Reinforcement Learning Applied to the Prognosis of Forced Shutdowns of Generating Units in Hydroelectric Power Plants
Reinforcement Learning, Machine Learning, Failure Diagnosis, Hydroelectric Plants, Forced Stops, Reliability Analysis.
Hydroelectric power plants are the main means of producing electricity in Brazil due to the low emission of polluting gases, their low operating cost, high efficiency and favorable climate conditions for their installation. In this context, the operation area is of great importance for the operation of the plants, and its workflow must occur aiming at the least amount of errors possible, in order to preserve the safety of people, assets and the environment. Furthermore, in addition to the difficulties inherent to the operation, operators are often responsible for different hydroelectric plants simultaneously, which can lead to human error. In this work, the development of a computational system to support decision-making in hydroelectric plants is proposed. The proposed solution covers the prognosis of forced stoppages of generating units in these plants, aiming at prediction and diagnosis via Reinforcement Learning methods. The model aims to inform the operator of the possible occurrences of a failure, to avoid or mitigate the impact of a forced stop on generating units. The proposed strategy should be compared with several Machine Learning algorithms to validate the results. A reliability analysis was also implemented on the equipment of the hydroelectric plant to obtain indicators related to their useful lives in order to verify the impact of the implementation of the approach.