A big data-based atmospheric pollutant rapid tracing system

Abstract

The application relates to the technical field of environmental monitoring and discloses a big data-based atmospheric pollutant rapid tracing system, which comprises a diffusion simulation module, a hybrid optimization module, a dynamic algorithm selection module, a multi-source analysis module and a verification and output module. The dynamic algorithm selection module is configured to build a dynamic matching rule of meteorological data features and a tracing algorithm, to adaptively switch an anti-interference algorithm model according to different meteorological conditions, to eliminate meteorological refraction deviation in pollutant concentration field simulation in real time, to reduce the intensity inversion error caused by optical interference in the traditional method, and to guarantee the physical authenticity of source intensity estimation. The hybrid optimization module is configured to use an axial detection and mode moving strategy to monitor the search path state of a genetic algorithm in real time, to automatically identify the parameter space deviation in the mathematical level, and to trigger a mode search correction mechanism and a genetic algorithm restart operation, so that the geometric accuracy and spatial reliability of the tracing result are improved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, specifically to a rapid source tracing system for air pollutants based on big data. Background Technology

[0002] Air pollutants refer to substances released into the atmosphere by human activities and natural processes that have harmful effects on humans and the environment. In a clean atmosphere, the composition of a constant amount of gas is negligible. However, the presence of trace amounts of substances in the atmosphere within a certain range, in both quantity and duration, can have adverse effects and harms on humans, animals, plants, and materials. When the concentration of pollutants in the atmosphere reaches a harmful level, thereby damaging the ecosystem and the conditions for normal human survival and development, and causing harm to people and things, the phenomenon is called air pollution.

[0003] Currently, in the field of big data-based air pollution source tracing, traditional tracing systems rely on fixed parameterized models to infer the location and emission intensity of pollution sources. However, these systems have significant drawbacks in practical applications: Due to the dynamic changes in meteorological fields and the spatial heterogeneity of underlying surface features, diffusion models cannot perceive and calibrate meteorological refraction deviations along the light transmission path in real time, leading to systematic color differences in the simulated pollutant concentration field. This optical interference poses a risk of misjudging the intensity of the tracing results. Furthermore, during the optimization algorithm execution process, a dynamic feedback mechanism for the pollution source search direction is not established. When the genetic algorithm gets stuck in a local optimum due to initial population bias and fitness deception, it cannot identify and correct path deviations in the mathematical search angle in real time, resulting in coordinate positioning errors exceeding 500 meters without automatic correction capabilities. In addition, facing multi-point source superposition scenarios in industrial parks, existing systems lack the ability to perform hierarchical analysis of multiple pollution areas. They cannot achieve layered segmentation of confidence intervals for the independent source strengths of the superimposed concentration field, leading to confusion between high-contribution and low-contribution emission intensities, reducing the accuracy of responsibility tracing and the effectiveness of law enforcement.

[0004] Therefore, a rapid source tracing system for air pollutants based on big data is proposed to solve the above problems. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a rapid source tracing system for air pollutants based on big data, which solves the problems mentioned in the background section.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution: a rapid source tracing system for air pollutants based on big data, comprising:

[0009] The diffusion simulation module constructs a Gaussian diffusion model algorithm to calculate the pollutant concentration field distribution based on the input meteorological parameters and underlying surface characteristic parameters.

[0010] The hybrid optimization module, coupled with the diffusion simulation module, uses a genetic-pattern search hybrid algorithm to iteratively optimize pollution source parameters, including:

[0011] The parameter encoding unit converts pollution source parameters into genotype individuals using decimal encoding;

[0012] The population evolution unit updates the source parameter population through selection, crossover, and mutation operations, and selects high-fitness individuals based on the fitness function;

[0013] The pattern search unit performs axial and pattern shift searches on the local optimal solutions output by the genetic algorithm to achieve global optimization.

[0014] The dynamic algorithm selection module matches the optimal source tracing algorithm type from the preset algorithm library based on meteorological data characteristics and observation point layout scheme;

[0015] The multi-source analysis module processes multi-point source scenarios in parallel and outputs the emission intensity confidence intervals of each pollution source.

