Hybrid ICA–Local Search for the Multi-Depot Vehicle Routing Problem
This paper proposes a two-layer hybrid Imperialist Competitive Algorithm combined with local search to simultaneously optimize customer-to-depot assignments and vehicle routes for the Multi-Depot Vehicle Routing Problem, achieving competitive results with gaps within approximately 2% on standard benchmarks.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a city where a single warehouse must deliver packages to hundreds of homes. The challenge is to figure out the most efficient way to send out a fleet of trucks so that every house gets a visit, no truck is overloaded, and the total distance driven is as short as possible. This is a classic puzzle known to mathematicians as the vehicle routing problem. But in the real world, logistics are rarely that simple. Often, goods do not come from one central hub but from several different depots scattered across a region. This adds a second, equally difficult layer to the puzzle: before a driver can even plan their route, someone must decide which depot is responsible for which customer. This expanded challenge, where the goal is to assign customers to the right depots and then plan the perfect driving paths for each, is called the multi-depot vehicle routing problem. It is a problem of immense complexity, where the number of possible combinations is so vast that finding the absolute best solution is computationally impossible for large cities. Because of this, researchers rely on smart shortcuts, known as metaheuristics, to find solutions that are very close to perfect without checking every single possibility.
In a recent study, researchers from North South University tackled this specific logistical headache by creating a new hybrid method that combines two distinct strategies. They built a system that separates the problem into two layers, much like a manager who first decides which team handles which territory, and then lets the team leaders figure out the best way to move around within that territory. The first layer of their system uses a technique called the Imperialist Competitive Algorithm. This approach mimics a form of social competition where a group of potential solutions, called countries, are ranked by how well they perform. The best solutions become imperialists, and the others become their colonies. Over time, the colonies try to become more like their imperialists by copying their decisions, while occasionally making random changes to keep the search fresh. In this specific study, the "decision" being copied is which depot serves which customer. The second layer of the system is a local-search router. Once the first layer has assigned customers to depots, this router steps in to build the actual driving routes. It starts by creating a basic path using a simple rule of adding the nearest available customer, and then it refines that path by testing small changes, such as swapping the order of two stops or moving a stop to a different part of the route, to see if the total distance shrinks.
The innovation in this work lies in how these two layers talk to each other. The local-search router acts as a judge for the Imperialist Competitive Algorithm. Every time the algorithm proposes a new way to assign customers to depots, the router instantly calculates the total driving distance for those assignments. This distance becomes the score, or fitness, that determines which assignments are kept and which are discarded. To make the system even sharper, the researchers added a final refinement step. After the main competition between the solutions has run its course, the system takes the best result found so far and performs a careful, manual check. It temporarily moves individual customers to different depots to see if a simple reassignment could squeeze out any remaining inefficiency. This entire process was tested against a standard set of difficult test cases known as the Cordeau benchmark instances, which are widely used by researchers to measure the performance of routing algorithms.
The results of this new hybrid method were impressive, particularly for smaller and medium-sized problems. On several test cases involving up to one hundred customers and multiple depots, the system found solutions that were within just a few percent of the best-known results ever recorded. For one specific instance with seventy-five customers and five depots, the method achieved a gap of only 1.16 percent from the best-known solution, meaning it was nearly perfect. The system also proved to be very stable; when the researchers ran the same test multiple times with different random starting points, the results remained consistent, with very little variation between runs. This suggests that the method is reliable and does not depend on luck to find a good answer. However, the study also revealed where the method faces limits. On the largest test case, which involved one hundred sixty customers, the gap between the new solution and the best-known solution widened to about 13.5 percent. The researchers noted that for the largest problems, the sheer size of the search space makes it harder for the local search to find deep improvements. Similarly, on instances with only two depots, the method struggled slightly more, likely because there are fewer opportunities to improve the solution by shuffling customers between different depots.
Ultimately, this research demonstrates that splitting a complex logistical problem into two distinct tasks—assigning customers to depots and then planning the routes—can be a highly effective strategy. By letting a competitive algorithm handle the big-picture assignments and a local search handle the fine-tuning of the routes, the researchers created a system that performs strongly across a range of scenarios. The work confirms that while finding the absolute mathematical best for every possible scenario remains out of reach for large-scale problems, this hybrid approach offers a practical and robust way to get very close to the ideal, ensuring that delivery networks can operate with greater efficiency and lower costs.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.