A method and system for monitoring particulate matter and ozone sources based on distributed sensors

Abstract

This invention provides a method and system for monitoring particulate matter and ozone sources based on distributed sensors, relating to the field of pollution control technology. The invention deploys geographically aware sensor nodes at high density within the monitoring area to collect pollutant concentrations and meteorological parameters in real time, uploading the data to a cloud platform for preprocessing and dynamic calibration. A machine learning model is constructed based on the joint characteristics of pollutants and meteorological factors to uncover driving relationships and extract pollutant influencing factors. Furthermore, influencing factors are jointly modeled with regional emission source data, using classification or clustering methods to identify the coupling strength between pollutants and emission sources, and to determine the main source categories. Combining backward trajectory simulation, source fingerprint analysis, and multi-source regression decomposition, pollution path inversion and source contribution rate quantification are achieved. Finally, a geographic information visualization platform is constructed to display pollution heat maps, contribution change maps, and evolution path maps, providing support for multi-source pollution tracing and scientific management.

Description

Technical Field

[0001] This invention relates to the field of pollution control technology, and in particular to a method and system for monitoring the sources of particulate matter and ozone based on distributed sensors. Background Technology

[0002] Against the backdrop of the growing demand for urban air pollution control and precise management, particulate matter (PM2.5) has become increasingly important. 2.5 PM 10 Pollution problems involving ozone (O3) are particularly prominent. Traditional methods for identifying pollutant sources mainly rely on limited data from fixed monitoring stations combined with atmospheric chemical models for source tracing analysis. However, the limited number of fixed stations and their poor spatial representativeness make it difficult to achieve refined tracking and dynamic monitoring of pollution sources at the urban scale. Furthermore, ozone is a secondary pollutant, and its concentration is not only related to primary emission sources but also highly coupled with photochemical reaction processes, further increasing the difficulty of identifying its sources.

[0003] In recent years, with the development of the Internet of Things (IoT) and sensor technology, low-cost, high-density distributed environmental sensor networks have gradually become an effective means to improve monitoring resolution and accuracy. By deploying numerous sensing units in various urban areas, high spatiotemporal resolution pollutant concentration data can be collected in real time, providing a new approach for data-driven pollutant source tracing. Simultaneously, the application of artificial intelligence and data mining methods provides technical support for the collaborative analysis of multiple pollutants, and is expected to enable automated identification and prediction of pollution causes.

[0004] Despite the advantages of distributed sensor networks, such as flexible deployment and rapid response, they still face challenges including insufficient data accuracy, poor generalization of algorithm models, and low precision in pollutant source analysis. On the one hand, existing sensor data calibration mechanisms are imperfect and susceptible to environmental interference. On the other hand, current source tracing methods are mostly at a rudimentary stage based on concentration spatial distribution inversion, lacking systematic analytical models that integrate meteorological data, emission inventories, and chemical transformation mechanisms, resulting in limited ability to analyze the co-sources of particulate matter and ozone. Therefore, there is an urgent need to propose a comprehensive method integrating distributed sensing, data fusion, and intelligent reasoning to achieve high-precision joint identification and dynamic monitoring of particulate matter and ozone sources. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for monitoring particulate matter and ozone sources based on distributed sensors. It constructs an integrated particulate matter and ozone source monitoring process that includes "distributed sensing, cloud processing, intelligent identification, simulated source tracing, contribution quantification, and visualization," thereby achieving full-chain, refined sensing and analysis of complex pollution source structures.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for monitoring particulate matter and ozone sources based on distributed sensors, comprising:

[0008] Multiple distributed sensor nodes for particulate matter and ozone are deployed within the monitoring area according to preset density and geographical features. These distributed sensor nodes are used to collect multi-source environmental data, which is then uploaded to a cloud data processing platform in real time. The multi-source environmental data includes pollutant concentration parameters and meteorological parameters.

[0009] In the cloud data processing platform, data preprocessing and dynamic calibration are performed on each of the pollutant concentration parameters and the meteorological parameters to obtain pollutant concentration datasets and meteorological preprocessing datasets.

[0010] The pollutant concentration dataset is fused with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographic coordinates to construct a pollutant-meteorological factor joint feature matrix. Then, based on machine learning methods, the driving relationship between pollutant concentration changes and meteorological factors in the pollutant-meteorological factor joint feature matrix is ​​mined to form a set of pollutant influencing factors.

[0011] The pollutant impact factor set is jointly modeled with the preset regional emission source related data, and the coupling strength between pollutants and various emission sources is identified by classification algorithms or clustering models in order to preliminarily determine the main emission source type corresponding to each pollutant.

[0012] Based on the emission source categories and coupling strength, combined with meteorological parameters and the spatiotemporal distribution of pollutant concentrations, the backward trajectory simulation method is applied to invert the pollutant source paths. Combined with source fingerprint analysis and multi-source regression decomposition technology, the contribution ratio of various emission sources to the target pollutants under different spatiotemporal conditions is quantified to form a source and formation structure map of particulate matter and ozone.

[0013] Based on the pollution source structure map, a visualization platform based on geographic information system is constructed to dynamically display the pollutant source heat map, the temporal contribution change map, and the evolution path map.

[0014] A particulate matter and ozone source monitoring system based on distributed sensors, comprising:

[0015] The distributed sensing and data acquisition module is used to deploy multiple distributed sensor nodes for particulate matter and ozone within the monitoring area according to preset density and geographical features, and to collect multi-source environmental data using the distributed sensor nodes, and upload the multi-source environmental data to the cloud data processing platform in real time; the multi-source environmental data includes pollutant concentration parameters and meteorological parameters;

[0016] The cloud-based data preprocessing and dynamic calibration module is used to perform data preprocessing and dynamic calibration on each of the pollutant concentration parameters and the meteorological parameters in the cloud-based data processing platform to obtain pollutant concentration datasets and meteorological preprocessing datasets.

