Traffic Adaptive Moving-window Service Patrolling for Real-time Incident Management during High-impact Events

TL;DR

This work tackles real-time incident management during high-impact events by introducing TAMPA, a traffic-adaptive moving-window patrolling framework. TAMPA combines a traffic pattern predictor, a complaints estimator, and a moving-horizon dynamic programming planner, with a DKW-based online adaptation that reconfigures the patrol graph when shift in complaint distributions is detected. The authors prove bounds on suboptimality under distribution shifts and demonstrate substantial performance gains over stationary and random patrol strategies in SUMO-based urban simulations. The approach promises improved responsiveness and routing efficiency for mega-events, with future directions including digital twin integration and broader field validation.

Abstract

This paper presents the Traffic Adaptive Moving-window Patrolling Algorithm (TAMPA), designed to improve real-time incident management during major events like sports tournaments and concerts. Such events significantly stress transportation networks, requiring efficient and adaptive patrol solutions. TAMPA integrates predictive traffic modeling and real-time complaint estimation, dynamically optimizing patrol deployment. Using dynamic programming, the algorithm continuously adjusts patrol strategies within short planning windows, effectively balancing immediate response and efficient routing. Leveraging the Dvoretzky-Kiefer-Wolfowitz inequality, TAMPA detects significant shifts in complaint patterns, triggering proactive adjustments in patrol routes. Theoretical analyses ensure performance remains closely aligned with optimal solutions. Simulation results from an urban traffic network demonstrate TAMPA's superior performance, showing improvements of approximately 87.5\% over stationary methods and 114.2\% over random strategies. Future work includes enhancing adaptability and incorporating digital twin technology for improved predictive accuracy, particularly relevant for events like the 2026 FIFA World Cup at MetLife Stadium.

Figures (4)

• Figure 1: An illustration of TAMPA: the camera captures traffic patterns on the road to assist in generating the prediction of traffic states given incidental situations, while crowd-sourcing collectors (e.g., mobile phone applications) generate user-reported complaint patterns, enabling the detection of shifts in complaints.
• Figure 2: The illustration of TAMPA that combines online solution of the planning window MDP problem (above) in \ref{['sec:pwmdp']} and the global adaptation process (below) in \ref{['sec:GAP']}. The TAMPA provides the patrol command to the service patrol vehicle at each time step. In solving the planning window MDP problem, the traffic predictor and the complaints estimator provide predicted conditions for the service patrol planner. The planner obtains an optimal action, and the service patroller executes it in the traffic network. The sensing via camera to observe traffic conditions and the crowd-sourcing collector to generate user-reported complaints in the network, then update the traffic predictor and the complaints estimator at each time step until the start of the next planning window. The global adaptation process helps the service patrol vehicle proactively adapt to the variation of the complaints conditions. The comparator will decide if complaints distribution shifting happens at each time step. If shifting happens, the patrol graph will be adapted, and a new planning window will be initiated immediately.
• Figure 3: On the left, we illustrate the urban traffic network covered by the patroller, consisting of 12-node, 19-edge. On the right, we compare the cost in \ref{['eq:global_cost']} of three patrolling strategies. The solid blue line represents our proposed method. The dotted yellow line represents solving \ref{['alg:patrol-decision']} under a stationary setting (i.e., fixed graph). The dotted green line illustrates a baseline strategy in which the patrol vehicle selects its next destination node at random. Before 360 minutes (corresponding to the 75th time step under a 5-minute interval), the blue and yellow lines coincide because the complaint distribution remains unchanged. After the distribution shifts at 360 minutes, our method (blue line), which adapts to the new conditions, outperforms both the stationary approach and the random baseline. Overall, our method achieves an approximate $87.5\%$ improvement over the non-adaptive version of our approach and a $114.2\%$ improvement compared to a random patrol strategy.
• Figure 4: The left side illustrates how the complaint hotspot shifts from node $1$ to node $4$. On the right, we compare the visit frequencies of three different strategies before and after this shift. Notable, $13, 14$ are two newly added nodes that split edges $2-1$ and $2-4$. Our method adapts to the changing hotspot (spend almost $90\%$ time patrolling around node $4$ after the shifting), whereas the other two strategies fail to detect or respond effectively.

Theorems & Definitions (2)

• Theorem 1
• Corollary 1