[0016] The verification and output module verifies the model accuracy based on simulation data and Prairie Grass field test data, and outputs the traceability results and convergence index.

[0017] Preferably, the diffusion simulation module further includes:

[0018] The reliability of the Gaussian diffusion model was verified based on the standardized average error, standardized average deviation, and root mean square error indices.

[0019] The model parameters are calibrated using measured concentration data from the Prairie Grass project to ensure the accuracy of the simulation of high-value areas and distribution ranges of pollutants.

[0020] Preferably, the fitness function of the hybrid optimization module is defined as:

[0021]

[0022] Where C 模拟,i Let C be the simulated concentration value at the i-th observation point. 观测,i Let be the measured concentration value at the i-th observation point, and n be the total number of observation points.

[0023] Preferably, the operation of the parameter encoding unit includes:

[0024] Construct a database of key emission sources and randomly select individuals to generate an initial population;

[0025] By comparing the errors between simulated and observed values ​​through parallel computing, individuals with fitness levels higher than a threshold are selected to enter the next generation of evolution.

[0026] Preferably, the dynamic algorithm selection module matches algorithms according to the following rules:

[0027] When the observation points are arranged in an arc shape, the near-source sparse point arrangement strategy is preferred.

[0028] When the observation points are arranged in a rectangular grid, a near-source dense point arrangement strategy is preferred.

[0029] When the atmospheric stability is unstable, increase the weight of observation points beyond 400 meters;

[0030] When the atmospheric stability is at the stable level, increase the weight of observation points within 200 meters.

[0031] Preferably, the verification process of the verification and output module includes:

[0032] Simulation evaluation: Based on the ideal concentration field generated by Prairie Grass meteorological conditions, the success rate and timeliness of source tracing under different sampling schemes, sampling densities, and stability conditions were tested;

[0033] Field verification: Compare the coordinate error distance and intensity deviation between the source tracing results and the actual source parameters to quantify the model uncertainty.

[0034] Preferably, it includes the following steps:

[0035] S1. Input meteorological data and pollutant concentration data at observation points;

[0036] S2. Calculate the concentration field distribution of each source parameter individual using the diffusion simulation module;

[0037] S3. Calculate the fitness value of individuals in the population based on the error between observed and simulated concentrations;

[0038] S4. Perform the genetic algorithm operation:

[0039] Selection action: Retain the top 30% of individuals by fitness;

[0040] Crossover operation: The remaining individuals undergo pairwise gene exchange, with the exchange ratio set at 60%;

[0041] Mutation operation: Randomly mutate an individual's gene value with a 5% probability;

[0042] S5. After the genetic algorithm has iterated 1000 times, the current optimal individual is used to search for the pattern:

[0043] Set the initial step size δ0 = 0.1 and the contraction factor α = 0.5;

[0044] The search range is dynamically narrowed using the step size update formula:

[0045] δ k+1 =α·δ k

[0046] Where δ k Let α be the search step size for the k-th iteration, α be the step size decay coefficient, and k be the iteration number index.

[0047] S6. When the error is less than 0.01 and the iteration reaches 2000, output the globally optimal source parameters (Q,x,y).

[0048] Preferably, the process includes the following steps before step S1:

[0049] Dynamically adjust the observation point deployment strategy:

[0050] If the atmospheric stability is level C, 50% of the observation points will be set up in an area 400 meters away from the source;

[0051] If the atmospheric stability is level E, 70% of the observation points will be located within 200 meters of the source.

[0052] Preferably, the specific implementation of the crossover operation in S4 is as follows:

[0053] Using a single-point crossover method, a cut point in the gene sequence is randomly selected, and the gene segments after the cut point in the parent individuals are exchanged.

[0054] After generating offspring, fitness is recalculated and individuals below the average fitness are culled.

[0055] Preferably, the termination condition for pattern search in S5 is:

[0056] Step size δ k Less than 0.001 and no better solution was found in 50 consecutive iterations;

[0057] Before outputting the final solution, calculate its correlation coefficient with the observed concentration. If it is lower than 0.95, trigger the genetic algorithm restart mechanism.