[0017] The emission source coupling identification and type determination module is used to combine the pollutant concentration dataset with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographic coordinates to construct a pollutant-meteorological factor joint feature matrix. Based on machine learning methods, the module mines the driving relationship between pollutant concentration changes and meteorological factors in the pollutant-meteorological factor joint feature matrix to form a set of pollutant impact factors.

[0018] The pollution source tracing simulation and contribution quantification module is used to jointly model the pollutant impact factor set with the preset regional emission source related data, and use classification algorithms or clustering models to identify the coupling strength between pollutants and various emission sources, so as to preliminarily determine the main emission source type corresponding to each pollutant.

[0019] The source map visualization module is used to invert the pollutant source path based on the emission source category and the coupling strength, combined with meteorological parameters and the spatiotemporal distribution of pollutant concentration, by applying the backward trajectory simulation method, and by combining source fingerprint analysis and multi-source regression decomposition technology to quantify the contribution ratio of various emission sources to the target pollutant under different spatiotemporal conditions, thus forming a source generative structure map of particulate matter and ozone.

[0020] Based on the pollution source structure map, a visualization platform based on geographic information system is constructed to dynamically display the pollutant source heat map, the temporal contribution change map, and the evolution path map.

[0021] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0022] This invention achieves high-resolution real-time acquisition of pollutants and meteorological parameters within the monitoring area by deploying high-density, geographically-aware distributed sensors, significantly improving the spatial representativeness and timeliness of the data. The cloud platform preprocesses and dynamically calibrates the raw data, and utilizes machine learning methods to construct a driving relationship model between pollutants and meteorological factors, thereby obtaining pollutant influencing factors with greater physical mechanism explanatory power. This invention jointly models pollutant influencing factors and emission source data, employing classification / clustering algorithms to identify coupling strength, transforming pollution source type determination from traditional empirical methods to data-driven methods, resulting in higher accuracy and scalability. Through the synergistic application of backward trajectory simulation, source fingerprint analysis, and multi-source regression decomposition, this invention not only achieves pollution source tracing but also quantitatively characterizes the contribution rate of various emission sources under different temporal and spatial conditions, enhancing the scientific rigor and decision-making reference value of pollution source tracing. This invention also constructs a geographic information visualization platform to dynamically display pollution heat maps, temporal trends, and evolution paths, effectively supporting the optimization of pollution control strategies and the formulation of source control recommendations for multi-source and multi-temporal pollution control. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0026] 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.

[0027] The purpose of this invention is to provide a method and system for monitoring particulate matter and ozone sources based on distributed sensors. It constructs an integrated particulate matter and ozone source monitoring process that includes "distributed sensing, cloud processing, intelligent identification, simulated source tracing, contribution quantification, and visualization," enabling full-chain, refined sensing and analysis of complex pollution source structures.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for monitoring particulate matter and ozone sources based on distributed sensors, comprising:

[0030] Step 100: Deploy multiple distributed sensor nodes for particulate matter and ozone within the monitoring area according to preset density and geographical features, collect multi-source environmental data using the distributed sensor nodes, and upload the multi-source environmental data to the cloud data processing platform in real time; the multi-source environmental data includes pollutant concentration parameters and meteorological parameters;

[0031] Step 200: In the cloud data processing platform, perform data preprocessing and dynamic calibration on each pollutant concentration parameter and meteorological parameter to obtain pollutant concentration dataset and meteorological preprocessing dataset;

[0032] Step 300: The pollutant concentration dataset is fused with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographic coordinates to construct a joint feature matrix of pollutants and meteorological factors. Based on machine learning methods, the driving relationship between pollutant concentration changes and meteorological factors in the joint feature matrix of pollutants and meteorological factors is mined to form a set of pollutant influencing factors.

[0033] Step 400: Jointly model the pollutant impact factor set with the preset regional emission source related data, and use classification algorithms or clustering models to identify the coupling strength between pollutants and various emission sources, so as to preliminarily determine the main emission source type corresponding to each pollutant;

[0034] Step 500: Based on the emission source category and coupling strength, combined with meteorological parameters and the spatiotemporal distribution of pollutant concentration, the backward trajectory simulation method is used to invert the pollutant source path. Combined with source fingerprint analysis and multi-source regression decomposition technology, the contribution ratio of various emission sources to the target pollutant under different spatiotemporal conditions is quantified to form the source and formation structure map of particulate matter and ozone.

[0035] Step 600: Construct a visualization platform based on a geographic information system based on the pollution source structure map to dynamically display the pollutant source heat map, time-series contribution change map, and evolution path map.

[0036] Preferably, the pollutant concentration parameter includes PM2.5. 2.5 Concentration, PM 10 Concentration and O3 concentration; the meteorological parameters include temperature data, humidity data, wind speed data, wind direction data, and air pressure data.

[0037] In this embodiment, the monitoring area can be a key area of ​​concern such as an urban built-up area, industrial park, or transportation hub. Based on geographical features, population density, and pollution sensitivity, multiple distributed environmental monitoring sensor nodes are deployed using a grid-like layout. The sensor nodes are installed on light poles, rooftops, or traffic signal poles within the monitoring area according to a preset spatial distribution density (e.g., 1 to 5 nodes per square kilometer) to achieve full coverage collection of environmental parameters within the target area.

[0038] Each sensor node integrates a particulate matter concentration sensor and an ozone gas sensor. Preferably, the particulate matter concentration sensor is a PM2.5 / PM10 detection module based on the laser scattering principle, and the ozone sensor is an O3 detection module based on the electrochemical principle. It also integrates a temperature and humidity sensor, an ultrasonic anemometer and wind direction sensor, and a barometric pressure sensor for synchronously collecting meteorological parameters. These sensor modules are controlled by a microprocessor to achieve multi-channel synchronous sampling, with a sampling time interval set from 1 minute to 10 minutes.

