Reinforcement learning for optimal protection coordination

The design of reliable protection systems is essential in order to guarantee the secure operation of power systems. The coordination of the response of protective equipment under fault conditions is a fundamental problem which is usually formulated as a non-linear optimization problem with the objective of minimizing the total operating time of the protection system. In general, optimal protection coordination can be regarded as a well-studied problem, with the existing literature featuring the application of a wide range of solution techniques. However, recent advances in the area of artificial intelligence and the increasing availability of near real-time measurements from distribution systems offer the possibility to envision adaptive protection systems capable of operating optimally under different power system operating conditions and, potentially, of being more resilient, by assigning local decision-making autonomy to relays, instead of relying on a centralized system to coordinate protective devices. To this end, in this study optimal protection coordination is cast as a reinforcement learning problem and relays are viewed as autonomous agents that can manipulate their time dial settings in order to optimally respond to signals from their environment, i.e., the power system. The reinforcement learning problem is solved by applying the Q-learning algorithm. The results of the case study indicate that this framework is capable of providing settings that achieve both a fast and coordinated protection system.