Experimental Analysis of Pheromone-Based Heuristic Column Generation Using irace

Pheromone-based heuristic column generation (ACO-HCG) is a hybrid algorithm that combines ant colony optimization and a MIP solver to tackle vehicle routing problems (VRP) with black-box feasibility. Traditionally, the experimental analysis of such a complex algorithm has been carried out manually by trial and error. Moreover, a full-factorial statistical analysis is infeasible due to the large number of parameters and the time required for each algorithm run. In this paper, we first automatically configure the algorithm parameters by using an automatic algorithm configuration tool. Then, we perform a basic sensitivity analysis of the tuned configuration in order to understand the significance of each parameter setting. In this way, we avoid wasting effort analyzing parameter settings that do not lead to a high-performing algorithm. Finally, we show that the tuned parameter settings improve the performance of ACO-HCG on the multipile VRP and the three-dimensional loading capacitated VRP.