[0039] The collected raw data is uploaded in real time to a pre-defined cloud data processing platform via the wireless communication module (such as 4G, NB-IoT, or LoRa) embedded in the sensor node. The uploaded data structure includes the sensor node's unique identifier, sampling timestamp, geographic location information, pollutant concentration parameters (PM2.5, PM10, O3), and meteorological parameters (temperature, humidity, wind speed, wind direction, air pressure). To ensure data integrity and stability, the sensor node supports breakpoint resume and data caching mechanisms, and has remote maintenance and fault alarm functions.

[0040] Through the above deployment methods and data upload mechanisms, continuous, stable, and real-time collection of particulate matter and ozone concentrations and related meteorological factors within the target monitoring area can be achieved, providing reliable data support for subsequent big data processing and pollution source analysis.

[0041] Preferably, in the cloud data processing platform, data preprocessing and dynamic calibration are performed on each of the pollutant concentration parameters and the meteorological parameters to obtain a pollutant concentration dataset and a meteorological preprocessing dataset, including:

[0042] Pollutant concentration parameters and meteorological parameters from different distributed sensor nodes are time-aligned at a uniform time interval. Based on the geographical coordinates of each distributed sensor node, a spatial interpolation algorithm is used to spatially reconstruct the pollutant concentration parameters and meteorological parameters of the missing area to obtain unified spatiotemporal data.

[0043] The spatiotemporal unified data is subjected to a sliding window statistical analysis algorithm to identify and remove outliers, and missing data in the spatiotemporal unified data is filled in using a time-series interpolation method to obtain the processed pollutant concentration parameters and processed meteorological parameters.

[0044] The treated pollutant concentration parameters were compared with data from authoritative standard environmental monitoring stations in the region to construct an error correction model based on linear regression.

[0045] Based on the error correction model, the processed pollutant concentration parameters are dynamically calibrated to generate a pollutant concentration dataset.

[0046] The meteorological parameter data is standardized by units, and the main statistical features are extracted to form a structured meteorological preprocessing dataset.

[0047] Preferably, the main statistical characteristics include daily average, extreme values, volatility, and trend.

[0048] In this embodiment, the cloud-based data processing platform receives pollutant concentration parameters and meteorological parameters uploaded in real time from multiple distributed sensor nodes. First, it performs time alignment processing according to a set uniform time interval (e.g., 5 minutes) to ensure that data at the same point in time can be compared horizontally. For sensor nodes with delayed or lost data uploads, a time window interpolation method based on historical data mean and sample deviation is used for time completion processing. Subsequently, based on the geographical coordinates of each distributed sensor node, inverse distance weighted (IDW) or Kriging spatial interpolation algorithms are used to spatially reconstruct pollutant concentration parameters and meteorological parameters in areas with missing data or sparse distribution, generating a complete environmental monitoring data grid with uniform spatiotemporal resolution.

[0049] Next, the platform uses a sliding window statistical analysis algorithm to identify and remove outliers from the aforementioned unified spatiotemporal dataset. Outlier identification methods include a combined moving mean and standard deviation method: the moving mean and standard deviation are calculated in units of a set window (e.g., 1 hour), and values ​​deviating from the moving mean by more than three times the standard deviation are considered outliers. For missing data points in the remaining data, time series interpolation methods such as linear interpolation or Lagrange interpolation are used to complete the data. After data cleaning, the platform pairs and compares the processed pollutant concentration parameters with measured data from authoritative standard environmental monitoring stations in the region, constructing an error correction model based on linear regression, in the form: Cadj = α·Craw + β; where Craw is the processed sensor data, Cadj is the corrected data, and α and β are regression coefficients obtained by least squares fitting. The model can be updated periodically according to time or nodes to achieve dynamic calibration of pollutant data, ultimately outputting a pollutant concentration dataset. At the same time, the collected meteorological parameter data are standardized in terms of units (e.g., temperature is standardized to degrees Celsius and wind speed to meters per second), and key statistical features are extracted, including: daily average (representing the basic level of the meteorological factor), extreme values ​​(used to identify extreme weather events), volatility (reflecting stability through standard deviation), and trend (obtaining the trend slope based on linear trend fitting), to construct a structured meteorological preprocessing dataset.

[0050] Preferably, the pollutant concentration dataset is fused with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographic coordinates to construct a pollutant-meteorological factor joint feature matrix. Then, based on machine learning methods, the driving relationship between pollutant concentration changes and meteorological factors in the pollutant-meteorological factor joint feature matrix is ​​mined to form a set of pollutant influencing factors, including:

[0051] Based on a unified timestamp and the geographic coordinates of the distributed sensor nodes, the pollutant concentration dataset is spatiotemporally aligned and matched with the meteorological preprocessing dataset to construct a pollutant-meteorological factor original fusion dataset containing the pollutant concentration value and corresponding meteorological variables at each observation time.

[0052] The original fusion dataset of pollutants and meteorological factors is subjected to feature engineering processing, and the structure is as follows: The pollutant-meteorological factor joint feature matrix is ​​obtained, where n is the number of samples and m is the total number of feature dimensions of pollutants and meteorological variables; the feature engineering process includes standardization, normalization, principal component extraction, time difference feature construction and variable encoding operations;

[0053] Using pollutant concentration as the target variable and each dimension of the pollutant-meteorological factor joint feature matrix as the input feature, the random forest regression method is used to evaluate the influence of each meteorological factor on the change of pollutant concentration, and the feature importance distribution is obtained.

[0054] Based on the importance ranking of each feature variable in the feature importance distribution, meteorological factors with significant influence and their weights are selected to form the pollutant influence factor set for the corresponding pollutant sample.

