WSNs Self-Calibration Approach for Smart City Applications Leveraging Incremental Machine Learning Techniques

The diffusion of the Internet of Things paradigm, in the last few years, has led to the need of deploying and managing large-scale Wireless Sensor Networks (WSNs), composed by a multitude of geographically distributed sensors, like the ones needed for Smart City applications. The traditional way to manage WSNs is not suitable for this type of applications, because manually managing and monitoring every single sensor would be too expensive, time consuming and error prone. Moreover, unattended sensors may suffer of several issues that progressively make their measures unreliable and consequently useless. For this reason, several automatically techniques have been studied and implemented for the detection and correction of measurements from sensors which are affected by errors caused by aging and/or drift. These methods are grouped under the name of self-calibration techniques. This paper presents a distributed system, which combines an incremental machine learning technique with a non-linear Kalman Filter estimator, which allows to automatically re-calibrate sensors leveraging the correlation with measurements made by neighbor sensors. After the description of the used model and the system implementation details, the paper describes also the proof-of-concept prototype that has been built for testing the proposed solution.