Dynamic power management of an embedded sensor network based on actor-critic reinforcement based learning

Wireless sensor networks (WSNs) have gained tremendous popularity in recent years due to the wide range of applications envisioned - ranging from aerospace and defense to industrial and commercial. Although limited by communication and energy constraints, the low cost, small sensor nodes lend themselves to be deployed in large numbers to form a network with high spatial distribution. The overall effectiveness of the sensor network depends on how well the mutually contradicting objectives of conserving the limited on-board battery power and keeping the sensors awake for stimuli, are managed. In this paper, we have proposed an actor-critic based reinforcement learning mechanism that can be practically implemented on an embedded sensor with limited memory and processing power. Specifically, the contribution of this paper is the development of the value function (or critic/reinforcement function) that is implemented on each sensor node which aids in dynamic power scheduling based on different situations. The effectiveness of the proposed method has been demonstrated with real world experiments.