[0058] (III) Beneficial Effects

[0059] Compared with existing technologies, this invention provides a rapid source tracing system for air pollutants based on big data, which has the following beneficial effects:

[0060] 1. In this invention, by setting a dynamic algorithm selection module, when performing source tracing calculations for atmospheric pollution sources, a dynamic matching rule between meteorological data features and source tracing algorithms is constructed. The anti-interference algorithm model is adaptively switched for different meteorological conditions, and meteorological refraction deviations in the pollutant concentration field simulation are eliminated in real time. This reduces the intensity inversion error caused by optical interference in traditional methods and ensures the physical authenticity of source strength estimation.

[0061] 2. In this invention, by setting up a hybrid optimization module, when searching for pollution source parameters, the axial detection and pattern movement strategies are used to monitor the search path status of the genetic algorithm in real time, automatically identify deviations in the mathematical parameter space, and trigger the pattern search correction mechanism and the genetic algorithm restart operation to solve the risk of coordinate positioning offset and improve the geometric accuracy and spatial reliability of the source tracing results.

[0062] 3. In this invention, by setting up a multi-source analysis module, in the scenario of multiple point sources in industrial parks, the independent contributing sources of the superimposed concentration field are separated by confidence interval layering and cutting technology. Based on the Bayesian probability model, the emission intensity of the main pollution source and the secondary pollution source is spatially decoupled and interval quantified, which solves the problem of confusion between high and low contribution sources in traditional methods and improves the accuracy of responsibility tracing and the effectiveness of law enforcement. Attached Figure Description

[0063] Figure 1 This is a schematic diagram of the overall system architecture of the present invention. Detailed Implementation

[0064] 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.

[0065] Please see Figure 1 This big data-based rapid source tracing system for air pollutants includes:

[0066] The diffusion simulation module constructs a Gaussian diffusion model algorithm to calculate the pollutant concentration field distribution based on the input meteorological parameters and underlying surface characteristic parameters.

[0067] The hybrid optimization module, coupled with the diffusion simulation module, uses a genetic-pattern search hybrid algorithm to iteratively optimize pollution source parameters, including:

[0068] The parameter encoding unit converts pollution source parameters into genotype individuals using decimal encoding;

[0069] The population evolution unit updates the source parameter population through selection, crossover, and mutation operations, and selects high-fitness individuals based on the fitness function;

[0070] The pattern search unit performs axial and pattern shift searches on the local optimal solutions output by the genetic algorithm to achieve global optimization.

[0071] The dynamic algorithm selection module matches the optimal source tracing algorithm type from the preset algorithm library based on meteorological data characteristics and observation point layout scheme;

[0072] The multi-source analysis module processes multi-point source scenarios in parallel and outputs the emission intensity confidence intervals of each pollution source.

[0073] The verification and output module verifies the model accuracy based on simulation data and Prairie Grass field test data, and outputs the traceability results and convergence index.

[0074] The diffusion simulation module also includes:

[0075] The reliability of the Gaussian diffusion model was verified based on the standardized average error, standardized average deviation, and root mean square error indices.

[0076] The model parameters are calibrated using measured concentration data from the Prairie Grass project to ensure the accuracy of the simulation of high-value areas and distribution ranges of pollutants.

[0077] The fitness function of the hybrid optimization module is defined as:

[0078]

[0079] Where C 模拟,i Let C be the simulated concentration value at the i-th observation point. 观测,i Let be the measured concentration value at the i-th observation point, and n be the total number of observation points.

[0080] The operations of the parameter encoding unit include:

[0081] Construct a database of key emission sources and randomly select individuals to generate an initial population;

[0082] By comparing the errors between simulated and observed values ​​through parallel computing, individuals with fitness levels higher than a threshold are selected to enter the next generation of evolution.

[0083] The dynamic algorithm selection module matches algorithms according to the following rules:

[0084] When the observation points are arranged in an arc shape, the near-source sparse point arrangement strategy is preferred.

[0085] When the observation points are arranged in a rectangular grid, a near-source dense point arrangement strategy is preferred.

[0086] When the atmospheric stability is unstable, increase the weight of observation points beyond 400 meters;

[0087] When the atmospheric stability is at the stable level, increase the weight of observation points within 200 meters.

