An environment-adaptive protection scheme with long-term reward for distribution networks

Increasing renewable penetration in the distribution brings uncertainties, raising concerns for reliable grid operation. For example, relatively regular topological changes in distribution grids and the frequent on/off status of some distributed generators (DGs) add ambiguity to the short circuit levels of distribution networks. Consequently, protective relays need to adapt their settings to protect different operation conditions on distribution systems. Without such capability, relays may false trip or be insensitive. Previous methods ignore the long-term relay setting effect in relay coordination design. To bridge the gap, this paper proposes an environment-adaptive protection scheme (E-APS) to solve the protection coordination issue from a sequential decision making perspective. The agent-environment interaction is designed with protection knowledge integrated to enable the protection agent's adaptivity. After defining the state, action, and reward in reinforcement learning for the relay settings, we prove the convergence of the value function for post-decision state protection setting. In the numerical results, different system operation scenarios are applied to validate the performance of the proposed E-APS. This scheme is also compared with other optimization-based protection schemes. Results show that the E-APS is more adaptive to environmental change and achieves high performance in protection coordination.