Environmental Sensor Drift Correction using Wavelet-mFCM-FIS Architecture: An Unsupervised Machine Learning Approach

Abstract

A novel hybrid sensor informatics architecture based on Discrete Wavelet Transform (DWT), Fuzzy logic based clustering (FCM) and Fuzzy Rule based Inference system (FIS) has been investigated and proposed to estimate dynamic generic sensor drift during physical sampling. DWT has been used for sensor pre-processing; data dimension reduction and feature extraction from sensor time series, where as DWT-FCM based approach has been used to estimate the cumulative drift in our sensory system. An intelligent generalized algorithm using multiple FCM maps (m-FCM) has been proposed to implement the drift correction on any future unknown data sets. In the next stage a newly proposed DWT-FCM-FIS architecture has been tested on multisensory environmental data sets covering three consecutive years (2009 -11) to predict probable month (environmental time of the year) with up to 93% accuracy. This novel hybrid data processing architecture have been implemented and tested upon a real time estuary sensory platform of temperature, conductivity and salinity sensors that have been deployed to monitor the Derwent Estuary in Hobart, Australia. This newly proposed approach could be an adaptive solution to tackle drifts in the sensor networks and improve overall monitoring quality.