[0088] The verification process for the verification and output module includes:

[0089] Simulation evaluation: Based on the ideal concentration field generated by Prairie Grass meteorological conditions, the success rate and timeliness of source tracing under different sampling schemes, sampling densities, and stability conditions were tested;

[0090] Field verification: Compare the coordinate error distance and intensity deviation between the source tracing results and the actual source parameters to quantify the model uncertainty.

[0091] Includes the following steps:

[0092] S1. Input meteorological data and pollutant concentration data at observation points;

[0093] S2. Calculate the concentration field distribution of each source parameter individual using the diffusion simulation module;

[0094] S3. Calculate the fitness value of individuals in the population based on the error between observed and simulated concentrations;

[0095] S4. Perform the genetic algorithm operation:

[0096] Selection action: Retain the top 30% of individuals by fitness;

[0097] Crossover operation: The remaining individuals undergo pairwise gene exchange, with the exchange ratio set at 60%;

[0098] Mutation operation: Randomly mutate an individual's gene value with a 5% probability;

[0099] S5. After the genetic algorithm has iterated 1000 times, the current optimal individual is used to search for the pattern:

[0100] Set the initial step size δ0 = 0.1 and the contraction factor α = 0.5;

[0101] The search range is dynamically narrowed using the step size update formula:

[0102] δ k+1 =α·δ k

[0103] Where δ k Let α be the search step size for the k-th iteration, α be the step size decay coefficient, and k be the iteration number index.

[0104] S6. When the error is less than 0.01 and the iteration reaches 2000, output the globally optimal source parameters (Q,x,y).

[0105] The steps preceding step S1 also include:

[0106] Dynamically adjust the observation point deployment strategy:

[0107] If the atmospheric stability is level C, 50% of the observation points will be set up in an area 400 meters away from the source;

[0108] If the atmospheric stability is level E, 70% of the observation points will be located within 200 meters of the source.

[0109] The specific implementation of the crossover operation in S4 is as follows:

[0110] Using a single-point crossover method, a cut point in the gene sequence is randomly selected, and the gene segments after the cut point in the parent individuals are exchanged.

[0111] After generating offspring, fitness is recalculated and individuals below the average fitness are culled.

[0112] The termination condition for pattern search in S5 is:

[0113] Step size δ k Less than 0.001 and no better solution was found in 50 consecutive iterations;

[0114] Before outputting the final solution, calculate its correlation coefficient with the observed concentration. If it is lower than 0.95, trigger the genetic algorithm restart mechanism.

[0115] Example 1: Source Tracing of Sudden Pollution Incidents in Industrial Zones

[0116] An abnormal surge in PM2.5 concentration occurred in a petrochemical park in the early morning. The system, through a dynamic algorithm selection module, monitored in real time that the wind direction pulsation frequency reached 2 times / minute and the atmospheric stability jumped to level E. According to the preset rules, the near-field dense deployment strategy was automatically activated: 17 mobile monitoring points were deployed within a 200-meter radius of the pollution source center, and the infrared absorption spectrum correction model in the anti-water vapor refraction algorithm library was activated at the same time.

[0117] The hybrid optimization module generates an initial population with an initial emission intensity of 500 kg / h and coordinates (X0, Y0). After 1200 genetic iterations, the suspected source is located at the A3 position in the tank area. At this time, the model search unit detects that the axial deviation exceeds the limit and immediately triggers the axial compensation mechanism to perform 5 model moves to correct the source coordinates to the leak point of the tank valve (X0+3.2m, Y0-1.7m).

[0118] The multi-source analysis module simultaneously separated the superimposed concentration fields of benzene series compounds and hydrogen sulfide, outputting that the main pollution source was valve leakage and the secondary pollution source was abnormal emissions from chimney No. 3. The final source tracing results were verified by the module by comparing the valve maintenance records with the chimney operating log, and the error was less than 1.7%.

[0119] Example 2: Delineation of Responsibilities in Multi-Source Industrial Parks

[0120] In response to the persistent ozone concentration exceeding standards in a certain chemical industrial cluster, the system initiated a multi-source collaborative tracing mode. The dynamic algorithm selection module identified that the observation network was a 2km×2km rectangular grid layout and that the atmospheric stability was in a C-level unstable state. It automatically matched a distant-source sparse weighting strategy: 40% of the monitoring points were deployed at least 800 meters away from the core area, and a distance weighting coefficient of 0.6 was assigned.

