Environmental sensors must be calibrated regularly in order to correctly take into account construction-related and external influences in the interpretation of the raw data and to correct them accordingly. Only then can any meaning be ascribed to the measurement data. Low-cost sensors are preferably used in Citizen Science applications, whereby a lengthy calibration with an expensive reference device usually contradicts the low-cost and simple conception of many Citizen Science applications. A measurement network, such as the one established in the SmartAQnet project, offers here the possibility of fully automated mobile calibration of sensors due to its spatial as well as temporal density of reference stations. On this unique dataset, we want to show that such an automated, mobile calibration, supported by a tightly meshed network of reference stations, can deliver qualitatively usable results. This allows SmartCities to closely monitor their air quality with low-cost, low-cost sensors by keeping them in permanent live contact with an underlying, much thinner network of high-quality reference stations.

We want to develop algorithms that automatically calibrate the sensors using machine learning. Due to the complex, non-linear influence of environmental factors, AI methods are excellently suited to achieve stable results, while the size of the data set protects against overfitting. At the same time, we want to develop a concept on how the calibration could be implemented in a live system.

Start/End

10/2020 – 05/2021

Contact

Paul Tremper (email: tremper(at)teco.edu)