Evolutionary generation of fuzzy knowledge bases for diagnosing monitored railway systems

Abstract

Classical approaches when building diagnosis and monitoring systems are rule-based systems, which allow the representation of existing knowledge by using rules. There are several techniques that facilitate this task, such as fuzzy logic, which allows knowledge to be modeled in an intuitive way. Nevertheless, sometimes it is not easy to define the fuzzy rule set that represents the knowledge from a certain domain. To overcome this drawback, an evolutionary system based on a grammar guided genetic programming technique for the automatic generation of fuzzy knowledge bases has been employed in diagnosing monitored railway networks. This system employs a grammar-based initialization method and both, grammar-based crossover and mutation operators, to achieve well balanced exploitation and exploration capabilities of the search space, assuring high convergence speed and low chance of getting trapped in local optima. Tests have been carried out in a real-world train monitoring domain, in which a sensor network is periodically monitoring critical train components. Results achieved show that this evolutionary system accomplishes an automatic knowledge discovery process, which is able to build a fuzzy rule base that represents the expert knowledge extracted from the domain of the detection of abnormal train conditions.