Drift Calibration Using Constrained Extreme Learning Machine and Kalman Filter in Clustered Wireless Sensor Networks

Abstract

Wireless sensor networks (WSNs) have been increasingly applied for environmental monitoring in recent years. However, the sensor data drift is a serious issue affecting the reliability of monitoring system. Based on the assumption that the neighboring nodes have correlated measurements, this paper presents a novel algorithm using constrained extreme learning machine and Kalman filter (CELM-KF) for tracking and calibrating drift of sensor data. CELM-KF has two phases: training phase and calibrating phase. In the training phase, the wavelet denoising method is used for data preprocessing. Then the cluster head trains the model of the constrained extreme learning machine (CELM) using the measurements of its neighbors to obtain the prediction data of the target sensor. In the calibrating phase, we track and correct the data drift of target sensor using Kalman filter. To evaluate the performance of CELM-KF, simulation experiments on different datasets are conducted, and two parameters including decision coefficient and mean square error (MSE) of CELM-KF are compared with those of existing algorithms. The simulation results show that CELM-KF can successfully calibrate the sensor data drifts.