[论文解读] Assessing the Value of Complex Refractive Index and Particle Density for Calibration of Low-Cost Particle Matter Sensor for Size-Resolved Particle Count and PM2.5 Measurements

Ching-Hsuan Huang, Jiayang He+3 more|arXiv (Cornell University)|Jun 6, 2021

Air Quality Monitoring and Forecasting参考文献 50被引用 18

一句话总结

本研究通过使用标准多分散颗粒物的受控气溶胶测试，开发并评估了Plantower PMS A003低成本PM传感器的校准模型。结果表明，不考虑颗粒物性质调整的线性校准模型可使数浓度的归一化平均绝对误差（NMAE）低于4.0%；而结合复折射率（CRI）、密度和非线性的模型显著降低了质量浓度（PM2.5、PM10）的误差，尤其在野火烟雾或工业环境等高浓度条件下效果显著。

ABSTRACT

Commercially available low-cost particulate matter (PM) sensors provide output as total or size-specific particle counts and mass concentrations. These quantities are not measured directly but are estimated by the original equipment manufacturers' (OEM) proprietary algorithms and have inherent limitations since particle scattering depends on their composition, size, shape, and complex index of refraction (CRI). Hence, there is a need to characterize and calibrate their performance under a controlled environment. We present calibration algorithms for Plantower PMS A003 sensor as a function of particle size and concentration. A standardized experimental protocol was used to control the PM level, environmental conditions and to evaluate sensor-to-sensor reproducibility. The calibration was based on tests when PMS A003 were exposed to different polydisperse standardized testing aerosols. The results suggested particle size distribution from PMS A003 was shifted compared to reference instrument measures. For calibration of number concentration, linear model without adjusting aerosol properties corrects the raw PMS A003 measurement for specific size bins with normalized mean absolute error within 4.0% of the reference instrument. Although the Bayesian Information Criterion suggests that models adjusting for particle optical properties and relative humidity are technically superior, they should be used with caution as the particle properties used in fitting were within a narrow range for challenge aerosols. The calibration models adjusted for particle CRI and density account for non-linearity in the OEM's mass concentrations estimates and demonstrated lower error. These results have significant implications for using PMS A003 in high concentration environments, including indoor air quality and occupational/industrial exposure assessments, wildfire smoke, or near-source monitoring scenarios.

研究动机与目标

• 为解决低成本PM传感器在测量粒径分辨的颗粒物数浓度和质量浓度方面缺乏可靠校准的问题。
• 评估颗粒物光学性质（特别是复折射率CRI和密度）对传感器精度的影响。
• 开发并验证可在不同颗粒物浓度和环境条件下提升传感器性能的校准模型。
• 评估在高PM水平下考虑OEM估算质量浓度非线性响应的必要性，尤其是在高浓度环境中。
• 为室内、职业和近源监测场景提供可标准化的传感器校准协议。

提出的方法

• 在气溶胶舱内开展受控实验室实验，使用四种标准多分散测试气溶胶（ATD、PSL、NaCl和S02），覆盖0–1000 #/cm³的浓度范围。
• 使用TSI空气动力学颗粒物分析仪（APS）作为基准标准，测量参考的颗粒物数浓度和质量浓度。
• 应用线性、多项式及混合模型，结合CRI、颗粒物密度和相对湿度（RH）对OEM传感器输出进行校正。
• 使用贝叶斯信息准则（BIC）比较模型拟合度并选择最优模型，同时计算归一化平均绝对误差（NMAE）用于验证。
• 评估传感器间的一致性及其在全浓度范围（0–1000 #/cm³）和低浓度范围（<100 #/cm³）下的性能表现。
• 通过回归分析生成PM1、PM2.5和PM10的校准方程，分别针对数浓度和质量浓度建立独立模型。

实验结果

研究问题

• RQ1 PMS A003传感器在不同颗粒物尺寸和浓度下，其原始数浓度和质量浓度输出与参考测量值相比如何？
• RQ2 颗粒物性质（如复折射率CRI和密度）在多大程度上影响低成本PM传感器测量的准确性？
• RQ3 包含CRI、密度和RH的校准模型是否能超越简单线性校正，显著提升质量浓度估算的准确性？
• RQ4 传感器在高浓度与低浓度环境下的性能表现有何差异，特别是在野火烟雾或室内污染等场景中？
• RQ5 哪种校准模型结构（线性、多项式或结合物理参数的混合模型）在实际部署中能实现准确性和可靠性的最佳平衡？

主要发现

• PMS A003传感器的原始数浓度测量值普遍低于APS参考值，且粒径分布出现偏移。
• 不考虑CRI或RH调整的简单线性校准模型，PM1的NMAE为3.11%，PM2.5的NMAE为4.53%，所有粒径区间的数浓度NMAE均低于4.0%。
• 结合CRI和颗粒物密度的模型，使PM2.5的NMAE从线性模型的4.53%降低至2.33%（多项式+ CRI + 密度），显著提升了质量浓度估算的准确性。
• 贝叶斯信息准则（BIC）偏好包含CRI、密度和RH的模型，表明其统计拟合更优，尽管这些模型基于的气溶胶类型范围较窄。
• 在高浓度下观察到OEM质量浓度估算存在非线性响应，且最佳校正效果由包含物理参数的多项式模型实现。
• 包含CRI、密度和非线性的PM2.5和PM10校准模型NMAE最低（分别为2.33%和2.61%），证实了在高浓度环境中进行此类调整的必要性。