[0055] In this embodiment, after receiving the dynamically calibrated pollutant concentration dataset and the meteorological preprocessing dataset, the cloud-based data processing platform first fuses and matches the two types of data according to a unified timestamp and the geographical coordinates of each distributed sensor node, following a two-dimensional "time-space" mapping relationship. Specifically, at each observation time point t... i At each sensor node location (xj, yj), the corresponding pollutant concentration value (e.g., PM2.5) is extracted. 2.5 The original pollutant-meteorological factor fusion sample is constructed using the pollutant (O3) and meteorological parameters (such as temperature, humidity, wind speed, wind direction, air pressure, etc.) recorded at the node. All samples are arranged in order of observation time, and the final pollutant-meteorological factor original fusion dataset is generated, which has a two-dimensional matrix structure [X]. n×m [The dataset is defined as follows, where n represents the number of samples and m represents the feature dimension (including combinations of pollutant types and meteorological parameters).] To improve modeling performance, feature engineering was applied to the fused dataset. First, the variables were normalized and standardized to eliminate the influence of dimensions. Then, principal component analysis (PCA) was used to extract principal component variables of potential meteorological factors to reduce dimensionality and improve efficiency. Finally, time differences were used to construct difference features representing the trend of pollutant changes (such as the first-order difference ΔC). t =C t -C t-1 For categorical variables such as wind direction, one-hot encoding or cosine-sine transformation is used to preserve the periodicity of direction. The processed feature vectors form the final pollutant-meteorological factor joint feature matrix, which is used for subsequent machine learning modeling. In the modeling stage, pollutant concentration values ​​are used as the target variable, and each meteorological variable and its derived features in the joint feature matrix are used as input features. A random forest regression model is used to train the feature influence model. This model can output the ranking of the importance of each feature variable in explaining changes in pollutant concentration. For example, for PM2.5... 2.5Important meteorological factors obtained after concentration modeling may include wind speed, temperature gradient, and humidity change rate. The platform sets a threshold based on feature importance scores (such as the cumulative weight of the top 10%) to screen out meteorological factors with significant impact, and uses their corresponding weight values ​​to form the pollutant impact factor set for the current pollutant sample, which is used for subsequent coupled modeling and analysis with emission source data.

[0056] Preferably, the set of pollutant impact factors is jointly modeled with preset regional emission source-related data, and classification algorithms or clustering models are used to identify the coupling strength between pollutants and various emission sources, so as to preliminarily determine the main emission source type corresponding to each pollutant, including:

[0057] The regional emission source data is standardized according to geographical units and mapped at the regional level to the corresponding geographical locations in the pollutant impact factor set to construct a regional emission source feature vector set; the regional emission source data includes traffic flow, industrial emission inventory and building energy consumption data.

[0058] The factor vectors of each pollutant in the pollutant impact factor set are concatenated and fused with the emission source feature vectors of the region to construct a pollutant-emission source joint feature space, which is used to represent the coupling situation between the spatiotemporal behavior of pollutants and regional emission activities.

[0059] Modeling is performed based on the joint feature space of the pollutants and emission sources to obtain the modeling results;

[0060] Based on the modeling results, the coupling strength index between pollutants and various emission sources is calculated, and the main emission source types of the corresponding pollutants in the current area are initially identified according to the preset ranking rules. The coupling strength index includes similarity coefficient, classification probability weight and cluster center distance.

[0061] In this embodiment, to identify the coupling relationship between pollutants and regional emission sources, the pre-defined regional emission source-related data is first standardized according to the administrative or grid geographical units of the monitoring area. The emission source-related data preferably includes traffic flow (e.g., road vehicle density), industrial emission inventories (e.g., VOCs or NOx emissions from key industries), and building energy consumption data (e.g., energy consumption per unit area), and their dimensions are unified through normalization and unit conversion. Then, based on the geographical locations of sensor nodes recorded in the pollutant impact factor set, they are mapped to the corresponding geographical units and matched with the emission source standard data of that unit, thereby constructing a regional emission source feature vector for each pollutant sample, forming a regional emission source feature vector set covering the target area.

[0062] When establishing the joint feature space, the influence factor vector corresponding to each pollutant sample is concatenated with the emission source feature vector of its region to form a unified pollutant-emission source joint feature vector. Using this joint feature set as input samples, a machine learning model is employed for joint modeling. Specifically, supervised classification algorithms (such as support vector machines and gradient boosting trees) can be used to predict pollutant source attribution, while unsupervised clustering algorithms (such as K-means and hierarchical clustering) can be used to analyze the similarity aggregation trend between pollutants and emission sources in the feature space. After model training, the platform calculates the coupling strength index between pollutants and various emission sources based on the probability of the sample belonging to a category, the distance from the sample to the cluster center, or vector similarity (such as cosine similarity). Finally, based on these coupling strength indices, the pollutants are ranked, and the possible main emission source types for each pollutant in the current region are selected, achieving preliminary source type identification and attribution.

[0063] Preferably, modeling is performed based on the joint feature space of the pollutant-emission source to obtain the modeling results, including:

[0064] Depending on whether the obtained samples have emission source labeling information, the following two methods are used for model training and output generation:

[0065] Given the known emission source categories of some pollutant samples, a pollutant-emission source classification model is established using the support vector machine algorithm, and the predicted emission source category and classification probability of each sample are output.

[0066] In the absence of labeled data, the K-means clustering model is used to perform joint feature clustering analysis, outputting the cluster number and distance to the cluster center for each pollutant sample.

[0067] In this embodiment, to achieve coupled identification and modeling between pollutants and emission sources, supervised classification and unsupervised clustering models are used for training and analysis, depending on whether known emission source label information for some pollutant samples is available. In the labeled case, Support Vector Machine (SVM) is selected as the classification model. The constructed pollutant-emission source joint feature vector is used as the input sample, and the emission source category corresponding to the sample is used as the supervision label to train and obtain the classification decision boundary function. After the model training is completed, inputting a new pollutant sample joint feature vector outputs its predicted emission source category and the corresponding classification probability distribution, which is used to measure the model's confidence in the sample's classification.

