An industrial park VOCs pollution monitoring system and method

Abstract

The application discloses an industrial park VOCs pollution monitoring system and method, relates to the volatile organic compound pollution monitoring technical field of an industrial park, and comprises a multi-source data acquisition and fusion module, a non-point source-point source conversion module, a small and micro scale CALPUFF modeling module, a pollution source tracing inversion module and a source tracing result verification module. By collecting and cleaning data, non-point sources are dispersed into virtual point sources, a CALPUFF model is constructed to simulate diffusion, pollution sources are inverted and verified. The application realizes high space-time resolution pollution monitoring with a 500m*500m grid precision, improves source tracing accuracy, solves the distortion problem of traditional non-point source processing, adapts to complex terrain and dynamic meteorological conditions, and provides accurate decision-making basis for park pollution control.

Description

Technical Field

[0001] This invention relates to the field of volatile organic compound (VOC) pollution monitoring technology in industrial parks, specifically a VOCs pollution monitoring system and method for industrial parks. Background Technology

[0002] Currently, VOCs pollution monitoring in industrial parks faces the following technical bottlenecks:

[0003] Insufficient accuracy in tracing non-point source pollution: Traditional methods are unable to effectively fit non-point source pollution in industrial parks into point sources. There is a lack of virtual point source layout algorithms that take into account factors such as terrain and wind direction, which makes it impossible to accurately invert the pollution source in a small-scale space of 500m×500m.

[0004] Weak small-scale diffusion simulation capability: When integrating WRF meteorological fields, existing CALPUFF models lack double nested grid technology and terrain correction mechanism, making it difficult to simulate the dry / wet deposition and chemical transformation processes of VOCs under complex terrain. The time resolution is generally less than 1 hour, which cannot meet the real-time monitoring requirements.

[0005] Multi-source data fusion is inefficient: online monitoring, mobile observation and meteorological data have spatiotemporal biases, and there is a lack of standardized processes for dynamic cleaning, spatiotemporal calibration and feature extraction, resulting in insufficient reliability of model input data.

[0006] The source tracing results lack closed-loop verification: Traditional techniques have not established a matching degree analysis of characteristic pollutant spectra and a parameter iterative correction mechanism, making it impossible to verify the accuracy of source tracing at a small scale and prone to misjudgment.

[0007] In addition, the existing technology has the following drawbacks:

[0008] Regarding the treatment of non-point source pollution: the combined effects of the geometric center of the sub-region, the prevailing wind direction, and the topographic relief were not considered, and the setting of the virtual point source location lacked scientific basis;

[0009] In terms of model simulation: high-precision coupling between the WRF meteorological field and the CALPUFF model has not been achieved, and the calibration of diffusion parameters relies on human experience, making it difficult to adapt to the dynamically changing atmospheric environment;

[0010] In terms of data application: The lack of spatiotemporal alignment technology for multi-source heterogeneous data leads to inaccurate extraction of pollution plume features, affecting the accuracy of source tracing and inversion.

[0011] To address the above problems, this invention proposes a VOCs pollution monitoring system and method for industrial parks. Summary of the Invention

[0012] The purpose of this invention is to provide a VOCs pollution monitoring system and method for industrial parks to solve the problems raised in the prior art.

[0013] To achieve the above objectives, the present invention provides the following technical solution:

[0014] A VOCs pollution monitoring system for industrial parks includes a multi-source data acquisition and fusion module, a surface-to-point source conversion module, a small-scale CALPUFF modeling module, a pollution source tracing and inversion module, and a source tracing result verification module. The multi-source data acquisition and fusion module is responsible for real-time acquisition of VOCs online monitoring, meteorological, pollution source emission, and mobile monitoring data from the industrial park, simultaneously performing dynamic cleaning, spatiotemporal calibration, and feature extraction to form a standardized dataset for subsequent modeling and analysis. The surface-to-point source conversion module uses a virtual point source processing algorithm to discretize the surface pollution area of ​​the industrial park into regularly distributed virtual point sources, establishing an equivalent relationship between surface and point source pollutant emissions, providing input parameters for CALPUFF modeling. The small-scale... The CALPUFF modeling module constructs a CALPUFF model based on a 500m×500m grid precision, integrating virtual point source parameters and a three-dimensional meteorological field generated by WRF to simulate the diffusion and transport process of VOCs in a small-scale space, outputting hourly concentration distribution data. The pollution source tracing and inversion module combines measured data from monitoring points with the model concentration field, matches pollution transport paths through inverse operations, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources. The source tracing result verification module verifies the accuracy of virtual point source tracing by analyzing the matching degree of characteristic pollutant spectra between mobile monitoring data and model inversion results, and iteratively corrects model parameters to improve the source tracing accuracy at the 500m×500m scale.

[0015] The multi-source data acquisition and fusion module includes a real-time data acquisition unit, a WRF data integration unit, a dynamic data cleaning unit, a spatiotemporal calibration unit, and a feature extraction unit.

[0016] The real-time data acquisition unit collects VOCs concentration, pollution source emission parameters, three-dimensional meteorological field data and mobile monitoring trajectory data in real time through online monitoring equipment, mobile monitoring vehicles and weather stations deployed in the park, and transmits the acquired multi-dimensional raw data to the dynamic data cleaning unit.

[0017] The WRF data integration unit acquires three-dimensional meteorological field data output by the mesoscale meteorological model WRF through an API interface. This data includes wind field, temperature and humidity field, and pressure field. The temporal resolution of the data is no less than 1 hour, and the spatial resolution is resampled using a bilinear interpolation algorithm until it matches the spatial resolution of the 500m×500m grid in the park.

[0018] The dynamic data cleaning unit uses an adaptive filtering algorithm to remove outliers and noise from the collected multi-source heterogeneous data, and at the same time, it fills in missing data through a data integrity verification mechanism to form a preliminary standardized dataset.

[0019] The spatiotemporal calibration unit is based on the 500m×500m grid coordinate system of the park. It unifies the timestamps and maps the spatial coordinates of data from different sources to eliminate clock deviation and spatial positioning error of monitoring equipment.

[0020] The feature extraction unit uses wavelet transform to remove spectral noise and principal component analysis to extract meteorological factor coupling features. It extracts VOCs feature component spectra, pollution plume diffusion features, and meteorological factor coupling relationships from the cleaned and calibrated data to generate a high spatiotemporal resolution dataset suitable for virtual point source modeling. The high spatiotemporal resolution dataset contains a VOCs feature spectrum matrix and a meteorological factor loading matrix, providing data input for the subsequent area source to point source conversion module.

[0021] The area source to point source conversion module includes an area source region division unit, a virtual point source generation unit, an emission parameter mapping unit, and a model adaptation and verification unit.

[0022] The non-point source pollution area division unit is based on the park's geographic information, land use type, and pollution source distribution data provided by the multi-source data acquisition and fusion module. It uses the Delaunay triangulation algorithm to divide the non-point source pollution area of ​​the industrial park into a preset number of sub-regions with regular shapes and similar areas.

[0023] The virtual point source generation unit uses a virtual point source placement algorithm based on pollutant diffusion characteristics to determine the location of virtual point sources within each sub-region. This algorithm comprehensively considers factors such as the geometric center of the sub-region, prevailing wind direction, and topographic relief to determine the coordinates of the virtual point sources. The calculation formula is as follows:

[0024] P i =f(C i W i ,T i );

[0025] Where P i Let C be the coordinates of the virtual point source within the i-th sub-region. i W represents the coordinates of the geometric center of the sub-region. i T is the prevailing wind direction vector for this region. i For terrain relief correction factor; C is the geometric center coordinate of the sub-region. i After dividing the non-point source pollution area of ​​the industrial park into sub-regions using the Delaunay triangulation algorithm, the geometric center is calculated based on the vertex coordinates of the sub-region polygons; the prevailing wind direction vector W is obtained. i Meteorological data from the park, collected by a multi-source data acquisition and fusion module, is used to statistically analyze wind direction and speed data within a preset time period. This analysis calculates wind direction frequency and a weighted average of wind speed to determine the direction and intensity of the prevailing wind in the area, representing this information in vector form. A terrain undulation correction coefficient T is also included.i Based on the digital elevation model data of the park, the standard deviation and range of terrain elevation within the sub-region are calculated, where the range of terrain elevation is the difference between the highest and lowest elevations within the sub-region. After normalizing the standard deviation and range of terrain elevation, coefficients are obtained to correct the location of virtual point sources.

[0026] The emission parameter mapping unit, based on the emission data of each pollution source within the area source sub-region, maps the emission parameters of each pollutant component of the area source to the corresponding virtual point source using the law of mass conservation and spatial weight allocation method. This establishes an equivalent relationship between the emission parameters of the area source and the virtual point source, enabling the virtual point source to reflect the pollution intensity of the area source. The parameters of the area source include, but are not limited to, pollutant emission rate and concentration. Specifically:

[0027] Based on the law of conservation of mass, the total pollution emissions of the area source sub-region are calculated using an additive method to ensure that the total emissions of the virtual point source are consistent with those of the area source. Then, a Gaussian distance decay function w is applied. j =exp(-d j 2 / 2ε 2 ) Calculate the weights, where d j Let ε be the Euclidean distance from the pollution source to the virtual point source, and ε be the characteristic scale of the sub-region, taken as 1 / 3 of the diagonal length of the sub-region. Pollution sources closer to the virtual point source have higher weights. Then, the emission parameters of each pollutant component from the area source are weighted and summed according to spatial weights to obtain a spatially weighted mapping of concentration parameters, generating a standardized emission parameter set for the virtual point source, including:

[0028] Quantitative parameters: emission rate and emission concentration of each pollutant component;

[0029] Spatial parameters: emission height and emission direction, determined by the average emission height and prevailing wind vector of the area source sub-region;

[0030] Time parameter: Emission period distribution, directly inheriting the measured emission cycle data of area source pollution sources;

[0031] The parameter set is stored in HDF5 format and includes a coordinate mapping table, a component concentration matrix and a time series table, which is output to the model adaptation and verification unit.

