A method for distinguishing meteorological conditions for atmospheric ozone pollution

Abstract

The present invention relates to air pollution monitoring technology, which discloses a method for distinguishing the meteorological conditions for atmospheric ozone pollution, which effectively reflects the impact of meteorological condition changes on the occurrence of ozone pollution, can be used for evaluating the meteorological conditions for the occurrence of historical ozone pollution and predicting the meteorological conditions for the occurrence of future ozone pollution, and provides a scientific basis for evaluating the effect of atmospheric ozone pollution prevention and control and formulating atmospheric ozone pollution prevention and control measures. The scheme can be summarized as follows: 1. Determine the sensitive physical parameters that affect the occurrence of atmospheric ozone pollution as model input parameters; 2. Collect and pre-process samples; 3. Normalize the model training samples; 4. Establish and train the meteorological condition level discrimination model for ozone pollution by region; 5. In actual application, the sensitive physical parameter data in the target area to be judged is used as the model input parameter, and the discrimination result of the meteorological condition level for atmospheric ozone pollution is obtained based on the trained discrimination model.

Description

Technical Field

[0001] The invention relates to air pollution monitoring technology, and in particular to a method for distinguishing the level of meteorological conditions for atmospheric ozone pollution. Background Art

[0002] Ozone in the atmosphere is a secondary pollutant, mainly generated by a series of photochemical reactions of nitrogen oxides and hydrocarbons in the atmosphere under special meteorological conditions. With the continuous acceleration of urbanization and industrialization, the number of motor vehicles has increased year by year, and the emission of ozone precursors has continued to increase. Atmospheric photochemical pollution characterized by ozone pollution has become a problem that cannot be ignored, especially in summer, when ozone has gradually replaced PM 2.5 The trend of becoming the primary pollutant has attracted widespread attention from society.

[0003] Accurate ozone pollution forecasts are beneficial to the general public and all sectors of society to arrange their lives, work and various social activities in a reasonable manner, and are beneficial to environmental management and decision-making departments to issue warnings in a timely manner and take effective measures to prevent serious pollution incidents. At present, the more popular atmospheric ozone pollution forecast methods are as follows:

[0004] 1. The ozone concentration forecast model established based on experience and statistical models. It mainly considers multiple meteorological factors that have a high impact on ozone pollution and establishes an ozone concentration forecast equation. This method is relatively simple and convenient to calculate, but it cannot meet the needs in terms of theory and timeliness.

[0005] 2. Use the environmental meteorological numerical forecast model to provide some ozone pollution forecast functions. It is mainly based on the forecast values ​​of meteorological elements and the emission status of pollution sources, and uses mathematical models to obtain the spatial and temporal distribution of ozone concentration. However, this method has a complex calculation process, high requirements on computer performance, high cost, and mainly predicts ozone concentration.

[0006] 3. The method of forecasting near-ground ozone concentration based on machine learning. This method mainly relies on the advantage of machine learning methods in effectively capturing the nonlinear characteristics hidden in the changes of atmospheric components, building a model through characteristic variables, and predicting atmospheric components.

[0007] It can be seen that the previous work mainly focused on forecasting ozone concentrations to determine whether ozone pollution occurs. In fact, the occurrence of atmospheric ozone pollution is mainly affected by factors such as meteorological conditions and pollution source emissions. In the short term, pollution sources are generally relatively stable, and meteorological conditions are the main factor affecting its occurrence. However, there is currently no relevant work that provides a method to reflect the impact of meteorological conditions on the occurrence of atmospheric ozone pollution. Summary of the invention

[0008] The technical problem to be solved by the present invention is to propose a method for distinguishing the level of meteorological conditions for the occurrence of atmospheric ozone pollution, which can effectively reflect the impact of changes in meteorological conditions on the occurrence of ozone pollution, and can be used for evaluating the meteorological conditions for the occurrence of historical ozone pollution and predicting the meteorological conditions for the occurrence of future ozone pollution, so as to provide a scientific basis for evaluating the effect of atmospheric ozone pollution prevention and control and formulating atmospheric ozone pollution prevention and control measures.

[0009] The technical solution adopted by the present invention to solve the above technical problems is:

[0010] A method for determining the meteorological condition level of atmospheric ozone pollution comprises the following steps:

[0011] S1. Determine the sensitive physical parameters that affect the occurrence of atmospheric ozone pollution as model input parameters;

[0012] S2. Obtain the ozone concentration data within a period of time in the target area and the historical data of sensitive physical parameters that affect the occurrence of atmospheric ozone pollution, perform preprocessing, and construct training samples;

[0013] S3, normalize the training samples;

[0014] S4. Establish a model for distinguishing the meteorological conditions for ozone pollution by region, and use normalized training samples to train the model;

[0015] S5. Using the sensitive physical parameter data in the target area to be identified as the model input parameters, based on the trained ozone pollution meteorological condition level identification model, the identification result of the atmospheric ozone pollution meteorological condition level is calculated.

