Ozone pollution cause path construction method based on causal inference and application

Abstract

The invention discloses an ozone pollution cause path construction method based on causal inference. The method comprises the following steps: step 1) taking conventional influence factors in ozone pollution cause analysis as explanatory variables of an ozone pollution cause path; 2) performing significance test on each explanatory variable, and defining a hybrid explanatory variable; 3) determining a key explanatory variable and an intermediary variable by using Granger causal relationship test, and preliminarily building an intermediary path; 4) checking and correcting the intermediary path; step 5) performing inter-group difference test to determine an adjustment variable, and preliminarily constructing an intermediary path with adjustment; and 6) carrying out statistical test and correction on the adjusted intermediate path, and evaluating the adjustment effect of the adjustment variable on the intermediate path. The ozone pollution cause path construction method provided by the invention provides theoretical guidance for clarifying an ozone pollution formation mechanism under a combined pollution system and proposing effective and feasible ozone pollution prevention and control measures.

Description

Technical Field

[0001] The present invention relates to the technical field of ozone pollution cause analysis, and in particular to a method for constructing an ozone pollution cause path based on causal inference, and an application of the ozone pollution cause path. Background Art

[0002] Ozone (O3) is a typical secondary pollutant. Analyzing the causes of ozone pollution is the key to solving ozone pollution and is also the main line for continuously improving air quality.

[0003] The causes of ozone pollution are very complex. The existing ozone pollution cause analysis model mainly describes the combined impact of multiple factors on ozone in a complex pollution system, and cannot accurately describe the interaction process and mechanism between multiple variables. Therefore, although various regions have carried out ozone pollution cause analysis research, due to the mutual influence and interdependence of multiple variables in affecting ozone, the key process of ozone pollution formation is still unclear, and thus it is impossible to propose new and efficient ideas for ozone pollution control. Therefore, it is an important development direction for ozone pollution research at this stage to establish a new ozone pollution cause path, identify factors that have a significant impact on the formation of ozone pollution, and reveal the causal mechanism of influencing factors on ozone in a complex pollution system. Summary of the invention

[0004] The present invention provides a method for constructing the causal path of ozone pollution based on causal inference, which is used to solve the problems in the prior art that the key path of ozone pollution formation is unclear due to the interaction between multiple factors and the targeted pollution control measures are unable to effectively control ozone pollution due to non-causal false correlations. It provides theoretical guidance for clarifying the ozone pollution formation mechanism under a complex pollution system and proposing effective and feasible ozone pollution prevention and control measures.

[0005] The method for constructing the path of ozone pollution causes based on causal inference of the present invention comprises the following steps:

[0006] Step 1) Using conventional influencing factors in ozone pollution cause analysis as explanatory variables of ozone pollution cause path, obtaining time series data of each explanatory variable;

[0007] Step 2) Performing a significance test on each explanatory variable by stepwise regression method, obtaining explanatory variables with significant impact on ozone, and using them as significant explanatory variables in the causal path of ozone pollution; defining explanatory variables other than the significant explanatory variables and other unconventional influencing factors that may affect ozone as confounding explanatory variables, wherein the confounding explanatory variables are characterized by the residuals of stepwise regression simulation of ozone;

[0008] Step 3) Granger causality test is used to analyze the causal relationship between significant explanatory variables and between significant explanatory variables and ozone, and key explanatory variables and mediating variables are selected. Based on the mediating effect model, a preliminary ozone pollution causal path with mediating effect is constructed, which is called the mediating path;

[0009] Step 4) Use the MCM box model to chemically test and correct the mediating pathway; Use the Bootstrap method to statistically test and correct the mediating pathway, and further use the Bootstrap method to evaluate the direct and indirect effects of the mediating pathway, that is, to evaluate the intensity of the impact of the mediating pathway on ozone and the impact proportion of each pathway;

[0010] Step 5) Other significant explanatory variables other than the key explanatory variables and mediating variables, as well as confounding explanatory variables, are tested for inter-group differences; explanatory variables with significant inter-group differences are identified as moderating variables, and a moderated mediation path is preliminarily established based on the moderating effect model;

[0011] Step 6) Use the MCM box model to chemically test and correct the moderated mediation path; use the Bootstrap method to statistically test and correct the moderated mediation path, and further use the Bootstrap method to evaluate the moderating effect of the moderating variable on the mediation path.

[0012] The present invention explores the causal path of ozone pollution under the joint action of multiple factors under the background of big data, determines key explanatory variables and mediating variables based on causal relationships and chemical mechanisms, uses a mediating effect model to evaluate the direct and indirect effects of the mediating path on ozone, determines the moderating variables through inter-group difference tests, and uses a moderating effect model to evaluate the influence of the moderating variables on the mediating path, thereby determining the possible causal mechanism and pollution causal path between the influencing factors and ozone, which improves the utilization efficiency of monitoring data in the background of the big data era, and can also effectively explore the potential impact of multiple factors on ozone, providing technical support for ozone pollution control.

