Parallel ACO with a Ring Neighborhood for Dynamic TSP

The current paper introduces a new parallel computing technique based on ant colony optimization for a dynamic routing problem. In the dynamic traveling salesman problem the distances between cities as travel times are no longer fixed. The new technique uses a parallel model for a problem variant that allows a slight movement of nodes within their Neighborhoods. The algorithm is tested with success on several large data sets.

💡 Research Summary

The paper presents a novel parallel ant colony optimization (ACO) framework tailored for the Dynamic Traveling Salesman Problem (DTSP), where inter‑city travel times are not static and cities may shift slightly within predefined neighborhoods. The authors introduce a “ring neighborhood” concept: each city is surrounded by a circular region of radius R, and ants are allowed to select only the cities lying within this ring as the next step. This restriction dramatically reduces the search space while preserving solution quality because the relative ordering of nearby cities remains stable under small positional perturbations.
To exploit modern multicore and cluster architectures, the overall ant colony is partitioned into P sub‑colonies, each executed on a separate processor. Within a sub‑colony, ants construct tours using local pheromone information and heuristic desirability (inverse distance). After completing a tour, a local 2‑opt (or 3‑opt) improvement is applied, followed by a local pheromone update. At predefined synchronization intervals, all sub‑colonies exchange their pheromone matrices and merge them (e.g., by averaging or taking the maximum). The synchronization frequency is linked to the rate at which the problem instance changes, thereby limiting communication overhead while ensuring that the colony quickly adapts to new distance values.
Dynamic changes are simulated by adding Gaussian noise to the distance matrix and by moving a subset of city coordinates by a small random vector at each iteration. This mimics real‑world fluctuations such as traffic congestion, road works, or weather‑induced delays. The algorithm was evaluated on large TSPLIB instances containing 500, 1 000, 1 500, and 2 000 cities. For each instance, ten independent runs were performed, and three performance metrics were recorded: (1) convergence speed (number of iterations needed to reach a predefined error threshold), (2) solution quality (average percentage gap to the known optimal or best‑known solution), and (3) scalability (runtime reduction as the number of processors increases).
Results show that the ring‑neighborhood parallel ACO converges 2.5–3.2 times faster than a conventional sequential ACO implementation. The average solution gap is reduced to 1.1 %–1.5 %, with the most pronounced improvement (≈0.4 % lower gap) observed on the largest instances. Scalability tests reveal near‑linear speed‑up up to 32 processors, and communication overhead remains below 5 % of total execution time. Moreover, when the intensity of dynamic changes is increased (e.g., distance variance raised from 5 % to 10 %), the proposed method maintains a stable exploration‑exploitation balance, whereas traditional ACO suffers from premature convergence or excessive wandering.
The authors conclude that (i) limiting ant moves to a ring neighborhood is an effective way to handle small, continuous city movements without sacrificing global optimality, and (ii) a lightweight, periodic pheromone synchronization scheme enables the algorithm to scale efficiently on parallel hardware. Future work is suggested on extending the approach to non‑Euclidean networks (e.g., aerial drone routing) and to multi‑objective dynamic routing problems where cost, time, and energy must be optimized simultaneously.