Skip to Main Content
Learning automata (LA) is an abstract model which can be used to guide action selection at any stage of a system by past actions and environment responses to improve some overall performance function. The use of intelligent algorithms based on learning automata can be efficient for traffic control. However, these learning schemes have been focused only to unimodal routing problem in connection oriented networks. The field of ant colony optimization (ACO) models real ant colony behavior using artificial ant algorithms and find its application in a whole range of optimization problems. Ant algorithms experimentally prove to work very well in static and dynamic optimization problems and match perfectly with some model of interconnected LA. In this paper, an adaptive multi-agent routing algorithm called LA-AntNet is proposed for both source and non-source routing in communication networks. In this algorithm, mobile ant agents form AntNet routing system (Dicaro & Dorigo 1998) are combined to a system of static distributed LA agents, statically connected to the network nodes and directly responsible for routing decisions. The mobile ant agents improve local decisions of LA and adapt it with network conditions by moving over the network and collecting information about traffic distribution. In this algorithm the decision policy of LA is modified to involve a heuristic parameter which is a suggestion coming from ACO field and can guide learning process and improve convergence results. The proposed algorithm is implemented on several topologies to obtain performance metrics namely, throughput and total delay. The results are compared to the ones obtained from AntNet and a learning automata technique.
Date of Conference: 22-24 Jan. 2009