Atmospheric pollution driving factor analysis method and device and storage medium

Abstract

The invention discloses an atmospheric pollution driving factor analysis method and device and a storage medium, and belongs to the field of environmental science and data intelligent analysis, and the method comprises the steps: training an ensemble learning model based on meteorological data and pollutant concentration observation data of a reference year; obtaining a pollutant concentration prediction model, a corresponding reference year training prediction value and a reference year model residual error; inputting the meteorological data of the target year into the pollutant concentration prediction model, and outputting a pollutant concentration prediction value of the target year under the emission assumption of the reference year; determining a meteorological driving item according to the pollutant concentration prediction value of the target year and the training prediction value of the reference year; according to the obtained pollutant concentration actual observation value of the target year, the pollutant concentration prediction value of the target year and the reference year model residual error, an artificial emission driving item is determined; according to the method, on the premise of not depending on a traditional chemical conveying model, efficient stripping of quantitative contribution of meteorological condition changes and emission changes in the pollution trend can be achieved.

Description

Technical Field

[0001] The present invention relates to an atmospheric pollution driving factor analysis method, device and storage medium, and belongs to the field of environmental science and data intelligent analysis. Background Art

[0002] With the gradual development and widespread use of ground-based observation platforms for atmospheric pollution, a vast amount of long-term observational data with high temporal and spatial resolution has been accumulated. In the actual atmospheric environment, changes in pollutant concentrations are often influenced by multiple factors, particularly the synergistic effects of anthropogenic emissions and meteorological conditions. However, relying solely on observational data makes it difficult to directly isolate the contributions of different driving factors, posing challenges for accurately assessing the effectiveness of pollution control and analyzing the mechanisms of pollution evolution.

[0003] In existing technologies, atmospheric chemical transport models are often used in conjunction with emission inventories and meteorological field data for simulation and analysis. However, such methods have high computational complexity, rely on server systems, have high operating costs, and have time lags and uncertainties in emission inventories, which limit their application in rapid response and decision support. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method, device, and storage medium for analyzing atmospheric pollution driving factors, which can efficiently separate the quantitative contributions of meteorological condition changes and emission changes to pollution trends without relying on traditional chemical transport models.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a method for analyzing driving factors of air pollution, comprising: Based on the meteorological data and pollutant concentration observation data of the base year, the ensemble learning model is trained to obtain the pollutant concentration prediction model and the corresponding base year training prediction value and base year model residual; Input the target year's meteorological data into the pollutant concentration prediction model, and output the predicted pollutant concentration value for the target year under the base year emission assumption; Determine the meteorological driving factors based on the predicted pollutant concentrations in the target year and the training predicted values ​​in the base year; Determine the anthropogenic emission driving factors based on the actual observed pollutant concentrations in the target year, the predicted pollutant concentrations in the target year, and the model residuals in the base year; Output meteorological driving items and anthropogenic emission driving items as analytical results of atmospheric pollution driving factors.

[0006] Furthermore, the ensemble learning model integrates the gradient boosting regression tree algorithm, the extreme gradient boosting algorithm and the random forest algorithm, and integrates the prediction results of the ensemble learning model through ridge regression.

[0007] Further, the target year in the benchmark year emission assumption under the pollutant concentration prediction value, the calculation formula is as follows: Pred i(m) = n 1 X RFi(m) + n 2 X GBRTi(m) + n 3 X XGBoosti(m) ; Wherein, Pred i(m) is the target year in the benchmark year emission assumption under the pollutant concentration prediction value, X RFi(m) is the prediction value of random forest algorithm, X GBRTi(m) is the prediction value of gradient boosting regression tree algorithm, X XGBoosti(m) is the prediction value of extreme gradient boosting algorithm, n 1 , n 2 , n 3 is the coefficient of regression model.

[0008] Further, the meteorological driving term is determined according to the pollutant concentration prediction value of the target year and the benchmark year training prediction value, and the formula is as follows: ; Wherein, MET is the meteorological driving term, Pred i(m) is the pollutant concentration prediction value of the target year, and Pred i(i) is the benchmark year training prediction value.

[0009] Further, the benchmark year model residual is determined by the difference between the observed value and the prediction value of the pollutant concentration of the benchmark year during the training process of the pollutant concentration prediction model.

[0010] Further, the artificial emission driving term is determined according to the obtained actual observation value of the pollutant concentration of the target year, the pollutant concentration prediction value of the target year and the benchmark year model residual, and the formula is as follows: ; Wherein, ANT is the artificial emission driving term, Obs m is the actual observation value of the pollutant concentration of the target year in the past, Pred i(m) is the pollutant concentration prediction value of the target year, and Res i is the benchmark year model residual.

