Urban air quality multi-source monitoring data analysis method

Abstract

This invention discloses a method for analyzing multi-source monitoring data of urban air quality, relating to the field of environmental monitoring multi-source heterogeneous data fusion and digital processing technology; it addresses the problems of inconsistent physical benchmarks, distortion in temporal breakpoint repair, and delayed early warning in multi-source monitoring data. First, based on meteorological parameters, differentiated physical benchmark transformations are performed on gaseous pollutants and particulate matter to generate a normalized concentration matrix aligned with physical states. Then, a temporal topology map is constructed to map breakpoints, and gradient compensation vectors are generated using the spatiotemporal first derivative of the neighboring sensor sequences to reconstruct the continuous temporal concentration matrix. Next, the discrete deviation between standard stations and microstations is calculated, and spatial heterogeneity weighting is performed using functional zoning to generate a dynamic calibration matrix, mapping and outputting a global calibration concentration field. Finally, the step-approximation difference vector is calculated by comparing two sets of limit matrices, and extreme nodes of abrupt gradients are extracted to generate a multi-dimensional pollution risk tracing map, achieving precise location and dynamic early warning of hidden pollution sources.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring multi-source heterogeneous data fusion and digital processing technology, specifically a method for analyzing multi-source monitoring data of urban air quality. Background Technology

[0002] With the acceleration of urbanization, the construction of urban air quality monitoring networks has been continuously improved, and monitoring methods have expanded from single standard monitoring stations to a multi-source heterogeneous node system encompassing micro-sensor networks and mobile monitoring. These heterogeneous monitoring devices generate massive amounts of ambient air quality measurement sequences and meteorological environmental parameters every day. How to utilize computer data processing technology to perform deep cleaning, benchmark alignment, and feature extraction on these massive multi-dimensional monitoring data streams, and then to refine the analysis of pollution topological evolution in the complex urban microenvironment, has become a core issue of great concern in the field of digital data processing and environmental information fusion.

[0003] Existing multi-source data fusion and analysis methods suffer from significant systemic biases and algorithmic limitations at the underlying processing logic level. On the one hand, traditional algorithms directly perform isomorphic accumulation and mean calculation when receiving raw concentration scalars of gaseous pollutants and particulate matter from heterogeneous devices, ignoring the underlying physical conversion benchmark differences between different molecular phases in actual measurements and reference states. This results in an irreversible systemic benchmark bias in the fused underlying data matrix. On the other hand, when faced with missing time-series data due to device power outages or malfunctions, existing interpolation algorithms mainly employ direct node removal or static mean coverage. This severely disrupts the continuous first-order derivative evolution characteristics of the spatiotemporal sequence itself, leading to severe topological distortion and drift errors in the subsequently reconstructed concentration field. Furthermore, conventional risk assessment models only determine exceedances based on static, single constraint thresholds, resulting in extremely flat convergence logic. The warning results output by the model can easily mask hidden pollution abrupt extremes within local grids. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for analyzing multi-source monitoring data of urban air quality, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing multi-source monitoring data of urban air quality, comprising the following steps: S1, acquiring multi-source monitoring data streams and parsing pollutant identifiers and meteorological parameters; determining physical measurement benchmarks based on pollutant identifiers; converting the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to meteorological parameters; and mapping the original measured concentration vector of particulate matter to an actual state concentration vector; cascading the reference state concentration vector and the actual state concentration vector to generate a normalized concentration matrix; S2, parsing the timestamp sequence of the normalized concentration matrix to construct a time-series topology graph; extracting breakpoints in the time-series topology graph and mapping them to nodes to be compensated; extracting the target dense sensing sequence within the spatial neighborhood of the nodes to be compensated and performing spatiotemporal first derivative operations to generate a gradient compensation vector; and injecting the gradient compensation vector into... S3. The continuous time-series concentration matrix is ​​reconstructed from the void nodes to be compensated. S4. The continuous time-series concentration matrix is ​​decomposed into a first-precision standard station sequence and a second-precision micro-station sequence. The discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window is calculated. The ambient air functional zoning category is introduced to perform spatial heterogeneity weighting on the discrete deviation array to generate a dynamic calibration matrix. The dynamic calibration matrix is ​​used to perform feature mapping on the continuous time-series concentration matrix to reconstruct the output global calibration concentration field. S5. The transition stage limit matrix and the basic item limit matrix are introduced. The global calibration concentration field is compared with the two sets of limit matrices. The step-approximation difference vector of the spatial grid in the global calibration concentration field converges from the transition stage limit matrix to the basic item limit matrix. The abrupt gradient extreme value node of the step-approximation difference vector is extracted and mapped to generate a multi-dimensional pollution risk source map.

