Evaluation Calibration Study And Application Of Low-cost Light Scattering Particulate Matter Sensor

In recent years,with the increase in public health and environmental awareness,there is a growing demand for monitoring particulate matter concentrations such as PM_2.5.The use of low-cost particulate matter concentration sensors based on the light scattering principle is a promising monitoring tool.However,it is questionable whether the monitoring data output from these sensors can meet the expectations of users compared to laboratory-level monitoring instruments.In addition,current research is relatively lacking in the analysis of the application characteristics of low-cost PM sensors in high concentration monitoring environments.This thesis takes the evaluation of low-cost particulate matter concentration sensors as a starting point,and summarizes the application characteristics of the Plantower PMSA003（referred to as G10）sensor in three concentration environments under the guidance of sensor data quality evaluation guidelines.To address the underestimation of PM_2.5 concentration by the G10 sensor in high concentration environments,this thesis establishes a data calibration model and analyzes the utility of four linear calibration methods and five machine learning calibration models,and finally selects the optimal calibration model.Then,this thesis uses the screening building of coal processing plant as a field application scenario and analyzes the particle concentration,particle size mass distribution and air quality variation characteristics at dust production points on different floors.Finally,this thesis validates the established calibration model with the field data of G10 sensors,and establishes a PM_2.5 monitoring network system based on the calibrated G10 sensors to study the spatial and temporal distribution patterns of PM_2.5 on different floors of the screening building of the coal processing plant,and this thesis draws the following conclusions:（1）The accuracy,inter-group correlation and linearity between the G10 sensor and the reference instrument Dust Trak were good in the monitoring environment at low concentrations,and the error performance was in accordance with the standard and the application performance was as expected;however,there was a monitoring error between the G10 sensor and the reference instrument at high concentrations（104-2360μg/m³）that was difficult to ignore（mean NRMSE of 103.5%）.However,the higher R²and intergroup correlations indicate the possibility of data calibration;the individual sensor（G10-2）was subjected to high ambient monitoring interference and out of calibration after use in the high concentration environment.（2）The accuracy of the predictions made by the multivariate linear model for different peak particulate concentrations was not stable.XGBoost machine learning regression model was the optimal calibration model in this thesis.The order of performance comparison of the five models is XGBoost>Light GBM>random forest>KNN>linear model B>support vector machine.（3）The characteristics of particle concentration changes at different dust-producing points on different floors show that the trends of particle concentration changes after the screening machine and belt operation are gradual increase and increase,then decrease,then increase,respectively.The particle size occupying the largest mass of particles is 5-10μm.the mass share of PM_2.5 particles is about 17-20%.After the calibration of the XGBoost model,the G10 sensor data quality can meet the demand.the results of the spatio-temporal pattern analysis of PM_2.5 mass concentration show that before the machine operation,the overall PM_2.5 mass concentration on the third floor of the screening building is close to that in outdoor air,while the negative floor of the screening building is much higher than outdoor.The peak PM_2.5concentrations at the two monitoring locations were about 950-1050μg/m³.the PM_2.5mass concentration located at the well-ventilated location was about one-third of the peak PM_2.5 concentration in the whole space.The results show that good ventilation effectively reduces the PM_2.5 concentration at local locations in the screening building to improve the environment.