[0013] The two commonly used models for analyzing the causes of ozone pollution are chemical transport models and statistical models. The chemical transport model can explore the process of ozone pollution produced by various influencing factors through physical and chemical reactions, but it is greatly limited by the chemical mechanism embedded in the model. For example, different simulation results often appear when different chemical mechanisms are embedded in the model, and common chemical mechanisms do not consider multiphase reactions. In addition, the data quality requirements of the chemical transport model are relatively high, and it is difficult to obtain high-resolution emission inventory data.

[0014] The statistical model is based on observational data, which effectively avoids the problem of difficulty in obtaining data. The statistical model mainly explores the statistical correlation between various influencing factors and ozone. According to the correlation, targeted emission reduction measures can be proposed to achieve the purpose of reducing ozone concentration. 10 It is usually negatively correlated with ozone, which is mainly due to the increase in particulate matter concentration, which increases the optical thickness of the aerosol, resulting in a decrease in the near-ground photolysis rate of NO2 and O3. Based on this correlation, PM2.5 can be adjusted by controlling the PM2.5 concentration. 10 However, in the complex atmospheric pollution system, judging the cause of ozone pollution only by correlation may lead to illusory correlation and misleading conclusions. 2.5 The significant positive correlation with ozone is not due to PM 2.5 It promotes the formation of ozone, but PM 2.5 The presence of a common precursor with ozone causes PM 2.5 It has little effect on controlling ozone pollution. Therefore, we need to introduce causal inference methods in the analysis of the causes of ozone pollution, and analyze the real causes of ozone pollution based on the causal relationship between explanatory variables and ozone changes.

[0015] The definition of causality is that the cause is a phenomenon that causes a certain phenomenon; the result is a phenomenon that is caused in series due to the action of the cause. In other words, the occurrence of A leads to the occurrence of B, A is the cause of B, and B is the result of A. However, this definition is not clear in the field of ozone. The occurrence of A leading to the occurrence of B can be understood as A generating B through chemical reactions (direct generation), or it can be understood as A causing a change in the concentration of B by affecting the reaction process (participating in the ozone generation process). Therefore, the introduction of causal inference must first clarify who is the cause and who is the effect.

[0016] According to the characteristics of causal relationship: the cause comes first and the result comes later, and the appearance of the former leads to the latter. Causal relationship is often judged according to the order of occurrence in time. In the field of ozone, cause and effect exist at the same time (in the actual atmosphere, ozone precursors and ozone exist at the same time), and there is a reverse causal effect (such as NO2 is a precursor of ozone, it can generate ozone through photolysis reaction, and ozone will also affect the generation of NO2 by affecting the atmospheric oxidizability). Therefore, the present invention uses the time sequence of occurrence in statistical causal test and the generation relationship in chemical generation mechanism as the standard for determining causal relationship. The causal relationship is determined by combining Granger causality test with MCM box model. Granger causality test can test the statistical time sequence of variables, that is, the causal relationship between A and B is determined by the time when A and B occur. This method has a simple algorithm and low data requirements. However, this relationship is not a true causal relationship, so it must be further combined with the chemical reaction mechanism in the MCM box model for determination. According to the chemical reactions that A and B participate in in the actual atmosphere, the reactants that generate ozone are determined as the cause (including reactants that directly generate ozone through photolysis reactions and reactants that indirectly generate ozone through chain reactions, etc.). It should be noted that the MCM box model alone can theoretically determine the cause-effect relationship, but in the actual atmospheric environment, multiple physical and chemical reaction processes occur simultaneously and are coupled to each other, and it is difficult to determine the cause-effect relationship using the MCM box model alone.

[0017] The core content of the ozone pollution causal pathway is to build an intermediary pathway and a moderated intermediary pathway. The construction of the intermediary pathway includes three parts: the determination of key explanatory variables and intermediary variables, the testing and correction of the intermediary pathway, and the evaluation of the intermediary effect; the moderated intermediary pathway includes three parts: the determination of the moderating variables, the testing and correction of the moderated intermediary pathway, and the evaluation of the moderating effect.

[0018] The method for determining key explanatory variables and mediating variables of the present invention is a combination of statistical Granger causality test and chemical mechanism, that is, the key explanatory variables and mediating variables are selected through the causal relationship determined by Granger causality test and MCM box model. The method for determining the regulating variable of the present invention is a combination of statistical difference test and reaction process in chemical generation mechanism. The inter-group difference test can test whether there is a significant difference in the effect of the high group and the low group of the explanatory variable on ozone, but this significant difference only proves that the effect of the explanatory variable on ozone is sometimes strong and sometimes weak, and cannot prove that the explanatory variable will affect the causal relationship. It is necessary to further combine with the MCM box model to determine whether there is a chemical reaction mechanism in which the explanatory variable affects the causal relationship. Due to the complexity of the chemical mechanism, the purpose of determining the regulating variable cannot be achieved by using the MCM box model alone.

[0019] The inspection and correction of the mediation path mainly includes chemical mechanism inspection and correction and statistical inspection and correction. The inspection and correction of the chemical mechanism is to use the MCM box model to determine whether the preliminary constructed mediation path conforms to the ozone generation mechanism, and to correct the path that does not conform to the ozone generation mechanism. The statistical inspection and correction is to use the Bootstrap test to determine whether the preliminary constructed mediation path meets the statistical requirements, and to correct the statistically insignificant path.