[0016] As a further optimization, in step S1, by analyzing the relationship between ozone and meteorological factors and precursors, sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are obtained as model input parameters. The sensitive physical parameters include: temperature, precipitation in the previous hour, relative humidity, mixing layer height and average NO2 mass concentration of the previous night.

[0017] As a further optimization, in step S2, the method for obtaining ozone concentration data and historical data of sensitive physical parameters affecting the occurrence of atmospheric ozone pollution is:

[0018] The meteorological data is used to obtain the temperature, precipitation in the previous hour, relative humidity and mixing layer height data; the air quality monitoring data is used to obtain the ozone mass concentration and the average NO2 mass concentration data of the previous night.

[0019] As a further optimization, the ozone mass concentration is the average of the hourly data of all effective environmental monitoring stations in the city, and the corresponding meteorological data is the data of the location closest to the center of the effective environmental monitoring station.

[0020] As a further optimization, in step S2, the preprocessing includes:

[0021] The meteorological conditions for ozone pollution are divided into levels, and the ozone mass concentration is associated with the level of meteorological conditions for ozone pollution. The number of samples for each level in the target area is obtained based on the statistical analysis of historical data. After the number of samples for each level in multiple target areas is obtained, the number of models to be modeled is determined based on the number of samples in each target area, and then sample screening is performed.

[0022] As a further optimization, the classification of meteorological conditions for ozone pollution and the association of ozone mass concentration with the meteorological conditions for ozone pollution are specifically including:

[0023] The meteorological conditions for ozone pollution are divided into three levels, each level corresponds to a different hourly ozone mass concentration. The lower the hourly ozone mass concentration, the lower the level. When the meteorological conditions for ozone pollution are level 1, it indicates that the meteorological conditions are very unfavorable for the occurrence of ozone pollution; when the meteorological conditions for ozone pollution are level 2, it indicates that the meteorological conditions are unfavorable for the occurrence of ozone pollution; when the meteorological conditions for ozone pollution are level 3, it indicates that the meteorological conditions are favorable for the occurrence of ozone pollution.

[0024] As a further optimization, the number of models to be modeled is determined based on the number of samples in each target area, specifically including:

[0025] According to the number of samples in the target area that contain high-level meteorological conditions for ozone pollution (for example, the meteorological conditions for ozone pollution are level 3), all city types in the target area are divided into two categories: "cities prone to ozone pollution" and "other cities". For cities classified as "cities prone to ozone pollution", a separate model is built. For cities classified as "other cities", samples of cities with similar pollution characteristics and close to each other are merged to build a model. If the pollution characteristics of a city in "other cities" are significantly different from those of surrounding cities and the sample size is sufficient, a separate model is also built. In this way, the number of models to be built and the candidate sample set corresponding to each model are counted.

[0026] As a further optimization, the sample screening specifically includes:

[0027] In the candidate sample sets corresponding to each model, random sampling is performed in proportion according to the frequency distribution of ozone concentration at each level to obtain the training sample sets corresponding to each model:

[0028] If the model corresponds to only one city, first determine the total number of samples N that need to be extracted at each level in the candidate sample set corresponding to the model, and then calculate the sample frequency P of different ozone concentration ranges at each level of the samples that need to be screened i, i is the number of ozone concentration intervals in each level, and finally by calculating P i The product of N is used to obtain the number of samples required for each level of ozone concentration interval, and random sampling is performed;

[0029] If the model corresponds to multiple cities, first determine the total number of samples N required to be extracted at each level in the candidate sample set of these multiple cities. The sample screening method for each city is: according to the proportion w of the number of samples of the corresponding level of the city to the total number of samples of the same level in these multiple cities, calculate the total number of samples w*N required to be extracted at the corresponding level, and then calculate the sample frequency P of different ozone concentration ranges of each level of the samples to be screened i , i is the number of ozone concentration intervals in each level, and finally by calculating P i The product of w*N is used to obtain the number of samples required for each level of ozone concentration range, and random sampling is performed.

[0030] As a further optimization, in step S4, the ozone pollution meteorological condition level discrimination model established in each region includes an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is 5, the number of nodes in the output layer is 3, and the number of hidden layers and the number of nodes are selected comprehensively based on the training error and the discrimination accuracy.

[0031] As a further optimization, in step S5, the sensitive physical parameter data in the target area to be identified is historical actual data or forecast data.

[0032] The beneficial effects of the present invention are:

[0033] (1) From the application perspective, the impact of the five physical parameters determined by the present invention on the occurrence of atmospheric ozone pollution will not change significantly with time and location, and they are all conventional parameters, and the acquisition method is relatively common and simple;

[0034] (2) Based on a large amount of historical data analysis, the ozone pollution characteristics of the target area are obtained, and a model for distinguishing the meteorological conditions for atmospheric ozone pollution is established by region, which is more theoretical and accurate;

[0035] (3) The ozone pollution meteorological condition level discrimination model is based on a neural network and can adapt to the complex nonlinear relationship between ozone pollution and factors such as temperature, precipitation, relative humidity, and mixing layer height. It can effectively discriminate the meteorological condition level when ozone pollution occurs, and has high computational efficiency, low requirements on computer performance, and low cost.

