كاربرد رگرسيون فازي در استنباط قوانين بهره برداري از مخازن سدها

One important issue in application of implicit stochastic or long-term planning models to reservoir operation problem is how the set of optimal solutions should be aggregated to derive general operating rules. In this paper, a dynamic programming (DP) model, whose optimal solution is used as input to models inferring general operating rules, has been developed for optimizing a long-term reservoir operation problem. Subsequently, traditional least squares regression (OLSR), fuzzy regression (FR), and statistical analysis of optimal releases (AS) have been evaluated and tested in the problem of deriving reservoir operating rules. The problem of finding the coefficients of an FR model yields in solving a linear programming (LP) optimization model. The objective function of the LP is to minimize total fuzziness of the FR model. This objective is fulfilled through minimizing the support of fuzzy coefficients of the regression model subject to having observed values of the dependent variable confined within those values predicted by FR model. The OLSR, FR, and AS based rules are then simulated and compared based on their performance when they are applied to Dez reservoir system as the case study. The results indicate that although OLSR performs better than FR in terms of fitting capability, the rules derived by FR performs better in simulation, especially in the situations where imperfect and partially informative data are addressed. This implies that in facing with a data set which does not include some behavioral aspects of a given phenomena to be modeled, trying to find a regression equation which best fits the data available will not necessarily be the best approach. Instead, it should search for an equation which performs more robust, where situations not represented in data set are likely to occur in real world.