Comparing hybrid metaheuristics for the bus driver rostering problem

Abstract

SearchCol is a recently proposed approach hybridizing column generation, problem specific algorithms and distinct well known metaheuristics (VNS, Tabu Search, Simulated Annealing, etc.). SearchCol allows to solve several combinatorial optimization problems by applying column generation to a given decomposition model, and using one of the available metaheuristics to search for an integer solution combining the previously generated columns, which are components of the problem. A new evolutionary algorithm (EA) was proposed as the first population based metaheuristic included in SearchCol. This EA uses a representation of individuals based on the generated columns and has been used to obtain integer solutions for a new model for the Bus Drivers Rostering problem (BDRP). Special features of this EA include local search and elitism. This paper presents a computational study evaluating the new population based heuristic (EA) versus two single solution heuristics: VNS and Simulated Annealing, exploiting different configurations of the framework on a set of benchmark instances for the BDRP.