Abstract
In this paper, a revised GMDH having a unified heuristic criteria and a new construction of a prediction model of water demand are developed. Water demand varies with social and economic activity and with rhythm of living. So the system which consumes water may be complex with many variables.
The GMDH algorithm is one of the strong methods to deal with modeling of such complex problems. However it is difficult to obtain such a model that has moderate complexity and non-overfitting by the basic GMDH. Accordingly, we present the two studies below. The one is the revised GMDH algorithm which automatically selects optimal polynomials, intermediate variables in each layer and decides to stop the calculation by Akaike's Information Criterion (AIC) evaluated by using all the parameters in the structure of the final expression. The other is the relative type prediction model of water demand which is not affected by increase or decrease of population and economic activity during long term.
The revised GMDH and the relative type model are applied to identify the daily water demand in one year and made sure the practical use by results of prediction for six successive months.
