Banca de DEFESA: JOSÉ FERNANDO DE TOLEDO

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

In Brazil, in order to carry out the planning of the operation of the generation of electric energy in the hydrothermal power system, computational models are used that provide information related to the energy resource, such as the forecast of the amount of water that may be available in the reservoirs of the Hydroelectric Power Plants (flow natural affluent), fuel and the availability of Thermal Power Plants, as well as the load forecast for the planning period. As output data, the models execute the programming of the operation of these Plants and the Marginal Cost of Operation, that is, cost per unit of electric energy produced to meet an increase in load in the system. It is noteworthy that the future availability of water in the reservoirs is associated with the randomness of the natural flows flowing into them, so that they become flow forecast models that provide increasingly reliable scenarios of influences, thus allowing the projection of service demand more reliable and safer. Therefore, it can be seen that there is a strong dependence on climate issues in this process, as variations in the average rainfall and temperature, which can directly impact the availability of water for electricity generation, and can influence in different ways the operational decision-making in systems with great hydroelectric predominance when adequate planning of the available energy resource is not carried out. Based on this context, this work investigated the influence of using climate indices of teleconnections as exogenous variables in prediction models of natural inflows in Brazil, using machine learning and autoregressive models for fourteen hydroelectric plants located in different regions in the territory. Brazilian. The achieved results indicated that there is a significantly adapted, among the adopted climatic indices, the locations of the hydroelectric plants and the adopted models. Finally, it is concluded that this methodology, which proposes to include climate indices in the prediction models of natural flows flowing to the power plants, is very promising and will be able to help the agents of the electricity sector in decision-making and will, as a consequence, have an operating direction that guarantees to the final consumer more energy security.