Dynamic cooperative co-evolutionary automated mobile sensor deployment via localized fitness evaluation.

摘要

A wireless sensor network is a self-organized network consisting of a large number of small sensor nodes distributed over an area of interests. Such networks are capable of observing and sensing the environment, and sending the collected data to a data sink for further processing. Sensors must be deployed before they can provide useful data. Therefore the deployment of static or mobile sensors is an important basis for sensor networking.;Automated mobile sensor deployment of a wireless sensor network has a significant impact on the network performance, such as network sensing coverage, communication or mobile costs, and connectivity. Due to the small size of sensors, they are equipped with small batteries and have low-power computing and communication resources. The lifetime of a sensor is determined by its battery life and it can not operate for an infinite amount of time. Therefore, a good deployment yields a high utilization of power resources.;In this thesis, we propose an innovative cooperative co-evolutionary computation framework, Localized Distributed Coevolution (LODICO), to optimize the automated sensor deployment with arbitrary initial positions. LODICO is a fully distributed and localized algorithm. It can be executed on all sensors of the network in parallel. Meanwhile the information exchange has to be done locally as each sensor can only communicate with those within a distance. Further, we extend LODICO to LODICO/D to provide dynamic interaction to neighboring computing agents during the evolutionary process. It models the potential local interactions between computing agents, and uses the the imaginary neighboring movements to improve its local fitness and to help escaping from local optima.;This thesis is a powerful extension work to the traditional Cooperative Coevolutionary Algorithm. One feature of it is the utilization of local fitness to achieve a global optimum, which makes co-evolutionary algorithms applicable to localized distributed environments, such as network computing. Another salient feature is that the proposed algorithms can adjust and adapt the frequent dynamic change of network structures due to the position changes or failures of computing agents. LODICO/D incorporates LODICO with mode D to help to escape local optima. Mode D creates the third feature of imaginary collaboration with the neighboring computing agents during the evolutionary process to improve its local fitness. Our experiments show that LODICO and LODICO/D are effective in obtaining good solutions under such dynamic, distributed, and localized condition constraints.