A cognitive framework for WSN based on weighted cognitive maps and Q-learning

Abstract

Due to the rising popularity of Wireless Sensor Networks (WSN), there has been an increasing demand for more challenging applications. Generally speaking, such applications demand the support of Quality of Service (QoS) constraints and often multiple conflicting objectives. To address these requirements, network management becomes a critical task. However, single layer and cross-layer techniques will probably not be sufficient. Fortunately, cognitive techniques have shown great potential in improving network performance and achieving the end-to-end goals. In this paper, cognitive solutions for resource management in WSN are proposed to address the needs of challenging applications. The solutions proposed include a reasoning machine designed for the first time using the mathematical tool known as Weighted Cognitive Maps (WCM), and a learning protocol designed using the Q-learning algorithm. The WCM has powerful inference properties and can model complex systems using their underlying causal relationships. Thus, it can consider multiple conflicting objectives with low complexity. On the other hand, the learning protocol uses Q-learning to build a knowledge base that is used to enhance the performance of the WCM reasoning machine and address the end-to-end goals of the applications directly. Extensive computer simulations are used to evaluate the performance of the cognitive framework. The results show that it can efficiently consider multiple conflicting goals and constraints, thus outperforming its state-of-the-art counterparts.