Design and implementation of adaptive online setting system for feeder automation based on reinforcement learning

Traditional feeder automation systems have protection setting mismatch and fault response lag problems in dynamic distribution network environments. This paper designs an adaptive relay protection and feeder automation setting online setting system, builds an adaptive setting model, designs a hierarchical software module, introduces a double-delay deep deterministic policy gradient (TD3) algorithm, and combines the graph attention network (GAT) to dynamically model the distribution network. At the same time, the CUDA-accelerated PyTorch framework is used for algorithm deployment and the lightweight packaging of the core module is achieved through Docker containerization technology. The maximum protection misoperation rate of the system proposed in this paper is 8.0%, the minimum is 1.2%, and the average fault isolation time is only 764.57 milliseconds, which provides a feasible solution for real-time control of smart distribution networks.