[0032] The model adaptation and verification unit inputs the virtual point source parameters generated by the emission parameter mapping unit into the small-scale CALPUFF model, and verifies the rationality of the parameters through the following process:

[0033] The pollution diffusion process of a virtual point source was simulated using the CALPUFF model, and the hourly concentration field was output. This data was compared with mobile monitoring and fixed-point monitoring data synchronized with the simulation period in the multi-source data acquisition and fusion module to calculate the receptor point concentration deviation E, as shown in the following formula:

[0034]

[0035] Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model The receptor site contaminant concentration simulated by the CALPUFF model;

[0036] If E exceeds the preset threshold, the virtual point source location and emission parameters are adaptively adjusted using an error backpropagation algorithm. For location adjustment: the virtual point source coordinates P are corrected based on the terrain elevation range and standard deviation. i =f(C i W i ,T i For emission parameter adjustment: remap the emission rate and concentration of the area source sub-region and update the virtual point source parameter set;

[0037] The model was iteratively adjusted until a spatial resolution of 500m×500m was achieved. The deviation between the model simulation results and the actual monitoring data was controlled within a preset threshold range, providing calibrated point source input data for the subsequent small-scale CALPUFF modeling module.

[0038] The small-scale CALPUFF modeling module includes a grid generation unit, a meteorological field processing unit, a virtual point source integration unit, a diffusion simulation unit, and an output calibration unit.

[0039] The grid division unit is based on a spatial resolution requirement of 500m×500m. A double nested grid technology is used to divide the industrial park and the surrounding buffer area. The outer buffer area refers to the outer grid coverage area surrounding the core simulation area, which provides boundary meteorological conditions for the inner high-resolution grid. The outer grid provides boundary conditions for the inner grid and controls the accuracy of small-scale simulation. The inner grid covers the core pollution area, and the grid spacing is limited to 500m×500m.

[0040] The meteorological field processing unit receives the WRF model output results provided by the multi-source data acquisition and fusion module, preprocesses the three-dimensional meteorological field data through the CALMET program module, and generates a meteorological input field suitable for the CALPUFF model by combining the park's topographic elevation data and land use type.

[0041] The virtual point source integration unit performs spatiotemporal matching between the virtual point source parameters calibrated by the model adaptation and verification unit from the area-to-point source conversion module and the WRF meteorological field data from the multi-source data acquisition and fusion module. Employing a double-nested grid technique, it first transforms the virtual point source coordinates from the park's geographic coordinate system to the CALPUFF model grid coordinate system, converting the emission direction parameters to the wind direction offset angle of the CALPUFF source attributes. Then, it performs linear interpolation between the intermittent emission parameters and the WRF meteorological field to generate an hourly emission rate matrix. Subsequently, it integrates virtual point sources, park fixed point sources, and volume sources to generate a mixed-source emission inventory in HDF5 format, containing coordinates, emission parameters, and source types. The inventory data structure is SourceInventory = (x,y,z,t,Q). i C i ,Type,H s ,θ), where H s Here, θ represents the emission height, θ represents the emission direction, and Type identifies the source type. Simultaneously, based on the virtual point source coordinates, the corresponding grid meteorological parameters are extracted to pre-calculate the diffusion coefficient. Based on the emission height Hs and the meteorological stability level from the WRF, the diffusion coefficient σ is pre-calculated using the Pasquill-Gifford classification method. x , σ y , σ z Ultimately, it provides spatiotemporally consistent source term inputs for diffusion simulation in the small-scale CALPUFF modeling module; among them, the HDF5 format contains three-level datasets, namely coordinate mapping, emission parameters and meteorological correlation, which are compatible with the CALPUFF model input format;

[0042] The diffusion simulation unit, based on the Gaussian smoke flow model algorithm of the CALPUFF model and the Lagrange particle tracking method, simulates the pollutant transport process within each 500m×500m grid hourly. The Lagrange particle tracking method is used to calculate the pollutant concentration distribution, and the calculation formula is as follows:

[0043]

[0044] Where C(x,y,z,t) is the pollutant concentration at coordinate (x,y,z) at time t; Q i Let x be the emission rate of the i-th virtual point source; i ,y i ,z i ) represents the coordinates of the virtual point source; σ x , σ y , σ z These are the diffusion coefficients in the x, y, and z directions, respectively;

[0045] The output calibration unit compares the simulated concentration field with the measured data from the multi-source data acquisition and fusion module, and uses the same gradient descent algorithm as the source tracing result verification module to calibrate the diffusion coefficient σ. x , σ y , σ z Adaptive iterative adjustments are performed; the calibrated concentration field needs to be re-input into the pollution source tracing and inversion module, and the source tracing result verification module is triggered to calculate the matching degree of characteristic pollutant spectra until the deviation between the simulated concentration and the measured data at the 500m×500m scale is controlled within the preset threshold. Finally, hourly high spatiotemporal resolution data containing the concentration of each VOCs component are output to provide model input for the pollution source tracing and inversion module. The iterative calibration trigger condition is that the spectral matching degree < the preset threshold.

[0046] The pollution source tracing and inversion module includes a data fusion unit, an inverse operation and inversion unit, a contribution rate calculation unit, and a spatial positioning unit.

[0047] The data fusion unit is responsible for spatiotemporally aligning the hourly pollutant concentration field data output by the small-scale CALPUFF modeling module with the measured data from the multi-source data acquisition and fusion module; specifically, it uses a Kalman filter algorithm to eliminate the bias between the two types of data and generates a fused dataset C. model (x,y,z,t) and C obs (x,y,z,t); where the model concentration field data C model (x,y,z,t) comes from the simulation results of the CALPUFF modeling module based on virtual point source parameters and WRF meteorological field, and the measured concentration data C. obs (x,y,z,t) comes from fixed-point monitoring and mobile monitoring equipment in the multi-source data acquisition and fusion module, including VOCs concentration, mobile trajectory and characteristic component spectrum information;

[0048] The inverse operation inversion unit employs an improved regularized inversion algorithm to solve for the pollution transport inverse trajectory based on the dataset generated by the data fusion unit. Specifically, by minimizing the mean square error between the measured concentration and the model-predicted concentration, and introducing a regularization constraint term, the optimal solution for the virtual point source emission rate vector q is obtained. The calculation formula is as follows:

[0049] min q ||C obs -C model (q)|| 2 +λ||Lq|| 2 ;

[0050] Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model(q) is the simulated value of q based on the CALPUFF modeling module; q is the virtual point source emission rate vector. The initial value of q is generated by the area source-point source conversion module through the law of mass conservation and mapping the measured emission data of the area source. It is then used as an optimization variable for iterative adjustment; the regularization parameter λ is determined based on historical monitoring data through cross-validation, and the constraint matrix L is a preset prior constraint on the spatial distribution of pollution sources.

[0051] The contribution rate calculation unit calculates the pollution contribution rate of each virtual point source to the fixed monitoring receiver point in the park based on the virtual point source emission parameters obtained by the inverse operation inversion unit and using the source allocation receptor model; the calculation formula is as follows:

[0052]

[0053] Among them, Q i For the emission rate of the i-th virtual point source obtained by inversion, C contrib,i f represents the contribution of this point source to the receptor concentration simulated by the CALPUFF model. i Q represents the contribution rate of the i-th virtual point source. j The pollutant emission rate of the j-th virtual point source is obtained by the inverse operation inversion unit of the pollution source tracing inversion module through a regularized inversion algorithm. Its initial value originates from the emission parameter mapping unit of the area source-point source conversion module and is generated based on the measured area source emission data through the law of mass conservation. contrib,j This represents the pollutant concentration contribution of the j-th virtual point source to the receiver point, simulated by the CALPUFF model. It characterizes the concentration contribution of pollutants emitted from this point source at the receiver point and is generated by the diffusion simulation unit of the small-scale CALPUFF modeling module, based on the virtual point source coordinates and emission rate Q. j The data, along with WRF meteorological field data, were obtained by simulating the pollutant diffusion process using a Gaussian smoke flow model and a Lagrange particle tracking method.

[0054] The spatial positioning unit spatially matches the coordinates of virtual point sources with contribution rates exceeding a threshold with the standardized parameter set of virtual point sources generated by the area source-point source conversion module and the area source region division results. It then combines this with the GIS geographic information system to generate a pollution source heat map. The virtual point source coordinates are determined by the virtual point source generation unit of the area source-point source conversion module. The standardized parameter set of virtual point sources includes the geometric center coordinates of the sub-regions output by the area source region division unit, the prevailing wind direction vector, and the emission rate and concentration generated by the emission parameter mapping unit. Through spatial matching and visualization processing, the location of the pollution source at a grid scale of 500m×500m is obtained.

[0055] The source tracing result verification module includes a spectral data acquisition unit, a feature spectral extraction unit, a matching degree calculation unit, a parameter iteration correction unit, and a verification report generation unit;

[0056] The spectral data acquisition unit synchronously acquires the virtual point source feature spectral data output by the contribution rate calculation unit in the pollution source tracing and inversion module and the mobile monitoring spectral data from the real-time data acquisition unit of the multi-source data acquisition and fusion module to establish a spatiotemporally matched verification dataset. The virtual point source feature spectral data is the VOCs component concentration distribution vector of each virtual point source, which is generated by diffusion simulation by the small-scale CALPUFF modeling module. The mobile monitoring spectral data is acquired by a flight mass spectrometer with a time resolution of 1 hour and spatial positioning matched with the 500m×500m grid of the park.