[0121] The diffusion simulation module loads wind profile parameters obtained from topographic wind tunnel experiments to construct a composite diffusion model including chloric acid and VOCs. The hybrid optimization module initially identifies four potential sources through genetic evolution: the reactor at plant A, the storage tank group at plant B, the incinerator at plant C, and the loading and unloading area at plant D. The pattern search unit then initiates independent axial probing for each source: when an oscillation is detected in the fitness function of the storage tank group at plant B, the genetic algorithm restart mechanism is activated to reinitialize the population.

[0122] After three rounds of collaborative optimization, the multi-source analysis module used Bayesian hierarchical sampling to segment out the primary and secondary contributing sources: the leakage of the storage tank at Plant B contributed 78.5%, with an intensity range of [2.3t / a, 2.5t / a]; the incinerator at Plant C contributed 12.1%, with an intensity range of [0.35t / a, 0.38t / a]; and the total contribution of the remaining sources was less than 9%. Based on this result, the environmental enforcement department implemented precise production restrictions on Plant B.

[0123] Example 3: Tracing the Source of Pollution Transmission Across Borders

[0124] In regional dust storm transboundary transport events, the system identifies the prevailing wind direction as northwest and the average wind speed as 8.4 m / s through meteorological big data. The dynamic algorithm selection module detects that the monitoring points are distributed in an arc shape spanning 200 kilometers, automatically switches the arc-shaped point layout optimization scheme and activates the aerosol ultraviolet scattering correction algorithm.

[0125] The hybrid optimization module first performs a coarse search using a 100km-level grid with a resolution of 10km×10km. After 800 genetic iterations, it identifies three potential dust source areas: the Mongolian Gobi Desert, the border coal mining belt, and local construction sites. Then, it initiates a three-stage refined source tracing process: the first stage performs a pattern search on the Mongolian Gobi Desert with a step size δ0 = 5km, shrinking the coordinates to 42.36°N + 1.2km through 12 axial movements; the second stage employs an adaptive mutation strategy on the border coal mining belt, increasing the mutation rate from 5% to 15% to overcome the fitness plateau caused by terrain shading; the third stage initiates near-field DNA comparison for local sources, calling upon historical pollution source gene banks to match the dust characteristic spectrum of construction sites.

[0126] The final multi-source analysis module outputs a responsibility allocation with a 95% confidence level: cross-border transmission contributes 91.2%, and the Mongolian source strength is [4.8 × 10].6 t, 5.1 × 10 6 [t], local contribution 8.8%, construction site source [0.4×10] 6 t, 0.45×10 6 t).

[0127] Example 4: Real-time tracking of mobile source pollution

[0128] To track sudden leaks from hazardous chemical transport vehicles, the system integrates road network monitoring and UAV mobile monitoring. The dynamic algorithm selection module matches the dynamic wind field reconstruction algorithm and loads the road underlying surface roughness parameter z0 = 0.8m based on the characteristics of wind speed pulsation greater than level 3 and pollution plumes spreading in a banded manner.

[0129] The hybrid optimization module establishes a dual-process collaborative mechanism: the main process uses a genetic algorithm to invert the leakage intensity and movement speed, and the initial population includes combinations of load capacities of 15t and 25t and vehicle speeds of 40km / h and 80km / h; the slave process corrects the vehicle position in real time through pattern search, and updates the axial detection vector every 5 seconds after receiving new coordinates from the UAV.

[0130] When an abnormal decrease in the fitness of the speed parameter was detected during an iteration, backpropagation verification was immediately triggered: the simulated concentration field was compared in real time with 17 road surface sensors. After a sudden change in the concentration gradient at K128+300 meters, the vehicle speed gene was forcibly reset to 50 km / h. After 8 minutes, the source of pollution was identified as a propylene tanker truck with license plate XYZ, with a leakage intensity range of [38 kg / min, 42 kg / min] and a location error of 12.3 meters. The verification module retrieved the vehicle's GPS trajectory and loading / unloading records, confirming a 100% match in the leakage time period.

