RESUMEN
OBJECTIVE: The aim was to evaluate and optimize the performance of sensor monitors in measuring PM2.5 and PM10 under typical emission scenarios both indoors and outdoors. METHOD: Parallel measurements and comparisons of PM2.5 and PM10 were carried out between sensor monitors and standard instruments in typical indoor (2 months) and outdoor environments (1 year) in Shanghai, respectively. The optimized validation model was determined by comparing six machining learning models, adjusting for meteorological and related factors. The intra- and inter-device variation, measurement accuracy, and stability of sensor monitors were calculated and compared before and after validation. RESULTS: Indoor particles were measured in a range of 0.8-370.7 µg/m3 and 1.9-465.2 µg/m3 for PM2.5 and PM10, respectively, while the outdoor ones were in the ranges of 1.0-211.0 µg/m3 and 0.0-493.0 µg/m3, correspondingly. Compared to machine learning models including multivariate linear model (ML), K-nearest neighbor model (KNN), support vector machine model (SVM), decision tree model (DT), and neural network model (MLP), the random forest (RF) model showed the best validation after adjusting for temperature, relative humidity (RH), PM2.5/PM10 ratios, and measurement time lengths (months) for both PM2.5 and PM10, in indoor (R2: 0.97 and 0.91, root-mean-square error (RMSE) of 1.91 µg/m3 and 4.56 µg/m3, respectively) and outdoor environments (R2: 0.90 and 0.80, RMSE of 5.61 µg/m3 and 17.54 µg/m3, respectively), respectively. CONCLUSIONS: Sensor monitors could provide reliable measurements of PM2.5 and PM10 with high accuracy and acceptable inter and intra-device consistency under typical indoor and outdoor scenarios after validation by RF model. Adjusting for both climate factors and the ratio of PM2.5/PM10 could improve the validation performance.
RESUMEN
With the development of monitoring technology, the variety of ozone precursors that can be detected by monitoring stations has been increased dramatically. And this has brought a great increment of information to ozone prediction and explanation studies. This study completes feature mining and reconstruction of multi-source data (meteorological data, conventional pollutant data, and precursors data) by using a machine learning approach, and built a cross-stacked ensemble learning model (CSEM). In the feature engineering process, this study reconstructed two VOCs variables most associated with ozone and found it works best to use the top seven variables with the highest contribution. The CSEM includes three base models: random forest, extreme gradient boosting tree, and LSTM, learning the parameters of the model under the integrated training of cross-stacking. The cross-stacked integrated training method enables the second-layer learner of the ensemble model to make full use of the learning results of the base models as training data, thereby improving the prediction performance of the model. The model predicted the hourly ozone concentration with R2 of 0.94, 0.97, and 0.96 for mild, moderate, and severe pollution cases, respectively; mean absolute error (MAE) of 4.48 µg/m3, 5.01 µg/m3, and 8.71 µg/m3, respectively. The model predicted ozone concentrations under different NOx and VOCs reduction scenarios, and the results show that with a 20 % reduction in VOCs and no change in NOx in the study area, 75.28 % of cases achieved reduction and 15.73 % of cases got below 200 µg/m3. In addition, a comprehensive evaluation index of the prediction model is proposed in this paper, which can be extended to any prediction model performance comparison and analysis. For practical application, machine learning feature selection and cross-stacked ensemble models can be jointly applied in ozone real-time prediction and emission reduction strategy analysis.