[0036] (4) By judging the level of meteorological conditions for the occurrence of atmospheric ozone pollution, the impact of changes in meteorological conditions on the occurrence of ozone pollution can be effectively reflected. In practical applications, the input data of the model can include historical actual data or forecast data, and different input data can be used according to application requirements. When the input data is historical actual data, the judgment result can be used to evaluate the meteorological conditions for the occurrence of historical ozone pollution. When the input data is forecast data, the judgment result can be used to forecast the meteorological conditions for the occurrence of future ozone pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the method for distinguishing the meteorological condition level of atmospheric ozone pollution in the present invention;

[0038] Figure 2 This is the daily variation trend chart of ozone mass concentration;

[0039] Figure 3 This is the monthly trend chart of ozone mass concentration. DETAILED DESCRIPTION

[0040] The present invention aims to propose a method for distinguishing the level of meteorological conditions for the occurrence of atmospheric ozone pollution, effectively reflect the impact of changes in meteorological conditions on the occurrence of ozone pollution, provide an assessment of the meteorological conditions for the occurrence of historical ozone pollution and a forecast of the meteorological conditions for the occurrence of future ozone pollution, and provide a scientific basis for evaluating the effects of atmospheric ozone pollution prevention and control and formulating atmospheric ozone pollution prevention and control measures.

[0041] The core idea is: first, by analyzing the relationship between ozone and meteorological factors and precursors, the sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are determined as model input parameters; then, the ozone concentration data over a period of time in the target area and the historical data of sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are obtained, and preprocessed to establish a connection between the ozone mass concentration and the level of meteorological conditions for ozone pollution, and the meteorological conditions for ozone pollution are divided into three levels to form model training samples; then, when the multi-dimensional input samples belong to different dimensions and the orders of magnitude differ greatly, the model training samples are normalized; then, a discrimination model for the level of meteorological conditions for ozone pollution is established and trained by region; finally, in practical applications, the sensitive physical parameter data in the target area to be discriminated are used as model input parameters, and based on the trained discrimination model, the discrimination results of the level of meteorological conditions for atmospheric ozone pollution are obtained.

[0042] The specific implementation process of the solution of the present invention is as follows Figure 1 As shown, it includes the following steps:

[0043] 1. Determine the model input physical quantity parameters:

[0044] By analyzing the relationship between ozone and meteorological factors and precursors, sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are obtained as model input parameters. The physical parameters include temperature, precipitation in the previous hour, relative humidity, mixing layer height, and average NO2 mass concentration of the previous night (the specific time period corresponding to the night is determined according to the local daily ozone pollution occurrence period, which will vary in different regions).

[0045] 2. Sample data collection and preprocessing:

[0046] (1) Sample data collection:

[0047] It includes historical air quality monitoring data and real-time meteorological data; wherein the temperature, precipitation, relative humidity, and mixing layer height are derived from meteorological data, and the ozone mass concentration and NO2 mass concentration data are derived from air quality monitoring data.

[0048] (2) Classification of meteorological conditions for ozone pollution:

[0049] According to the Technical Provisions on Ambient Air Quality Index (AQI) (Trial) (HJ 633-2012) issued by the Ministry of Environmental Protection of China, the hourly ozone mass concentration is linked to the meteorological conditions for ozone pollution. The meteorological conditions for ozone pollution are divided into three levels, as shown in Table 1:

[0050] Table 1: Schematic representation of the relationship between hourly ozone mass concentration and meteorological conditions for ozone pollution

[0051]

[0052] From Table 1 we can see that:

[0053] The hourly ozone mass concentration is 0~160μg / m 3 When the air quality is excellent (level 1), the corresponding meteorological condition level for ozone pollution is defined as level 1, indicating that the meteorological conditions are very unfavorable for the occurrence of ozone pollution;

[0054] The hourly ozone mass concentration is 160~200μg / m 3 When the air quality is good (level 2), the corresponding meteorological condition level for ozone pollution is defined as level 2, indicating that the meteorological conditions are not conducive to the occurrence of ozone pollution;

[0055] Hourly ozone mass concentration is greater than 200 μg / m 3 When ozone pollution (mild pollution or above) occurs, the corresponding meteorological condition level for ozone pollution is defined as Level 3, indicating that the meteorological conditions are conducive to the occurrence of ozone pollution.