[0020] The test and correction of moderated mediation paths mainly include chemical mechanism test and correction and statistical test and correction. The test and correction of chemical mechanism is to use the MCM box model to determine whether there are differences in the chemical reactions between the high and low groups of the mediation variable, and to correct the paths without differences. The statistical test and correction is to use the Bootstrap test to determine whether the initially constructed moderated mediation path meets the statistical requirements, and to correct the statistically insignificant paths.

[0021] As a limitation of the above technical solution, conventional influencing factors in ozone pollution cause analysis are factors that are generally recognized as having a greater impact on ozone in the field of ozone pollution cause analysis, including meteorological factors such as ultraviolet radiation intensity, relative humidity, temperature, air pressure, wind speed, wind direction, etc.; and atmospheric pollutant factors such as PM 10 、PM 2.5 , NO2, SO2, CO, NMVOCs, etc.; unconventional influencing factors are factors that are generally recognized in the field of ozone pollution cause analysis and have relatively little impact on ozone or are difficult to obtain data, including meteorological elements such as rainfall, cloud cover, boundary layer height, etc., pollutant factors such as HONO, VOCs species, etc., and other factors such as population density, ground roughness, land use type, etc.

[0022] As a limitation of the above technical solution, the selection of the significant explanatory variables must meet the following conditions:

[0023] The explanatory variables were selected one by one according to their normalized coefficients from large to small, and the explanation degree of ozone by significant explanatory variables was required to be higher than the threshold of the stepwise regression model; at least three significant explanatory variables were selected, including one key explanatory variable, one mediating variable and one moderating variable; the number of significant explanatory variables should be appropriately adjusted according to the path test results and actual application needs.

[0024] As a limitation of the above technical solution, the calculation formula of the confounding explanatory variable is:

[0025]

[0026] Among them, x k represents the time series data of significant explanatory variables; β k represents the regression coefficient; represents the time series data of ozone simulated by stepwise regression; O3 represents the time series data of ozone; ε represents the time series data of confounding explanatory variables.

[0027] As a limitation of the above technical solution, the mediation effect model is as follows:

[0028] Y=cX+e1

[0029] M=aX+e2

[0030] Y=c′X+bM+e3

[0031] Among them, X is the key explanatory variable, Y is ozone, M is the mediating variable, a, b, c, c′ are the coefficients of the mediation effect test results, among which c is the total effect, c′ is the direct effect, and ab is the indirect effect; e1, e2, e3 are the constants of the test results.

[0032] As a limitation of the above technical solution, the causal relationship between significant explanatory variables and between significant explanatory variables and ozone is tested through Granger causality, the position of significant explanatory variables in the causal path of ozone pollution is judged, and the key explanatory variables X and mediating variables M are determined.

[0033] As a limitation of the above technical solution, X and M must meet the following requirements: X is the cause of M and Y, and M is the cause of Y.

[0034] Determining key explanatory variables, mediating variables, and moderating variables is the basis for building the causal path of ozone pollution. The basis for determining key explanatory variables and mediating variables is the definition of mediating effect. Mediating effect means that the cause affects the result through one or several intermediate variables. Such intermediate variables are called mediating variables. Figure 1 The schematic diagram of the mediation effect is shown in the figure. Y represents ozone, X represents the key explanatory variable, and M represents the mediating variable. The determination of X and M should meet the following conditions: X is the cause of M and Y, and M is the cause of Y.

[0035] The intermediary path must satisfy both statistical and chemical tests.

[0036] As a limitation of the above technical solution, the direct and indirect effects in the mediation path are evaluated as follows:

[0037] Step 4.1) Use the Bootstrap method to test the total effect c. If it is significant, the argument is based on the mediation effect. Otherwise, the argument is based on the masking effect. Regardless of whether it is significant or not, the subsequent steps are tested.

[0038] Step 4.2) Test coefficients a and b by the Bootstrap method. If both are significant, the indirect effect is significant, and go to step 4.4); if at least one is not significant, proceed to step 4.3);

[0039] Step 4.3) Test the indirect effect ab by the Bootstrap method; if it is significant, the indirect effect is significant, and proceed to step 4.4); otherwise, the indirect effect is not significant, and the analysis is stopped;

[0040] Step 4.4) After confirming the existence of the mediating effect, the direct effect c' is tested by the Bootstrap method. If it is not significant, that is, the direct effect is not significant, indicating that there is only a mediating effect; if it is significant, that is, the direct effect is significant, proceed to step 4.5);

[0041] Step 4.5) Compare the signs of ab and c. If they have the same sign, it is a partial mediation effect, and report the ratio of the mediation effect to the total effect as ab / c. If they have different signs, it is a masking effect, and report the absolute value of the ratio of the indirect effect to the direct effect |ab / c|.

[0042] As a limitation of the above technical solution, the regulatory effect model is:

[0043] Y=β0+β1X+β2W+β3WX+ε

[0044] Among them, X is the key explanatory variable, Y is ozone, W is the adjustment variable, β is the coefficient of the adjustment effect test result, and ε is the constant of the test result.