[0057] The feature spectrum extraction unit performs wavelet transform denoising on the original spectrum data to eliminate instrument noise and environmental interference, completes baseline correction through cubic polynomial fitting, and then matches feature peaks based on the NIST mass spectrometry database to extract pollutant concentration values ​​and generate a normalized feature pollutant spectrum vector, providing standardized data input for subsequent matching degree calculation.

[0058] The matching degree calculation unit uses a weighted cosine similarity algorithm to calculate the matching degree index WeightefdSimilarity between the virtual point source spectrum and the mobile monitoring spectrum, in order to quantify the accuracy of the source tracing results. The calculation formula is as follows:

[0059]

[0060] Among them, s model,i The concentration of the i-th characteristic pollutant in the virtual point source spectrum is generated by the CALPUFF model; s obs,i ω represents the concentration of the i-th characteristic pollutant in the mobile monitoring spectrum; i is the weighting coefficient for the i-th pollutant, set according to the degree of hazard of the pollutant in the industry standard; n is the number of characteristic pollutant groups;

[0061] When the parameter iteration correction unit detects that the matching degree is lower than the preset threshold, the unit triggers the inverse operation inversion unit of the pollution source tracing inversion module to perform parameter iteration optimization.

[0062] The verification report generation unit integrates the matching degree calculation results and parameter correction records to generate a visual verification report that includes a matching degree time series, parameter iteration trajectory, GIS comparison maps of pollution hotspot areas before and after correction, and optimization suggestions. The matching degree time series reflects the source tracing accuracy at each time period, the parameter trajectory records the iteration process of q and λ, and the GIS comparison map intuitively shows the differences in the location of pollution sources.

[0063] The parameter iteration correction unit further includes the following:

[0064] When the matching degree index between the virtual point source spectrum and the mobile monitoring spectrum is lower than a preset threshold, the matching degree is transformed into an optimizable error index using the error function Loss = 1 - Weighted Similarity as the optimization objective. The gradient descent algorithm is then used to iteratively update the virtual point source emission rate vector q and the regularization parameter λ until the matching degree meets the standard or the maximum number of iterations is reached. The calculation formula is as follows:

[0065]

[0066] Where q is the virtual point source emission rate vector, q k Let q be the virtual point source emission rate vector for the k-th iteration, with initial values ​​generated by the area-to-point source conversion module through the law of mass conservation; k+1 The emission rate vector is used in the (k+1)th iteration to gradually approximate the optimal solution; α is the learning rate, which controls the iteration step size; ▽ q Loss is the gradient of the error function with respect to the emission rate vector q, representing the direction and magnitude of the influence of changes in q on the error; λ is the regularization parameter. k λ is the regularization parameter for the k-th iteration, and its initial value is determined through cross-validation of historical monitoring data from the multi-source data acquisition and fusion module; k+1 The regularization parameter for the (k+1)th iteration; β is the learning rate, and the parameter update step size is less than α; The gradient of the error function with respect to the regularization parameter λ characterizes the effect of changes in λ on the error;

[0067] The iterated q and λ are fed back to the inverse operation inversion unit of the pollution source tracing module to recalculate the pollution contribution rate of the virtual point source to the receiver point until the matching degree between the source tracing results and the measured data meets the 500m×500m scale requirement.

[0068] A method for monitoring VOCs pollution in industrial parks includes the following steps:

[0069] S1. The system first collects VOCs online monitoring data, meteorological data, pollution source emission parameters and mobile observation trajectory data in the park in real time through the multi-source data acquisition and fusion module. Simultaneously, the data is dynamically cleaned, spatiotemporally calibrated and feature extracted to form a standardized dataset.

[0070] S2. Using the area source to point source conversion module, the area source pollution area in the industrial park is discretized into virtual point sources. The location of the virtual point source is determined by combining the geometric center coordinates of the sub-region, the prevailing wind direction vector, and the terrain undulation correction coefficient. The emission equivalence relationship between area source and point source is established through the law of mass conservation.

[0071] S3. Construct a small-scale CALPUFF model based on a 500m×500m grid precision, integrate the coordinates, emission rate, pollutant composition of virtual point sources and wind field, temperature and humidity field, and pressure field data generated by WRF, simulate the diffusion and transport process of VOCs in a small-scale space, and output hourly concentration distribution data.

[0072] S4. The pollution source tracing and inversion module integrates the concentration field output by the CALPUFF model with the VOCs concentration from fixed-point monitoring and the characteristic component spectrum data from mobile monitoring. It solves the reverse trajectory of pollution transmission through a regularized inversion algorithm, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources.

[0073] S5. The source tracing result verification module obtains the VOCs component concentration distribution vector of the virtual point source and the dynamic spectrum data of the mobile monitoring. After preprocessing, the weighted cosine similarity algorithm is used to calculate the matching degree. When the matching degree is lower than the threshold, the emission rate of the virtual point source and the regularization parameter are iteratively corrected.

[0074] S6. Finally, based on the calibrated source tracing results and the pollution source list, a GIS heat map of the pollution hotspot area, a matching degree time series curve, and a visualization verification report of the parameter iteration trajectory are generated to provide a basis for decision-making in the park's pollution control.

[0075] Compared with the prior art, the beneficial effects of the present invention are:

[0076] 1. Improved accuracy of small-scale source tracing: By using a virtual point source processing algorithm to discretize the surface source into regularly distributed virtual point sources, and combined with the 500m×500m grid accuracy simulation of the micro-scale CALPUFF model, high-precision pollution source tracing with a spatial resolution of 500m×500m and a time resolution of 1 hour is achieved, which improves the positioning accuracy by more than 30% compared with traditional methods.

[0077] 2. Innovative Area-to-Point Source Equivalent Conversion: The area-to-point source conversion module uses the Delaunay triangulation algorithm to divide the area source region, combines the prevailing wind direction vector and terrain undulation correction coefficient to determine the virtual point source location, and maps emission parameters through the law of mass conservation to establish the equivalent relationship between area source and point source, thus solving the distortion problem of traditional area source processing.

[0078] 3. Enhanced adaptability to complex scenarios: The Lagrange particle tracking method, which integrates the WRF three-dimensional meteorological field and the CALPUFF model, takes into account dry / wet deposition, chemical transformation and topographic effects, and realizes VOCs diffusion simulation under complex terrain and dynamic meteorological conditions. The simulation accuracy in complex scenarios is effectively improved compared with traditional models. Attached Figure Description

[0079] Figure 1This is a flowchart illustrating the workflow of a VOCs pollution monitoring system for industrial parks according to the present invention. Detailed Implementation

[0080] 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.

[0081] Example: Figure 1 As shown, the present invention provides a technical solution.

[0082] A VOCs pollution monitoring system for industrial parks includes a multi-source data acquisition and fusion module, a surface-to-point source conversion module, a small-scale CALPUFF modeling module, a pollution source tracing and inversion module, and a source tracing result verification module. The multi-source data acquisition and fusion module is responsible for real-time acquisition of VOCs online monitoring, meteorological, pollution source emission, and mobile monitoring data from the industrial park, simultaneously performing dynamic cleaning, spatiotemporal calibration, and feature extraction to form a standardized dataset for subsequent modeling and analysis. The surface-to-point source conversion module uses a virtual point source processing algorithm to discretize the surface pollution area of ​​the industrial park into regularly distributed virtual point sources, establishing an equivalent relationship between surface and point source pollutant emissions, providing input parameters for CALPUFF modeling. The small-scale... The CALPUFF modeling module constructs a CALPUFF model based on a 500m×500m grid precision, integrating virtual point source parameters and a three-dimensional meteorological field generated by WRF to simulate the diffusion and transport process of VOCs in a small-scale space, outputting hourly concentration distribution data. The pollution source tracing and inversion module combines measured data from monitoring points with the model concentration field, matches pollution transport paths through inverse operations, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources. The source tracing result verification module verifies the accuracy of virtual point source tracing by analyzing the matching degree of characteristic pollutant spectra between mobile monitoring data and model inversion results, and iteratively corrects model parameters to improve the source tracing accuracy at the 500m×500m scale.

[0083] The multi-source data acquisition and fusion module includes a real-time data acquisition unit, a WRF data integration unit, a dynamic data cleaning unit, a spatiotemporal calibration unit, and a feature extraction unit.

[0084] The real-time data acquisition unit collects VOCs concentration, pollution source emission parameters, three-dimensional meteorological field data and mobile monitoring trajectory data in real time through online monitoring equipment, mobile monitoring vehicles and weather stations deployed in the park, and transmits the acquired multi-dimensional raw data to the dynamic data cleaning unit.

[0085] The WRF data integration unit acquires three-dimensional meteorological field data output by the mesoscale meteorological model WRF through an API interface. This data includes wind field, temperature and humidity field, and pressure field. The temporal resolution of the data is no less than 1 hour, and the spatial resolution is resampled using a bilinear interpolation algorithm until it matches the spatial resolution of the 500m×500m grid in the park.

[0086] The dynamic data cleaning unit uses an adaptive filtering algorithm to remove outliers and noise from the collected multi-source heterogeneous data, and at the same time, it fills in missing data through a data integrity verification mechanism to form a preliminary standardized dataset.

[0087] The spatiotemporal calibration unit is based on the 500m×500m grid coordinate system of the park. It unifies the timestamps and maps the spatial coordinates of data from different sources to eliminate clock deviation and spatial positioning error of monitoring equipment.

[0088] The feature extraction unit uses wavelet transform to remove spectral noise and principal component analysis to extract meteorological factor coupling features. It extracts VOCs feature component spectra, pollution plume diffusion features, and meteorological factor coupling relationships from the cleaned and calibrated data to generate a high spatiotemporal resolution dataset suitable for virtual point source modeling. The high spatiotemporal resolution dataset contains a VOCs feature spectrum matrix and a meteorological factor loading matrix, providing data input for the subsequent area source to point source conversion module.

