Despacho econômico de sistemas elétricos de potência utilizando aprendizagem por reforço

Abstract

Performing economic dispatch through numerical methods may not be the most efficient way when considering the unpredictable and dynamic nature of renewable energy sources, such as the loss of efficiency of generating units, or contingencies that can lead to a value far from optimal. In this work, economic dispatch will be performed through a reinforcement learning (RL) method called Q-learning. Since Q-learning is a method that can estimate an optimal policy solely with data from the environment (power system), it aims to achieve good accuracy and high performance in executing optimal economic dispatch. The proposed method will be compared with methods addressed in other published works to verify if the expected results were achieved.