[0056] (3) Hourly ozone concentration data and meteorological elements temporal and spatial matching:

[0057] The occurrence of ozone pollution is closely related to solar radiation. The sample data selection period is the main ozone pollution period, which is determined according to the historical monthly average and daily average distribution characteristics of ozone pollution in the target area. Since there are fewer environmental monitoring stations in each city and they are mainly concentrated in urban areas, one city is taken as the research object, that is, the hourly ozone concentration is the average of the hourly data of all valid environmental monitoring stations in the city, and the corresponding meteorological elements are the data at the location closest to the center of the effective environmental monitoring station (such as the nearest grid data or site data).

[0058] (4) Regional division:

[0059] Affected by topography, climate conditions and distribution of precursors, ozone pollution characteristics vary in different regions, and the meteorological conditions for occurrence are also different. It is more in line with theory and practice to establish a regional discrimination model for the meteorological conditions for ozone pollution. Considering the geographical location of the city and the number of samples, a discrimination model is established by region. The zoning method is as follows: the number of samples of three levels in the target area is obtained based on a large amount of historical data analysis, and the three-level sample numbers are divided into cities prone to ozone pollution and other cities. Among them, cities prone to ozone pollution are built separately, and cities with similar pollution characteristics and close to each other in other cities are combined to build a model. If the pollution characteristics of a city are very different from those of surrounding cities and the sample size is sufficient, a separate model can also be built. In this way, the number of models to be built and the candidate sample set corresponding to each model are counted.

[0060] (5) Sample screening:

[0061] When training the model, the number of samples of the three levels should be basically consistent. The level with a larger number of samples needs to be randomly screened and extracted. The number of extractions is based on the level with the least number of samples, but not less than 30. Otherwise, the level with more samples will continue to be used as the basis. When only one level meets the conditions, directly take 100.

[0062] In order to ensure that the model training samples conform to the actual distribution law, random sampling is performed in proportion to the distribution frequency of ozone concentration at each level (that is, the proportion of the number of samples in different concentration ranges in each level to the number of samples in the corresponding level). The specific extraction method is as follows:

[0063] 1) If the model corresponds to only one city, first determine the total number of samples N required to be extracted at each level in the candidate sample set corresponding to the model, and then calculate the sample frequency P of each level of different ozone concentration ranges for the samples to be screened i , i is the number of ozone concentration intervals in each level, and finally by calculating P i The product of N is used to obtain the number of samples required for each level of ozone concentration interval, and random sampling is performed;

[0064] For example, for a model corresponding to only one city, 50 samples need to be drawn for each level, that is, N = 50. For the meteorological conditions of ozone pollution, the level 1 corresponds to an hourly ozone concentration of 0 to 160 μg / m 3 , at 10 μg / m 3 As the interval division range, there are 16 concentration intervals, namely: C1 = 0 ~ 10μg / m 3 , C2=10~20μg / m 3 , C3=20~30μg / m 3 …C 16 =150~160μg / m 3 .

[0065] Assume that there are 4501 level 1 samples in the candidate sample set of this city, with ozone concentration ranging from 0 to 10 μg / m 3 There are 45 samples, accounting for 1%, that is, P1 = 1%, and the ozone concentration range is 10-20μg / m 3 There are 135 samples, accounting for 3%, that is, P2 = 3%, and the ozone concentration range is 20-30μg / m 3 There are 248 samples, accounting for 5.5%, that is, P3 = 5.5%... The ozone concentration range is 150-160μg / m 3 There are 171 samples, accounting for 3.8%, that is, P 16 =5.5%

[0066] Then it can be calculated:

[0067] The ozone concentration range is 0-10 μg / m 3 50*1%=0.5 samples are selected from the 45 samples, and rounded off to the nearest integer, and it is finally determined that 1 sample needs to be selected;

[0068] The ozone concentration range is 10-20 μg / m 3 50*3%=1.5 samples are selected from the 135 samples, and by rounding, it is finally determined that 2 samples need to be selected;

[0069] The ozone concentration range is 20-30 μg / m 3 50*5.5%=2.75 samples are selected from the 248 samples. By rounding, it is determined that 3 samples need to be selected.

[0070] …needs to be within the ozone concentration range of 150-160 μg / m 3 50*3.8%=1.9 samples are drawn from the 171 samples. By rounding up, it is finally determined that 2 samples need to be drawn.

[0071] It should be noted that since the number of samples calculated for each concentration interval was processed by rounding, the number of samples drawn in a few intervals also needs to be fine-tuned so that the total number of samples drawn in all concentration intervals of level 1 is 50.

[0072] 2) If the model corresponds to multiple cities, first determine the total number of samples N required to be extracted at each level in the candidate sample set of these multiple cities. The sample screening method for each city is: according to the proportion w of the number of samples of the corresponding level of the city to the total number of samples of the same level in these multiple cities, calculate the total number of samples w*N required to be extracted at the corresponding level, and then calculate the sample frequency P of different ozone concentration ranges of each level of the samples to be screened i Finally, by calculating P i The product of w*N is used to obtain the number of samples required for each level of ozone concentration range, and random sampling is performed.

