Calibrating Low-Cost Sensors by a Two-Phase Learning Approach for Urban Air Quality Measurement

Abstract

Urban air quality information, e.g., PM2.5 concentration, is of great importance to both the government and society. Recently, there is a growing interest in developing low-cost sensors, installed on moving vehicles, for fine-grained air quality measurement. However, low-cost mobile sensors typically suffer from low accuracy and thus need careful calibration to preserve a high measurement quality. In this paper, we propose a two-phase data calibration method consisting of a linear part and a nonlinear part. We use MLS (multiple least square) to train the linear part, and use RF (random forest) to train the nonlinear part. We propose an automatic feature selection algorithm based on AIC (Akaike information criterion) for the linear model, which helps avoid overfitting due to the inclusion of inappropriate features. We evaluate our method extensively. Results show that our method outperforms existing approaches, achieving an overall accuracy improvement of 16.4% in terms of PM2.5 levels compared with state-of-the-art approach.

Evaluation and calibration of a low-cost particle sensor in ambient conditions using machine-learning methods

“…The reference method used in this study was a Dylos DCI1700 device, which is not a US EPA federal reference method (FRM) or FEM. Loh and Choi (2019) trained and tested the SVM, K-nearest neighbor, RF, and XGBoost machine-learning algorithms to calibrate PM 2.5 sensors using 319 hourly data points. XGBoost archived the best performance with an RMSE of 5.0 µg m −3 .…”

“…Most previous studies focused on calibrating gas sensors (Maag et al, 2018). There are two studies on PM sensor calibrations using machine learning, but they used a short-term dataset that did not include seasonal changes in ambient conditions (Lin et al, 2018;Loh and Choi, 2019). The shortcomings of the two studies were discussed above.…”

Evaluation and Calibration of a Low-cost Particle Sensor in Ambient Conditions Using Machine Learning Technologies

“…Calibrations of PM2.5 sensors were also reported in recent studies. Lin et al (2018) performed two-step calibrations for PM2.5 sensors using 236 hourly data collected on buses and road cleaning vehicles. The first step was to construct a linear model, and the second step used RF machine learning for further calibration.…”

An improved low-power measurement of ambient NO<sub>2</sub> and O<sub>3</sub> combining electrochemical sensor clusters and machine learning

“…Significant uncertainty in measurements is introduced because individual sensors each have a unique response to simple environmental conditions such as humidity and temperature Moltchanov et al, 2015). This can lead to a relatively high degree of inter-sensor variability and response drift (Lewis et al, 2016;Spinelle et al, 2017) over durations as short as a few hours (Jiao et al, 2016;Masson et al, 2015), rendering in-laboratory calibrations (where the interfering variables are controlled or non-existent) ineffective . Electrochemical (EC) sensors can display some chemical cross-interferences with other pollutants that are likely to be present (Mead et al, 2013;Lewis et al, 2016;Masson et al, 2015), and accounting for these can be difficult when the relative concentration ratios of the target measurand and interferences change.…”

“…This can lead to a relatively high degree of inter-sensor variability and response drift (Lewis et al, 2016;Spinelle et al, 2017) over durations as short as a few hours (Jiao et al, 2016;Masson et al, 2015), rendering in-laboratory calibrations (where the interfering variables are controlled or non-existent) ineffective . Electrochemical (EC) sensors can display some chemical cross-interferences with other pollutants that are likely to be present (Mead et al, 2013;Lewis et al, 2016;Masson et al, 2015), and accounting for these can be difficult when the relative concentration ratios of the target measurand and interferences change. Metal oxide sensors (MOSs) often lack selectivity and provide only a rough bulk measure of a particular pollutant class such as VOCs, and the responses generated can depend on the chemical content of the mixture presented to the sensor.…”