Chemical mechanism verification optimization method based on peroxide free radical constraint

Abstract

The invention discloses a chemical mechanism verification optimization method based on peroxy free radical constraint, and relates to the technical field of atmospheric chemical simulation and numerical model optimization, and the method comprises the following steps: S1, obtaining and preprocessing observation data; s2, initial model setting and simulation operation; s3, performing closure degree analysis and deviation diagnosis; s4, carrying out global sensitivity analysis and key reaction identification; s5, performing mechanism optimization and iterative optimization circulation; and S6, optimizing effect evaluation and mechanism generalization ability testing. Based on an observation and simulation closed analysis method, model mechanism deviation and causes thereof are identified and diagnosed by comparing and analyzing observation data and simulation results of peroxy free radicals. On the basis, global sensitivity analysis is introduced, a key chemical reaction which has the most significant influence on simulation uncertainty is positioned, then dynamic parameters of the key reaction are systematically optimized, the simulation result and observation data are optimally matched, and finally a set of parameter-optimized chemical mechanism is obtained. And the simulation accuracy of the model on the peroxy free radical concentration and the atmospheric oxidation capacity is obviously improved.

Description

Technical Field

[0001] This invention relates to the field of atmospheric chemical simulation and numerical model optimization technology, and in particular to a chemical mechanism verification and optimization method based on peroxy radical constraints. Background Technology

[0002] Peroxy radicals (HO2 and RO2) play a central role in atmospheric oxidation, controlling the regeneration of hydroxyl radicals (OH), the formation of ozone (O3), and the formation of secondary pollutants. Currently, widely used atmospheric chemical mechanisms (such as MCM and RACM) all involve complex peroxy radical chemical reactions. However, these mechanisms often exhibit significant deviations when simulating real atmospheric environments, especially in the simulation of HO2 and RO2 concentrations.

[0003] Traditional mechanistic verification methods largely rely on simulation and observational comparisons of stable species (such as O3, NOx, and VOCs), while direct constraints on key intermediates—peroxy radicals—are relatively weak. This is mainly because peroxy radicals are highly reactive and have low concentrations, making accurate measurement difficult. In recent years, the maturity of techniques such as peroxidase-peroxidase-chemical-scale (PERCA) has made high-temporal-resolution online observation of HO2 and RO2 possible, providing a data foundation for direct verification and optimization of their chemical mechanisms.

[0004] Current technologies lack a systematic, closed-loop framework capable of effectively utilizing observational data to diagnose mechanistic defects, identify key uncertainties, and guide targeted optimization of mechanisms. Therefore, developing a method that combines observation, simulation, sensitivity analysis, and iterative mechanistic upgrades is crucial for improving the accuracy of atmospheric chemistry model predictions. Summary of the Invention

[0005] The purpose of this application is to provide a chemical mechanism verification and optimization method based on peroxy radical constraints, aiming to fundamentally improve the problem of insufficient credibility of chemical mechanisms caused by peroxy radical simulation bias.

[0006] The chemical mechanism verification and optimization method based on peroxy radical constraint includes the following steps:

[0007] S1, Observational data acquisition and preprocessing: Simultaneously acquire high temporal resolution observational concentration data of peroxy free radicals in the target area or experimental environment, as well as synchronous auxiliary parameter observational data, and perform quality control, unified timestamp and average time processing.

[0008] S11, Collaborative Observation Data Acquisition: At the same time and location, multiple environmental element monitoring equipment are integrated to conduct continuous observation and acquire basic data;

[0009] S12, Data Quality Control and Dataset Construction: Through data quality control, data alignment and resampling, the problems of noise, drift, missing values ​​and inconsistent temporal resolution in multi-source raw data are solved;

[0010] S2, Initial Model Setup and Simulation Run: Run an unmodified, accepted baseline chemical mechanism under conditions as close as possible to real observation conditions to generate a set of "ideal" peroxy radical simulation data, which will be compared with actual observation data;

[0011] S21, Model Selection: Based on the purity of the chemical process and computational efficiency, a zero-dimensional box model is selected;

[0012] S22, Model constraint settings: Constrain the model using concentration, meteorological parameters, and radiation parameters;

[0013] S23, Model Run and Output: Integrate all constraints into the box model, set the time range and step size consistent with the observation data, run the model, and then output the simulated concentration time series of peroxy free radicals.