[0131] Example 5: Precise Source Tracing of Odor Pollution

[0132] A sudden high-intensity odor complaint occurred near a sewage treatment plant in a certain city. The system used a dynamic algorithm to select a module that captured the atmospheric stability jumping from level D (neutral) to level F (strongly stable) within 2 hours, and the wind speed dropped to below 0.8 m / s. Based on the amine characteristic spectrum of the odor pollutants and the low wind speed diffusion characteristics, the system automatically activated the valley inversion compensation algorithm and loaded the terrain elevation data.

[0133] The hybrid optimization module divides the initial population into dual-objective optimization groups for hydrogen sulfide and ammonia: the H2S population generates the initial source strength [5ppm, 15ppm] based on the monitoring concentration at the plant boundary, while the NH3 population sets the search range based on the coordinates of the sludge treatment area. When the genetic algorithm iterates to generation 900, the pattern search unit detects a "pseudo-peak trap" in the NH3 objective function—a false high fitness value is formed near the coordinates of the sludge dewatering workshop. This immediately triggers the three-dimensional axial detection mechanism: a positive and negative bidirectional search is performed along the vertical direction with a step size δz = 2m. It is found that the fitness at elevation +8m actually increases by 12.7%, thus determining that the leak point is located at the roof exhaust valve rather than the ground equipment.

[0134] The multi-source analysis module used odor activity value weighted cutting technology to separate the main odor source as the sludge anaerobic digester, with H2S contributing 73%, and the secondary source as the composting workshop, with NH3 contributing 21%. Based on the results, the environmental protection department modified the sealing system of the digester, and odor complaints decreased by 92% after 48 hours.

[0135] Example 6: Source Tracing of Radionuclide Leaks

[0136] An abnormal gamma-ray energy spectrum was detected by the environmental monitoring network around the nuclear facility, with a characteristic peak of 1.46 MeV. The system initiated an emergency source tracing for radioactive contamination. The dynamic algorithm selection module identified that the monitoring points were distributed in concentric circles and that there was atmospheric precipitation with a rainfall of 4 mm / h. The system automatically matched the wet deposition correction algorithm and loaded the radionuclide migration model.

[0137] The hybrid optimization module faces three challenges for the first time:

[0138] 1. The half-life of a radionuclide leads to a dynamic decay in its concentration;

[0139] 2. Radon background noise;

[0140] 3. The protective wall's shielding effect distorts the diffusion field.

[0141] The system creates a spatiotemporal four-dimensional optimization framework: the time dimension is injected with the cesium-137 half-life function, and the spatial dimension is added with a shielding attenuation coefficient, where the concrete wall μ = 0.2cm. -1 The genetic algorithm excludes nuclides with a half-life of less than 1 hour during the population initialization stage. After 600 generations of evolution, it locks down two potential leak points: the spent fuel pool, coordinate A, and the maintenance passage, coordinate B. At this point, the pattern search unit detects that the gradient at coordinate B is abnormally flat and initiates the quantum tunneling detection strategy: virtual observation points are set up on both sides of the shielding wall to perform penetration calculations. Finally, coordinate A is confirmed as the real leak point, and the fitness is improved by 8.3 times.

[0142] The multi-source analysis module utilizes Bayesian Monte Carlo layering to output the cesium-137 intensity range [3.2 × 10⁻⁶] of the spent fuel pool leak. 4Bq, 3.5 × 10 4 [Bq], with a confidence probability of 99.7%, the result is only 0.75 meters away from the component damage location recorded in the accident log.