[0073] 3. Normalization of model input data:

[0074] When multidimensional input samples belong to different dimensions and their orders of magnitude differ greatly, each input parameter must be normalized to avoid affecting the network recognition accuracy due to the difference in orders of magnitude. The maximum and minimum value normalization method is used to normalize the input parameters to between 0 and 1.

[0075] 4. Establish a model to distinguish the meteorological conditions for ozone pollution:

[0076] The ozone pollution meteorological condition level discrimination model established by region includes input layer, hidden layer and output layer:

[0077] The input layer inputs the model training samples. The input physical parameters are five items: temperature, precipitation in the previous hour, relative humidity, mixing layer height, and average NO2 mass concentration in the previous night. Therefore, the number of input layer nodes is 5.

[0078] The output layer contains three characteristic quantities of meteorological conditions for ozone pollution, which are 1, 2, and 3 respectively, so the number of nodes in the output layer is 3.

[0079] The selection of the number of hidden layers and nodes is also very important. Too few nodes will easily lead to poor training results, and too many nodes will easily lead to long training time. Use the training error and discrimination accuracy to select the appropriate number of hidden layer nodes. The training error refers to the mean square error between the network training results and the expected results. The network training results are correct when they are consistent with the expected results, and wrong when they are inconsistent. The discrimination accuracy refers to the percentage of correct samples to the total number of samples.

[0080] By selecting appropriate transfer functions, learning and training functions, number of training times, target error, and target error measurement indicators, the signal is forward propagated and the error is back propagated, and the adjustment of weights and thresholds is repeated until the preset number of learning and training times is reached, or the output error is reduced to an allowable level, the required discriminant model is finally obtained.

[0081] 5. Obtain the results of the meteorological conditions for ozone pollution:

[0082] The historical actual data or forecast data of sensitive physical parameters that have a significant impact on atmospheric ozone pollution in the target area are used as model input parameters to obtain the results of the meteorological conditions for atmospheric ozone pollution. Single-station data or grid data can be selected according to application requirements. The results of historical actual data can be used to evaluate the meteorological conditions for historical ozone pollution, and the results of forecast data can be used to predict the meteorological conditions for future ozone pollution.

[0083] Example:

[0084] This embodiment selects Sichuan Province as an example study area and explains the technical solution in detail. The specific steps are as follows:

[0085] 1. Determine the model input physical quantity parameters:

[0086] By analyzing the relationship between ozone and meteorological factors and precursors in Chengdu, the provincial capital city in the target area, the sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are obtained as model input parameters. The physical parameters include temperature, precipitation in the previous hour, relative humidity, mixing layer height, and NO2 mass concentration in the previous night; according to the distribution characteristics of the daily average value of historical ozone pollution in the target area (such as Figure 2 The night time period is from 22:00 to 08:00 the next day (Beijing time, the same below), and the data fluctuates slightly.

[0087] Table 2 shows the correlation coefficients of air temperature, relative humidity, mixing layer height, NO2 mass concentration of the previous night and ozone mass concentration, all of which are above 0.5; Table 3 shows the removal efficiency of ozone by hourly maximum precipitation. Precipitation has a positive removal effect on ozone, and the removal rates at different levels are large and not much different. This is mainly because when precipitation occurs, there are generally more clouds, and the clouds will absorb solar radiation, which is not conducive to photochemical reactions.

[0088] Table 2: Correlation coefficients between various physical parameters and ozone

[0089] Physical parameters Temperature Relative humidity Mixing layer height <![CDATA[NO2 concentration the previous night]]> Correlation coefficient 0.73 -0.79 0.51 0.52

[0090] Table 3: Maximum hourly precipitation for ozone removal efficiency

[0091] Maximum hourly precipitation / mm 0.1～0.9 1.0～4.9 5.0～9.9 ≥10.0 Average clearance rate / % 25.42 24.46 23.00 34.60

[0092] It can be seen from Table 2 and Table 3 that the five parameters selected in the present invention, namely, air temperature, precipitation in the previous hour, relative humidity, mixing layer height, and NO2 mass concentration in the previous night, are all physical parameters that have a significant impact on the occurrence of atmospheric ozone pollution.

[0093] 2. Sample data collection and preprocessing:

[0094] (1) Sample data collection:

[0095] It includes historical air quality monitoring data from 2018 to 2020 and the national smart grid real-time fusion analysis product (CLDAS). The temperature, precipitation, relative humidity, and mixing layer height come from CLDAS data, and the ozone mass concentration and NO2 mass concentration data come from air quality monitoring data.

[0096] (2) Classification of meteorological conditions for ozone pollution:

[0097] In this embodiment, the meteorological conditions for ozone pollution are divided into three levels according to the method shown in Table 1.