[0014] S3, Closure analysis and deviation diagnosis: Quantitatively assess the consistency between the initial simulation and observation data, and qualitatively determine the patterns and potential causes of inconsistencies, thus providing direction for subsequent sensitivity analysis;

[0015] S31, Closure analysis: Calculate the ratio of simulation to observation, correlation, and root mean square error to analyze the deviation between the model and the observation;

[0016] S32, Deviation Analysis: By dividing the observed and simulated data into bins using key environmental parameters, the simulation closure of each bin is evaluated, thereby identifying systematic deviations of the model under specific conditions. These deviation patterns can directly point to deficiencies in the relevant chemical reaction mechanisms, providing clear targets for subsequent sensitivity analysis and mechanism optimization.

[0017] S4, Global Sensitivity Analysis and Key Response Identification: Varying all input parameters across the entire reasonable parameter space and evaluating the contribution of each parameter individually and the interactions between parameters to the output uncertainty;

[0018] S41, Global Sensitivity Analysis: Screen key response parameters, generate parameter sample sets using the Monte Carlo method within their defined range, input the samples into the box model and run it, calculate the first-order and total-order sensitivity indices of each parameter based on the model output results, thereby quantitatively evaluating the contribution of individual parameters and their interactions to the uncertainty of the simulation results;

[0019] S42, Key Reaction Identification: Select the top 5-10 parameters with the highest total sensitivity index, which are the key reactions that contribute the most to the uncertainty of peroxy radical simulation results under specific conditions;

[0020] S5, Mechanism Optimization and Iterative Optimization Cycle: Based on the list of key reactions selected in S4, within a reasonable physicochemical range, the kinetic parameters of the key reactions are systematically optimized to achieve the best fit between the simulation results and the observation data;

[0021] S6, Evaluation of optimization effect and test of mechanism generalization ability: Re-simulate under the same conditions using the optimized parameters. By comparing and analyzing with the baseline simulation and observation data, verify whether the simulation effect has been improved. Then, use an independent observation dataset to test the generalization ability of the optimized mechanism. If its performance is significantly better than the baseline mechanism, it indicates that the optimization has successfully captured universal chemical laws; otherwise, it is necessary to backtrack and replace the data.

[0022] S61, Optimization effect evaluation: Under the same conditions as S2, rerun the model using the optimization parameters and compare the new simulation results with the baseline simulation and observation data before optimization. By calculating indicators such as S / O Ratio, R2, and RMSE, verify whether each indicator shows significant and consistent improvement. If the improvement is not obvious or the results worsen, the optimization process needs to be backtracked and checked.

[0023] S62, Mechanism Generalization Ability Test: The baseline mechanism and the optimized mechanism are tested using independent observation data, and the key indicators of the two are compared. If the RMSE of the optimized mechanism is significantly reduced, it indicates that it has successfully captured universal chemical laws; otherwise, it indicates that the optimization may overfit the training set noise and needs to be re-validated with different data.

[0024] As a further improvement of the present invention, in step S11, the observation of the basic data includes core observation and auxiliary parameter observation. The core observation is the observation of peroxy free radicals, and the auxiliary parameter observation is the observation of nitrogen oxides, volatile organic compounds, photolysis rate, ozone, carbon monoxide, sulfur dioxide, nitrous acid, and meteorological parameters such as wind, temperature, humidity, and pressure. The inlet of the observation instrument should be as close as possible to each other. Data acquisition and preprocessing are achieved through multi-parameter collaborative observation. Collaborative observation data acquisition is achieved by integrating and deploying multiple monitoring devices at the same spatiotemporal location, and by adopting a close-proximity layout of the inlet and synchronization with the NTP timestamp, thus realizing the synchronous capture of the same air mass by multiple parameters (core and auxiliary). Its core advantage lies in ensuring, from the source of observation, the high degree of consistency of all input data in terms of spatial representativeness and temporal series, establishing a reliable and unified realistic benchmark for subsequent observation-simulation comparison.

[0025] As a further improvement of the present invention, in step S12, the data quality control includes regularly performing zero-point and range calibration of the equipment, identifying and eliminating invalid periods due to instrument maintenance and human interference, removing obviously abnormal data points based on physical probability and statistical methods, and smoothing high-frequency data. The data alignment and resampling include unifying the timestamps of all instrument data to the same time zone and unifying data with different time resolutions to the same time grid through interpolation methods. The dataset construction involves data merging through QC and resampling. The merged data is then organized into a data table, which is used as the input file to drive and constrain the box model. In the data quality control and dataset construction, through a systematic quality control process (such as calibration, outlier removal, and smoothing) and strict data alignment and resampling operations, multi-source and heterogeneous raw observation data are integrated into a time-uniform and formatted driving file. Its core advantage lies in effectively eliminating instrument noise and errors, solving the problems of time misalignment and resolution differences between data, thereby generating a high-quality fusion dataset that can be directly used for model driving, ensuring the accuracy and reliability of the model's initial and boundary conditions.