[0089] The area source to point source conversion module includes an area source region division unit, a virtual point source generation unit, an emission parameter mapping unit, and a model adaptation and verification unit.

[0090] The non-point source pollution area division unit is based on the park's geographic information, land use type, and pollution source distribution data provided by the multi-source data acquisition and fusion module. It uses the Delaunay triangulation algorithm to divide the non-point source pollution area of ​​the industrial park into a preset number of sub-regions with regular shapes and similar areas.

[0091] The virtual point source generation unit uses a virtual point source placement algorithm based on pollutant diffusion characteristics to determine the location of virtual point sources within each sub-region. This algorithm comprehensively considers factors such as the geometric center of the sub-region, prevailing wind direction, and topographic relief to determine the coordinates of the virtual point sources. The calculation formula is as follows:

[0092] P i =f(C i W i ,T i );

[0093] Where P i Let C be the coordinates of the virtual point source within the i-th sub-region. i W represents the coordinates of the geometric center of the sub-region. iT is the prevailing wind direction vector for this region. i For terrain relief correction factor; C is the geometric center coordinate of the sub-region. i After dividing the non-point source pollution area of ​​the industrial park into sub-regions using the Delaunay triangulation algorithm, the geometric center is calculated based on the vertex coordinates of the sub-region polygons; the prevailing wind direction vector W is obtained. i Meteorological data from the park, collected by a multi-source data acquisition and fusion module, is used to statistically analyze wind direction and speed data within a preset time period. This analysis calculates wind direction frequency and a weighted average of wind speed to determine the direction and intensity of the prevailing wind in the area, representing this information in vector form. A terrain undulation correction coefficient T is also included. i Based on the digital elevation model data of the park, the standard deviation and range of terrain elevation within the sub-region are calculated, where the range of terrain elevation is the difference between the highest and lowest elevations within the sub-region. After normalizing the standard deviation and range of terrain elevation, coefficients are obtained to correct the location of virtual point sources.

[0094] The emission parameter mapping unit, based on the emission data of each pollution source within the area source sub-region, maps the emission parameters of each pollutant component of the area source to the corresponding virtual point source using the law of mass conservation and spatial weight allocation method. This establishes an equivalent relationship between the emission parameters of the area source and the virtual point source, enabling the virtual point source to reflect the pollution intensity of the area source. The parameters of the area source include, but are not limited to, pollutant emission rate and concentration. Specifically:

[0095] Based on the law of conservation of mass, the total pollution emissions of the area source sub-region are calculated using an additive method to ensure that the total emissions of the virtual point source are consistent with those of the area source. Then, a Gaussian distance decay function w is applied. j =exp(-d j 2 / 2ε 2 ) Calculate the weights, where d j Let ε be the Euclidean distance from the pollution source to the virtual point source, and ε be the characteristic scale of the sub-region, taken as 1 / 3 of the diagonal length of the sub-region. Pollution sources closer to the virtual point source have higher weights. Then, the emission parameters of each pollutant component from the area source are weighted and summed according to spatial weights to obtain a spatially weighted mapping of concentration parameters, generating a standardized emission parameter set for the virtual point source, including:

[0096] Quantitative parameters: emission rate and emission concentration of each pollutant component;

[0097] Spatial parameters: emission height and emission direction, determined by the average emission height and prevailing wind vector of the area source sub-region;

[0098] Time parameter: Emission period distribution, directly inheriting the measured emission cycle data of area source pollution sources;

[0099] The parameter set is stored in HDF5 format and includes a coordinate mapping table, a component concentration matrix and a time series table, which is output to the model adaptation and verification unit.

[0100] The model adaptation and verification unit inputs the virtual point source parameters generated by the emission parameter mapping unit into the small-scale CALPUFF model, and verifies the rationality of the parameters through the following process:

[0101] The pollution diffusion process of a virtual point source was simulated using the CALPUFF model, and the hourly concentration field was output. This data was compared with mobile monitoring and fixed-point monitoring data synchronized with the simulation period in the multi-source data acquisition and fusion module to calculate the receptor point concentration deviation E, as shown in the following formula:

[0102]

[0103] Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model The receptor site contaminant concentration simulated by the CALPUFF model;

[0104] If E exceeds the preset threshold, the virtual point source location and emission parameters are adaptively adjusted using an error backpropagation algorithm. For location adjustment: the virtual point source coordinates P are corrected based on the terrain elevation range and standard deviation. i =f(C i W i ,T i For emission parameter adjustment: remap the emission rate and concentration of the area source sub-region and update the virtual point source parameter set;

[0105] The model was iteratively adjusted until a spatial resolution of 500m×500m was achieved. The deviation between the model simulation results and the actual monitoring data was controlled within a preset threshold range, providing calibrated point source input data for the subsequent small-scale CALPUFF modeling module.

[0106] The small-scale CALPUFF modeling module includes a grid generation unit, a meteorological field processing unit, a virtual point source integration unit, a diffusion simulation unit, and an output calibration unit.

[0107] The grid division unit is based on a spatial resolution requirement of 500m×500m. A double nested grid technology is used to divide the industrial park and the surrounding buffer area. The outer buffer area refers to the outer grid coverage area surrounding the core simulation area, which provides boundary meteorological conditions for the inner high-resolution grid. The outer grid provides boundary conditions for the inner grid and controls the accuracy of small-scale simulation. The inner grid covers the core pollution area, and the grid spacing is limited to 500m×500m.

[0108] The meteorological field processing unit receives the WRF model output results provided by the multi-source data acquisition and fusion module, preprocesses the three-dimensional meteorological field data through the CALMET program module, and generates a meteorological input field suitable for the CALPUFF model by combining the park's topographic elevation data and land use type.

[0109] The virtual point source integration unit performs spatiotemporal matching between the virtual point source parameters calibrated by the model adaptation and verification unit from the area-to-point source conversion module and the WRF meteorological field data from the multi-source data acquisition and fusion module. Employing a double-nested grid technique, it first transforms the virtual point source coordinates from the park's geographic coordinate system to the CALPUFF model grid coordinate system, converting the emission direction parameters to the wind direction offset angle of the CALPUFF source attributes. Then, it performs linear interpolation between the intermittent emission parameters and the WRF meteorological field to generate an hourly emission rate matrix. Subsequently, it integrates virtual point sources, park fixed point sources, and volume sources to generate a mixed-source emission inventory in HDF5 format, containing coordinates, emission parameters, and source types. The inventory data structure is SourceInventory = (x,y,z,t,Q). i C i ,Type,H s ,θ), where H s Here, θ represents the emission height, θ represents the emission direction, and Type identifies the source type. Simultaneously, based on the virtual point source coordinates, the corresponding grid meteorological parameters are extracted to pre-calculate the diffusion coefficient. Based on the emission height Hs and the meteorological stability level from the WRF, the diffusion coefficient σ is pre-calculated using the Pasquill-Gifford classification method. x , σ y , σ z Ultimately, it provides spatiotemporally consistent source term inputs for diffusion simulation in the small-scale CALPUFF modeling module; among them, the HDF5 format contains three-level datasets, namely coordinate mapping, emission parameters and meteorological correlation, which are compatible with the CALPUFF model input format;

[0110] The diffusion simulation unit, based on the Gaussian smoke flow model algorithm of the CALPUFF model and the Lagrange particle tracking method, simulates the pollutant transport process within each 500m×500m grid hourly. The Lagrange particle tracking method is used to calculate the pollutant concentration distribution, and the calculation formula is as follows:

[0111]

[0112] Where C(x,y,z,t) is the pollutant concentration at coordinate (x,y,z) at time t; Q i Let x be the emission rate of the i-th virtual point source; i ,y i ,z i ) represents the coordinates of the virtual point source; σx , σ y , σ z These are the diffusion coefficients in the x, y, and z directions, respectively;

[0113] Furthermore, the diffusion simulation unit of this invention is developed based on the commercial CALPUFF model. Its built-in dry / wet deposition, chemical transformation and topographic impact modules can be automatically activated by inputting meteorological parameters, topographic data and VOCs reaction parameters generated by this system, without the need to modify the core algorithm of the model. Therefore, the patent does not impose additional restrictions on the treatment of environmental factors.

[0114] The output calibration unit compares the simulated concentration field with the measured data from the multi-source data acquisition and fusion module, and uses the same gradient descent algorithm as the source tracing result verification module to calibrate the diffusion coefficient σ. x , σ y , σ z Adaptive iterative adjustments are performed; the calibrated concentration field needs to be re-input into the pollution source tracing and inversion module, and the source tracing result verification module is triggered to calculate the matching degree of characteristic pollutant spectra until the deviation between the simulated concentration and the measured data at the 500m×500m scale is controlled within the preset threshold. Finally, hourly high spatiotemporal resolution data containing the concentration of each VOCs component are output to provide model input for the pollution source tracing and inversion module. The iterative calibration trigger condition is that the spectral matching degree < the preset threshold.

[0115] The pollution source tracing and inversion module includes a data fusion unit, an inverse operation and inversion unit, a contribution rate calculation unit, and a spatial positioning unit.

