Development of Energy-efficient Sensor Networks by Minimizing Sensors Numbers with a Machine Learning Model

With the increasing demand to construct sensor networks for a smart IoT (Internet of Things) world, numerous sensors with sensing and communication capabilities are expected to be deployed in the future. Thanks to the development of hardware manufacture technology, relatively small IoT smart sensors are now commercially available and cost-effective. However, the total power required by operating these sensors is expected to be enormous, due to their large number and frequent activity. Removing “unneeded sensors” is the most direct way to reduce the power consumption of sensor networks. Here, “unneeded sensors” refers to those that can be placed in sleep mode, or even be removed from the network topology entirely, without serious impact on the overall networks data processing performance. In this paper, we report the development of an energy-efficient sensor network by using a machine learning model to determine the actual necessity of all the sensors in a sensor network. Machine learning model is introduced to identify unneeded sensors by comparing the data from neighboring sensors to that from the potentially unneeded ones. For identifying unneeded sensors, different strategies with different computational complexity are also proposed. Numerical experiments conducted in two real indoor environments verify that our proposed scheme can reduce the total number of active sensors by around 1/3, while maintaining more than 90% of the original high monitoring performance of the sensor network.