[0045] The moderated mediation path consists of three parts: determination of the moderating variable, testing and correction of the moderated mediation path, and evaluation of the moderating effect.

[0046] As a limitation of the above technical solution, the inter-group difference test includes the following steps:

[0047] Step 5.1) Import the time series data of other significant explanatory variables except the key explanatory variables and mediating variables and the corresponding ozone time series data, import the time series data of the confounding explanatory variables and the corresponding ozone time series data, and analyze each explanatory variable one by one;

[0048] Step 5.2) Sort the explanatory variables to be analyzed by their numerical values ​​and divide them into two groups according to the sample size. That is, using the median of all data as the standard, the values ​​greater than or equal to the standard are defined as the high value group, referred to as the high group, and the values ​​less than the standard are defined as the low value group, referred to as the low group;

[0049] Step 5.3) Perform a Mann-Whitney U test on the ozone time series data corresponding to the high group and the low group. If the test result is significant, it indicates that there is a significant difference between the ozone in the high group and the ozone in the low group; if the test result is not significant, it indicates that there is no significant difference between the ozone in the high group and the ozone in the low group.

[0050] As a limitation of the above technical solution, the adjustment variable W needs to meet the following requirements: there is a significant difference between the Y corresponding to the W high group and the W low group.

[0051] The moderated mediation path must satisfy both statistical and chemical tests.

[0052] The evaluation method of the moderation effect is: if the effect value of the moderator variable β3>0, the positive impact of X on Y increases with the increase of W, or the negative impact weakens with the increase of M; if the effect value of the moderator variable β3<0, the positive impact of X on Y weakens with the increase of W, or the negative impact strengthens with the increase of M.

[0053] The basis for determining the moderating variable is the definition of the moderating effect, which means that the strength of the influence of the cause on the result will vary depending on individual characteristics or environmental conditions. Such characteristics or conditions are called moderating variables. Figure 2 The schematic diagram of the moderating effect is shown in the figure, where Y represents ozone, X represents the key explanatory variable, and W represents the moderating variable. The determination of W must meet the following conditions: the causal relationship of X→Y is moderated by W, that is, there is a significant difference in the effect of X on Y between the high W group and the low W group.

[0054] As a limitation of the above technical solution, the stepwise regression is implemented by the stepwise method in SPSS; the Granger causality test is obtained by fitting using STATA; the difference test is implemented by the nonparametric test in SPSS; and the construction of mediation paths and moderated mediation paths is implemented by using the Process plug-in in SPSS.

[0055] The method for constructing the causal path of ozone pollution based on causal inference of the present invention is different from the traditional statistical model for analyzing the correlation between influencing factors and ozone, and the chemical transmission model for analyzing influencing factors and ozone chemical processes. This application is based on causal inference and combines statistical algorithms and chemical mechanisms. It overcomes the defects of chemical transmission models with high basic data requirements, and overcomes the problem of hidden numerical relationships caused by the statistical model not fully considering the chemical reaction mechanism, and deeply analyzes the real impact of influencing factors on ozone.

[0056] At the same time, the present invention also provides an application of the ozone pollution causal pathway based on causal inference as described above, which is to use the above-obtained ozone pollution causal pathway based on causal inference to guide emission reduction, which can effectively reduce ozone concentration.

[0057] Specifically, the mediation path explains how the key explanatory variables affect ozone. When the mediation path is fully mediated, the impact of the key explanatory variables on ozone is completely transmitted by the mediating variables. At this time, controlling the mediating variables can effectively control the occurrence of ozone pollution. When the mediation path is partially mediated, the key explanatory variables affect ozone through direct effects and indirect effects through mediating variables. At this time, it is necessary to control both the key explanatory variables and the mediating variables to control the occurrence of ozone pollution. The regulated mediation path selects the optimal conditions for ozone pollution control measures. When the regulating variable has a positive effect on the mediation path, increasing the regulating variable can enhance the emission reduction effect. When the regulating variable has a negative effect on the mediation path, increasing the regulating variable can weaken the emission reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 , Diagram of mediation effect;

[0059] Figure 2 , schematic diagram of the regulatory effect;

[0060] Figure 3 , Summary of the SPSS stepwise regression variable model screening process;

[0061] Figure 4 , there may be paths for the formation of ozone pollution, with path 1 on the left and path 2 on the right;

[0062] Figure 5 , SPSS mediation effect model operation result interface;

[0063] Figure 6 , the causal path of ozone pollution in Shijiazhuang based on the mediating effect, ** in the figure indicates p<0.05;

[0064] Figure 7 , SPSS moderation effect model operation result interface;

[0065] Figure 8 , the causal pathway of ozone pollution in Shijiazhuang based on the moderated mediation effect;

[0066] Fig. 9 , a flowchart of the method for constructing the causal pathway of ozone pollution based on causal inference; DETAILED DESCRIPTION

[0067] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Example

[0069] Step 1) Obtain the time series data of explanatory variables related to the change of ozone concentration during the ozone pollution process in winter 2020 (January 25, 2020 to February 9, 2020), including meteorological factors: UVB (ultraviolet radiation intensity), RH (relative humidity), T (temperature), P (atmospheric pressure), WS (wind speed), WD (wind direction), etc.; pollutant factors: inhalable particulate matter (PM 10 ), fine particulate matter (PM 2.5 ), nitrogen dioxide (NO2), sulfur dioxide (SO2), carbon monoxide (CO), and non-methane hydrocarbons (NMVOCs), a total of 12 types.

