Real-time multipurpose reservoir system management using probabilistic balancing rule models

Emmanuel Uzochukwu Nzewi, Purdue University

Abstract

Real-time, daily, reservoir operations optimization of a multipurpose reservoir system is addressed. Hydropower production as well as other uses of the reservoir system are considered in the hydrosystems studied. A comparison of the performance of a proposed set of reservoir operating rules called the Probabilistic Balancing Rule model (Nzewi and Houch, 1987a, 1987b) and a traditional penalty-based model called the Penalty Scheme model (Sigvaldason, 1976) is presented. The Balancing Rule model defines the optimal set of decisions as those which minimize the maximum non-exceedance probability of the levels of the decisions within a specified forecast horizon. In contrast, the Penalty Scheme model defines the optimal decisions as the set of operations which minimize the assessed penalties, for non-ideal operations, within the forecast horizon. Forecasts are assumed to be perfectly made in both cases. The Balancing Rule model was shown to be superior to the Penalty Scheme model after the analysis of the results of a carefully designed numerical experiment. 45 cases of daily operations were simulated. In each case, the operations over a period of about 550 days were simulated and compared. The results of the numerical experiment also revealed a counter-intuitive conclusion about the performance of penalty-based models dispelling formerly held notions about their performance as more (perfect) forecasts are used. A review of a classical Inventory Theory problem called the Newsboy or Christmas Tree problem was exploited to provide some analytical motivation for the proposed Balancing Rule model.

Degree

Ph.D.

Advisors

Houck, Purdue University.

Subject Area

Civil engineering

COinS