[0011] Furthermore, the method further includes: a step of quantifying the contribution of short-term anthropogenic emission reduction events, specifically: Select base year i The pollutant concentration prediction model is trained based on the observation data of the target period and applied to the meteorological data of the target period to predict the pollutant concentration Pred during the emission reduction period under the baseline emission conditions. i(LD) and the pollutant concentration Pred during the non-abatement period i(NLD) ; The contribution of short-term emission reduction events to pollutant concentrations is calculated using the following formula: ; Among them ANT short represents the short-term anthropogenic emission reduction driving term, C LD Obs is the proportion of samples during the emission reduction period in all samples. LD Obs is the measured value of pollutant concentration during the emission reduction period, NLD It is the actual measured value of pollutant concentration during the non-emission reduction period.

[0012] Furthermore, the pollutant concentration is specifically PM 2.5 concentration, and the method is applicable to the analysis of driving factors of different regions and multiple pollutants.

[0013] Furthermore, the performance of the pollutant concentration prediction model was evaluated by the uncertainty of the integrated learning model using the root mean square error (RMSE), the standardized mean deviation (NMB), the goodness of fit index (IOA), and the correlation coefficient (R).

[0014] In a second aspect, the present invention provides an atmospheric pollution driving factor analysis device, comprising: Memory, used to store computer programs / instructions; A processor is configured to execute the computer program / instructions to implement the steps of any of the aforementioned methods.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the aforementioned methods when executed by a processor.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method, device and storage medium for analyzing atmospheric pollution driving factors. Starting from observation data, by constructing a pollutant concentration prediction model, it is possible to efficiently separate the quantitative contributions of changes in meteorological conditions and emissions to pollution trends without relying on traditional chemical transport models. The present invention significantly reduces model operation time and computing resource consumption, improves the accuracy of prediction and separation results through an integrated learning model, and has good consistency with the results of traditional atmospheric models. The present invention is applicable to the analysis of atmospheric observation data in various regions and pollution types, and has good generalizability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the performance evaluation of the method model provided by the embodiment of the present invention; Figure 2 Schematic diagram of comparative analysis between the method of the present invention and the WRF / CMAQ model simulation method in pollutant driving factor analysis provided by an embodiment of the present invention; Figure 3 This is a flow chart of a method for analyzing driving factors of atmospheric pollution provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] Example 1, as Figure 3 As shown, this embodiment introduces a method for analyzing driving factors of air pollution, including: Based on the meteorological data and pollutant concentration observation data of the base year, the ensemble learning model is trained to obtain the pollutant concentration prediction model and the corresponding base year training prediction value and base year model residual; Input the target year's meteorological data into the pollutant concentration prediction model, and output the predicted pollutant concentration value for the target year under the base year emission assumption; Determine the meteorological driving factors based on the predicted pollutant concentrations in the target year and the training predicted values ​​in the base year; Determine the anthropogenic emission driving factors based on the actual observed pollutant concentrations in the target year, the predicted pollutant concentrations in the target year, and the model residuals in the base year; Output meteorological driving items and anthropogenic emission driving items as analytical results of atmospheric pollution driving factors.

[0020] The present invention provides an analytical method for atmospheric pollution driving factors, which is used to distinguish atmospheric pollutants (PM 2.5The relative contributions of meteorological conditions and anthropogenic emissions to interannual variability. The core idea is to train a machine learning model based on multiple years of observational data under the assumption that anthropogenic emissions remain unchanged, cross-validate the impact of meteorological condition changes on pollutant concentrations, and back-calculate the driving effect of emission changes. At the same time, the method can assess the uncertainty range introduced due to different training sample selection.

[0021] The specific steps are as follows: First, select a certain reference year i (where i is any year in the target period), use the observation data of this year to train a machine learning model, where 70% of the data is used for model training and 30% of the data is used for model testing, and use RMSE, NMB, IOA and R as evaluation indicators to evaluate the model performance. In order to improve the prediction performance and robustness, an ensemble learning strategy is adopted, which integrates three mainstream algorithms: gradient boosting regression tree (GBRT), extreme gradient boosting algorithm (XGBoost) and random forest algorithm (RF). After obtaining the prediction results of each base model, ridge regression is used for ensemble optimization, and then the trained model is applied to the meteorological input data of the target year m (where m is any year in the target period, or the entire target period), the specific formula is as follows: Pred i(m) = n 1 X RFi(m) + n 2 X GBRTi(m) + n 3 X XGBoosti(m) ; where the dependent variable Pred i(m) is the result of ridge regression on the predicted variable PM 2.5 , and the predicted value represents the trend of pollutant change under the premise of meteorological condition change but unchanged emissions. By cross-modeling and predicting data of different years, a double nested cross-validation framework is established. First, the n reference year data is constructed in the outer loop, , and for each i , the inner loop of the prediction year m is established, , so as to minimize the uncertainty brought by training sample selection. The independent variables X represent the predicted values from the three algorithms, and the coefficients of the regression model n 1 , n 2 , n3 The contribution of each algorithm prediction is optimized, the multicollinearity problem between different machine learning model predictions is effectively solved, and the reliability and accuracy of the prediction results are ensured.