[0068] In the unlabeled case, K-means clustering is used to perform unsupervised clustering analysis on samples in the joint feature space of pollutants and emission sources. The specific steps include: first, setting the number of clusters K (which can be determined based on the silhouette coefficient or elbow rule); then, randomly initializing K cluster centers and iteratively performing sample assignment and center update processes until the cluster centers stabilize. Finally, the cluster number to which each pollutant sample belongs and its Euclidean distance to its cluster center are output. This is used to measure the similarity between the sample's belonging features and the center features in the joint feature space, thereby assisting in subsequent coupling strength quantification and source category attribution.

[0069] Preferably, based on the emission source category and the coupling strength, combined with meteorological parameters and the spatiotemporal distribution of pollutant concentrations, a backward trajectory simulation method is applied to invert the pollutant source path. Furthermore, source fingerprint analysis and multi-source regression decomposition techniques are combined to quantify the contribution ratio of various emission sources to the target pollutant under different spatiotemporal conditions, forming a source and formation structure map of particulate matter and ozone, including:

[0070] Based on the emission source category and the coupling strength, pollutant concentration data, meteorological parameters, and geographical location data of the distributed sensor nodes within the monitoring area are integrated to construct a multi-source input set for source tracing simulation.

[0071] Based on the spatiotemporal distribution characteristics and meteorological parameters of the pollutant concentration data, the backward trajectory simulation method is used to invert the transport path and potential source areas of pollutants, forming a set of pollutant flow trajectories.

[0072] The spatial overlap between the pollutant flow trajectory set and the regions of typical emission sources is calculated to obtain the corresponding trajectory coverage probability matrix, which is used to characterize the contribution probability of each typical emission source region in the pollution transport path; the calculation formula for the parameters in the trajectory coverage probability matrix is ​​as follows: Among them, P i,j Let N be the probability that the inversion trajectory of distributed sensor node i covers emission source region j. i,j T represents the number of trajectories passing through emission source region j among all inversion trajectories of node i. i The total number of trajectories for node i is the total number of backward trajectories simulated within the selected time window.

[0073] Source composition spectral information of typical emission sources within the monitoring area is collected, and chemical or physical fingerprint features of each type of typical emission source are extracted to form a source fingerprint feature set.

[0074] The source fingerprint feature set is matched item by item with the observable component features in the pollutant concentration data to identify the potential source structure and generate a potential matching relationship matrix between pollutant components and source types; the calculation formula for the parameters of the potential matching relationship matrix is ​​as follows: Among them, M k,l The potential match score between pollutant component l and source type l, where n is the number of monitoring points, and C is the value of C. k,s Let F be the measured concentration of pollutant component k in the s-th sample. l,s Let be the source fingerprint feature concentration value of emission source type l in the s-th sample, and sim(·,·) be the cosine similarity function;

[0075] Under the constraints of the pollutant flow trajectory set and trajectory coverage probability matrix, a multiple linear regression model is used, with the source fingerprint feature set and potential matching relationship matrix as input, to quantitatively calculate the concentration contribution rate of various typical emission sources to particulate matter and ozone under different spatiotemporal conditions, and output the source contribution rate distribution matrix; the calculation formula is as follows: Among them, Y i,t β represents the particulate matter / ozone concentration value measured by distributed sensor node i at time t; m represents the number of emission source types; β j S represents the regression coefficient for emission source type j, i.e., the concentration contribution per unit source strength; j,i,t ∈ represents the source intensity index of emission source type j at node i and time t; ∈ represents the model residual term, characterizing unmodeled factors or background concentration.

[0076] The source contribution rate distribution matrix of different time periods and regions is visualized and integrated to generate a source causal structure map of particulate matter and ozone in multiple sources and multiple spatiotemporal dimensions, which can be used to support the optimization of pollution control strategies and the formulation of source control recommendations.

[0077] In this embodiment, to achieve the inversion of pollutant source paths and the construction of a pollution causal structure map, firstly, based on the identified emission source categories and their corresponding coupling strength information, the pollutant concentration time-series data, real-time meteorological parameters (such as wind speed, wind direction, temperature, etc.), and the geographical location information of the nodes collected by each distributed sensor node within the monitoring area are integrated to construct a multi-source input data set for backward trajectory simulation. Subsequently, a backward trajectory simulation method based on the Lagrange particle diffusion model is adopted, combined with meteorological element driving, to invert the pollutant transport path of each node at a specific time point and form a pollutant flow trajectory set to characterize the potential source area.

[0078] Next, spatial overlap analysis was performed on the aforementioned pollutant flow trajectory set and the spatial distribution of known typical emission sources within the region. The coverage frequency of pollution trajectories in each emission source region was statistically analyzed, and a trajectory coverage probability expression for the emission source region was constructed. Simultaneously, source component spectral data of various typical emission sources within the region were collected, including information on major chemical components, physical characteristics, or gaseous markers, and source fingerprint feature sets were extracted accordingly. By matching each observable component in the pollutant concentration observation data item by item, combined with indicators such as cosine similarity, possible pollution source-component matching relationships were identified, generating a potential matching relationship matrix between pollution components and source types. This matrix characterizes the contribution tendency of various emission sources to specific pollutant components.

[0079] Finally, supported by trajectory simulation inversion results, source fingerprint feature sets, and matching relationship matrices, a multiple linear regression model is constructed. Using source fingerprint features as predictors, and under trajectory constraints and source strength data support, the spatiotemporal contribution ratios of various typical emission sources to pollutant concentrations are quantitatively calculated, outputting a source contribution rate distribution matrix. To support the formulation of pollution prevention and control strategies, the source contribution rate results from multiple time periods and regions are integrated and visualized to form a source causal structure map of particulate matter and ozone under multi-source driving. The source strength index can be dynamically generated by combining regional emission inventories, unit energy consumption statistics, and real-time traffic flow information, thereby improving the accuracy and context adaptability of contribution rate estimation.

