Deep and Reinforcement Learning for Active Distribution Network Protection

The protection system is an important component for microgrid operation and stability. The protection system must be capable of identifying faults, isolating faulty sections, and ensuring continuity of power supply to unaffected loads. Microgrids introduce major challenges to the protection system because of low short circuit levels and changing network topologies. The objective of this chapter is to propose and develop a novel deep reinforcement learning (DRL)-based distance protection capable of detecting low- and high-resistance faults in AC microgrids with renewable energy resources (RES). The proposed model includes a convolutional neural network and gated recurrent unit CNN-GRU hybrid for fault classification. The proposed approach adopts a protection approach for accuracy, stability against reverse direction faults, and noise immunity for the evaluation of the proposed model. In this study, the performance of the proposed model is evaluated by conducting prototype hardware in the loop (HIL) simulation using Raspberry Pi microcontroller and MATLAB Simulink. The results demonstrate the potential of the proposed DRL-based distance relay to have higher detection sensitivity for the power system faults compared to the conventional one. The experimental results are in full agreement with the simulation results showing the high accuracy of the proposed DRL-based distance relay in detecting high-resistance faults. Moreover, it confirms the implementability of the proposed model for real-time applications and achieved 93.52% overall accuracy.