[0026] As a further improvement of the present invention, in step S2, the concentration constraint involves inputting the initial concentrations of various observed chemical species into the model, setting the concentrations of species with long lifespans and slow changes to fixed values, and directly inputting the observation time series of species with drastic concentration changes into the model to simulate the current concentration. The meteorological parameter constraint includes inputting the observed temperature, humidity, and pressure time series, which are used to constrain the reaction process. The radiation parameter constraint is achieved by inputting the actual observed time series of photolysis rates of each species into the model. The initial model setup and simulation operation employ a zero-dimensional box model framework and strictly constrain all observable environmental driving factors (concentration, physical parameters, and radiation parameters) to actual observed values, completing the core design from complex realistic simulation to isolated testing of chemical mechanisms. Its core advantage lies in maximally eliminating the influence of external uncertainties such as physical transport, meteorological simulation, and radiation calculation on the simulation results, forcing the model to operate under boundary conditions completely consistent with reality. The resulting difference between simulation and observation can be more directly and reliably attributed to defects in the chemical mechanism itself (reaction pathway, kinetic parameters), thus providing a pure signal for subsequent deviation diagnosis and optimization.

[0027] As a further improvement of the present invention, in step S3, the binning process involves data types including NOx concentration, VOCs composition, concentration of dominant VOCs species, photolysis intensity, diurnal cycle, and temperature parameters. The deviation diagnosis is used to pinpoint problems in a certain type of chemical reaction. Closure analysis and deviation diagnosis include quantitative assessment using the simulation / observation ratio, correlation coefficient, and root mean square error, and identification of systematic deviation patterns by binning data for key environmental parameters (such as NOx, VOCs, and light intensity). This method achieves a technological leap from overall error assessment to condition-specific deviation tracing. Its core advantage lies in its ability to not only quantitatively reveal the overall deviation between simulation and observation, but also to correlate deviations with specific environmental conditions (such as high NOx) and chemical reaction types. This transforms the vague problem of model "inaccuracy" into a clear scientific hypothesis of "under what conditions and in what chemical mechanisms problems might exist," providing precise directional targets for subsequent sensitivity analysis.

[0028] As a further improvement of the present invention, in step S4, the reaction with uncertainty is used to extract uncertain reaction parameters. K uncertain reaction parameters are extracted, and a reasonable range of variation is defined for K according to the reasonable parameter space. The range of variation is used to generate a sample set. The sample set is generated using the Monte Carlo method to establish a K-dimensional parameter space for K. The number of parameters in the K-dimensional parameter space is set to k. N sets of parameter combinations are extracted from the K-dimensional parameter space, and the number of samples is set to N. Two independent N×k sampling matrices A and B are generated through the K-dimensional parameter space. The formulas for A and B are as follows:

[0029] , ;

[0030] The x represents parameter values ​​for different rows and columns in the matrix. One of the K parameters is set to i. The exponent of each i is calculated, and a transformation matrix A is constructed based on i. B (i) The transformation matrix A B (i) Specifically, it means replacing the i-th column of matrix A with the i-th column of matrix B, where matrix A... B (i) The formula is as follows:

[0031] ;

[0032] Substituting matrices A and B into the box model, and using the same constraints in S2 to select key time periods for S3, we obtain the model output vectors ƒ(A), ƒ(B), and ƒ(A...). B (i) ), the ƒ(A), ƒ(B), ƒ(AB (i) All are n-dimensional column vectors.

[0033] As a further improvement of the present invention, the model output vector is used to calculate the first-order exponent and the total-order exponent, wherein the first-order exponent is set to S. i The total order exponent is set to S. Ti The S i and S Ti The calculation formula is as follows:

[0034] ;

[0035] ;

[0036] The j represents the j-th dimension of the output. A Monte Carlo method is used to generate a sample matrix, and the independent and interactive contributions of each reaction parameter to simulation uncertainty are quantified by calculating first-order and total-order sensitivity indices. This achieves a shift from empirical parameter selection to systematic quantitative attribution. Its core advantage lies in its ability to objectively and quantitatively identify the critical few reactions that have the most significant impact on the simulation results from dozens of uncertain reactions. This effectively overcomes the subjectivity and limitations of relying on expert experience, ensuring that subsequent optimization work focuses on the most core sources of mechanistic uncertainty.

