CN-120539259-B - Online detection system and method for acidity of atmospheric aerosol

CN120539259BCN 120539259 BCN120539259 BCN 120539259BCN-120539259-B

Abstract

The application discloses an online detection method and system for the acidity of atmospheric aerosol, which belong to the field of atmospheric pollutant detection and comprise the steps of acquiring aerosol data and meteorological data of an atmospheric observation station, preprocessing the acquired data, calculating the preprocessed data by adopting a Metastable-Forward mode thermodynamic model ISORROPIA II to obtain an aerosol acidity theoretical value pH, constructing a plurality of machine learning models by taking preset measurement parameters as characteristic variables and taking the aerosol acidity theoretical value pH as target variables, selecting a machine learning model with optimal performance through cross verification, acquiring real-time aerosol component data by adopting mass spectrum analysis, acquiring current meteorological data, and calculating the real-time aerosol acidity by utilizing the optimal machine learning model. Aiming at the poor real-time performance of atmospheric aerosol acid detection by physical sampling, the real-time performance is improved by combining a thermodynamic model with machine learning and the like without replacing detection consumables.

Inventors

Zhan Puning
WANG JIAPING

Assignees

南京大学

Dates

Publication Date: 20260508
Application Date: 20250526

Claims (7)

1. An on-line detection method of the acidity of atmospheric aerosol is characterized by comprising the following steps: The method comprises the steps of acquiring aerosol data and meteorological data of an atmosphere observation site, preprocessing the acquired data, wherein the aerosol data comprise mass concentrations of ammonium salt, sulfate, nitrate and chloride, the meteorological data comprise atmospheric temperature and relative humidity, and preprocessing comprises the steps of sequencing the acquired aerosol data and the meteorological data according to time stamps of all the data in time sequence and aligning time nodes to obtain a data time sequence with consistent time; Calculating the preprocessed data by adopting a Metastable-Forward model thermodynamic model ISORROPIA II to obtain an aerosol acidity theoretical value pH; Taking preset measurement parameters as characteristic variables, taking an aerosol acidity theoretical value pH as a target variable, constructing a plurality of machine learning models, and selecting the machine learning model with optimal performance through cross verification, wherein the preset measurement parameters comprise local time, atmospheric temperature, relative humidity, ammonium salt mass concentration, sulfate mass concentration, nitrate mass concentration and chloride mass concentration; Acquiring real-time aerosol component data by mass spectrometry, acquiring current meteorological data, and calculating real-time atmospheric aerosol acidity by using an optimal machine learning model; taking a preset measurement parameter as a characteristic variable, taking an aerosol acidity theoretical value pH as a target variable, constructing a plurality of machine learning models, and selecting the machine learning model with optimal performance through cross verification, wherein the method comprises the following steps of: Establishing a time sequence data set of which the characteristic variables correspond to the target variables one by one; constructing a plurality of machine learning models according to the time sequence data set; performing N-fold cross validation on each constructed machine learning model; verifying the performance of each machine learning model; selecting a model with optimal performance as an optimal machine learning model; acquiring real-time aerosol component data by mass spectrometry, acquiring current meteorological data, and calculating real-time atmospheric aerosol acidity by using an optimal machine learning model, wherein the method comprises the following steps of: Acquiring real-time aerosol data and meteorological data by using a mass spectrometer, and preprocessing the acquired data; taking the preprocessed data as input, calculating the acidity of the real-time atmospheric aerosol by using an optimal machine learning model through the following formula : Wherein, the Real-time atmospheric aerosol acidity for the ith data point, f is the optimal machine learning model, The j-th eigenvalue of the i-th data point.
2. The method for on-line detection of atmospheric aerosol acidity according to claim 1, characterized in that: machine learning models include linear regression, fine trees, integrated trees, neural networks, and gaussian process regression models.
3. The method for on-line detection of atmospheric aerosol acidity according to claim 2, characterized in that: The integrated tree comprises a bagged tree and a lifting tree, and the Gaussian process regression comprises a plurality of basic functions and kernel functions.
4. The method for on-line detection of atmospheric aerosol acidity according to claim 1, characterized in that: preprocessing the acquired data, including: identifying a time period without data record according to the data time sequence, and marking the positions of data points in the data period as non-number NaN; And detecting abnormal values by adopting a moving window method according to the data time sequence after marking the non-number NaN, calculating local median and local conversion absolute median MAD in N1 moving windows, and marking data points with the local median difference being greater than or equal to N2 times of the local conversion absolute median MAD as the non-number NaN to obtain preprocessed data.
5. The method for on-line detection of atmospheric aerosol acidity according to claim 4, characterized in that: The preprocessed data is calculated by adopting a Metastable-Forward model thermodynamic model ISORROPIA II to obtain an aerosol acidity theoretical value pH, which comprises the following steps: The preprocessed data is used as input, and the mass concentration of hydrogen ions in the aerosol is calculated by utilizing a Metastable-Forward model thermodynamic model ISORROPIA II And liquid water mass concentration LWC; by means of hydrogen ion mass concentration And liquid water mass concentration LWC, the aerosol acidity theoretical value pH is calculated by the following formula: 。
6. the method for on-line detection of atmospheric aerosol acidity according to any one of claims 1 to 5, characterized in that: After calculating the real-time atmospheric aerosol acidity, further comprising: (1) Analyzing the space-time distribution of the aerosol acidity according to preset measurement parameters and the calculated real-time atmospheric aerosol acidity; (2) Analyzing the transmission path of the atmospheric aerosol by using a Lagrange traceability model, and calculating the contributions of different air group source areas to the acidity of the aerosol at the observation station by using a concentration weight track method : Wherein, the For the contribution of grid i to the detected site j data, Is a track The size of the j data when it arrives at the detection site, Is a track The residence time of the j data in the grid i is determined, and N is the total track number; (3) Calculating a Shapley value by using a SHAP method based on a game theory; quantitative assessment of the contribution of preset measurement parameters to the acidity of the atmosphere by means of Shapley values : Wherein, the The ith aerosol acidity value obtained for the model inversion, Is a value of Shapley, and the value is a value of Shapley, Is a constant value, and is used for the treatment of the skin, For the j-th characteristic variable pair S refers to a subset of the set of input parameters in the ith calculation and contains no j variables, The influence on the model result when j feature variables are included in the model calculation and are not included; (4) And (3) analyzing the time-space change rule of the acidity of the atmospheric aerosol according to the analysis results of the steps (1) to (3).
7. An online detection system of the acidity of an atmospheric aerosol for carrying out the method of any one of claims 1 to 6, comprising: The data acquisition module acquires aerosol data and meteorological data of an atmospheric observation site and preprocesses the acquired data; The theoretical value module is used for calculating the preprocessed data by adopting a Metastable-Forward thermodynamic model ISORROPIA II to obtain an aerosol acidity theoretical value pH; the machine learning module is used for constructing various machine learning models by taking preset measurement parameters as characteristic variables and taking an aerosol acidity theoretical value pH as a target variable, and selecting a machine learning model with optimal performance through cross verification; The detection module is used for acquiring real-time aerosol data and meteorological data through the mass spectrum analyzer and detecting the acidity value of the real-time atmospheric aerosol by utilizing the optimal machine learning model ; Deduction module according to real-time atmospheric aerosol acidity value And obtaining the time-space variation rule of the acidity of the atmospheric aerosol by analyzing the time-space distribution, lagrange traceability analysis and SHAP analysis based on game theory.

