Data-driven virtual sensing for spatial distribution of temperature and humidity

Visibility of thermal conditions at various locations in a building opens up several avenues in building automation and control. For instance, spatial thermal relationships can be exploited in large spaces to optimize occupant comfort at minimum energy. However, pervasive sensing can significantly increase the deployment and maintenance costs for building monitoring systems. Prior work shows that thermal measurements in indoor environments observe certain spatiotemporal relationships that can be leveraged to observe the spatial thermal distribution with less number of sensors. This paper discusses the study of the temperature and humidity measurements collected across three different settings in an academic campus: a laboratory, a conference hall and an office. A systematic, simple, practical approach is presented to identify locations to deploy physical sensors in each of these set-ups and design virtual sensors that estimate temperature and humidity in the locations without sensors. The proposed approach leverages correlation and appropriately defined accuracy metrics to design effective virtual sensors and thus minimize the required sensor interventions. Results demonstrate how the virtual sensors are viable alternatives to allow for maintenance free monitoring of indoor environments. Meaningful insights pertaining to actual field implementation of the approach are also discussed.