[Paper Review] A New Shortest Path Algorithm Generalized on Dynamic Graph for Commercial Intelligent Navigation for Transportation Management

Yong Kiam Tan|arXiv (Cornell University)|May 11, 2019

Data Management and Algorithms6 references4 citations

TL;DR

This paper proposes a novel dynamic shortest path algorithm optimized for time-dependent traffic networks in intelligent transportation systems. By introducing a time-continuous 'Hot Zone' overlay graph model, it enables fast, scalable, and accurate route computation with sub-second query times even on large urban networks, significantly improving upon traditional methods like Bellman-Ford through offline pre-processing and localized Dijkstra computation.

ABSTRACT

Dynamic graph research is an essential subject in Computer Science. The shortest path problem is still the central in this field; moreover there is a variety of applications in practical projects. In this paper, we select the transportation activity as a paradigm to study. Of course there exists the most clear and appealing practical application over the intelligent navigation, also which is an important part in intelligent transportation system (ITS). Totally speaking, this activity is fairly simple and well-behavior under our consideration, for which, there is a lot of sophisticated theories and techniques to motivate our researches although this problem relates to interdisciplinary; so that we can be dedicated to computing aspect. Below, a host of practical problems will by the way be discussed in theory and empiricism including correctness, sampling, accuracy, compatibility, quick query and the application into other research; therein some have been scarcely argued intentionally or not in literatures before. Through these contents, we are able to have a smart system built which easy to incorporate other modules to realize a sophisticated industrial system.

Motivation & Objective

• To address the limitations of existing dynamic shortest path algorithms in handling real-time, time-varying traffic data efficiently.
• To develop a scalable and modular framework for commercial intelligent navigation systems in urban transportation.
• To reduce online query time and computational overhead by pre-processing traffic data into compact, time-continuous Hot Zones.
• To enable efficient solving of the Best Opportunity Problem (BOP) for optimal departure time selection.
• To provide a flexible, compatible, and accurate solution that integrates well with industrial ITS modules.

Proposed method

• Proposes a time-dependent graph model where edge weights vary over time, represented by discrete time-dependent tables.
• Introduces the concept of a 'Hot Zone' (Z), a compact, time-continuous overlay graph that clusters nodes with similar dynamic behavior to reduce search space.
• Applies Dijkstra’s algorithm (DCA) within each Hot Zone, enabling fast online queries with minimal overhead.
• Uses offline pre-processing to generate zone models that represent traffic states over a given time period, ensuring query flexibility.
• Employs a hierarchical clustering approach to group paths with identical or similar trajectories, reducing redundancy in storage and computation.
• Leverages the property that time-dependent shortest paths are non-decreasing and uses a modified labeling method for dynamic weight updates.

Experimental results

Research questions

• RQ1 How can dynamic shortest path computation be made efficient and scalable for large urban transportation networks with time-varying traffic?
• RQ2 What is an effective way to represent time-varying traffic states in a compact and query-friendly form?
• RQ3 Can localized, time-continuous zones (Hot Zones) reduce online computation time without sacrificing accuracy?
• RQ4 How can the Best Opportunity Problem (BOP) — finding the optimal departure time — be solved efficiently using pre-processed data?
• RQ5 What is the trade-off between zone size, pre-processing cost, and online query performance in dynamic pathfinding?

Key findings

• The proposed Hot Zone model reduces the effective search space, enabling Dijkstra’s algorithm to run in under 250 ms on a 3.5k-node zone using a standard laptop.
• For metropolitan-scale networks (e.g., ~100k nodes), the maximum zone size is approximately 30, with query times in the microsecond range.
• The algorithm achieves sub-second query performance even on long-distance paths (e.g., 6-day journeys), with a maximum time deviation of 45 minutes at the 1-minute resolution.
• Offline pre-processing of Hot Zones allows for efficient handling of the Best Opportunity Problem (BOP), reducing per-query cost compared to repeated Bellman-Ford runs.
• The method supports arbitrary time-point queries due to the time-continuous nature of the zones, enhancing flexibility for real-world navigation systems.
• The computational complexity of the zone-based approach is bounded by O(λ~n log~n + 2M~n), where ~n = √(cn) and c ≥ 1, demonstrating scalability.