Heuristic Hybridization for CaRSP, a multilevel decision problem

This study investigates algorithm hybridizations for an NP-hard problem in which different decisions regarding its variables must be made. Hybridizations were implemented in a multi-agent architecture inspired by the particle swarm optimization metaheuristic. In this architecture, particles are agents able to move in the solution space, learn, and make decisions. The problem addressed in this study is the Traveling Car Renter with Passengers, a problem derived from the Traveling Salesman, in which the salesman can use different vehicles to build a Hamiltonian circuit among $n$ cities. Different cars are available for salesmen in each city. Different cars assign different costs to the arcs. The vehicles used by the salesman were offered to the mobility system. The driver knows the ride requests in advance and chooses which to meet. Therefore, decisions are made regarding route, car, and ride requests. The architecture hybridized twelve heuristic procedures. This paper reports tests performed on 72 instances in up to 130 cities, 5 cars, and 375 ride requests. The architecture results were compared against a memetic algorithm presented in the literature.