[0070] In this example, the time series data of meteorological factors (T, P, RH, WS and WD) were obtained from the National Ground Meteorological Monitoring Station (https: / / q-weather.info / weather), and the hourly observation data of UVB were obtained from the Shijiazhuang UV1000-H solar radiation observation system. 2.5 、PM 10 The hourly observation data of VOCs (sulfuric acid, sulfur dioxide, and sulfur dioxide) are from the China National Environmental Monitoring Center (http: / / www.cnemc.cn), and the NMVOCs data are obtained by the Shijiazhuang VOCs automatic online monitoring instrument AirmOzone Analysis System (ASS). Some of the data are shown in Table 1.

[0071] Table 1: Data display of pollution process in Shijiazhuang (partial)

[0072]

[0073]

[0074] Step 2) Use the stepwise method of stepwise regression to conduct significance tests on the obtained meteorological and pollutant factors, and determine the explanatory variables that have a significant impact on ozone changes from a variety of explanatory variables, that is, the significant explanatory variables in the causal path of ozone pollution.

[0075] There is no unique method for screening significant explanatory variables. The stepwise method used in the present invention combines accuracy and simplicity. In addition, this embodiment uses the mapminmax function to normalize the explanatory variables when using the stepwise method, ensuring that the effects of significant explanatory variables on ozone are comparable. The larger the normalization coefficient, the greater its importance to ozone.

[0076] In order to ensure the simplicity of the causal path of ozone pollution, the selection of significant explanatory variables should meet the following conditions: the selection of significant explanatory variables should be selected one by one according to the normalized coefficient of the explanatory variables from large to small; the explanation degree of significant explanatory variables for ozone should be higher than the threshold of the stepwise regression model, R 2 =0.6; at least three significant explanatory variables are required, including one key explanatory variable, one mediating variable and one moderating variable; the number of significant explanatory variables should be adjusted appropriately according to the path test results and actual application needs.

[0077] In this embodiment, this step is implemented by the stepwise method in linear regression in SPSS. The process of SPSS screening variables is summarized as follows: Figure 3 A total of 10 variables were screened in the pollution process in Shijiazhuang City, which have significant effects on ozone, namely RH, T, NO2, SO2, PM 10 , P, WD, WS, PM 2.5 In order to establish the most concise path of ozone pollution, this example first selected the three variables that have the greatest impact on ozone: NO2, PM 2.5 and T. However, since T met the selection criteria for adjustment variables in the subsequent actual modeling process, it did not pass the statistical test of the adjusted mediating path. Therefore, the variables were selected in turn according to the normalization coefficient ranking and tested again until the established adjusted mediating path could pass the test. The results of the significant explanatory variables finally selected in this embodiment are shown in Table 2. The significant explanatory variables of the ozone pollution process in Shijiazhuang are CO, NO2, PM 2.5 , T and RH, among which NO2 has the greatest impact on ozone and RH has the smallest impact on ozone. These five explanatory variables can jointly explain 75% of the ozone changes during the study period.

[0078] Table 2 Results of screening of significant explanatory variables for ozone pollution process in Shijiazhuang City

[0079]

[0080] However, in the actual atmospheric environment, in addition to the above five significant explanatory variables, other influencing factors will also have an impact on ozone. In order to comprehensively consider the impact of various influencing factors on ozone, this embodiment incorporates explanatory variables other than significant explanatory variables and unconventional influencing factors into the construction of pollution cause path in the form of mixed explanatory variables. The calculation results of some mixed explanatory variables are shown in Table 3.

[0081] Table 3 Calculation results of confounding explanatory variables for ozone pollution process in Shijiazhuang (partial)

[0082]

[0083]

[0084] Step 3) Based on the Granger causality test, determine the causal relationship between significant explanatory variables and between significant explanatory variables and ozone, and select key explanatory variables and mediating variables based on this. The key explanatory variables, mediating variables and ozone should satisfy the causal chain: the key explanatory variables are the cause of the mediating variables, and the key explanatory variables and mediating variables are the causes of ozone. Based on the mediating effect model, a preliminary ozone pollution causal path with a mediating effect is constructed, called a mediating path; taking the simplest scenario as an example, one key explanatory variable and one mediating variable are selected based on the causal chain relationship. At this time, the mediating path constructed in combination with the mediating effect model is as follows Figure 1 shown.

[0085] In this example, the Granger causality test results between significant explanatory variables and between significant explanatory variables and ozone are shown in Table 4. CO and NO2 are the causes of ozone (p<0.05), so CO and NO2 may be the key explanatory variables and mediating variables in the causal path of ozone pollution. However, since CO and NO2 are mutually causal (p<0.05), and ozone has a reverse causal relationship with NO2 (p<0.05), it shows that the position of the significant explanatory variable in the mediating effect model is not uniquely determined. It is speculated that the possible paths are as follows Figure 4 As shown in Figure 1, Path 1: CO is the key explanatory variable and NO2 is the mediating variable; Path 2: NO2 is the key explanatory variable and CO is the mediating variable.