[0098] (3) Hourly ozone concentration data and meteorological elements temporal and spatial matching:

[0099] Ozone pollution is closely related to solar radiation. The sample data is selected during the main ozone pollution period. According to the historical distribution of daily average values ​​of ozone pollution in the target area (such as Figure 2 ) and monthly mean distribution (as shown in Figure 3 The main ozone pollution period is from 09:00 to 21:00 every day from April to September each year. Since there are fewer environmental monitoring stations in each city and they are mainly concentrated in urban areas, one city is taken as the research object, that is, the hourly ozone concentration is the average of the hourly data of all effective environmental monitoring stations in the city, and the corresponding meteorological elements are the grid data closest to the center of the effective environmental monitoring station.

[0100] (4) Regional division:

[0101] Affected by topography, climate conditions and distribution of precursors, ozone pollution characteristics vary in different regions, and the meteorological conditions for occurrence are also different. It is more in line with theory and practice to establish a regional discrimination model for meteorological conditions for ozone pollution. Considering the geographical location of the city and the number of samples, the discrimination model is established by region. The zoning method is as follows: according to a large amount of historical data analysis, the number of samples of the three levels in the target area is obtained (see Table 4), and the three-level sample numbers are divided into cities prone to ozone pollution (Chengdu, Deyang, Meishan, Zigong) and other cities. Among them, cities prone to ozone pollution have a separate model (Chengdu, Deyang, Meishan, Zigong), and other cities with similar pollution characteristics and close to each other are combined to build a model. If the pollution characteristics of a city are quite different from those of the surrounding cities and the sample size is sufficient, a separate model is also built (such as Mianyang). The final 12 models and the number of samples at each level are shown in Table 5.

[0102] Table 4: Number of samples of each level in 21 cities and prefectures in Sichuan Province from 09:00 to 21:00 from April to September 2018 to 2020

[0103] City Level 1 Level 2 Level 3 Chengdu 4501 415 266 Mianyang 4887 261 33 Yibin 4802 298 91 Panzhihua 4898 88 16 Luzhou 4793 308 87 Zigong 4649 347 169 Deyang 4695 357 130 Nanchong 5056 123 10 Guangyuan 5132 44 3 Suining 4946 213 28 Neijiang 4871 248 59 Leshan 4959 170 54 Meishan 4610 441 127 Guangan 4922 244 20 Dazhou 4985 178 26 Ya'an 5003 152 26 Bazhong 4307 25 0 Ziyang 4837 309 36 Aba Prefecture 5135 0 0 Ganzi Prefecture 5126 13 0 Liangshan Prefecture 4885 110 14

[0104] Table 5: Model number, included cities and number of statistical samples at each level

[0105] Model Number City Level 1 Level 2 Level 3 1 Chengdu 4501 415 266 2 Mianyang 4887 261 33 3 Deyang 4695 357 130 4 Zigong 4649 347 169 5 Meishan 4610 441 127 6 Nanchong, Dazhou, Guang'an 14963 545 56 7 Suining, Ziyang, Neijiang 14654 770 123 8 Yibin, Luzhou 9595 606 178 9 Ya'an, Leshan 9962 322 80 10 Liangshan Prefecture, Panzhihua 9783 198 30 11 Guangyuan, Bazhong 9439 69 3 12 Aba Prefecture, Ganzi Prefecture 10261 13 0

[0106] (5) Sample screening:

[0107] When training the model, the number of samples of the three levels should be basically consistent. The level with a larger number of samples needs to be randomly screened and extracted. The number of extractions is based on the level with the least number of samples, but not less than 30. Otherwise, the level with more samples will continue to be used as the basis. When only one level meets the conditions, directly take 100.

[0108] In order to ensure that the model training samples conform to the actual distribution law, random sampling is carried out in proportion according to the distribution frequency of ozone concentration at each level. The specific extraction method is as follows:

[0109] 1) If the model corresponds to only one city, first determine the total number of samples N required to be extracted at each level in the candidate sample set corresponding to the model, and then calculate the sample frequency P of each level of different ozone concentration ranges for the samples to be screened i , i is the number of ozone concentration intervals in each level, and finally by calculating P i The product of N is used to obtain the number of samples required for each level of ozone concentration interval, and random sampling is performed;

[0110] 2) If the model corresponds to multiple cities, first determine the total number of samples N required to be extracted at each level in the candidate sample set of these multiple cities. The sample screening method for each city is: according to the proportion w of the number of samples of the corresponding level of the city to the total number of samples of the same level in these multiple cities, calculate the total number of samples w*N required to be extracted at the corresponding level, and then calculate the sample frequency P of different ozone concentration ranges of each level of the samples to be screened i Finally, by calculating P i The product of w*N is used to obtain the number of samples required for each level of ozone concentration range, and random sampling is performed.

