Banca de QUALIFICAÇÃO: JOSÉ FERNANDO DE TOLEDO

MACHINE LEARNING MODELS WITH CLIMATE INDICATORS FOR THE GENERATION OF NATURAL FLOW SCENARIOS

    In order to plan the operation of the Brazilian Electric System, computational models are used, which provide input data, on the future availability of water in the reservoirs of the Hydroelectric Power Plants, the availability of thermal power plants and the forecast load for the period. As output data, the execution models include the scheduling of the Plants (Dispatch) and the Marginal Operating Cost - CMO (cost per unit of energy adequate to meet an increase in the load on the system). It should be noted that the future availability of water in the reservoirs is associated with the randomness of the flows affluent to them, with forecast models that provide inflow scenarios available. There is a strong dependence on the climate in this process, as variations in average preference and temperature directly impact the availability of water available for energy generation. In this work, the influence of the use of regional and teleconnections climate indicators in flow forecasting models was investigated, using Machine Learning techniques and autoregressive models with exogenous variables, for twelve Hydroelectric Power Plants in the Brazilian territory. The results indicated that there is a correlation of some indicators for preference, with the adoption of the method in predictive models appearing to be promising.