SINGER: System for implementing new genetic evasion rules.

摘要

Genetics-Based Machine Learning (GBML) systems are broadly applicable to many complex search domains. SINGER is a program using one form of GBML architecture, a production rule classifier system, to discover new strategies in the domain of differential pursuit games. The classifier system uses a genetic algorithm as its core for rule discovery, and uses the genetic operators of reproduction, crossover, and mutation as search operators to assist in finding the best rule strategies to maximize performance of an evader in the pursuit game. Both agents, the pursuer and evader, are given realistic physical constraints such that the evader simulates a high performance jet aircraft, and the pursuer represents a much higher speed missile attempting to destroy the evader. The SINGER system is described in detail, followed by the results of the system's trials and their statistical significance.