Comparing algorithms, representations and operators for the multi-objective knapsack problem

This paper compares the performance of three evolutionary multi-objective algorithms on the multi-objective knapsack problem. The three algorithms are SPEA2 (strength Pareto evolutionary algorithm, version 2), MOGLS (multi-objective genetic local search) and SEAMO2 (simple evolutionary algorithm for multi-objective optimization, version 2). For each algorithm, we try two representations: bit-string and order-based. Our results suggest that a bit-string representation works best for MOGLS, but that SPEA2 and SEAMO2 perform better with an order-based approach. Although MOGLS outperforms the other algorithms in terms of solution quality, SEAMO2 runs much faster than its competitors and produces results of a similar standard to SPEA2.