[0116] The data fusion unit is responsible for spatiotemporally aligning the hourly pollutant concentration field data output by the small-scale CALPUFF modeling module with the measured data from the multi-source data acquisition and fusion module; specifically, it uses a Kalman filter algorithm to eliminate the bias between the two types of data and generates a fused dataset C. model (x,y,z,t) and C obs (x,y,z,t); where the model concentration field data C model (x,y,z,t) comes from the simulation results of the CALPUFF modeling module based on virtual point source parameters and WRF meteorological field, and the measured concentration data C. obs (x,y,z,t) comes from fixed-point monitoring and mobile monitoring equipment in the multi-source data acquisition and fusion module, including VOCs concentration, mobile trajectory and characteristic component spectrum information;

[0117] The inverse operation inversion unit employs an improved regularized inversion algorithm to solve for the pollution transport inverse trajectory based on the dataset generated by the data fusion unit. Specifically, by minimizing the mean square error between the measured concentration and the model-predicted concentration, and introducing a regularization constraint term, the optimal solution for the virtual point source emission rate vector q is obtained. The calculation formula is as follows:

[0118] min q ||C obs -C model (q)|| 2 +λ||Lq|| 2 ;

[0119] Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model (q) is the simulated value of q based on the CALPUFF modeling module; q is the virtual point source emission rate vector. The initial value of q is generated by the area source-point source conversion module through the law of mass conservation and mapping the measured emission data of the area source. It is then used as an optimization variable for iterative adjustment; the regularization parameter λ is determined based on historical monitoring data through cross-validation, and the constraint matrix L is a preset prior constraint on the spatial distribution of pollution sources.

[0120] The contribution rate calculation unit calculates the pollution contribution rate of each virtual point source to the fixed monitoring receiver point in the park based on the virtual point source emission parameters obtained by the inverse operation inversion unit and using the source allocation receptor model; the calculation formula is as follows:

[0121]

[0122] Among them, Q i For the emission rate of the i-th virtual point source obtained by inversion, C contrib,i f represents the contribution of this point source to the receptor concentration simulated by the CALPUFF model. i Q represents the contribution rate of the i-th virtual point source. j The pollutant emission rate of the j-th virtual point source is obtained by the inverse operation inversion unit of the pollution source tracing inversion module through a regularized inversion algorithm. Its initial value originates from the emission parameter mapping unit of the area source-point source conversion module and is generated based on the measured area source emission data through the law of mass conservation. contrib,j This represents the pollutant concentration contribution of the j-th virtual point source to the receiver point, simulated by the CALPUFF model. It characterizes the concentration contribution of pollutants emitted from this point source at the receiver point and is generated by the diffusion simulation unit of the small-scale CALPUFF modeling module, based on the virtual point source coordinates and emission rate Q. j The data, along with WRF meteorological field data, were obtained by simulating the pollutant diffusion process using a Gaussian smoke flow model and a Lagrange particle tracking method.

[0123] The spatial positioning unit spatially matches the coordinates of virtual point sources with contribution rates exceeding a threshold with the standardized parameter set of virtual point sources generated by the area source-point source conversion module and the area source region division results. It then combines this with the GIS geographic information system to generate a pollution source heat map. The virtual point source coordinates are determined by the virtual point source generation unit of the area source-point source conversion module. The standardized parameter set of virtual point sources includes the geometric center coordinates of the sub-regions output by the area source region division unit, the prevailing wind direction vector, and the emission rate and concentration generated by the emission parameter mapping unit. Through spatial matching and visualization processing, the location of the pollution source at a grid scale of 500m×500m is obtained.

[0124] The source tracing result verification module includes a spectral data acquisition unit, a feature spectral extraction unit, a matching degree calculation unit, a parameter iteration correction unit, and a verification report generation unit;

[0125] The spectral data acquisition unit synchronously acquires the virtual point source feature spectral data output by the contribution rate calculation unit in the pollution source tracing and inversion module and the mobile monitoring spectral data from the real-time data acquisition unit of the multi-source data acquisition and fusion module to establish a spatiotemporally matched verification dataset. The virtual point source feature spectral data is the VOCs component concentration distribution vector of each virtual point source, which is generated by diffusion simulation by the small-scale CALPUFF modeling module. The mobile monitoring spectral data is acquired by a flight mass spectrometer with a time resolution of 1 hour and spatial positioning matched with the 500m×500m grid of the park.

[0126] The feature spectrum extraction unit performs wavelet transform denoising on the original spectrum data to eliminate instrument noise and environmental interference, completes baseline correction through cubic polynomial fitting, and then matches feature peaks based on the NIST mass spectrometry database to extract pollutant concentration values ​​and generate a normalized feature pollutant spectrum vector, providing standardized data input for subsequent matching degree calculation.

[0127] The matching degree calculation unit uses a weighted cosine similarity algorithm to calculate the matching degree index WeightefdSimilarity between the virtual point source spectrum and the mobile monitoring spectrum, in order to quantify the accuracy of the source tracing results. The calculation formula is as follows:

[0128]

[0129] Among them, s model,i The concentration of the i-th characteristic pollutant in the virtual point source spectrum is generated by the CALPUFF model; s obs,i ω represents the concentration of the i-th characteristic pollutant in the mobile monitoring spectrum; i is the weighting coefficient for the i-th pollutant, set according to the degree of hazard of the pollutant in the industry standard; n is the number of characteristic pollutant groups;

[0130] When the parameter iteration correction unit detects that the matching degree is lower than the preset threshold, the unit triggers the inverse operation inversion unit of the pollution source tracing inversion module to perform parameter iteration optimization.

[0131] The verification report generation unit integrates the matching degree calculation results and parameter correction records to generate a visual verification report that includes a matching degree time series, parameter iteration trajectory, GIS comparison maps of pollution hotspot areas before and after correction, and optimization suggestions. The matching degree time series reflects the source tracing accuracy at each time period, the parameter trajectory records the iteration process of q and λ, and the GIS comparison map intuitively shows the differences in the location of pollution sources.

[0132] The parameter iteration correction unit further includes the following:

[0133] When the matching degree index between the virtual point source spectrum and the mobile monitoring spectrum is lower than a preset threshold, the matching degree is transformed into an optimizable error index using the error function Loss = 1 - Weighted Similarity as the optimization objective. The gradient descent algorithm is then used to iteratively update the virtual point source emission rate vector q and the regularization parameter λ until the matching degree meets the standard or the maximum number of iterations is reached. The calculation formula is as follows:

[0134]

[0135] Where q is the virtual point source emission rate vector, q k Let q be the virtual point source emission rate vector for the k-th iteration, with initial values ​​generated by the area-to-point source conversion module through the law of mass conservation; k+1 Let be the emission rate vector for the (k+1)th iteration, which is used to gradually approximate the optimal solution through iteration; α is the learning rate, which controls the iteration step size; λ is the gradient of the error function with respect to the emission rate vector q, representing the direction and magnitude of the influence of changes in q on the error; λ is the regularization parameter. k λ is the regularization parameter for the k-th iteration, and its initial value is determined through cross-validation of historical monitoring data from the multi-source data acquisition and fusion module; k+1 The regularization parameter for the (k+1)th iteration; β is the learning rate, and the parameter update step size is less than α; ▽ λ The gradient of the Loss error function with respect to the regularization parameter λ represents the effect of changes in λ on the error.

[0136] The iterated q and λ are fed back to the inverse operation inversion unit of the pollution source tracing module to recalculate the pollution contribution rate of the virtual point source to the receiver point until the matching degree between the source tracing results and the measured data meets the 500m×500m scale requirement.

[0137] A method for monitoring VOCs pollution in industrial parks includes the following steps:

[0138] S1. The system first collects VOCs online monitoring data, meteorological data, pollution source emission parameters and mobile observation trajectory data in the park in real time through the multi-source data acquisition and fusion module. Simultaneously, the data is dynamically cleaned, spatiotemporally calibrated and feature extracted to form a standardized dataset.

[0139] S2. Using the area source to point source conversion module, the area source pollution area in the industrial park is discretized into virtual point sources. The location of the virtual point source is determined by combining the geometric center coordinates of the sub-region, the prevailing wind direction vector, and the terrain undulation correction coefficient. The emission equivalence relationship between area source and point source is established through the law of mass conservation.

[0140] S3. Construct a small-scale CALPUFF model based on a 500m×500m grid precision, integrate the coordinates, emission rate, pollutant composition of virtual point sources and wind field, temperature and humidity field, and pressure field data generated by WRF, simulate the diffusion and transport process of VOCs in a small-scale space, and output hourly concentration distribution data.

[0141] S4. The pollution source tracing and inversion module integrates the concentration field output by the CALPUFF model with the VOCs concentration from fixed-point monitoring and the characteristic component spectrum data from mobile monitoring. It solves the reverse trajectory of pollution transmission through a regularized inversion algorithm, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources.

[0142] S5. The source tracing result verification module obtains the VOCs component concentration distribution vector of the virtual point source and the dynamic spectrum data of the mobile monitoring. After preprocessing, the weighted cosine similarity algorithm is used to calculate the matching degree. When the matching degree is lower than the threshold, the emission rate of the virtual point source and the regularization parameter are iteratively corrected.

[0143] S6. Finally, based on the calibrated source tracing results and the pollution source list, a GIS heat map of the pollution hotspot area, a matching degree time series curve, and a visualization verification report of the parameter iteration trajectory are generated to provide a basis for decision-making in the park's pollution control.

[0144] Assume a 6km × 6km chemical industrial park located at 111.68°E, 40.79°N in the southwestern suburbs of Hohhot, comprising 3 chemical workshops (area source), 2 chimneys (30m / 20m high), and a tank area (volume source). The terrain elevation is 1000-1050m, with prevailing northwest winds (315°, frequency 28%). During the monitoring period from 8:00 to 18:00 on June 15, 2025, the average wind speed was 5.2m / s, the temperature ranged from 22-28℃, and the atmospheric stability was alternating between CD and other categories. The system deployed 10 GC-2030 gas chromatographs (detection limit 0.1ppm), 5 meteorological stations, and 1 TOF-MS mobile monitoring vehicle. Data was stored in HDF5 format with a spatial resolution of 500m × 500m and a temporal resolution of 1 hour.

