Paralelização assíncrona da programação dinâmica dual com agrupamento ótimo de nós para o planejamento da operação de sistemas hidrotérmicos

Abstract

The hydrothermal coordination problem of real systems is complex due to the need to con sider the uncertainties inherent in the inflow to hydroelectric plants. To solve this problem, several methodologies can be applied, including the Dual Dynamic Programming (DDP), which is a decomposition strategy capable of solving multi-stage stochastic optimization problems. For real problems such as Brazil’s National Interconnected System (SIN), this methodology may require a high computational effort due to the number of reservoirs and study horizon.This masters dissertation presents a new methodology that aims to reduce the computational time of DDP. The implemented methodology proposes an enhancement in asynchronous parallel processing algorithm for DDP, including an optimal grouping of scenario tree nodes into subproblems by genetic algorithm. These strategies make the best use of available resources bypassing some DDP synchronization restrictions that can be very detrimental to parallelism. The efficiency of the proposed methodology is shown for a set of SIN hydroelectric plants.