[0086] Table 4 Granger causality test results of ozone pollution process in Shijiazhuang City

[0087]

[0088] Step 4) Use the MCM box model to chemically test and correct the mediating path, and use the Bootstrap test to statistically test and correct the mediating path to determine the tested and corrected mediating path. At this time, the mediating path not only conforms to the chemical mechanism of ozone generation, but also meets the statistical requirements. The direct and indirect effects of the mediating path are further evaluated in the Bootstrap test process.

[0089] According to the possible mediation path preliminarily inferred, the corresponding mediation effect model is the mediation effect model 4 in SPSS-Process. The SPSS mediation effect model operation result interface is as follows Figure 5 After sorting, in this embodiment, the intermediate path after verification and correction is as follows Figure 6As shown in the figure, the key explanatory variable affecting the change of ozone concentration is NO2, and the mediating variable is CO. There are two main ways to affect ozone. Pathway 1 is that NO2 directly affects ozone (p<0.05, R1-R3), and Pathway 2 is that NO2 affects ozone by affecting CO (p<0.05, NO2→CO is R5-R7 in Table 5, CO→O3 is R10).

[0090] Table 5 Main chemical reaction processes involved in the path of ozone pollution (from MCM box model)

[0091]

[0092]

[0093] The evaluation results of the mediation path are shown in Table 6. The total effect c, direct effect c' and indirect effect ab of the mediation path are all significant, so the mediation effect is partial mediation. And ab and c have opposite signs, and the mediation effect is a masking effect. This shows that the impact of NO2 on ozone in the ozone pollution process in Shijiazhuang City is transmitted through two pathways. The first is that NO2 directly affects ozone, and the second is that NO2 indirectly affects ozone by affecting CO. Among them, the indirect effect accounts for 28.87% of the direct effect.

[0094] Table 6 Results of evaluation of intermediary paths in Shijiazhuang

[0095]

[0096] Step 5) From other significant explanatory variables (specifically T, RH, P, WS, WD, UVB, PM in this example) except the key explanatory variables and mediating variables, 2.5 、PM 10 , SO2, NMVOCs) and confounding explanatory variables, a total of 11 explanatory variables, were determined by the inter-group difference test method. The moderating variable W must meet the following conditions: the causal relationship of X→Y is moderated by W, that is, there is a significant difference in the effect of X on Y between the high W group and the low W group. Based on the moderating effect model, a moderated mediation path was initially established; taking the simplest scenario as an example, one moderating variable was selected based on the inter-group difference test results. At this time, the moderating path built in combination with the moderating effect model is as follows Figure 2 shown.

[0097] Here is PM 2.5 Let's take this as an example. First, based on PM 2.5The time series data of were sorted by numerical value and divided into two groups according to the sample size. That is, the median of all data was used as the standard. The values ​​greater than or equal to the standard were classified as the high value group, referred to as the high group, and the values ​​less than the standard were classified as the low value group, referred to as the low group. The ozone time series data corresponding to the high group and the low group were subjected to the Mann-Whitney U test. The test results are shown in Table 7. High PM 2.5 Ozone and low PM 2.5 There were significant differences in ozone between the groups (p = 0.000), namely PM 2.5 It may be a moderating variable for the causal pathway of ozone pollution. In this example, the moderating variables with significant differences in the impact of ozone in the high group and the low group were initially determined to be T, RH, WS, UVB, PM 2.5 and NMVOCs.

[0098] Table 7 PM 2.5 Summary of the Mann-Whitney U Test

[0099]

[0100] Step 6) Use the MCM box model to chemically test and correct the mediating path, use the Bootstrap test to statistically test and correct the mediating path, and finally determine the mediating path with adjustment after testing and correction. At this time, the mediating path with adjustment not only conforms to the chemical mechanism of ozone generation, but also meets the statistical requirements. In the Bootstrap test process, further evaluate the moderating effect of the moderating variable on the mediating path.

[0101] The moderated mediation path initially constructed corresponds to the mediation effect model 21 in SPSS-Process. The SPSS moderated effect model operation result interface is as follows: Figure 7 After sorting, in this embodiment, the adjusted intermediate path after inspection and correction is as follows Figure 8 As shown, the adjustment variable is PM 2.5 and RH. The results of the evaluation of the moderating effect are shown in Table 8. PM 2.5 The regulating effect of RH on NO2→CO is significant, and the regulating effect of RH on CO→O3 is significant. 2.5 The effect values ​​β3 of PM and CO are both positive, indicating that PM 2.5 The regulating effect of RH on NO2→CO and the regulating effect of RH on CO→O3 are both positive.