[0145] Online monitoring in Workshop A showed that the benzene emission rate was 1.2 kg / h and the toluene emission rate was 0.8 kg / h from 8:00 to 9:00, stored in VOCs_Emission.csv; at 8:30, the mobile monitoring vehicle measured 52 ppm of benzene and 35 ppm of toluene at grid (2,3) (geographic coordinates 1000m, 1500m), stored in Airsampling_TOFMS.h5; weather station #1 (coordinates (1,1)) recorded a wind speed of 5.1 m / s, a wind direction of 312°, and a temperature of 23°C at 8:00, stored in Meteorolog y_Station1.txt.

[0146] WRF obtains 1-hour resolution data of the WRF model through API. The wind speed at the center of the park is 4.9 m / s and the wind direction is 310° from 8:00 to 9:00. The original 1km×1km grid is resampled to 500m×500m through bilinear interpolation. After grid (2,2) interpolation, the wind speed is 4.85 m / s and the wind direction is 311°. The data is converted to CALPUFF format and stored in WRF_CA LPUFF.h5.

[0147] The outlier of 1.8 kg / h at 8:15 in the benzene emission rate sequence of Workshop A (mean 1.2 kg / h, standard deviation 0.15 kg / h) was corrected to 1.3 kg / h according to the 3σ principle; the missing data from 9:00 to 10:00 in Workshop B were supplemented using the forward filling method to obtain a benzene concentration of 48 ppm and an emission rate of 20,000 m³ / h. 3 / h, the cleaned data is stored in Cleaned_Data.h5; the online device clock deviation +5 minutes is synchronized via NTP, the time stamp of the navigation track is aligned with the weather station (error ≤ ±30 seconds), and the geographic coordinates (1000m, 1500m) are converted to grid (1,2).

[0148] The feature-extracted benzene spectrum was de-denoised using a three-level db4 wavelet decomposition, increasing the signal-to-noise ratio from 15dB to 25dB. Principal component analysis extracted benzene and toluene as feature components (cumulative variance contribution rate of 85%), generating feature matrices of 49.5ppm benzene and 33.2ppm toluene, which were stored in Feature_Spectra.csv.

[0149] The area source region division is based on the Delaunay triangulation algorithm, dividing the pollution source distribution points (workshops A(1,2), B(2,3), C(3,1)) in the park into 4 sub-regions. The vertices of sub-region 1 are (1,2), (2,2), and (2,3), and the geometric center is calculated as: C. i =({1+2+2} / {3},{2+2+3} / {3}=(1.667,2.333),(grid coordinates);

[0150] Virtual point source generates prevailing wind direction 315°, which is converted into a unit vector W. i = (0.707, -0.707), sub-region 1 elevation 1020m, 1030m, 1025m, calculated standard deviation σ = 4.08m, range 10m, normalized topographic correction factor: T i =0.608 The initial virtual point source coordinates are calculated as: P i = (44.667, -40.667);

[0151] Due to exceeding the park area, T was revised. i =0.2 then P i = (15.807, -11.807), actual coordinates are (8403.5m, -5403.5m). Emission parameter mapping and verification: Total emission rate of sub-region 1: Benzene 1.2 + 1.0 + 0.8 = 3.0 kg / h, Toluene 0.8 + 0.6 + 0.5 = 1.9 kg / h; Gaussian weighting function w j =exp(-d_j 2 / (2ε 2 In the simulation, the diagonal of the sub-region ε = 500√2 / 3 ≈ 235.7m, and the weight at 50m from the source is calculated using the weighted average of the benzene concentration from the virtual point source: C = 49.0ppm. The input CALPUFF simulation at 8:30 shows the concentration C at receptor point (2,3). mode l = 0.115, {mg / m 3}, compared with the measured C obs =0.123 mg / m 3 The deviation was 6.5%. After adjusting the emission height to 12m, the deviation decreased to 4%. The calibration parameters are stored in Calibrated_Source.h5.

[0152] A double-grid nesting method was used, with an outer layer of 2000m×2000m (3×3 grid) and an inner layer of 500m×500m (4×4 grid) covering the core area; WRF data was preprocessed using CALMET and combined with an elevation of 1020m to generate a meteorological input field with stability class C. The pre-calculated diffusion parameters were: o' x =105,o' y =85,o' z =55, Virtual point source integration and diffusion simulation: Virtual point source coordinates are transformed to the inner mesh (2,2), emission direction offset angle is 0°, and linear interpolation is used to generate the emission matrix of 3.0 kg / h for the period of 9:00-10:00; Concentration calculation at receptor point (3,3): C = 1.46 × 10 -18 ,{kg / m 3 Measured concentration: 0.10 mg / m³ 3 Compared with the simulated value of 0.09 mg / m³ 3 The error function Loss = (0.10 - 0.09)2 =0.0001, adjust o' y =85m→80m After the loss drops to 0.000025.

[0153] During the 8:30 monitoring period, the system first fuses the model concentration field with the measured data using Kalman filtering. The model concentration field Cmodel = 0.11 mg / m³. 3 Based on CALPUFF simulations, the measured value of Cobs is 0.12 mg / m³. 3 The data was obtained from mobile monitoring and iterative calculations based on state-space equations. The process noise covariance Q = 0.01 and the observation noise covariance R = 0.02. The final filtered result yielded an estimated true concentration of x = 0.116 mg / m³. 3 This effectively eliminates the bias between the two types of data. The inverse operation inversion unit starts the regularized inversion with the initial emission rate vector q = (3.0, 1.9) kg / h (benzene, toluene), and the objective function is...

[0154] min||0.12-C model (q)|| 2 +2×0.05×|Lq|| 2 ,

[0155] Where L is a second-order difference matrix; after 5 iterations, q converges to (2.92, 1.83) kg / h, at which point Loss = 0.001, which meets the inversion accuracy requirement.

[0156] Based on the inversion results, the contribution rate calculation unit determined that the concentration contribution of virtual point source 1 to the receptor point was 0.08 mg / m³. 3 It accounts for 0.10 mg / m³ of the total contribution. 3 The contribution rate of virtual point source 2 is 20%, while the contribution rate of virtual point source 2 is 80%. The spatial positioning unit matches the coordinates (2,2) of the virtual point source with a contribution rate of over 70% with the GIS layer and finds that it precisely corresponds to the location of workshop A. In the generated heat map, the red area (contribution rate > 70%) completely covers the workshop, and the yellow area (50-70%) is distributed on the surrounding roads.

[0157] The virtual point source spectrum showed benzene concentrations of 49.0 ppm and toluene concentrations of 32.0 ppm. The walkthrough spectrum at 8:30 showed benzene concentrations of 50 ppm and toluene concentrations of 33 ppm. After performing three polynomial baseline corrections on the toluene spectrum, the characteristic spectrum extraction unit corrected the concentration to 32.5 ppm. By matching benzene (m / z = 78, 98% matching) and toluene (m / z = 92, 95% matching) against the NIST database, a normalized feature vector (0.8, 0.2) was generated.

[0158] The matching degree was calculated using a weighted cosine similarity algorithm, with weight coefficients wi = (0.6, 0.4). Substituting these values ​​into the formula, the numerator was calculated as 1470 + 422.4 = 1892.4, and the denominator as 3596. The final matching degree was approximately 0.526, which was lower than the preset threshold of 0.9, triggering parameter iteration. The parameter iteration correction unit used Loss = 1 - 0.526 = 0.474 as the optimization objective, adjusting q to (2.95, 1.85) kg / h and λ to 0.06 using a gradient descent algorithm. After 10 iterations, the matching degree improved to 0.92.

[0159] The verification report shows that the matching degree time series from 8:00 to 9:00 is 0.85→0.92→0.89, the parameter trajectory record q converges from (3.0,1.9) to (2.95,1.85), and the GIS comparison map clearly shows that the pollution hotspot has been corrected from (2,3) to (2,2) in workshop A area. Finally, it is recommended that the workshop improve the waste gas collection efficiency to over 90%.

[0160] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A VOCs pollution monitoring system for industrial parks, characterized in that: The system includes a multi-source data acquisition and fusion module, a surface-to-point-source conversion module, a small-scale CALPUFF modeling module, a pollution source tracing and inversion module, and a source tracing result verification module. The multi-source data acquisition and fusion module is responsible for real-time acquisition of VOCs online monitoring, meteorological, pollution source emission, and mobile monitoring data from the industrial park, simultaneously performing dynamic cleaning, spatiotemporal calibration, and feature extraction to form a standardized dataset for subsequent modeling and analysis. The surface-to-point-source conversion module uses a virtual point source processing algorithm to discretize the surface-source pollution area of ​​the industrial park into regularly distributed virtual point sources, establishing an equivalent relationship between surface-source and point-source pollutant emissions, providing input parameters for CALPUFF modeling. The small-scale CALPUFF modeling module... The model module constructs a CALPUFF model based on a 500m×500m grid precision, integrating virtual point source parameters and a three-dimensional meteorological field generated by WRF to simulate the diffusion and transport process of VOCs in a small-scale space, and outputs hourly concentration distribution data. The pollution source tracing and inversion module combines the measured data of monitoring points with the model concentration field, matches the pollution transport path through inverse operation, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources. The source tracing result verification module verifies the accuracy of virtual point source tracing by analyzing the matching degree of characteristic pollutant spectra between mobile monitoring data and model inversion results, and iteratively corrects model parameters to improve the source tracing accuracy at the 500m×500m scale.

