Transactions on Machine Learning and Data Mining (ISSN: 1865-6781)
Volume 8 - Number 2 - October 2015 - Pages 77-100
Traceable Measurements using Sensor Networks
Alistair B. Forbes
National Physical Laboratory, UK
Abstract
National Metrology Institutes such as the National Physical Laboratory provide the scientific and technical framework to ensure measurements are traceability to standard units or references. Many of the procedures developed for ensuring traceability involve measurements in laboratory conditions, whereby environmental factors such as temperature are closely controlled, and using instruments whose responses are well-characterised by validated models. However, many societal challenges relating climate and environment, energy and sustainability, health and well-being necessarily involve measurements outside laboratory conditions. In such circumstances, the mathematical and statistical modelling tools need to be strengthened so that concepts of traceability, uncertainty evaluation and calibration can also be applied outside the laboratory. In particular, we need to develop tools that account for uncertainties associated with the models of the system, for we can no longer be certain that we understand all aspects of the response of a potentially complex system. One such challenge is the measurement of air quality. Air quality has a significant impact on quality of life and many regulations now apply to controlling pollution. In order to ensure compliance to regulations, it is necessary to monitor air quality. However, measurement of air quality can only be made at a finite number of sites while the regulations apply to the complete air quality field. In this paper, we consider approaches for accounting for model uncertainty and the use of Gaussian processes to model the temporal and spatial and correlation in order to estimate the air quality field and its associated uncertainty.
Keywords:traceability, sensor networks, uncertainty, Gaussian processes
Download Paper
(1232 KB)