[0080] Preferably, the source strength indicators include emission inventory values, energy consumption, and traffic flow.

[0081] Furthermore, in the post-processing stage of this embodiment, to achieve a graphical representation of the causes of pollutants, a unified data fusion framework is first constructed based on the source contribution rate distribution results of multiple distributed sensor nodes at different time periods and spatial locations. This framework standardizes the source contribution rate data at different times and in different regions, and groups and archives them according to pollutant type, time dimension (e.g., hour, day, month), and spatial resolution (e.g., street, grid). Subsequently, a graphics rendering engine is used to generate structured graphical data, and multi-dimensional graphical results, including pollutant source heatmaps, contribution time-series change maps, and pollution evolution path maps, are output according to user needs, realizing multi-angle visual analysis of pollution causes.

[0082] To improve the stability and fitting accuracy of the source contribution rate estimation model, a cross-validation mechanism was introduced during model building. Sample data was analyzed by time or geographical region, and the regression model was repeatedly trained and validated. The optimal combination of model parameters was selected based on evaluation metrics such as the sum of squared residuals, coefficient of determination, and mean absolute error. Furthermore, to address the potential multicollinearity problem among multiple input variables, principal component analysis was used to reduce the dimensionality of fingerprint feature variables. A penalty factor was also set during regression training to prevent overfitting, ensuring a balance between explanatory power and generalization ability in the source contribution rate results.

[0083] The visualization platform is built on a WebGIS architecture. The front end uses JavaScript and a map engine (such as Leaflet or Mapbox) to dynamically display pollutant source maps, while the back end receives source contribution rate results and metadata through a high-performance data interface. The platform provides interactive heat map rendering, time-series sliding analysis windows, and pollution path playback functions. It supports users to filter and switch layers by pollutant type, date range, and source type, further meeting the needs of management departments and researchers to understand the evolution process of pollution sources and identify intervention paths.

[0084] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides a particulate matter and ozone source monitoring system based on distributed sensors, including:

[0085] The distributed sensing and data acquisition module is used to deploy multiple distributed sensor nodes for particulate matter and ozone within the monitoring area according to preset density and geographical features, and to collect multi-source environmental data using the distributed sensor nodes, and upload the multi-source environmental data to the cloud data processing platform in real time; the multi-source environmental data includes pollutant concentration parameters and meteorological parameters;

[0086] The cloud-based data preprocessing and dynamic calibration module is used to perform data preprocessing and dynamic calibration on each of the pollutant concentration parameters and the meteorological parameters in the cloud-based data processing platform to obtain pollutant concentration datasets and meteorological preprocessing datasets.

[0087] The joint feature construction and driving factor extraction module is used to fuse the pollutant concentration dataset with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographic coordinates to construct a pollutant-meteorological factor joint feature matrix, and to mine the driving relationship between pollutant concentration changes and meteorological factors in the pollutant-meteorological factor joint feature matrix based on machine learning methods to form a set of pollutant impact factors.

[0088] The emission source coupling identification and type determination module is used to jointly model the pollutant impact factor set with the preset regional emission source related data, and use classification algorithms or clustering models to identify the coupling strength between pollutants and various emission sources, so as to preliminarily determine the main emission source type corresponding to each pollutant.

[0089] The pollution source tracing simulation and contribution quantification module is used to invert the pollutant source path based on the emission source category and the coupling strength, combined with meteorological parameters and the spatiotemporal distribution of pollutant concentration, by applying the backward trajectory simulation method, and by combining source fingerprint analysis and multi-source regression decomposition technology to quantify the contribution ratio of various emission sources to the target pollutant under different spatiotemporal conditions, and form a source and formation structure map of particulate matter and ozone.

[0090] The source map visualization module is used to construct a visualization platform based on the geographic information system according to the pollution source structure map, so as to dynamically display the pollutant source heat map, the time-series contribution change map and the evolution path map.

[0091] The beneficial effects of this invention are as follows:

[0092] This invention proposes a method and system for monitoring particulate matter and ozone sources based on distributed sensors. It enables the acquisition and fusion of multi-source data on pollutant concentrations and meteorological parameters with high spatiotemporal resolution, breaking through the limitations of traditional monitoring methods in terms of coverage, data granularity, and real-time performance, effectively improving the accuracy and completeness of pollution monitoring. By constructing a joint feature matrix of pollutants and meteorological factors and introducing a machine learning model to explore the driving relationship between pollutant concentration changes and meteorological factors, this invention can accurately identify key influencing factors in pollution evolution, providing a data foundation and methodological support for pollution causal analysis. Combining regional emission source data, a joint feature space of pollutants and emission sources is constructed, and the coupling strength between pollutants and different types of emission sources is evaluated through classification and clustering algorithms, achieving a preliminary determination of the main source types of different pollutants and improving the pertinence and explanatory power of source apportionment. Based on this, trajectory simulation, source fingerprint analysis, and regression decomposition techniques are integrated to further quantitatively analyze the source contributions of particulate matter and ozone at different spatiotemporal scales. The pollutant source structure and evolution characteristics are visually displayed through a GIS visualization platform, providing strong technical support for precise pollution control and scientific decision-making. The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0093] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A distributed sensor-based particulate matter and ozone source monitoring method, characterized by, The method comprises the following steps: Deploying a plurality of particulate matter and ozone distributed sensor nodes according to a preset density and geographical features in a monitoring area, collecting multi-source environmental data by using the distributed sensor nodes, and uploading the multi-source environmental data to a cloud data processing platform in real time; the multi-source environmental data includes pollutant concentration parameters and meteorological parameters; In the cloud data processing platform, data preprocessing and dynamic calibration are performed on each of the pollutant concentration parameters and the meteorological parameters to obtain a pollutant concentration dataset and a meteorological preprocessing dataset; Fusing the pollutant concentration dataset and the synchronously recorded meteorological preprocessing dataset according to time stamps and geographical coordinates to construct a pollutant-meteorological factor joint feature matrix, and based on a machine learning method, mining the driving relationship between the pollutant concentration change and the meteorological factor in the pollutant-meteorological factor joint feature matrix to form a pollutant influence factor set; Joint modeling of the pollutant influence factor set and the preset regional emission source related data, using a classification algorithm or a clustering model to identify the coupling strength between pollutants and various types of emission sources to preliminarily determine the corresponding emission source type of each pollutant; Based on the emission source type and the coupling strength, combining the meteorological parameters and the spatial and temporal distribution of the pollutant concentration, applying a backward trajectory simulation method to inverse the pollutant source path, and combining source fingerprint analysis and multi-source regression decomposition technology to quantify the contribution proportion of each type of emission source to the target pollutant under different spatiotemporal conditions to form a source cause structure map of particulate matter and ozone; According to the source cause structure map, a geographic information system-based visualization platform is constructed to dynamically display the pollutant source heat map, the time series contribution change map and the evolution path map.

