Evolutionary Adaptation in Complex Systems Using the Example of a Logistics Problem

There is a trend towards networked and distributed systems, complicating the design, construction and maintenance of complex software systems without any central control instance. This paper provides a solution to the problem of modeling and adaptation of complex adaptive systems (CAS). Here an agent based model is introduced to describe complex adaptive systems in order to study self-organization and adaptivity. An evolutionary computation (EC) enabled multi-agent system (MAS) is used that exploits the flow of money or energy in order to realize distributed fitness calculation. Our approach uses fully decentralized operators for reproduction like mutation, recombination and selection, regulated by market mechanisms. This paper presents two general outcomes of our model: how adaptation occurs in the number and strategies of agents leading to an improvement at the system level. The novelty of this approach lies in the biology-inspired bottom-up adaptation method for decentralized systems that is intended to be general for a wide variety of artificial systems. As an example of complex adaptive systems, a logistics network is introduced and used to study the expected results.