[0037] As a further improvement of the present invention, in step S5, the uncertainty range is used to determine the adjustment boundary. The iterative optimization loop proposes a new set of parameter values ​​according to the optimization algorithm. These parameter values ​​are substituted into a box model with the same constraints as S2 and S4. The iterative parameters are set with the objective function as the guide, and the iterative parameters are iterated until the stopping condition is met. Mechanism optimization and iterative loop take the key reaction parameters identified in S4 as the optimization object. By setting the objective function (such as minimizing RMSE) and using the optimization algorithm for automatic iterative optimization, the parameter set that minimizes the difference between simulation and observation is finally obtained, realizing a technological leap from manual trial and error adjustment to algorithm-driven efficient optimization. Its core advantage is that by establishing an automated iterative optimization framework, the optimal parameter combination can be searched efficiently and systematically within a reasonable materialized boundary, significantly improving the efficiency and objectivity of the optimization process and avoiding the blindness and inefficiency of manual parameter tuning.

[0038] Compared with the prior art, the beneficial effects of this invention are as follows:

[0039] 1. A complete closed-loop framework from data acquisition to mechanism verification and optimization has been constructed, significantly improving the systematicness and reproducibility of the research. This invention begins with multi-parameter collaborative observation and rigorous quality control, separates the chemical mechanism from the physical process through a zero-dimensional model, and then, through closed-loop analysis, sensitivity identification, parameter optimization, and dual verification, forms a standardized and complete process. This solves the problems of disconnected links and reliance on subjective experience in traditional research, providing a scientific, transparent, and reproducible method for the evaluation and improvement of chemical mechanisms.

[0040] 2. This invention enables precise tracing of errors from overall data to condition-specific biases, greatly enhancing the depth and guiding value of problem diagnosis. Traditional methods often only provide overall deviations between simulation and observation. By introducing data binning technology based on key environmental parameters, this invention can systematically identify differences in model performance under specific chemical conditions (such as high NOx), thereby transforming the general conclusion of "inaccurate model" into scientific hypotheses pointing to specific reaction types or pathways, providing clear and precise targets for subsequent mechanism correction.

[0041] 3. This invention introduces objective and quantitative sensitivity analysis and an algorithm-driven optimization process, overcoming the subjectivity and inefficiency of traditional experience-based parameter tuning. By utilizing global sensitivity analysis, this invention objectively and quantitatively identifies the key response parameters that contribute most to simulation uncertainty, replacing subjective selection based on expert experience. Furthermore, by constructing an automated iterative optimization loop guided by the objective function, it can efficiently and systematically search for optimal parameter combinations within reasonable materialized boundaries. This significantly improves the scientific rigor, efficiency, and objectivity of mechanism optimization work, avoiding the blindness of manual trial and error.

[0042] 4. A dual verification mechanism, including generalization ability testing, has been established to effectively ensure the reliability and universality of the optimization results. Unlike common practices that only focus on improving the fit of the training set, this invention emphasizes rigorous testing of the optimized mechanism using completely independent observational data. This mechanism can effectively distinguish whether the optimization truly improves the physicochemical characterization ability of the mechanism (generalization ability) or merely overfits the random noise of a specific dataset. This ensures that the optimization results have stronger reliability and can be applied to a wider range of atmospheric environment simulation scenarios.

[0043] In summary, this invention advances the assessment and optimization of atmospheric chemical mechanisms from a traditional model that relies on expert intuition, involves fragmented processes, and has high uncertainty in conclusions, into a modern scientific process that is data-driven, precise in diagnosis, quantitatively optimized, and rigorously verified, providing a systematic solution for improving the accuracy and reliability of chemical mechanisms. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example: A chemical mechanism verification and optimization method based on peroxy radical constraint, such as... Figure 1 As shown, it includes the following steps:

[0048] S1. Data Acquisition and Preprocessing: Simultaneously acquire high temporal resolution observational concentration data of peroxy free radicals (HO2 and RO2) in the target area or experimental environment, along with synchronous auxiliary parameter observational data. Perform quality control, timestamp unification, and averaging processing on the data, including but not limited to: nitrogen oxides (NO, NO2), ozone (O3), carbon monoxide (CO), volatile organic compounds (VOCs), photolysis rate (J value), and meteorological parameters (temperature, pressure, humidity). Perform quality control, timestamp unification, and averaging processing on the aforementioned observational data.

[0049] S11, Collaborative Observation Data Acquisition: Simultaneous and continuous observation of multiple environmental element monitoring equipment at the same time and location to acquire basic data, including: core observations: peroxide free radicals (HO2+RO2), nitrogen oxides (NOx), volatile organic compounds (VOCs), photolysis rate (J-Values), ozone (O3), carbon monoxide (CO), sulfur dioxide (SO2), nitrous acid (HONO), and meteorological parameters such as wind, temperature, humidity, and pressure. All instrument inlets should be as close as possible to ensure the collection of data from the same air mass; and a unified Network Time Protocol (NTP) server should be used for data timestamp calibration and synchronization.