2. The VOCs pollution monitoring system for industrial parks according to claim 1, characterized in that: The multi-source data acquisition and fusion module includes a real-time data acquisition unit, a WRF data integration unit, a dynamic data cleaning unit, a spatiotemporal calibration unit, and a feature extraction unit. The real-time data acquisition unit collects VOCs concentration, pollution source emission parameters, three-dimensional meteorological field data and mobile monitoring trajectory data in real time through online monitoring equipment, mobile monitoring vehicles and weather stations deployed in the park, and transmits the acquired multi-dimensional raw data to the dynamic data cleaning unit. The WRF data integration unit acquires three-dimensional meteorological field data output by the mesoscale meteorological model WRF through an API interface. This data includes wind field, temperature and humidity field, and pressure field. The temporal resolution of the data is no less than 1 hour, and the spatial resolution is resampled using a bilinear interpolation algorithm until it matches the spatial resolution of the 500m×500m grid in the park. The dynamic data cleaning unit uses an adaptive filtering algorithm to remove outliers and noise from the collected multi-source heterogeneous data, and at the same time, it fills in missing data through a data integrity verification mechanism to form a preliminary standardized dataset. The spatiotemporal calibration unit is based on the 500m×500m grid coordinate system of the park. It unifies the timestamps and maps the spatial coordinates of data from different sources to eliminate clock deviation and spatial positioning error of monitoring equipment. The feature extraction unit uses wavelet transform to remove spectral noise and principal component analysis to extract meteorological factor coupling features. It extracts VOCs feature component spectra, pollution plume diffusion features, and meteorological factor coupling relationships from the cleaned and calibrated data to generate a high spatiotemporal resolution dataset suitable for virtual point source modeling. The high spatiotemporal resolution dataset contains a VOCs feature spectrum matrix and a meteorological factor loading matrix, providing data input for the subsequent area source to point source conversion module.

3. The VOCs pollution monitoring system for industrial parks according to claim 2, characterized in that: The area source to point source conversion module includes an area source region division unit, a virtual point source generation unit, an emission parameter mapping unit, and a model adaptation and verification unit. The non-point source pollution area division unit is based on the park's geographic information, land use type, and pollution source distribution data provided by the multi-source data acquisition and fusion module. It uses the Delaunay triangulation algorithm to divide the non-point source pollution area of ​​the industrial park into a preset number of sub-regions with regular shapes and similar areas. The virtual point source generation unit uses a virtual point source placement algorithm based on pollutant diffusion characteristics to determine the location of virtual point sources within each sub-region. This algorithm comprehensively considers factors such as the geometric center of the sub-region, prevailing wind direction, and topographic relief to determine the coordinates of the virtual point sources. The calculation formula is as follows: P i =f(C i ,W i ,T i ); Where P i Let C be the coordinates of the virtual point source within the i-th sub-region. i W represents the coordinates of the geometric center of the sub-region. i T is the prevailing wind direction vector for this region. i For terrain relief correction factor; C is the geometric center coordinate of the sub-region. i After dividing the non-point source pollution area of ​​the industrial park into area source region division units using the Delaunay triangulation algorithm, the geometric center is obtained based on the coordinates of the vertices of the sub-region polygons. Dominant wind vector W i Meteorological data from the park, collected by a multi-source data acquisition and fusion module, is used to statistically analyze wind direction and speed data within a preset time period. This analysis calculates wind direction frequency and a weighted average of wind speed to determine the direction and intensity of the prevailing wind in the area, representing this information in vector form. A terrain undulation correction coefficient T is also included. i Based on the digital elevation model data of the park, the standard deviation and range of terrain elevation within the sub-region are calculated, where the range of terrain elevation is the difference between the highest and lowest elevations within the sub-region. After normalizing the standard deviation and range of terrain elevation, coefficients are obtained to correct the location of virtual point sources. The emission parameter mapping unit, based on the emission data of each pollution source within the area source sub-region, maps the emission parameters of each pollutant component of the area source to the corresponding virtual point source using the law of mass conservation and spatial weight allocation method. This establishes an equivalent relationship between the emission parameters of the area source and the virtual point source, enabling the virtual point source to reflect the pollution intensity of the area source. The parameters of the area source include, but are not limited to, pollutant emission rate and concentration. Specifically: Based on the law of conservation of mass, the total pollution emissions of the area source sub-region are calculated using an additive method to ensure that the total emissions of the virtual point source are consistent with those of the area source. Then, a Gaussian distance decay function w is applied. j =exp(-d j 2 / 2ε 2 ) Calculate the weights, where d j Let ε be the Euclidean distance from the pollution source to the virtual point source, and ε be the characteristic scale of the sub-region, taken as 1 / 3 of the diagonal length of the sub-region. Pollution sources closer to the virtual point source have higher weights. Then, the emission parameters of each pollutant component from the area source are weighted and summed according to spatial weights to obtain a spatially weighted mapping of concentration parameters, generating a standardized emission parameter set for the virtual point source, including: Quantitative parameters: emission rate and emission concentration of each pollutant component; Spatial parameters: emission height and emission direction, determined by the average emission height and prevailing wind vector of the area source sub-region; Time parameter: Emission period distribution, directly inheriting the measured emission cycle data of area source pollution sources; The parameter set is stored in HDF5 format and includes a coordinate mapping table, a component concentration matrix and a time series table, which is output to the model adaptation and verification unit. The model adaptation and verification unit inputs the virtual point source parameters generated by the emission parameter mapping unit into the small-scale CALPUFF model, and verifies the rationality of the parameters through the following process: The pollution diffusion process of a virtual point source was simulated using the CALPUFF model, and the hourly concentration field was output. This data was compared with mobile monitoring and fixed-point monitoring data synchronized with the simulation period in the multi-source data acquisition and fusion module to calculate the receptor point concentration deviation E, as shown in the following formula: Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model The receptor site contaminant concentration simulated by the CALPUFF model; If E exceeds the preset threshold, the virtual point source location and emission parameters are adaptively adjusted using an error backpropagation algorithm. For location adjustment: the virtual point source coordinates P are corrected based on the terrain elevation range and standard deviation. i =f(C i W i ,T i For emission parameter adjustment: remap the emission rate and concentration of the area source sub-region and update the virtual point source parameter set; The model was iteratively adjusted until a spatial resolution of 500m×500m was achieved. The deviation between the model simulation results and the actual monitoring data was controlled within a preset threshold range, providing calibrated point source input data for the subsequent small-scale CALPUFF modeling module.

4. The VOCs pollution monitoring system for industrial parks according to claim 3, characterized in that: The small-scale CALPUFF modeling module includes a grid generation unit, a meteorological field processing unit, a virtual point source integration unit, a diffusion simulation unit, and an output calibration unit. The grid division unit is based on a spatial resolution requirement of 500m×500m. A double nested grid technique is used to divide the industrial park and the surrounding buffer area. The outer buffer area refers to the outer grid coverage area surrounding the core simulation area, which provides boundary meteorological conditions for the inner high-resolution grid. The outer grid provides boundary conditions for the inner grid, controlling the accuracy of small-scale simulation. The inner grid covers the core pollution area, with a grid spacing of 500m × 500m. The meteorological field processing unit receives the WRF model output results provided by the multi-source data acquisition and fusion module, preprocesses the three-dimensional meteorological field data through the CALMET program module, and generates a meteorological input field suitable for the CALPUFF model by combining the park's topographic elevation data and land use type. The virtual point source integration unit performs spatiotemporal matching between the virtual point source parameters calibrated by the model adaptation and verification unit from the area-to-point source conversion module and the WRF meteorological field data from the multi-source data acquisition and fusion module. Employing a double-nested grid technique, it first transforms the virtual point source coordinates from the park's geographic coordinate system to the CALPUFF model grid coordinate system, converting the emission direction parameters to the wind direction offset angle of the CALPUFF source attributes. Then, it performs linear interpolation between the intermittent emission parameters and the WRF meteorological field to generate an hourly emission rate matrix. Subsequently, it integrates virtual point sources, park fixed point sources, and volume sources to generate a mixed-source emission inventory in HDF5 format, containing coordinates, emission parameters, and source types. The inventory data structure is SourceInventory = (x,y,z,t,Q). i C i ,Type,H s ,θ), where H s Here, θ represents the emission height, θ represents the emission direction, and Type identifies the source type. Simultaneously, based on the virtual point source coordinates, the corresponding grid meteorological parameters are extracted to pre-calculate the diffusion coefficient. Based on the emission height Hs and the meteorological stability level from the WRF, the diffusion coefficient σ is pre-calculated using the Pasquill-Gifford classification method. x , σ y , σ z Ultimately, it provides spatiotemporally consistent source term inputs for diffusion simulation in the small-scale CALPUFF modeling module; among them, the HDF5 format contains three-level datasets, namely coordinate mapping, emission parameters and meteorological correlation, which are compatible with the CALPUFF model input format; The diffusion simulation unit, based on the Gaussian smoke flow model algorithm of the CALPUFF model and the Lagrange particle tracking method, simulates the pollutant transport process within each 500m×500m grid hourly. The Lagrange particle tracking method is used to calculate the pollutant concentration distribution, and the calculation formula is as follows: Where C(x,y,z,t) is the pollutant concentration at coordinate (x,y,z) at time t; Q i Let x be the emission rate of the i-th virtual point source; i ,y i ,z i ) represents the coordinates of the virtual point source; σ x , σ y , σ z These are the diffusion coefficients in the x, y, and z directions, respectively; The output calibration unit compares the simulated concentration field with the measured data from the multi-source data acquisition and fusion module, and uses the same gradient descent algorithm as the source tracing result verification module to calibrate the diffusion coefficient σ. x , σ y , σ z Adaptive iterative adjustments are performed; the calibrated concentration field needs to be re-input into the pollution source tracing and inversion module, and the source tracing result verification module is triggered to calculate the matching degree of characteristic pollutant spectra until the deviation between the simulated concentration and the measured data at the 500m×500m scale is controlled within the preset threshold. Finally, hourly high spatiotemporal resolution data containing the concentration of each VOCs component are output to provide model input for the pollution source tracing and inversion module. The iterative calibration trigger condition is that the spectral matching degree < the preset threshold.

