Inver-over Operator for the TSP

Abstract. In this paper we investigate the usefulness of a new operator, inver-over, for an evolutionary algorithm for the TSP. Inver-over is based on simple inversion, however, knowledge taken from other individuals in the population in uences its action. Thus, on one hand, the proposed operator is unary, since the inversion is applied to a segment of a single individual, however, the selection of a segment to be inverted is population driven, thus the operator displays some characterictics of recombination. This operator outperforms all other `genetic' operators, whether unary or binary, which have been proposed in the past for the TSP in connection with evolutionary systems and the resulting evolutionary algorithm is very fast. For test cases, where the number of cities is around 100, the algorithm reaches the optimum in every execution in a couple of seconds. For larger instances (e.g., 10,000 cities) the results stay within 3% from the estimated optimum.

• Type: Conference
• Year: 1998
• Where: PPSN
• Authors: Guo Tao, Zbigniew Michalewicz
• Keywords: Distributed And Parallel Computing, Evolutionary Algorithm, Evolutionary Systems, PPSN 1998, Simple Inversion