[0006] Further, the process of acquiring multi-source monitoring data streams and parsing pollutant identifiers and meteorological parameters, determining physical measurement benchmarks based on pollutant identifiers, converting the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to meteorological parameters, and mapping the original measured concentration vector of particulate matter into an actual state concentration vector is as follows: Parsing the multi-source monitoring data streams extracts the pollutant identifiers and environmental meteorological parameters encapsulated in the data packet header. The molecular phase attributes corresponding to the pollutant identifiers are identified. When the molecular phase attribute indicates gaseous pollutants, the atmospheric temperature and atmospheric pressure parameters within the environmental meteorological parameters are extracted. Preset standard reference temperature and pressure constants are introduced to establish a state transition equation. The original measured concentration vector of gaseous pollutants is substituted into the state transition equation to perform a physical benchmark alignment operation, outputting the reference state concentration vector. When the molecular phase attribute indicates particulate matter, the real-time monitoring temperature and real-time monitoring pressure parameters within the environmental meteorological parameters are locked to construct an actual state benchmark. The original measured concentration vector of particulate matter is projected in the same dimension according to the real-time monitoring temperature and real-time monitoring pressure parameters, outputting the actual state concentration vector.

[0007] Furthermore, the specific process of generating a normalized concentration matrix by cascading the reference state concentration vector and the actual state concentration vector is as follows: extract the sampling timestamp sequence and spatial geographic coordinate sequence carried by the reference state concentration vector and the actual state concentration vector; perform time granularity alignment operation on the sampling timestamp sequence and perform geographic grid discretization processing on the spatial geographic coordinate sequence to generate a spatiotemporal alignment feature index set; extract the reference state concentration vector and the actual state concentration vector, perform tensor concatenation operation according to the spatiotemporal alignment feature index set, and output a normalized concentration matrix that integrates the physical benchmark features of the multiphase state.

[0008] Furthermore, the specific process of constructing a temporal topology graph by parsing the timestamp sequence of the normalized concentration matrix and extracting the breakpoints in the temporal topology graph as nodes to be compensated is as follows: The timestamp sequence of the normalized concentration matrix is ​​divided according to a preset time granularity to generate a discrete time axis vector. A sliding verification time window is set along the discrete time axis vector, and the frequency of continuous valid samples within the sliding verification time window is counted. The frequency of continuous valid samples is compared with the preset statistical validity baseline condition. Time window intervals that do not meet the preset statistical validity baseline condition are extracted to construct a discrete time segment set. The discrete time segment set is projected onto the temporal network structure to generate a temporal topology graph. Breakpoint features that are interrupted in the temporal topology graph are located. Null value marking is performed on the breakpoint features to assign missing labels, and nodes with missing labels are output as nodes to be compensated.

[0009] Furthermore, the specific process of extracting the target dense sensing sequence within the spatial neighborhood of the void node to be compensated and performing spatiotemporal first derivative operations to generate a gradient compensation vector, and injecting the gradient compensation vector into the void node to be compensated to reconstruct the continuous temporal concentration matrix is ​​as follows: The latitude and longitude coordinate features bound to the void node to be compensated are read to establish a spatial buffer neighborhood. The target dense sensing sequence output by the micro-sensor nodes deployed within the spatial buffer neighborhood is captured. Partial derivative calculations are performed simultaneously on the target dense sensing sequence along the spatial distance evolution dimension and the temporal evolution dimension to extract the concentration change rate and physical diffusion direction features in the local space, generating a multi-source temporal feature tensor. A nonlinear feature vector synthesis operation is performed on the multi-source temporal feature tensor to output a gradient compensation vector. Based on the timestamp index sequence, the gradient compensation vector is written into the void node to be compensated to perform a missing state overwrite operation, reconstructing a continuous temporal concentration matrix without discontinuities.

[0010] Furthermore, the continuous time-series concentration matrix is ​​decomposed into a first-precision standard station sequence and a second-precision micro-station sequence. The specific process for calculating the discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window is as follows: The device attribute tags bound to the continuous time-series concentration matrix are parsed to perform a classification and decomposition operation. Data nodes with national control benchmark precision attributes are extracted to construct the first-precision standard station sequence, and data nodes with spatially dense deployment attributes are extracted to construct the second-precision micro-station sequence. The timestamp features carried by the first-precision standard station sequence and the second-precision micro-station sequence are aligned according to a preset synchronization time span to establish a time slice alignment vector. Discrete deviation calculation is performed on the benchmark concentration nodes associated with the first-precision standard station sequence and the drift concentration nodes associated with the second-precision micro-station sequence that are in the same spatial buffer neighborhood within the time slice alignment vector. The differential feature component sets output by the discrete deviation calculation are summarized to generate a discrete deviation array.

