Wireless sensor networks (WSNs) are an attractive platform for monitoring and measuring physical phenomena. WSNs typically\nconsist of hundreds or thousands of battery-operated tiny sensor nodes which are connected via a low data rate wireless network. A\nWSNapplication, such as object tracking or environmentalmonitoring, is composed of individual taskswhichmust be scheduled on\neach node. Naturally the order of task execution influences the performance of theWSN application. Scheduling the tasks such that\nthe performance is increased while the energy consumption remains low is a key challenge. In this paper we apply online learning to\ntask scheduling in order to explore the tradeoff between performance and energy consumption. This helps to dynamically identify\neffective scheduling policies for the sensor nodes. The energy consumption for computation and communication is represented\nby a parameter for each application task. We compare resource-aware task scheduling based on three online learning methods:\nindependent reinforcement learning (RL), cooperative reinforcement learning (CRL), and exponential weight for exploration\nand exploitation (Exp3). Our evaluation is based on the performance and energy consumption of a prototypical target tracking\napplication.We further determine the communication overhead and computational effort of these methods.\n1. Introduction\nA wireless sensor network (WSN) is an attractive platformfor\nvarious applications including target tracking, environmental\nmonitoring, data aggregation, and smart environments. The\napplication is composed of tasks which need to be executed\nduring the operation on the sensor nodes. The sensor nodes\nare typically supplied by batteries and thus pose strong\nlimitations not only on energy but also on computation,\nstorage, and communication capabilities [1ââ?¬â??4].\nThe scheduling of the individual tasks has a strong\ninfluence on the achievable performance and energy consumption.\nWSNs operate in a dynamic environment where\nthe need for adaptive and autonomous task scheduling is\nwell recognized [5]. Since it is not possible to schedule the\ntasks a priori, online, and resource-aware task scheduling\nis required for a WSN. For determining the next task to\nexecute, the sensor nodes need to consider the impact of each\navailable task on the energy budget and the applicationââ?¬â?¢s performance.\nThere is tradeoff between application performance\nand resource consumption, and the task scheduler of the\nnode should be able to adapt to changes in the environment
Loading....