Abstract. The reservoir management problem for a hydrothermal power system is well suited to modeling via dynamic programing. In this paper we describe a dual approach which we term "constructive dynamic programming" (CDP) which has been successfully applied to optimize releases in a stochastic two-reservoir model of the New Zealand power system. That model ignores serial correlations of inflows, though, and hence assumes that current inflow observations do not have any impact on future release decisions. Tests show, however, that better decision rules can be produced by accounting for inflow correlation. Hence we have developed an extension to the standard CDP to explicitly deal with serial correlation of reservoir inflows, and we report on those extensions also.
IntroductionThe reservoir management problem for a hydrothermal power system is to decide how much water should be released in each period so as to minimize the expected operational cost, including the fuel cost of thermal plants, and the cost of shortages, should they occur. In economic terms the basic principle of reservoir management is to continue generation until the marginal value of releasing water is equal to the fuel cost of the most expensive thermal station used to meet residual demand after hydro generation or during a shortage the value of meeting loads which would otherwise not be serviced. The marginal value of the water in stock depends on the optimal utilization of that stock in future periods. Because all reservoirs have limited storage capacities and because the inflows are uncertain, water management becomes quite complex.Dynamic programming (DP) has been widely used in reservoir management. This is mainly due to its ability to handle nonlinear and stochastic features which characterize many reservoir systems. A drawback with traditional DP is that it can only provide a discrete approximation for problems with a continuous state space. Problems involving several "state variables" (in this case reservoir storage) also run into serious computational difficulties, the so called "curse of dimensionality."A dual approach to dynamic programming was formulated by Read and George [1986, 1990] and has the advantage of constructing the optimal solution directly and hence reducing the computational requirements, and a very similar technique was developed by Bannister and Kaye [1991] and further developed by Kaye and Travers [1997]. Kaye and Travers refer to this approach as a "constructive dynamic programming" (CDP) technique, and we will use that term here in order to distinguish it from the rather different technique which has been developed by Pereira and Pinto [1991] under the name of stochastic dual DP. CDP is more accurate than primal DP when finding a solution using the same number of grid points, be- In CSDP the optimal release rules are expressed by a set of guidelines consisting of the locus over time of storage levels (S t) whose marginal water values (MWV) are equal to the fuel costs of particular thermal stations. This discussi...