[0111] 3. Normalization of model input data:

[0112] When multidimensional input samples belong to different dimensions and have large differences in magnitude, each input parameter must be normalized to avoid affecting the network recognition accuracy due to differences in magnitude. The maximum and minimum value normalization method is used to normalize the input parameters to between 0 and 1. The formula is:

[0113]

[0114] Where X is the original data, X max , X min are the maximum and minimum values ​​of the original data respectively.

[0115] 4. Establish a model to distinguish the meteorological conditions for ozone pollution:

[0116] This embodiment establishes a model for distinguishing the meteorological conditions for ozone pollution by region based on the BP (Back Propagation) neural network, which includes an input layer, two hidden layers and an output layer:

[0117] The input layer inputs the model training samples. The input physical parameters are five items: temperature, precipitation in the previous hour, relative humidity, mixing layer height, and average NO2 mass concentration in the previous night. Therefore, the number of input layer nodes is 5.

[0118] The output layer contains three characteristic quantities of meteorological conditions for ozone pollution, which are 1, 2, and 3 respectively, so the number of nodes in the output layer is 3.

[0119] The selection of the number of hidden layer nodes is also very important. Too few nodes will easily lead to poor training results, and too many nodes will easily lead to long training time. Use the training error and discrimination accuracy to select the appropriate number of hidden layer nodes. The training error refers to the mean square error between the network training results and the expected results. The network training results are correct when they are consistent with the expected results, and wrong when they are inconsistent. The discrimination accuracy refers to the percentage of correct samples to the total number of samples.

[0120] In this example, as the number of hidden layer nodes increases, the training error decreases. However, when the number of hidden layer nodes is too large, the network training results will be overly consistent. Therefore, in order to save computing resources and achieve ideal discrimination ability, it is more appropriate to set the number of nodes in the first hidden layer between [7,9], and the number of nodes in the second hidden layer can be 1 to 2 more than the first layer.

[0121] The transfer function used in model training is the logarithmic S-type transfer function (logsig), the learning function is the gradient descent method with momentum and adaptive learning rate (traingdx), the number of training times is 10,000, the target error measurement indicator is the mean square error (MSE), and the target error is selected according to specific requirements. In other examples, these parameters can be adjusted according to specific circumstances. After the signal forward propagation and the error back propagation, the adjustment of the weights and thresholds is repeated until the preset number of learning training times or the output error is reduced to an allowable level, and finally the required discriminant model is obtained.

[0122] 5. Obtain the results of the meteorological conditions for ozone pollution:

[0123] The historical actual data of physical parameters such as temperature, precipitation in the previous hour, relative humidity, mixing layer height, and NO2 mass concentration of the previous night in the target area from April 1 to August 16, 2021 were used as model input parameters to obtain the results of the meteorological conditions for the occurrence of atmospheric ozone pollution in Sichuan Province.

[0124] In order to verify the effectiveness of the model, we statistically analyzed the accuracy of the meteorological conditions for ozone pollution in various cities, as shown in Table 6:

[0125] Table 6: Accuracy of meteorological condition classification for atmospheric ozone pollution in Sichuan Province from April 1 to August 16, 2021

[0126] City Discrimination accuracy (%) City Discrimination accuracy (%) Chengdu 67.66 Leshan 65.05 Mianyang 79.10 Meishan 73.15 Yibin 78.22 Guangan 85.68 Panzhihua 72.99 Dazhou 87.34 Luzhou 75.11 Ya'an 67.53 Zigong 73.26 Bazhong 90.17 Deyang 59.43 Ziyang 68.55 Nanchong 86.52 Aba Prefecture 93.24 Guangyuan 84.90 Ganzi Prefecture 95.62 Suining 72.65 Liangshan Prefecture 62.87 Neijiang 74.97 — ——

[0127] It can be seen from Table 6 that the discrimination accuracy of all cities is above 50%, and the discrimination accuracy of 15 cities is above 70%. Therefore, the ozone pollution meteorological condition level discrimination model established by the present invention has a relatively accurate recognition and judgment ability, and is expected to become a new evaluation and forecasting method for the prevention and control of atmospheric ozone pollution.

[0128] It should be noted that the modeling in this embodiment adopts BP neural network. In actual operation, other machine learning methods can also be used to replace BP neural network, such as random forest, long short-term memory neural network (LSTM), GRU (Gated Recurrent) neural network, etc. In addition, the area division method, sample processing method, etc. can also be adjusted accordingly according to actual needs. Therefore, the scheme for which protection is requested in this application includes but is not limited to the above-mentioned embodiments. For ordinary technicians in this field, without departing from the principles of the present invention, several improvements and modifications can be made on the basis of the above-mentioned embodiments, and these improvements and modifications are also regarded as the scope of protection of the present invention.