Description

Online detection system and method for acidity of atmospheric aerosol Technical Field The application relates to the field of atmospheric pollutant detection, in particular to an online detection system and method for the acidity of atmospheric aerosol. Background Atmospheric aerosols are small solid or liquid particles suspended in the atmosphere, whose acidity (pH) is one of the most basic properties of an aerosol, with significant differences in different time-space environments. The acidity of the aerosol has various important effects on the environment, climate and human health, namely, the acidity can obviously promote the formation of secondary organic aerosol, change the light absorption characteristic of brown carbon aerosol, further influence the radiation balance, and change the hygroscopic growth characteristic and the surface tension of the aerosol by influencing the gas-liquid distribution balance and the phase separation process. The change rule of the acidity of the atmospheric aerosol is accurately measured and understood, and the method has important scientific significance and application value for environmental quality evaluation, climate change prediction and health risk evaluation. However, direct detection of aerosol pH in the ambient atmosphere presents a number of technical challenges. In the traditional research, the detection of the acidity of the atmospheric aerosol mainly depends on the extraction of deionized water in a laboratory after the aerosol is collected to a filter membrane, and then the analysis is carried out by utilizing the color development phenomenon of an acid-base indicator. The time resolution of the off-line method is generally low, a data point is usually obtained every 24 hours in actual normal state observation, the detection of the average acidity of the atmosphere is mainly carried out, and the real-time monitoring requirement is not met. In the partial encryption observation, the sampling time is controlled to switch the filter membranes to improve the time resolution by adding a plurality of groups of filter membranes, but the sampling filter membranes usually have longer interval time from the acquisition to the final detection, and a low-temperature environment is generally needed to be added to ensure the relative stability of volatile substances in the sample. Currently, the existing online in-situ detection technology of aerosol acidity is very few, and is mainly based on indicator color development, raman spectrum and other optical identification technologies (such as CN 116202919A and CN 119246489A). The technology has the main problems of low time resolution, high consumption of consumable materials, high measurement uncertainty, high uncertainty of the measurement based on an acid-base indicator, high relative humidity environment of a sampling interface, high detection condition limit, limit of a detection range, limit of a color change range of a single acid-base indicator (such as methyl yellow only covers pH=2.9 to 4.0), wide distribution of atmospheric aerosol acidity between pH=1 to 8, insufficient resolution, incapacity of meeting the accurate measurement requirement due to the fact that the measurement of the acid-base indicator is influenced by factors such as the configuration quality of the indicator, the detection precision of an optical signal, the ambient temperature and the like, limit of the detection condition, limit of the detection range, and incapacity of providing explanation of the acidity formation and change mechanism due to the fact that the prior art stays only at an acidity value measurement level. Taking the aerosol acidity detection device and method based on an imaging system disclosed by CN 116202919A as an example, the technology realizes on-line detection to a certain extent, but still adopts the traditional acid-base indicator principle, samples are required to be collected through components such as a sampler, a gas flow sensor, a micro air pump, a particulate matter detection sensor and the like, and then the detection is carried out by using indicator color development and LED light source irradiation. The method is limited by various factors such as sampling film replacement, environmental humidity control, indicator color change range and the like, and is difficult to realize high-precision, wide-range and real-time continuous aerosol acidity monitoring. Disclosure of Invention Aiming at the problem that the real-time performance of atmospheric aerosol acid detection is poor through physical sampling, the application provides the online detection system and the online detection method for the atmospheric aerosol acid, and the real-time performance is improved through combining a thermodynamic model with machine learning and the like without replacing detection consumables. The application provides a detection method of atmospheric aerosol acidity, which comprises the steps of obtaining aerosol data and met