5. The VOCs pollution monitoring system for industrial parks according to claim 4, characterized in that: The pollution source tracing and inversion module includes a data fusion unit, an inverse operation and inversion unit, a contribution rate calculation unit, and a spatial positioning unit. The data fusion unit is responsible for spatiotemporally aligning the hourly pollutant concentration field data output by the small-scale CALPUFF modeling module with the measured data from the multi-source data acquisition and fusion module; specifically, it uses a Kalman filter algorithm to eliminate the bias between the two types of data and generates a fused dataset C. model (x,y,z,t) and C obs (x,y,z,t); where the model concentration field data C model (x,y,z,t) comes from the simulation results of the CALPUFF modeling module based on virtual point source parameters and WRF meteorological field, and the measured concentration data C. obs (x,y,z,t) comes from fixed-point monitoring and mobile monitoring equipment in the multi-source data acquisition and fusion module, including VOCs concentration, mobile trajectory and characteristic component spectrum information; The inverse operation inversion unit employs an improved regularized inversion algorithm to solve for the pollution transport inverse trajectory based on the dataset generated by the data fusion unit. Specifically, by minimizing the mean square error between the measured concentration and the model-predicted concentration, and introducing a regularization constraint term, the optimal solution for the virtual point source emission rate vector q is obtained. The calculation formula is as follows: my q ||C obs -C model (q)|| 2 +λ||Lq|| 2 ; Among them, C obs C represents the measured value from the multi-source data acquisition and fusion module. model (q) is the simulated value of q based on the CALPUFF modeling module; q is the virtual point source emission rate vector. The initial value of q is generated by the area source-point source conversion module through the law of mass conservation and mapping the measured emission data of the area source. It is then used as an optimization variable for iterative adjustment; the regularization parameter λ is determined based on historical monitoring data through cross-validation, and the constraint matrix L is a preset prior constraint on the spatial distribution of pollution sources. The contribution rate calculation unit calculates the pollution contribution rate of each virtual point source to the fixed monitoring receiver point in the park based on the virtual point source emission parameters obtained by the inverse operation inversion unit and using the source allocation receptor model; the calculation formula is as follows: Among them, Q i For the emission rate of the i-th virtual point source obtained by inversion, C contrib,i f represents the contribution of this point source to the receptor concentration simulated by the CALPUFF model. i Q represents the contribution rate of the i-th virtual point source. j The pollutant emission rate of the j-th virtual point source is obtained by the inverse operation inversion unit of the pollution source tracing inversion module through a regularized inversion algorithm. Its initial value originates from the emission parameter mapping unit of the area source-point source conversion module and is generated based on the measured area source emission data through the law of mass conservation. contrib,j This represents the pollutant concentration contribution of the j-th virtual point source to the receiver point, simulated by the CALPUFF model. It characterizes the concentration contribution of pollutants emitted from this point source at the receiver point and is generated by the diffusion simulation unit of the small-scale CALPUFF modeling module, based on the virtual point source coordinates and emission rate Q. j The data, along with WRF meteorological field data, were obtained by simulating the pollutant diffusion process using a Gaussian smoke flow model and a Lagrange particle tracking method. The spatial positioning unit spatially matches the coordinates of virtual point sources with contribution rates exceeding a threshold with the standardized parameter set of virtual point sources generated by the area source-point source conversion module and the area source region division results. It then combines this with the GIS geographic information system to generate a pollution source heat map. The virtual point source coordinates are determined by the virtual point source generation unit of the area source-point source conversion module. The standardized parameter set of virtual point sources includes the geometric center coordinates of the sub-regions output by the area source region division unit, the prevailing wind direction vector, and the emission rate and concentration generated by the emission parameter mapping unit. Through spatial matching and visualization processing, the location of the pollution source at a grid scale of 500m×500m is obtained.

6. The VOCs pollution monitoring system for industrial parks according to claim 5, characterized in that: The source tracing result verification module includes a spectral data acquisition unit, a feature spectral extraction unit, a matching degree calculation unit, a parameter iteration correction unit, and a verification report generation unit; The spectral data acquisition unit establishes a spatiotemporal matching verification dataset by synchronously acquiring the virtual point source feature spectral data output by the contribution rate calculation unit in the pollution source tracing and inversion module and the mobile monitoring spectral data from the real-time data acquisition unit of the multi-source data acquisition and fusion module. The virtual point source feature spectrum is the VOCs component concentration distribution vector of each virtual point source, which is generated by diffusion simulation by the small-scale CALPUFF modeling module; the mobile monitoring spectrum is obtained by a flight mass spectrometer with a time resolution of 1 hour and spatial positioning matched with the 500m×500m grid of the park. The feature spectrum extraction unit performs wavelet transform denoising on the original spectrum data to eliminate instrument noise and environmental interference, completes baseline correction through cubic polynomial fitting, and then matches feature peaks based on the NIST mass spectrometry database to extract pollutant concentration values ​​and generate a normalized feature pollutant spectrum vector, providing standardized data input for subsequent matching degree calculation. The matching degree calculation unit uses a weighted cosine similarity algorithm to calculate the matching degree index WeightefdSimilarity between the virtual point source spectrum and the mobile monitoring spectrum, in order to quantify the accuracy of the source tracing results. The calculation formula is as follows: Among them, s model,i The concentration of the i-th characteristic pollutant in the virtual point source spectrum is generated by the CALPUFF model; s obs,i ω represents the concentration of the i-th characteristic pollutant in the mobile monitoring spectrum; i is the weighting coefficient for the i-th pollutant, set according to the degree of hazard of the pollutant in the industry standard; n is the number of characteristic pollutant groups; When the parameter iteration correction unit detects that the matching degree is lower than the preset threshold, the unit triggers the inverse operation inversion unit of the pollution source tracing inversion module to perform parameter iteration optimization. The verification report generation unit integrates the matching degree calculation results and parameter correction records to generate a visual verification report that includes the matching degree time series, parameter iteration trajectory, GIS comparison map of pollution hotspot areas before and after correction, and optimization suggestions. The matching degree time series reflects the source tracing accuracy at each time period, the parameter trajectory records the iteration process of q and λ, and the GIS comparison map intuitively shows the differences in the location of pollution sources.

7. The VOCs pollution monitoring system for industrial parks according to claim 6, characterized in that: The parameter iteration correction unit further includes the following: When the matching degree index between the virtual point source spectrum and the mobile monitoring spectrum is lower than the preset threshold, the matching degree of the spectrum is transformed into an optimizable error index with the error function Loss = 1 - Weighted Similarity as the optimization objective; the gradient descent algorithm is used to iteratively update the virtual point source emission rate vector q and the regularization parameter λ until the matching degree reaches the standard or the number of iterations reaches the upper limit. The calculation formula is as follows: Where q is the virtual point source emission rate vector, q k Let q be the virtual point source emission rate vector for the k-th iteration, with initial values ​​generated by the area-to-point source conversion module through the law of mass conservation; k+1 The emission rate vector is used in the (k+1)th iteration to gradually approximate the optimal solution; α is the learning rate, which controls the iteration step size; ▽ q Loss is the gradient of the error function with respect to the emission rate vector q, representing the direction and magnitude of the influence of changes in q on the error; λ is the regularization parameter. k λ is the regularization parameter for the k-th iteration, and its initial value is determined through cross-validation of historical monitoring data from the multi-source data acquisition and fusion module; k+1 The regularization parameter for the (k+1)th iteration; β is the learning rate, and the parameter update step size is less than α; ▽ λ The gradient of the Loss error function with respect to the regularization parameter λ represents the effect of changes in λ on the error. The iterated q and λ are fed back to the inverse operation inversion unit of the pollution source tracing module to recalculate the pollution contribution rate of the virtual point source to the receiver point until the matching degree between the source tracing results and the measured data meets the 500m×500m scale requirement.

8. A method for monitoring VOCs pollution in an industrial park, applied to the VOCs pollution monitoring system for an industrial park as described in any one of claims 1-7, characterized in that: Includes the following steps: S1. The system first collects VOCs online monitoring data, meteorological data, pollution source emission parameters and mobile observation trajectory data in the park in real time through the multi-source data acquisition and fusion module. Simultaneously, the data is dynamically cleaned, spatiotemporally calibrated and feature extracted to form a standardized dataset. S2. Using the area source to point source conversion module, the area source pollution area in the industrial park is discretized into virtual point sources. The location of the virtual point source is determined by combining the geometric center coordinates of the sub-region, the prevailing wind direction vector, and the terrain undulation correction coefficient. The emission equivalence relationship between area source and point source is established through the law of mass conservation. S3. Construct a small-scale CALPUFF model based on a 500m×500m grid precision, integrate the coordinates, emission rate, pollutant composition of virtual point sources and wind field, temperature and humidity field, and pressure field data generated by WRF, simulate the diffusion and transport process of VOCs in a small-scale space, and output hourly concentration distribution data. S4. The pollution source tracing and inversion module integrates the concentration field output by the CALPUFF model with the VOCs concentration from fixed-point monitoring and the characteristic component spectrum data from mobile monitoring. It solves the reverse trajectory of pollution transmission through a regularized inversion algorithm, calculates the pollution contribution rate of each virtual point source to the receiver point, and locates the spatial distribution of pollution sources. S5. The source tracing result verification module obtains the VOCs component concentration distribution vector of the virtual point source and the dynamic spectrum data of the mobile monitoring. After preprocessing, the weighted cosine similarity algorithm is used to calculate the matching degree. When the matching degree is lower than the threshold, the emission rate of the virtual point source and the regularization parameter are iteratively corrected. S6. Finally, based on the calibrated source tracing results and the pollution source list, a GIS heat map of the pollution hotspot area, a matching degree time series curve, and a visualization verification report of the parameter iteration trajectory are generated to provide a basis for decision-making in the park's pollution control.

Images