[0050] S12, Data Quality Control and Dataset Construction: Through data quality control (QC), data alignment, and resampling, this section addresses noise, drift, missing values, and inconsistent temporal resolution in multi-source raw data. This includes: Data Quality Control (QC): Regularly performing zero-point and range calibration of equipment; identifying and eliminating invalid time periods such as instrument maintenance, power outages, and human interference; removing obviously abnormal data points based on physical probability (e.g., concentration cannot be negative) and statistical methods (e.g., the 3σ criterion); smoothing high-frequency data (e.g., meteorological parameters) using moving averages to reduce random noise while balancing temporal resolution; Data Alignment and Resampling: Unifying the timestamps of all instrument data to the same time zone (e.g., UTC); unifying data with different temporal resolutions (e.g., minute-level NOx, second-level meteorological parameters) onto the same time grid using averaging or interpolation; Dataset Construction: Merging all data after QC and resampling into a unified, well-formatted data table or file (e.g., CSV) as input for driving and constraining box models.

[0051] S2, Initial Model Setup and Simulation Run: Run an unmodified, recognized baseline chemical mechanism under conditions as close as possible to actual observations to generate a set of "ideal" peroxy radical simulation data, which will be compared with actual observation data.

[0052] S21, Model Selection: Based on the model's purity (no diffusion, transport, or other processes) and efficiency (running speed), a zero-dimensional box model is selected. This model assumes the air mass under study is a uniformly mixed "box," free from the influence of complex physical transport processes such as advection and diffusion. It allows for the isolated study of the chemical mechanisms themselves, eliminating uncertainties introduced by meteorological transport.

[0053] S22, Model Constraint Settings, including concentration constraints: Input the initial concentrations (at t=0) of various chemical species observed into the model, including O3, CO, NO, NO2, SO2, and various VOCs (such as alkanes, alkenes, aromatics, aldehydes, ketones, etc.). For species with long lifespans and slow changes (such as CO), set the concentration to a fixed value. For species with drastic concentration changes (such as NO, NO2, O3, and specific VOCs), the observed time series are directly input into the model and forced to simulate the current concentration. Meteorological parameter constraints: Input the observed temperature, humidity, and pressure time series to constrain the reaction process (reaction rate and heterogeneous reactions, etc.). Radiation parameter constraints: Input the actual observed photolysis rates of various species (such as J(O¹D), J(NO2), J(HCHO), etc.) into the model, replacing the internal calculations using the radiative transfer formula, eliminating radiation simulation errors, and ensuring accurate driving of photochemical processes.

[0054] S23, Model Run and Output, Integrated Run: Integrate all constraints into the box model (except for the chemical mechanism, all other driving conditions are fixed to the observed values), set the time range and step size consistent with the observed data (e.g., output once every 10 minutes), and run the model. Output Results: Output the simulated concentration time series results of peroxy free radicals (HO2+RO2).

[0055] S3, Closure Analysis and Deviation Diagnosis, quantitatively assesses the consistency (closure) between the initial simulation and observation data, and qualitatively determines the patterns and potential causes of inconsistencies (deviations), thus providing direction for subsequent sensitivity analysis.

[0056] S31, Closure Analysis, Simulation / Observation Ratio (S / ORatio): Calculate the simulated concentration / observed concentration of peroxy radicals (S / O Ratio) at each time point or time period. If S / O≈1, it indicates that the simulation matches the observation; if S / O>1, it indicates that the model overestimates, possibly indicating overproduction or underconsumption of peroxy radicals; if S / O<1, it indicates that the model underestimates, possibly indicating underproduction or overconsumption of peroxy radicals. Correlation Analysis: Simulated and Observed Time Series of Peroxy Radicals: Linear regression will be performed to calculate the correlation coefficient (r) and the coefficient of determination (R²). 2 The coefficient of determination R0 is given by the equation ), where r ∈ [-1, 1]. A larger absolute value indicates a stronger linear correlation, while the sign indicates the direction of the correlation (positive or negative). 2 ∈[0, 1], the closer to 1, the better the model fits the data. Root mean square error: Calculate the root mean square error (RMSE) between the simulated and observed values ​​of peroxy free radicals. The smaller the value, the smaller the deviation between the simulated and observed values, i.e., the higher the simulation accuracy. If there is a deviation between the model and the observation, further deviation analysis is performed.