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0144] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A big data based fast atmospheric pollutant tracing system, characterized in that: The method comprises the following steps: a diffusion simulation module, which constructs a Gaussian diffusion model algorithm and calculates a pollutant concentration field distribution according to input meteorological parameters and underlying surface characteristic parameters; a hybrid optimization module, which is coupled with the diffusion simulation module and iteratively optimizes pollution source parameters by using a genetic-mode search hybrid algorithm, including: a parameter coding unit, which converts pollution source parameters into genotypic individuals by using decimal coding, including: constructing a key emission source information database and randomly extracting individuals to generate an initial population; comparing simulated values with observed values by parallel computing and screening individuals with a fitness value higher than a threshold value to enter the next generation evolution; a population evolution unit, which updates source parameter populations by selection, crossover and mutation operations and selects high-fitness individuals based on a fitness function; a mode search unit, which performs axial and mode movement searches on local optimal solutions output by the genetic algorithm to realize global optimization; a dynamic algorithm selection module, which matches an optimal source tracing algorithm type from a preset algorithm library according to meteorological data characteristics and an observation point arrangement scheme and matches the algorithm according to the following rules: when the observation point arrangement is an arc scheme, a near-source sparse observation point arrangement strategy is preferentially selected; when the observation point arrangement is a rectangular grid scheme, a near-source dense observation point arrangement strategy is preferentially selected; when atmospheric stability is unstable, the weight of an observation point 400 meters away is increased; when atmospheric stability is stable, the weight of an observation point within 200 meters is increased; a multi-source analysis module, which processes a multi-point source scene in parallel and outputs confidence intervals of emission intensities of each pollution source; a verification and output module, which verifies model accuracy based on simulation data and Prairie Grass field test data, outputs source tracing results and convergence indexes, and the verification process comprises the following steps: simulation evaluation: ideal concentration fields are generated based on Prairie Grass meteorological conditions to test source tracing success rates and time efficiency under different observation point arrangement schemes, observation point densities and stability conditions; field verification: the coordinate error distance and intensity deviation of source tracing results and true source parameters are compared to quantify model uncertainty.

2. The big data based atmospheric pollutant rapid tracing system according to claim 1, characterized in that: The diffusion simulation module further comprises the following steps: the reliability of the Gaussian diffusion model is verified based on standardized mean error, standardized mean deviation and root mean square error indexes; the model parameters are calibrated by using measured concentration data of the Prairie Grass project to ensure the simulation accuracy of high-value and distribution areas of pollutants.

3. The big data based atmospheric pollutant rapid tracing system according to claim 1, characterized in that: The fitness function of the hybrid optimization module is defined as follows: ； wherein is the simulated concentration value for the i-th observation point, is the measured concentration value for the i-th observation point, is the total number of observation points.

4. The big data based atmospheric pollutant rapid tracing system according to claim 1, characterized in that: The method comprises the following steps: S1. Input meteorological data and pollutant concentration data of observation points; S2. Calculate concentration field distributions of each source parameter individual by using the diffusion simulation module; S3. Calculate fitness values of population individuals based on errors between observed concentrations and simulated concentrations; S4. Perform genetic algorithm operations: selection operation: retain individuals with a top 30% fitness value; crossover operation: perform gene exchange between two individuals, and the exchange ratio is set to 60%; mutation operation: randomly mutate individual gene values with a probability of 5%; S5. When the genetic algorithm is iterated for 1000 times, input the current optimal individual into the mode search algorithm: Setting initial step size , shrink factor ; dynamically reduce the search range by using a step update formula: ； wherein is the search step size for the kth iteration, is the step size decay coefficient, is the iteration number index; S6. When the error is less than 0.01 and the iteration reaches 2000 times, output the global optimal source parameter .

5. The big data based atmospheric pollutant rapid tracing system according to claim 4, characterized in that: Before the step S1, the following steps are further included: dynamically adjust the observation point arrangement strategy: If the atmospheric stability is C, 50% of the observation points are arranged in the area 400 meters away from the source; If the atmospheric stability is E, 70% of the observation points are arranged in the area 200 meters away from the source.

6. The big data based atmospheric pollutant rapid tracing system according to claim 4, characterized in that: The specific implementation of the cross operation in S4 is: Single-point crossover method is adopted, and a cutting point in the gene sequence is randomly selected to exchange the gene fragments after the cutting point of the parent individual; After the offspring individual is generated, the fitness is recalculated, and the individual below the average value is eliminated.

7. The big data based atmospheric pollutant rapid tracing system according to claim 4, characterized in that: The termination condition of the mode search in S5 is: Step size Less than 0.001 and no better solution found for 50 consecutive iterations; Before outputting the final solution, the correlation coefficient between it and the observed concentration is calculated, and if it is lower than 0.95, the genetic algorithm restart mechanism is triggered.

Images