Heuristic operators, redundant mapping and other issues in genetic algorithms

This paper uses the 0-1 knapsack problems (KPs) to investigate such issues as early convergence, exploration versus exploitation, redundant mapping and the role of heuristic operators etc. in genetic algorithms (GAs) with (mu+lambda)-strategy. We use the order-based representation for chromosome and propose two different decoding approaches, the Order-Decoding (preserving redundancy) and the Cycle-Decoding (eliminating redundancy), to decode it. A new crossover and two new mutation operators are also proposed in this paper. The KPs with various kinds of item numbers, capacities, and correlations between profits and weights are tested with a wide range of possible combinations of genetic operators. Computer simulation results show that heuristic operators must be used appropriately to achieve better results; exploration operators must be used with care; super individuals, early convergence and redundant mapping are not harmful for GAs.