Active distribution networks (ADNs), characterized by the presence of distributed generation (DG) units and dynamic topology changes, introduce new protection challenges in addition to well-known issues related to control, management, and demand response. Moreover, the existence of switching points including tie-points in active distribution networks adds further complexity to network protection. Aiming to develop an advanced protection framework, this study proposes a two-stage hybrid protection framework based on both non-deep neural network (DNN)-based and local DNN-based agents for active distribution networks. The test system used in this research study is the IEEE 34-bus standard system, equipped with DGs and predefined tie-points to enable network reconfiguration. These switching points allow for load redistribution, loss reduction, and voltage improvement; however, they also necessitate a new protection strategy that adapts to changes in tie-point status and relay settings. The proposed framework consists of two main components: (1) A topology identification agent (TIA) that detects the network configuration using the status of tie-point switches and adaptively updates relay settings to maintain readiness under changing in tie-points status and post-fault conditions; and (2) A local backup protection agent (LBPA) based on DNNs, designed to operate as a complementary backup to the intelligent electronic devices (IEDs) installed in the network without any need for communications to transmit the measurement parameters. Given the challenges associated with single-phase-to-ground faults detection by conventional protection devices in the distribution networks, the backup protection agent in this study is specifically designed to detect single-phase-to-ground faults in minimum one stage by IEDs or in maximum two stages by the LBPA. Simulations and coding were carried out in DIgSILENT, MATLAB, and Python environments.
