Identificación de reglas de operación óptimas de embalses para el control de inundaciones a partir de modelos de operación. caso de estudio: Cuenca del Río Yuna en República Dominicana
Tami Riveros, Carlos Alfredo | 2020-11-12
This study addresses the use of computational tools in order to obtain optimal operating rules for the Hatillo reservoir (Dominican Republic), mainly considering the purpose of reducing flooding downstream of the dam without affecting its other uses (hydroelectric generation and irrigation for agriculture). Because it is a multipurpose reservoir, the problem is posed under a multiobjective approach, where the use of evolutionary algorithms (optimizers) is explored, together with approximation functions of Artificial Neural Networks, Radial Basis Networks and linear functions (Operational Models parametric) for the direct search of the operational rules. The proposed operational models were developed for the available information that includes a period of 10 years (2009-2019), on a daily level, the controlled discharges from the reservoir were defined from the approximation functions, which receive as inputs the system status variables (reservoir level, inflows, previous outflows), likewise, the system's own physical components are used to define the restrictions of controlled discharges, reservoir operation limits, and to define uncontrolled releases. On the operational models, the optimization algorithms were applied to obtain the optimal operational rules, the parameters of the approximation functions being the decision variables of each model. The optimization algorithms used were the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) and Multiobjective Evolutionary Algorithm Based on Decomposition (MOEA / D), the previous process was carried out in JMETALPY, a multiobjective optimization environment developed in Python. The results obtained show that it is possible to reduce the peaks of the reservoir discharge hydrographs and therefore the magnitude of downstream floods by applying the operational rules obtained from the optimization of the operational models. For this particular case, it was found that the approximation functions of Neural Networks and Radial Basis Networks allow to adequately parameterize the reservoir's operating rules since they can generate patterns or complex shapes that normally cannot be built by other functions, such as, linear functions. The optimization results show that artificial Neural Networks are better adjusted than the other methods for this case study, being the NSGAII the optimization algorithm that has the best performance in terms of computational time and optimization results.