2. The distributed sensor-based particulate matter and ozone source monitoring method of claim 1, wherein, The pollutant concentration parameters include PM 2.5 concentration, PM 10 concentration, and O3 concentration; and the meteorological parameters include temperature data, humidity data, wind speed data, wind direction data, and barometric pressure data.

3. The distributed sensor-based particulate matter and ozone source monitoring method of claim 1, wherein, In the cloud data processing platform, data preprocessing and dynamic calibration are performed on each of the pollutant concentration parameters and the meteorological parameters to obtain a pollutant concentration dataset and a meteorological preprocessing dataset, including: Time-aligning the pollutant concentration parameters and the meteorological parameters from different distributed sensor nodes according to a unified time interval, and based on the geographical coordinates of each distributed sensor node, reconstructing the pollutant concentration parameters and the meteorological parameters in the missing area through spatial interpolation algorithm to obtain spatiotemporally unified data; Applying a sliding window statistical analysis algorithm to the spatiotemporally unified data to identify and eliminate outliers, and using a time series interpolation method to complete the missing data in the spatiotemporally unified data to obtain processed pollutant concentration parameters and processed meteorological parameters; Comparing the processed pollutant concentration parameters with the data of the standard environmental monitoring site in the region with authority to construct an error correction model based on linear regression; Based on the error correction model, dynamically calibrating the processed pollutant concentration parameters to generate a pollutant concentration dataset; Unit standardizing the meteorological parameter data and extracting statistical features to form a structured meteorological preprocessing dataset.

4. The distributed sensor-based particulate matter and ozone source monitoring method of claim 3, wherein, The statistical features include daily average, extreme value, volatility and change trend.

5. The distributed sensor-based particulate matter and ozone source monitoring method of claim 1, wherein, The pollutant concentration dataset is fused with the synchronously recorded meteorological preprocessing dataset according to timestamps and geographical coordinates to construct a pollutant-meteorological factor combined feature matrix, and a driving relationship between pollutant concentration changes and meteorological factors in the pollutant-meteorological factor combined feature matrix is mined based on a machine learning method to form a pollutant influencing factor set, including: Based on the unified timestamps and geographical coordinates of the distributed sensor nodes, the pollutant concentration dataset is spatiotemporally aligned and matched with the meteorological preprocessing dataset to construct a pollutant-meteorological factor original fusion dataset containing pollutant concentration values and corresponding meteorological variables at each observation time; The feature engineering processing is performed on the pollution-weather factor original fusion data set to construct a pollution-weather factor combined feature matrix in the form of wherein n is the sample quantity, and m is the total dimension of pollution and weather variable features; the feature engineering processing includes standardization, normalization, principal component extraction, time difference feature construction and variable coding operation; With the pollutant concentration as a target variable and each dimension of the pollutant-meteorological factor combined feature matrix as an input feature, a random forest regression method is used to evaluate the influence degree of each meteorological factor on the pollutant concentration change to obtain a feature importance distribution; According to the importance ranking results of each feature variable in the feature importance distribution, meteorological factors with significant influence degree and weights are screened out to constitute the pollutant influencing factor set corresponding to the pollutant sample.

6. The distributed sensor-based particulate matter and ozone source monitoring method of claim 5, wherein, The pollutant influencing factor set is jointly modeled with preset regional emission source related data, and a classification algorithm or a clustering model is used to identify the coupling strength between pollutants and various emission sources to preliminarily determine the emission source type corresponding to each pollutant, including: The regional emission source related data is standardized according to geographical units and regionally mapped to the corresponding geographical positions in the pollutant influencing factor set to construct a regional emission source feature vector set; the regional emission source related data includes traffic flow, industrial source emission inventory and building energy consumption data; The factor vector of each pollutant in the pollutant influencing factor set is spliced and fused with the emission source feature vector of the region to construct a pollutant-emission source combined feature space for representing the coupling situation of pollutant spatiotemporal behavior and regional emission activity; Modeling is performed based on the pollutant-emission source combined feature space to obtain a modeling result; Based on the modeling result, a coupling strength index between pollutants and various emission sources is calculated, and the emission source type of the corresponding pollutant in the current region is preliminarily identified according to a preset ranking rule; the coupling strength index includes a similarity coefficient, a classification probability weight and a clustering center distance.

7. The distributed sensor-based particulate matter and ozone source monitoring method of claim 6, wherein, Modeling is performed based on the pollutant-emission source combined feature space to obtain a modeling result, including: According to whether the obtained sample has emission source label information, the following two types of methods are used for model training and output generation respectively: In the case that part of the pollutant samples are known to correspond to emission source types, a support vector machine algorithm is used to establish a pollutant-emission source classification model to output the predicted emission source type and classification probability corresponding to each sample; In the case of no labeled data, a K-means clustering model is used for joint feature clustering analysis to output the cluster number corresponding to each pollutant sample and the distance to the cluster center.