[0022] The core calculation formula is as follows: PM 2.5 Observation value decomposition of concentration: Obs=Pred+Res; Wherein, Obs is the actual observation value, Pred is the model prediction value, and Res is the model residual.

[0023] The separation expression of meteorological driving term and anthropogenic emission driving term is: DeltaObs=MET+ANT; Wherein, DeltaObs represents the total amount of PM 2.5 Concentration change between different years, MET represents the concentration change caused by meteorological change, and ANT represents the concentration change caused by anthropogenic emission change.

[0024] Meteorological driving term estimation formula: ; That is, under the condition that the emission keeps unchanged in the benchmark year, the difference between the predicted values in different years is the contribution of the meteorological factor.

[0025] Anthropogenic emission driving term estimation formula: ; That is, the difference between the measured value and the predicted value under the fixed emission assumption, after eliminating the model error, is the influence caused by the emission change.

[0026] In addition, the method of the present application can also be extended to identify the influence of short-term anthropogenic emission reduction events on air pollution, further expanding the cross-matrix modeling and prediction framework. The specific method is as follows: selecting a historical year i As the benchmark year, the observation data of the year is used to train the model, and the trained model is applied to the meteorological condition input during the emission reduction period to predict the pollutant concentration level under the benchmark emission assumption, so as to realize the identification and quantification of the influence of short-term unconventional emission change.

[0027] The contribution of short-term emission reduction events to pollutant concentration is calculated, and the formula is as follows: ; Wherein, Obs LD and Pred i(LD) are the measured value and the predicted value during the emission reduction period, Obs NLD and Pred i(NLD) are the data during the same period but not during the emission reduction period, and C LDThe proportion of the sample during the emission reduction period in all samples.

[0028] Figure 1 The method model performance evaluation. Based on the test set data of the integrated learning model developed independently in the Yangtze River Delta (YRD) region from 2013 to 2020, the figure is the summary of the performance indicators of each annual model. The root mean square error (RMSE) of each annual test set is between 8.0-14.1 μg m −3 , the normalized mean bias (NMB) is between 8%-18%, the goodness of fit index (IOA) is stable between 0.96-0.97, and the correlation coefficient (R) reaches the level of 0.92-0.95. All indicators show that the model maintains excellent prediction accuracy and robustness in cross-year validation.

[0029] Figure 2 . The method of the present application and the simulation method of WRF / CMAQ model in the analysis of pollutant driving factor. a indicates the decomposition results of meteorological changes, emission changes and emission reduction period obtained by using the simulation method of WRF / CMAQ model; b indicates the quantitative identification results of meteorological changes, emission changes and emission reduction period obtained by using the "cross matrix modeling and prediction method" of the present application.

[0030] The embodiment provides an atmospheric pollution driving factor analysis method based on an integrated learning algorithm. The method starts from observation data, constructs a nonlinear response model between atmospheric pollutant concentration and meteorological conditions, innovatively introduces a cross matrix prediction mechanism, and can realize efficient separation of the quantitative contribution of meteorological condition changes and emission changes in pollution trend without relying on traditional chemical transport models.

[0031] The method has the following advantages: Efficiency: greatly reduces model operation time and computing resource consumption, suitable for conventional computing platforms (such as notebook computers); High precision: improve the accuracy of prediction and separation results through integrated learning model, and have good consistency with traditional atmospheric model results; Strong adaptability: suitable for atmospheric observation data analysis of various regions and pollution types, and has good generalizability.

[0032] Embodiment 2, the embodiment provides an atmospheric pollution driving factor analysis device, comprising: a memory for storing computer programs / instructions; a processor for executing the computer programs / instructions to realize the steps of the method of any one of embodiments 1.

[0033] Embodiment 3 provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method of any one of Embodiments 1.

[0034] The above merely provides the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and variations can be made, which should be considered as the protection scope of the present application.

[0035] Those skilled in the art will appreciate that embodiments of the present disclosure can be provided as methods, systems or computer program products. Therefore, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer usable program code.

[0036] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more functions specified in the flowchart or flows and / or blocks. Figure 1 an apparatus that implements one or more functions specified in the flowchart or flows and / or blocks.