[0102] Table 8 Moderating effect evaluation results

[0103]

[0104] In summary, the ozone pollution cause pathway constructed by the present invention (see Figure 8 ), NO2 is the key explanatory variable affecting ozone, CO is the mediating variable, PM 2.5 and RH are the adjustment variables. The changes of NO2 and CO are the main causes of ozone pollution in Shijiazhuang. 2.5 The level of RH can significantly adjust the effect of NO2 on CO (the adjustment direction is positive), and RH can adjust the effect of CO on ozone (the adjustment direction is positive). This shows that in terms of guiding ozone emission reduction, controlling the concentration of NO2 and CO can effectively reduce the ozone concentration level, and adjusting PM on the basis of controlling NO2 and CO 2.5 and RH levels can significantly improve the efficiency of ozone pollution prevention and control.

[0105] In order to verify the effectiveness of the present invention in preventing and controlling ozone pollution, the emission reduction effects are compared based on the results of different ozone pollution cause analysis models. Here, the statistical model takes stepwise regression as an example, and the chemical transport model takes the WRF-CMAQ model as an example for comparison. For the same ozone pollution process, different pollution cause analysis models obtain different causes of ozone pollution. The ozone pollution causes obtained by the ozone pollution cause path of the present invention are CO and NO2, and the ozone pollution causes obtained by the statistical model are CO, NO2, PM 2.5 and RH, the ozone pollution causes obtained by the chemical transport model are T, NO2 and VOCs. In view of the particularity of the environmental field, in order to compare the effects of different models in guiding ozone pollution prevention and control, the observation-based box model (OBM) is used here to simulate the emission reduction scenarios for the ozone pollution causes obtained by different models, and the results are shown in Table 9.

[0106] Table 9 Simulation of different emission reduction scenarios

[0107]

[0108] According to the ozone pollution cause pathway constructed by the present invention, emission reduction can effectively reduce ozone concentration. Controlling the key explanatory variables and mediating variables can effectively control the ozone concentration level. Controlling the key explanatory variables can achieve the best prevention and control effect of ozone pollution (ozone concentration reduced by 7.50 μg / m 3 ), and the effect of controlling the mediating variable was second (ozone concentration decreased by 3.50 μg / m 3 ). On this basis, adjusting the adjustment variables can effectively improve the efficiency of ozone pollution prevention and control (ozone concentrations were reduced by 11.75 and 4.10 μg / m 3 ).

[0109] Emission reduction simulations based on the results of statistical models and chemical transport models show that PM 2.5The chemical transport process in which VOCs participate has a significant impact on the ozone concentration value, but it is not the real cause of ozone pollution. The ozone prevention effect obtained by controlling it is not high, and even rebound may occur (ozone concentration increased by 1.52 μg / m 3 and decrease by 0.20 μg / m 3 ).

[0110] Obviously, due to the complex relationship between influencing factors, there are both simple one-way influence relationships and mutual influence relationships. The traditional statistical model reflects the direct relationship between independent variables and dependent variables, and it is difficult to express the complex relationship between independent variables. The present invention is based on the path of ozone pollution caused by causal inference. Fig. 9 The construction process shown can effectively analyze this complex relationship step by step, and is more scientific and effective in guiding the prevention and control of ozone pollution.

Claims

1. A method for constructing the causal pathway of ozone pollution based on causal inference, characterized by: The construction method includes the following steps: Step 1) Using conventional influencing factors in ozone pollution cause analysis as explanatory variables of ozone pollution cause path, obtaining time series data of each explanatory variable; Step 2) Performing a significance test on each explanatory variable by stepwise regression method, obtaining explanatory variables with significant impact on ozone, and using them as significant explanatory variables in the causal path of ozone pollution; defining explanatory variables other than the significant explanatory variables and other unconventional influencing factors that may affect ozone as confounding explanatory variables, wherein the confounding explanatory variables are characterized by the residuals of stepwise regression simulation of ozone; Step 3) Use Granger causality to test the causal relationship between significant explanatory variables and between significant explanatory variables and ozone, determine the key explanatory variables and mediating variables, and preliminarily build an ozone pollution causal path with a mediating effect based on the mediating effect model, which is called the mediating path; Step 4) Use the MCM box model to chemically test and correct the mediating pathway; Use the Bootstrap method to statistically test and correct the mediating pathway, and further use the Bootstrap method to evaluate the direct and indirect effects of the mediating pathway, which is used to evaluate the intensity of the impact of the mediating pathway on ozone and the impact proportion of each pathway; Step 5) Other significant explanatory variables other than the key explanatory variables and mediating variables, as well as confounding explanatory variables, are tested for inter-group differences; explanatory variables with significant inter-group differences are identified as moderating variables, and a moderated mediation path is preliminarily established based on the moderating effect model; Step 6) Use the MCM box model to chemically test and correct the moderated mediation path; use the Bootstrap method to statistically test and correct the moderated mediation path, and further use the Bootstrap method to evaluate the moderating effect of the moderating variable on the mediation path.

2. The method for constructing the causal path of ozone pollution based on causal inference according to claim 1 is characterized by: The conventional influencing factors in the analysis of the causes of ozone pollution are the factors that are recognized in the field of ozone pollution cause analysis as having a greater impact on ozone, including meteorological factors and atmospheric pollutant factors; the unconventional influencing factors are the factors that are recognized in the field of ozone pollution cause analysis as having relatively little impact on ozone or are difficult to obtain data, including meteorological elements, pollutant factors and other factors.

