A role-based imitation algorithm for the optimisation in dynamic fitness landscapes

Organic Computing (OC) deals with technical systems consisting of a large number of system elements that can adapt their structure and behaviour to the operational environment in order to accomplish a given goal. In this context, self-adaptation is a key aspect that allows a system to perform in (possibly dynamic) environments without intervention from outside. Establishing self-adaptation in technical systems requires adequate optimisation algorithms that can find high-quality solutions in an acceptable period of time. In this paper, we present a new population-based optimisation algorithm (Role Based Imitation algorithm - RBI) that can be used to establish self-adaptation in OC systems with dynamic fitness landscapes. RBI proposes a novel role assignment strategy for exploring and exploiting agents to find high-quality solutions within a short period of time (i.e., with high convergence speed). We compare RBI with Differential Evolution (DE), Particle Swarm Optimisation (PSO), Evolutionary Algorithm (EA) and Simulated Annealing (SA) in static and dynamic fitness landscapes. Our experiments show that RBI performs better than the competing algorithms especially in noisy and highly dynamic environments.