.
Three kinds of GMDH-programings are studied by comparison. The results were evaluated by prediction errors and comparison of the predicted values with the measured values in graphs, and the following were cleared. 1) The difference in the prediction errors of the three kinds of GMDH-programings was a little. 2) The prominant difference by the methods appeared in the case of TRS. The R. filter GMDH (correlation coefficient was used as a criterion function for selection.) was the best among the three in fitting with the predicted values for large variation of the measured values.