[0037] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more functions specified in the flowchart or flows and / or blocks. Figure 1 an apparatus that implements one or more functions specified in the flowchart or flows and / or blocks.

[0038] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more functions specified in the flowchart or flows and / or blocks. ​ an apparatus that implements one or more functions specified in the flowchart or flows and / or blocks.

[0039] It should be pointed out finally that the above embodiments are only used for illustrating the technical solutions of the present disclosure but not for limiting the protection scope thereof, and although the present disclosure has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that after reading the present disclosure, various modifications, changes or equivalent replacements of the specific embodiments of the present disclosure can be made by those of ordinary skill in the art, but these modifications, changes or equivalent replacements are within the protection scope of the disclosed claims.

Claims

1. A method for analyzing driving factors of air pollution, characterized in that: include: Based on the meteorological data and pollutant concentration observation data of the base year, the ensemble learning model is trained to obtain the pollutant concentration prediction model and the corresponding base year training prediction value and base year model residual; Input the target year's meteorological data into the pollutant concentration prediction model, and output the predicted pollutant concentration value for the target year under the base year emission assumption; Determine the meteorological driving factors based on the predicted pollutant concentrations in the target year and the training predicted values ​​in the base year; Determine the anthropogenic emission driving factors based on the actual observed pollutant concentrations in the target year, the predicted pollutant concentrations in the target year, and the model residuals in the base year; Output meteorological driving items and anthropogenic emission driving items as analytical results of atmospheric pollution driving factors.

2. The method for analyzing driving factors of air pollution according to claim 1, characterized in that: The ensemble learning model integrates the gradient boosting regression tree algorithm, the extreme gradient boosting algorithm and the random forest algorithm, and integrates the prediction results of the ensemble learning model through ridge regression.

3. The method for analyzing driving factors of air pollution according to claim 2, characterized in that: The predicted pollutant concentration in the target year under the base year emission assumption is calculated using the following formula: Pred i(m) = n 1 X RFi(m) + n 2 X GBRTi(m) + n 3 X XGBoosti(m) ; Among them, Pred i(m) is the predicted pollutant concentration in the target year under the assumption of base year emissions, X RFi(m) is the predicted value of the random forest algorithm, X GBRTi(m) is the predicted value of the gradient boosting regression tree algorithm, X XGBoosti(m) is the predicted value of the extreme gradient boosting algorithm, n 1 , n 2 , n 3 are the coefficients of the regression model.

4. The method for analyzing driving factors of air pollution according to claim 1, characterized in that: The base year model residual is determined by the difference between the observed value and the predicted value of the pollutant concentration in the base year during the pollutant concentration prediction model training process.

5. The method for analyzing driving factors of air pollution according to claim 3, characterized in that: The meteorological driving term is determined based on the predicted pollutant concentration value of the target year and the training predicted value of the base year. The formula is as follows: ； Among them, MET is the meteorological driving term, Pred i(i) The predicted values ​​are trained for the base year.

6. The method for analyzing driving factors of air pollution according to claim 3, characterized in that: The anthropogenic emission driving term is determined based on the actual observed values ​​of pollutant concentrations in the target year, the predicted values ​​of pollutant concentrations in the target year, and the model residuals in the base year. The formula is as follows: ； Among them, ANT is the anthropogenic emission driving term, Obs m is the actual observed value of the pollutant concentration in the target year in advance, Res i is the base year model residual.

7. The method for analyzing driving factors of air pollution according to claim 1, characterized in that: The method further includes: a step of quantifying the contribution of short-term anthropogenic emission reduction events, specifically: Select base year i The pollutant concentration prediction model is trained based on the observation data of the target period and applied to the meteorological data of the target period to predict the pollutant concentration Pred during the emission reduction period under the baseline emission conditions. i(LD) and the pollutant concentration Pred during the non-abatement period i(NLD) ; The contribution of short-term emission reduction events to pollutant concentrations is calculated using the following formula: ； Among them ANT short represents the short-term anthropogenic emission reduction driving term, C LD Obs is the proportion of samples during the emission reduction period in all samples. LD Obs is the measured value of pollutant concentration during the emission reduction period, NLD It is the actual measured value of pollutant concentration during the non-emission reduction period.

8. The method for analyzing driving factors of air pollution according to claim 1, characterized in that: The performance of the pollutant concentration prediction model was evaluated by ensemble learning model uncertainty and cross-year cross-validation through root mean square error (RMSE), standardized mean deviation (NMB), goodness-of-fit index (IOA) and correlation coefficient (R).

9. An atmospheric pollution driving factor analysis device, characterized in that: include: Memory, used to store computer programs / instructions; A processor configured to execute the computer program / instructions to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images