[0057] S32, Bias Analysis: First, select one or more parameters such as NOx concentration, VOCs composition, dominant VOCs species concentration, photolysis intensity, diurnal cycle, and temperature for independent data grouping, or combine multiple parameters for screening (high NOx + high VOCs). Then, define binning intervals for the selected environmental parameters, such as NOx (low concentration [0,1] ppb, medium concentration (1.5] ppb, high concentration (5,10] ppb, extremely high (10,100] ppb). Then, for each bin, analyze the closure (same as step 3.1), identify bias patterns, and discover under what specific environmental conditions the model performs poorly / well.

[0058] Bias patterns can pinpoint problems in specific chemical reactions. For example, bias under high NOx conditions is likely related to NOx-related reactions, while bias under specific VOC conditions points to the oxidation mechanism of that VOC. This helps to define key time periods and targets for subsequent sensitivity analysis. Screening examples are as follows:

[0059]

[0060] S4, Global Sensitivity Analysis and Key Response Identification, involves varying all input parameters across a reasonable parameter space (such as the uncertainty range recommended by NASA) and assessing the contribution of each parameter individually and the interactions between parameters to the output uncertainty.

[0061] S41, Global Sensitivity Analysis: Key response parameters are screened, and a parameter sample set is generated within its defined range using the Monte Carlo method. After the sample is input into the box model and run, the first-order and total-order sensitivity indices of each parameter are calculated based on the model output results. This quantitatively assesses the contribution of individual parameters and their interactions to the uncertainty of the simulation results.

[0062] S42, Key Reaction Identification: Select the top 5-10 parameters with the highest total sensitivity index, which are the key reactions that contribute the most to the uncertainty of peroxy radical simulation results under specific conditions.

[0063] First, determine the key parameters and define their uncertainty range: Select the reactions diagnosed by S3, which involve the generation, circulation, and consumption of HO2+RO2, and are known to have significant uncertainties. Extract K uncertain parameters (K1, k2, k3, ..., Ki), and define a reasonable range of variation for Ki based on the uncertainty ranges recommended by NASA and others. i_min ,K i_max ].

[0064] Then, a sample set is generated: using the Monte Carlo method, N sets of parameter combinations are extracted from the above K-dimensional parameter space to generate two independent N×k sampling matrices (N is the number of samples, and k is the number of parameters) A and B.

[0065] ,

[0066] Redefine the mixing matrix: To calculate the exponent for each parameter i, establish the transformation matrix A. B (i) Replace the i-th column of matrix A with the i-th column of matrix B.

[0067]

[0068] Then run the model to obtain the output vector: matrix A, B, AB (i) Substitute the constraints into the box model and use the same constraints as in step S2, but only for the key time period selected in step S3, to obtain the model output vectors ƒ(A), ƒ(B), and ƒ(A). B (i) ), all of which are n-dimensional column vectors.

[0069] Then calculate the sensitivity index: use the model output vector to calculate the first-order exponent S. i (This represents the uncertainty caused by the individual parameter changing on its own; the larger the value, the greater the influence of that parameter on the output.) Total order exponent S Ti (This indicates the uncertainty caused by the interaction of parameters; the larger the value, the more the parameter affects the output through its interaction with other parameters.)

[0070] ① First-order exponent S i calculate:

[0071] ② Total order index S Ti calculate:

[0072] Finally, key reactions are identified and screened: the sensitivity indices of all calculated parameters are sorted, and the top 5-10 parameters (reactions) with the highest total sensitivity indices are selected. These are the key reactions that contribute the most to the uncertainty of the peroxy radical simulation results under specific conditions, and then proceed to the next optimization step.

[0073] S5, the mechanism optimization and iterative optimization loop, is based on the list of key reactions selected in S4. Within a reasonable physicochemical range, it systematically optimizes the kinetic parameters of the key reactions to achieve the best fit between the simulation results and the observation data. First, the basic conditions are: using the key reactions and parameters generated in S4, and determining the adjustment boundary according to the uncertainty range; defining the objective function (e.g., minimizing RMSE), setting the optimization algorithm (e.g., adjusting the number of parameters, adjustment ratio), and setting the iteration parameters (e.g., maximum number of iterations, convergence tolerance). Then, an iterative optimization loop is performed: based on the optimization algorithm, a new set of parameter values ​​is proposed (e.g., K1 increases by 10%, K2 decreases by 10%), and these are substituted into the box model, where the model constraints are the same as S2 and S4; the objective function value (e.g., RMSE) between the simulation results and the observed data is calculated; the new objective function value is compared with the historical value, and if the RMSE decreases, the set of parameters is accepted as the new parameter iteration starting point; if the RMSE increases, the parameter is rejected as the new parameter iteration starting point and the loop iteration continues; the above steps are repeated until the stopping condition is met (e.g., the number of iterations exceeds 1000, and the objective function improves by less than 1% in 10 consecutive iterations); finally, a set of optimized parameter values ​​and chemical mechanisms are output.

