The proliferation of illegal parking has a major impact on traffic safety and efficiency. The problem of studying how to capture as many parking infringements as possible in a limited time is called Traveling Officer Problem (TOP), which is a variant of Traveling Salesman Problem (TSP). Compared with TSP, the state of the nodes in the path of TOP changes (0 or 1) over time and the time is limited. The purpose of TOP is to capture as many points with the state of 1 as possible in a limited time. According to the parking information provided by the traffic management system, we propose a Dynamic Graph Attention Network (DGAT) to address this problem. The results on real datasets from Melbourne, Australia show that our model has clear advantages over other optimization algorithms in many metrics, especially the capture rate.