A prediction-based data collection method in wireless sensor network using Kalman Filter

Abstract

Since data collecting is the primary functionality of sensor networks, it is crucial to provide an appropriate method for information gathering in face of poor hardware equipments. The core challenge in these applications is to minimize energy consumption during the periodic data collection process. We then propose a distinctive prediction-based approach profiting from the Kalman filter to deal with the situation. In this scheme, we first take advantage of the Kalman smoothing operation to obtain the optimal historical system state according to the feasible measurements, based on which we reasonably evaluate the future estimation followed by variable numbers of prediction steps. By means of the proposed algorithm, we need not pay a hefty premium for continuous data transmission from sources to the center in a time-varying network system, whereas we just adopt the estimations within limited error tolerance as the actual sensed values. With trading relatively small amounts of computation cost for communication saving, this proposed algorithm demonstrates significant energy conservation performance in comparison with traditional methods and caching static data approach. Experiments built upon the real-world streams verify the effectiveness and validity of proposed methods. © 2011 ICIC International.