8. The distributed sensor-based particulate matter and ozone source monitoring method of claim 1, wherein, Based on the emission source type and the coupling strength, the source contribution rate distribution matrix in different time periods and different regions is visualized and integrated to generate a source cause structure atlas of particulate matter and ozone in multiple sources and multiple spatio-temporal dimensions, which is used to support pollution control strategy optimization and source control suggestion making. The source intensity index includes emission inventory value, energy consumption, and traffic flow. It includes: The pollution flow trajectory set is calculated with the spatial overlap degree of the area of the typical emission source, to obtain a corresponding trajectory coverage probability matrix, which is used to represent the contribution possibility of each typical emission source area in the pollution transmission path; the calculation formula of the parameters in the trajectory coverage probability matrix is: ; wherein, is the coverage probability of the inversion trajectory of the distributed sensor node in the emission source area , is the number of trajectories passing through the emission source area in all inversion trajectories of the node , is the total number of trajectories of the node , that is, the total number of backward trajectories simulated in the selected time window. A distributed sensing and data acquisition module is used to deploy a plurality of distributed sensor nodes of particulate matter and ozone in the monitoring area according to the preset density and geographical features, and to collect multi-source environmental data using the distributed sensor nodes and upload the multi-source environmental data to a cloud data processing platform in real time. The multi-source environmental data includes pollutant concentration parameters and meteorological parameters. The source fingerprint feature set is matched item by item with the observable component features in the pollutant concentration data to identify the potential source structure and generate a potential matching relationship matrix between pollutant components and source types; the calculation formula for the parameters of the potential matching relationship matrix is ​​as follows: ;in, pollutant components With source type Potential match score, For the number of monitoring points, For the first Pollutant components in each sample The measured concentration, For the first Emission source types in each sample The source fingerprint feature concentration value, The cosine similarity function; Under the constraints of the pollutant flow trajectory set and trajectory coverage probability matrix, a multiple linear regression model is adopted to quantitatively calculate the concentration contribution rates of various typical emission sources to particulate matter and ozone under different space-time conditions by taking the source fingerprint feature set and the potential matching relationship matrix as inputs, and to output a source contribution rate distribution matrix; the calculation formula is: ; wherein, is a distributed sensor node The particulate matter / ozone concentration value measured at time ; is the number of emission source types; is the regression coefficient of the emission source type , that is, the concentration contribution under a unit source intensity; is the source intensity index of the emission source type at node , time ; is a model residual term, representing unmodeled factors or background concentrations; A cloud data preprocessing and dynamic calibration module is used to preprocess and dynamically calibrate each pollutant concentration parameter and meteorological parameter in the cloud data processing platform to obtain a pollutant concentration data set and a meteorological preprocessing data set.

9. The distributed sensor-based particulate matter and ozone source monitoring method of claim 8, wherein, A joint feature construction and driving factor extraction module is used to fuse the pollutant concentration data set and the meteorological preprocessing data set recorded synchronously according to time stamp and geographical coordinates to construct a pollutant-meteorological factor joint feature matrix, and to mine the driving relationship between pollutant concentration change and meteorological factors in the pollutant-meteorological factor joint feature matrix based on a machine learning method to form a pollutant influence factor set.

10. A distributed sensor-based particulate matter and ozone source monitoring system, characterized in that, An emission source coupling identification and type determination module is used to jointly model the pollutant influence factor set and the preset regional emission source related data, and to identify the coupling strength between pollutants and various emission sources using a classification algorithm or a clustering model to preliminarily determine the emission source type corresponding to each pollutant. A pollution source tracing simulation and contribution quantification module is used to combine meteorological parameters and spatio-temporal distribution of pollutant concentration based on the emission source type and the coupling strength, apply a backward trajectory simulation method to reverse the source path of pollutants, and combine source fingerprint analysis and multi-source regression decomposition technology to quantify the contribution proportion of various emission sources to target pollutants under different spatio-temporal conditions to form a source cause structure atlas of particulate matter and ozone. The source intensity index includes emission inventory value, energy consumption, and traffic flow. It includes: A distributed sensing and data acquisition module is used to deploy a plurality of distributed sensor nodes of particulate matter and ozone in the monitoring area according to the preset density and geographical features, and to collect multi-source environmental data using the distributed sensor nodes and upload the multi-source environmental data to a cloud data processing platform in real time. The multi-source environmental data includes pollutant concentration parameters and meteorological parameters. A cloud data preprocessing and dynamic calibration module is used to preprocess and dynamically calibrate each pollutant concentration parameter and meteorological parameter in the cloud data processing platform to obtain a pollutant concentration data set and a meteorological preprocessing data set. A joint feature construction and driving factor extraction module is used to fuse the pollutant concentration data set and the meteorological preprocessing data set recorded synchronously according to time stamp and geographical coordinates to construct a pollutant-meteorological factor joint feature matrix, and to mine the driving relationship between pollutant concentration change and meteorological factors in the pollutant-meteorological factor joint feature matrix based on a machine learning method to form a pollutant influence factor set. An emission source coupling identification and type determination module is used to jointly model the pollutant influence factor set and the preset regional emission source related data, and to identify the coupling strength between pollutants and various emission sources using a classification algorithm or a clustering model to preliminarily determine the emission source type corresponding to each pollutant. A pollution source tracing simulation and contribution quantification module is used to combine meteorological parameters and spatio-temporal distribution of pollutant concentration based on the emission source type and the coupling strength, apply a backward trajectory simulation method to reverse the source path of pollutants, and combine source fingerprint analysis and multi-source regression decomposition technology to quantify the contribution proportion of various emission sources to target pollutants under different spatio-temporal conditions to form a source cause structure atlas of particulate matter and ozone. The source map visualization display module is configured to construct a geographic information system-based visualization platform based on the source cause structure map to dynamically display a pollutant source heat map, a time-series contribution change map, and an evolution path map.

Images