Temporal-Correlation-Aware Dynamic Self-Management of Wireless Sensor Networks

In wireless sensor networks (WSNs), sensor observations are spatiotemporally correlated, and that correlation signifies redundancy among the observations. Spatial correlation is primarily employed to estimate the minimum number of event-monitoring nodes. However, an event-monitoring node can intelligently exploit the temporal correlation between its observations to adapt with its dynamic surroundings. This self-adaptation helps resource-constrained nodes to enhance their performance by saving battery power and maintaining the quality of transmitted data. In WSNs, the sensor nodes switch between the active and sleep states to conserve energy. Using temporal correlation, a node can dynamically estimate the appropriate sleep duration, which is an important parameter for a node to adapt with its dynamic surroundings in an energy-efficient manner. In this paper, dynamic Bayesian network and entropy are used to estimate utility of observations. Moreover, a node estimates temporal correlation between its consecutive observations by mutual information. Further, the sensor nodes calculate appropriate sleep duration and control their communications at a particular time instant on the basis of estimated temporal correlation. A reinforcement-learning-based approach is used, in a distributed manner, to calculate the optimum sleep duration. Extensive simulation studies show that the proposed approach performs more efficiently in terms of energy conservation, energy utilization, and data accuracy than the benchmark schemes.