[0074] S6, Optimization effect evaluation and mechanism generalization ability test, wherein the optimization effect evaluation: using the optimized parameters, under the exact same settings and constraints as S2, the box model is rerun to obtain a new simulated peroxy radical time series; the simulation results are compared with the baseline simulation results before optimization and the observation data, and the S / O Ratio and R are calculated. 2 Indicators such as RMSE; at this stage, all key indicators should show significant and consistent improvement. For example, a 20% decrease in RMSE, an increase in the average S / O ratio from 0.7 to 0.8, and an improvement in R... 2 Improve from 0.6 to 0.7. If the result worsens or the improvement is not significant, it is necessary to backtrack and check steps S4 and S5.

[0075] Finally, the mechanism generalization ability was tested: using independent observation data that did not participate in any optimization process, the baseline mechanism and the optimized mechanism were driven to run, and the S / O Ratio and R were calculated. 2 The optimization mechanism is evaluated using metrics such as RMSE. If the RMSE of the optimized mechanism is significantly lower than that of the baseline mechanism on the independent test set, it indicates that the optimization was successful and that a more universal chemical law has been captured. If the RMSE of the optimized mechanism is higher than that of the baseline mechanism, it indicates that the optimization failed and that the optimized mechanism "remembered" the noise of the training set rather than its inherent laws. In this case, the data needs to be replaced and the above operations need to be repeated.

[0076] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A chemical mechanism verification and optimization method based on peroxy radical constraint, characterized in that, Specifically, the steps include the following: S1, Observational data acquisition and preprocessing: Simultaneously acquire high temporal resolution observational concentration data of peroxy free radicals in the target area or experimental environment, as well as synchronous auxiliary parameter observational data, and perform quality control, unified timestamp and average time processing. S11, Collaborative Observation Data Acquisition: At the same time and location, multiple environmental element monitoring equipment are integrated to conduct continuous observation and acquire basic data; S12, Data Quality Control and Dataset Construction: Through data quality control, data alignment and resampling, the problems of noise, drift, missing values ​​and inconsistent temporal resolution in multi-source raw data are solved; S2, Initial Model Setup and Simulation Run: Run an unmodified, accepted baseline chemical mechanism under conditions as close as possible to real observation conditions to generate a set of "ideal" peroxy radical simulation data, which will be compared with actual observation data; S21, Model Selection: Based on the purity of the chemical process and computational efficiency, a zero-dimensional box model is selected; S22, Model constraint settings: Constrain the model using concentration, meteorological parameters, and radiation parameters; S23, Model Run and Output: Integrate all constraints into the box model, set the time range and step size consistent with the observation data, run the model, and then output the simulated concentration time series of peroxy free radicals. S3, Closure analysis and deviation diagnosis: Quantitatively assess the consistency between the initial simulation and observation data, and qualitatively determine the patterns and potential causes of inconsistencies, thus providing direction for subsequent sensitivity analysis; S31, Closure analysis: Calculate the ratio, correlation, and root mean square error between simulated and observed values ​​to analyze the deviation between the model and the observations; S32, Deviation Analysis: By dividing the observed and simulated data into bins using key environmental parameters, the simulation closure of each bin is evaluated, thereby identifying systematic deviations of the model under specific conditions. These deviation patterns can directly point to deficiencies in the relevant chemical reaction mechanisms, providing clear targets for subsequent sensitivity analysis and mechanism optimization. S4, Global Sensitivity Analysis and Key Response Identification: Varying all input parameters across the entire reasonable parameter space and evaluating the contribution of each parameter individually and the interactions between parameters to the output uncertainty; S41, Global Sensitivity Analysis: Screen key response parameters, generate parameter sample sets using the Monte Carlo method within their defined range, input the samples into the box model and run it, calculate the first-order and total-order sensitivity indices of each parameter based on the model output results, thereby quantitatively evaluating the contribution of individual parameters and their interactions to the uncertainty of the simulation results; S42, Key Reaction Identification: Select the top 5-10 parameters with the highest total sensitivity index, which are the key reactions that contribute the most to the uncertainty of peroxy radical simulation results under specific conditions; S5, Mechanism Optimization and Iterative Optimization Cycle: Based on the list of key reactions selected in S4, within a reasonable physicochemical range, the kinetic parameters of the key reactions are systematically optimized to achieve the best fit between the simulation results and the observation data; S6, Evaluation of optimization effect and test of mechanism generalization ability: Re-simulate under the same conditions using the optimized parameters. By comparing and analyzing with the baseline simulation and observation data, verify whether the simulation effect has been improved. Then, use an independent observation dataset to test the generalization ability of the optimized mechanism. If its performance is significantly better than the baseline mechanism, it indicates that the optimization has successfully captured universal chemical laws; otherwise, it is necessary to backtrack and replace the data. S61, Optimization Effect Evaluation: Under the same conditions as S2, rerun the model using the optimized parameters and compare the new simulation results with the baseline simulation and observation data before optimization. Calculate the S / O Ratio and R... 2 Indicators such as RMSE are used to verify whether each indicator shows significant and consistent improvement. If the improvement is not obvious or the results worsen, the optimization process needs to be backtracked and checked. S62, Mechanism Generalization Ability Test: The baseline mechanism and the optimized mechanism are tested using independent observation data, and the key indicators of the two are compared. If the RMSE of the optimized mechanism is significantly reduced, it indicates that it has successfully captured universal chemical laws; otherwise, it indicates that the optimization may overfit the training set noise and needs to be re-validated with different data.

2. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 1, characterized in that: In step S11, the observation of the basic data includes core observation and auxiliary parameter observation. The core observation is the observation of peroxy free radicals, and the auxiliary parameter observation is the observation of nitrogen oxides, volatile organic compounds, photolysis rate, ozone, carbon monoxide, sulfur dioxide, nitrous acid, and meteorological parameters such as wind, temperature, humidity, and pressure. The sample inlet of the observation instrument should be as close as possible to the sample inlet.

3. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 2, characterized in that: In step S12, the data quality control includes periodically performing zero-point and range calibration of the equipment, identifying and eliminating invalid periods due to instrument maintenance and human interference, removing obviously abnormal data points based on physical probability and statistical methods, and smoothing high-frequency data. The data alignment and resampling include unifying the timestamps of all instrument data to the same time zone and unifying data with different time resolutions to the same time grid through interpolation methods. The dataset construction involves data merging by QC and resampling. After merging, the data is organized into a data table, which is used to drive and constrain the input file of the box model.

4. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 3, characterized in that: In step S2, the concentration constraint involves inputting the initial concentrations of various chemical species observed into the model, setting the concentrations of species with long lifespans and slow changes to fixed values, and directly inputting the observation time series of species with drastic concentration changes into the model to simulate the current concentration. The meteorological parameter constraint includes inputting the observed temperature, humidity, and pressure time series, which are used to constrain the reaction process. The radiation parameter constraint is achieved by inputting the actual observed time series of photolysis rates of each species into the model.

5. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 4, characterized in that: In step S3, the binning process involves data types including NOx concentration, VOCs composition, concentration of dominant VOCs species, photolysis intensity, daily cycle, and temperature parameters. The deviation diagnosis is used to pinpoint problems in a certain type of chemical reaction.

6. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 5, characterized in that: In step S4, the uncertain reaction is used to extract uncertain reaction parameters. K uncertain reaction parameters are extracted, and a reasonable range of variation is defined for K based on the reasonable parameter space. The range of variation is used to generate a sample set. The sample set is generated using the Monte Carlo method to establish a K-dimensional parameter space for K. The number of parameters in the K-dimensional parameter space is set to k. N sets of parameter combinations are extracted from the K-dimensional parameter space, and the number of samples is set to N. Two independent N×k sampling matrices A and B are generated through the K-dimensional parameter space. The formulas for A and B are as follows: , ； The x represents parameter values ​​for different rows and columns in the matrix. One of the K parameters is set to i. The exponent of each i is calculated, and a transformation matrix A is constructed based on i. B (i) The transformation matrix A B (i) Specifically, it means replacing the i-th column of matrix A with the i-th column of matrix B, where matrix A... B (i) The formula is as follows: ； Substituting matrices A and B into the box model, and using the same constraints in S2 to select key time periods for S3, we obtain the model output vectors ƒ(A), ƒ(B), and ƒ(A...). B (i) ), the ƒ(A), ƒ(B), ƒ(A B (i) All are n-dimensional column vectors.

7. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 6, characterized in that: The model output vector is used to calculate the first-order exponent and the total-order exponent, with the first-order exponent set as S. i The total order exponent is set to S. Ti The S i and S Ti The calculation formula is as follows: ； ； The j represents the j-th dimension of the output.

8. The chemical mechanism verification and optimization method based on peroxy radical constraint as described in claim 7, characterized in that: In step S5, the uncertainty range is used to determine the adjustment boundary. The iterative optimization loop proposes a new set of parameter values ​​based on the optimization algorithm. The parameter values ​​are substituted into the box model with the same constraints as S2 and S4. The iterative parameters are set with the objective function as the guide. The iterative parameters are iterated cyclically until the stopping condition is met.

Images