Evolving Petri Nets with a Genetic Algorithm

Holger Mauch

University of Hawaii at Manoa
Dept. of Information and Computer Science
1680 East-West Road,
Honolulu, HI 96822
hmauch@hawaii.edu

Abstract. In evolutionary computation many different representations ("genomes") have been suggested as the underlying data structures, upon which the genetic operators act. Among the most prominent examples are the evolution of binary strings, real-valued vectors, permutations, finite automata, and parse trees. In this paper the use of place-transition nets, a low-level Petri net (PN) class [1,2], as the structures that undergo evolution is examined. We call this approach "Petri Net Evolution" (PNE). Structurally, Petri nets can be considered as specialized bipartite graphs. In their extended version (adding inhibitor arcs) PNs are as powerful as Turing machines. PNE is therefore a form of Genetic Programming (GP). Preliminary results obtained by evolving variable-size place-transition nets show the success of this approach when applied to the problem areas of boolean function learning and classification.¹

¹This research was supported in part by DARPA grant NBCH1020004.

LNCS 2724, p. 1810 ff.

Full article in PDF

lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2003