Operação de reservatórios usando Otimização de Monte Carlo e Floresta Aleatória.

Abstract

Water systems are susceptible to a variety of challenges, such as problems associated with water excess and scarcity. The use of practical operation rules for proper water resources management emerges as an alternative to mitigate such difficulties. In this study, a protection rule based on Monte Carlo Optimization with Random Forest (MCORF) was applied for the monthly operation of reservoirs. To obtain the rule, the following steps were followed: (1) generation of synthetic inflows by the Fragment Method; (2) application of a deterministic model (DET) to optimize the system operation, minimizing its vulnerability; and (3) association of the optimal water allocations and availabilities to generate the MCO-RF rule. We validated the methodology by applying the procedure to the Epitácio Pessoa reservoir in Brazil. The generated rule was evaluated by using independent inflow series. The MCO-RF model proved to be promising, defining an allocation behavior similar to that produced by the DET model under perfect inflow prediction. The MCO-RF rule is also practical, requiring only the current month’s water availability to determine the allocation.