Use of genetic algorithms for finite-state automata discovery : Another approach to adaptive process control
The genetic algorithm method (GA) is an iterative search algorithm based on an analogy with the process of natural selection (Darwinism) and natural genetics. It is assumed that the process dynamics are such that it is feasible to describe the state space as being partitioned into a finite number of sub-regions through which the system moves in a deterministic fashion, according to the input signals (chaotic systems do not exhibit this property). The process behaviour can, therefore, be described by means of transition rules which can be represented adequately by a finite-state automaton. A detailed example is presented in the full paper, consisting of the discovery of an 8-state automaton with binary input and output symbols. A series of experiments was performed to test the ability of GA to discover a finite-state automaton. In each experiment, after several hundred generations, a correct automaton was found ("correct" meaning either strictly equivalent to the original one or obtained by a permutation of the states). Analysis of these results shows that, for this problem, GA performs much more efficiently than a simple random search or a systematic enumeration.
Bibliographic Reference: Paper presented: IFAC International Symposium on Distributed Intelligence Systems, Arlington, Virginia (US), August 13-15, 1991
Availability: Available from (1) as Paper EN 35984 ORA
Record Number: 199110420 / Last updated on: 1994-12-02
Original language: en
Available languages: en