Claims

1. A method for determining the meteorological condition level of atmospheric ozone pollution, characterized in that: The following steps are involved: S1. By analyzing the relationship between ozone and meteorological factors and precursors, the sensitive physical parameters that affect the occurrence of atmospheric ozone pollution are obtained as model input parameters. The sensitive physical parameters include: temperature, precipitation in the previous hour, relative humidity, mixing layer height, and average NO2 mass concentration in the previous night; S2. Obtain ozone concentration data within a period of time in the target area and historical data of sensitive physical parameters that affect the occurrence of atmospheric ozone pollution, and perform preprocessing to construct training samples; the preprocessing includes: Classify the meteorological conditions for ozone pollution and establish a correlation between the ozone mass concentration and the meteorological conditions for ozone pollution; obtain the number of samples of each level in the target area based on the statistical analysis of historical data; after obtaining the number of samples of each level in multiple target areas, determine the number of models to be built based on the number of samples in each target area, and then perform sample screening; The method of determining the number of models to be modeled based on the number of samples in each target area specifically includes: According to the number of samples with high ozone pollution meteorological conditions in the samples of the target area, the types of cities in all target areas are divided into two categories: "cities prone to ozone pollution" and "other cities". For cities classified as "cities prone to ozone pollution", a separate model is built. For cities classified as "other cities", samples of cities with similar pollution characteristics and close to each other are combined to build a model. If the pollution characteristics of a city in "other cities" are significantly different from those of the surrounding cities and the sample size is sufficient, a separate model is also built. In this way, the number of models to be built and the candidate sample set corresponding to each model are counted. The sample screening specifically includes: In the candidate sample sets corresponding to each model, random sampling is performed in proportion according to the frequency distribution of ozone concentration at each level to obtain the training sample sets corresponding to each model: If the model corresponds to only one city, first determine the total number of samples N that need to be extracted at each level in the candidate sample set corresponding to the model, and then calculate the sample frequency P of different ozone concentration ranges at each level of the samples that need to be screened i , i is the number of ozone concentration intervals in each level, and finally by calculating P i The product of N is used to obtain the number of samples required for each level of ozone concentration interval, and random sampling is performed; If the model corresponds to multiple cities, first determine the total number of samples N required to be extracted at each level in the candidate sample set of these multiple cities. The sample screening method for each city is: according to the proportion w of the number of samples of the corresponding level of the city to the total number of samples of the same level in these multiple cities, calculate the total number of samples w*N required to be extracted at the corresponding level, and then calculate the sample frequency P of different ozone concentration ranges of each level of the samples to be screened i Finally, by calculating P i The product of w*N is used to obtain the number of samples required for each level of ozone concentration interval, and random sampling is performed; S3, normalize the training samples; S4. Establish a model for distinguishing the meteorological conditions for ozone pollution by region, and use normalized training samples to train the model; S5. Using the sensitive physical parameter data in the target area to be identified as the model input parameters, based on the trained ozone pollution meteorological condition level identification model, the identification result of the atmospheric ozone pollution meteorological condition level is calculated.

2. The method for distinguishing the meteorological conditions for atmospheric ozone pollution according to claim 1, characterized in that: In step S2, the method for obtaining ozone concentration data and historical data of sensitive physical parameters that affect the occurrence of atmospheric ozone pollution is: obtaining temperature, precipitation in the previous hour, relative humidity and mixing layer height data through meteorological data; obtaining ozone mass concentration and average NO2 mass concentration data of the previous night through air quality monitoring data.

3. The method for distinguishing the meteorological condition level of atmospheric ozone pollution according to claim 2, characterized in that: The ozone mass concentration is the average hourly data of all effective environmental monitoring stations in the city, and the corresponding meteorological data is the data at the location closest to the center of the effective environmental monitoring station.

4. The method for determining the meteorological condition level of atmospheric ozone pollution according to claim 1, characterized in that: The classification of meteorological conditions for ozone pollution and the association of ozone mass concentration with the meteorological conditions for ozone pollution specifically include: The meteorological conditions for ozone pollution are divided into three levels, each level corresponds to a different hourly ozone mass concentration. The lower the hourly ozone mass concentration, the lower the level. When the meteorological conditions for ozone pollution are level 1, it indicates that the meteorological conditions are very unfavorable for the occurrence of ozone pollution; when the meteorological conditions for ozone pollution are level 2, it indicates that the meteorological conditions are unfavorable for the occurrence of ozone pollution; when the meteorological conditions for ozone pollution are level 3, it indicates that the meteorological conditions are favorable for the occurrence of ozone pollution.

5. The method for determining the meteorological condition level of atmospheric ozone pollution according to claim 1, characterized in that: In step S4, the ozone pollution meteorological condition level discrimination model established by region includes an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is 5, the number of nodes in the output layer is 3, and the number of hidden layers and the number of nodes are comprehensively selected based on the training error and the discrimination accuracy.

6. The method for determining the meteorological condition level of atmospheric ozone pollution according to any one of claims 1 to 5, characterized in that: In step S5, the sensitive physical parameter data in the target area to be determined is historical actual data or forecast data.

Images