3. The method for constructing the causal path of ozone pollution based on causal inference according to claim 1 is characterized in that: The selection of the significant explanatory variables must meet the following conditions: Select the explanatory variables one by one according to their normalized coefficients from large to small, and at the same time require that the explanation degree of the significant explanatory variables for ozone is higher than the threshold of the stepwise regression model; Select at least 3 significant explanatory variables.

4. The method for constructing the causal path of ozone pollution based on causal inference according to claim 1 is characterized in that: The calculation formula of the confounding explanatory variable is: In the formula, x k represents the time series data of significant explanatory variables; β k represents the regression coefficient; represents the time series data of ozone simulated by stepwise regression; O3 represents the time series data of ozone; ε represents the time series data of confounding explanatory variables.

5. The method for constructing the causal path of ozone pollution based on causal inference according to claim 1 is characterized in that: The mediation effect model is as follows: Y=cX+e1 M=aX+e2 Y=c′X+bM+e3 Among them, X is the key explanatory variable, Y is ozone, M is the mediating variable, a, b, c, c′ are the coefficients of the mediation effect test results, among which c is the total effect, c′ is the direct effect, and ab is the indirect effect; e1, e2, e3 are the constants of the test results.

6. The method for constructing the causal path of ozone pollution based on causal inference according to claim 5 is characterized by: Through the Granger causality test, the causal relationship between significant explanatory variables and between significant explanatory variables and ozone is tested to determine the position of significant explanatory variables in the causal path of ozone pollution and to determine the key explanatory variables X and mediating variables M.

7. The method for constructing the causal path of ozone pollution based on causal inference according to claim 6 is characterized by: X and M must satisfy the following requirements: X is the cause of M and Y, and M is the cause of Y.

8. The method for constructing the path of ozone pollution causes based on causal inference according to claim 5 is characterized in that: The direct and indirect effects in the mediation path were assessed as follows: Step 4.1) Use the Bootstrap method to test the total effect c. If it is significant, the argument is based on the mediation effect. Otherwise, the argument is based on the masking effect. Regardless of whether it is significant or not, the subsequent step test is carried out; Step 4.2) Test coefficients a and b by the Bootstrap method. If both are significant, the indirect effect is significant, and go to step 4.4); If at least one is not significant, proceed to step 4.3); Step 4.3) Test the indirect effect ab by the Bootstrap method; if significant, the indirect effect is significant, and proceed to step 4.4); Otherwise, the indirect effect is not significant and the analysis is stopped; Step 4.4) After confirming the existence of the mediating effect, the direct effect c' is tested by the Bootstrap method. If it is not significant, that is, the direct effect is not significant, indicating that there is only a mediating effect; if it is significant, that is, the direct effect is significant, proceed to step 4.5); Step 4.5) Compare the signs of ab and c. If they have the same sign, it is a partial mediation effect, and report the ratio of the mediation effect to the total effect as ab / c. If they have different signs, it is a masking effect, and report the absolute value of the ratio of the indirect effect to the direct effect |ab / c|.

9. The method for constructing the path of ozone pollution causes based on causal inference according to claim 1, characterized in that: The moderation effect model is: Y=β0+β1X+β2W+β3WX+ε Among them, X is the key explanatory variable, Y is ozone, W is the adjustment variable, β is the coefficient of the adjustment effect test result, and ε is the constant of the test result.

10. The method for constructing the causal path of ozone pollution based on causal inference according to claim 9, characterized in that: The inter-group difference test includes the following steps: Step 5.1) Import the time series data of other significant explanatory variables except the key explanatory variables and mediating variables and the corresponding ozone time series data, import the time series data of the confounding explanatory variables and the corresponding ozone time series data, and analyze each explanatory variable one by one; Step 5.2) Sort the explanatory variables to be analyzed by their numerical values ​​and divide them into two groups according to the sample size. That is, using the median of all data as the standard, the values ​​greater than or equal to the standard are defined as the high value group, referred to as the high group, and the values ​​less than the standard are defined as the low value group, referred to as the low group; Step 5.3) Perform a Mann-Whitney U test on the ozone time series data corresponding to the high group and the low group. If the test result is significant, it indicates that there is a significant difference between the ozone in the high group and the ozone in the low group; if the test result is not significant, it indicates that there is no significant difference between the ozone in the high group and the ozone in the low group.

11. The method for constructing the path of ozone pollution causes based on causal inference according to claim 10, characterized in that: The adjustment variable W must meet the following requirements: the Y corresponding to the W high group and the W low group has significant differences.

12. The method for constructing the path of ozone pollution causes based on causal inference according to claim 9, characterized in that: The evaluation method of the moderation effect is: if the effect value of the moderator variable β3>0, the positive impact of X on Y increases with the increase of W, or the negative impact weakens with the increase of M; if the effect value of the moderator variable β3<0, the positive impact of X on Y weakens with the increase of W, or the negative impact strengthens with the increase of M.

13. The application of the ozone pollution causal pathway based on causal inference, using the ozone pollution causal pathway based on causal inference obtained in any one of claims 1 to 12 to guide emission reduction, can effectively reduce ozone concentration.

Images