[0011] Furthermore, the process of introducing ambient air functional zoning categories to perform spatial heterogeneity weighting on the discrete deviation array to generate a dynamic calibration matrix, and then using the dynamic calibration matrix to perform feature mapping reconstruction on the continuous time-series concentration matrix to output the global calibration concentration field is as follows: The ambient air functional zoning category parameters associated with the monitoring nodes are retrieved to divide the spatial grid into Class I and Class II feature vectors. The discrete deviation array is mapped to the Class I and Class II feature vectors, and spatial heterogeneity weighting is performed to generate region-specific offset weight parameters. The corresponding time evolution parameters within the continuous time-series concentration matrix are extracted, and the region-specific offset weight parameters are tensor-extended along the time evolution parameters to construct the dynamic calibration matrix. The continuous time-series concentration matrix is ​​retrieved, and a tensor inner product operation is performed between the continuous time-series concentration matrix and the dynamic calibration matrix to perform feature mapping reconstruction, outputting the global calibration concentration field.

[0012] Furthermore, by introducing the transition phase limit matrix and the basic project limit matrix, and comparing the global calibration concentration field with the two sets of limit matrices, the specific process of calculating the step approximation difference vector of the spatial grid within the global calibration concentration field converging from the transition phase limit matrix to the basic project limit matrix is ​​as follows: The time buffer between the implementation of the transition phase limit matrix association and the regular limit application interval for the implementation of the basic project limit matrix association are read from the preset standard switching time node. The real-time dynamic concentration scalar carried by a single spatial grid is extracted by analyzing the global calibration concentration field. The initial environmental redundancy scalar is calculated by comparing the real-time dynamic concentration scalar with the first-level constraint threshold contained in the transition phase limit matrix. The ultimate bearing approximation scalar is calculated by comparing the real-time dynamic concentration scalar with the second-level constraint threshold contained in the basic project limit matrix. Spatiotemporal evolution dimensionality reduction analysis is performed on the initial environmental redundancy scalar and the ultimate bearing approximation scalar to calculate the convergence rate of the spatial grid evolving towards the regular limit application interval. The convergence rate is integrated to output the step approximation difference vector.

[0013] Furthermore, the specific process of extracting the extreme value nodes of the abrupt gradient from the step approximation difference vector to generate a multidimensional pollution risk source map is as follows: The step approximation difference vector is divided according to a preset step size to construct a local spatial evolution feature matrix. Second-order partial derivative operations are performed on the local spatial evolution feature matrix to obtain local curvature peak attributes. Data nodes whose spatial gradient changes exceed a preset variation threshold and deviate from the overall background concentration evolution trend of the region are selected as extreme value nodes of the abrupt gradient. Spatial density clustering is performed on the extreme value nodes of the abrupt gradient to generate a cluster of abrupt target activity coordinates. The cluster of abrupt target activity coordinates is then combined with the geographic information system coordinate system to perform a multidimensional spatial feature projection operation, reconstructing and outputting a multidimensional pollution risk source map.

[0014] The present invention has the following beneficial effects: (1) A method for analyzing multi-source monitoring data of urban air quality, which acquires multi-source monitoring data streams and analyzes pollutant identifiers and meteorological parameters, performs differentiated physical measurement benchmark conversion operations on gaseous pollutants and particulate matter based on pollutant identifiers, and generates a strictly aligned normalized concentration matrix, thus completely eliminating the systematic benchmark deviation of the physical measurement state of heterogeneous equipment from the underlying logic of data processing; at the same time, the time stamp sequence of the normalized concentration matrix is ​​analyzed to construct a time-series topology graph, and the extracted breakpoints are mapped as void nodes to be compensated. The spatiotemporal first derivative operation of the dense sensor sequence of the target in the spatial neighborhood is applied to generate a gradient compensation vector for accurate reconstruction, avoiding the temporal destruction caused by conventional static mean interpolation, and ensuring the topological integrity and high-frequency evolution fidelity of the continuous time-series concentration matrix output by reconstruction.

[0015] (2) A method for analyzing multi-source monitoring data of urban air quality, which decomposes the continuous time-series concentration matrix into a first-precision standard station sequence and a second-precision micro-station sequence, and introduces the environmental air functional zoning category to perform spatial heterogeneity weighting on the synchronous discrete deviation array of the two to generate a dynamic calibration matrix. The dynamic calibration matrix is ​​used for feature mapping reconstruction, which effectively smooths out the system drift error of densely deployed micro-station nodes and outputs a high-benchmark-precision global calibration concentration field. On this basis, two sets of progressive limit matrices are introduced to calculate the step-approximation difference vector of the global grid from the transition stage limit matrix to the basic item limit matrix, and extracts the extreme value node of the mutation gradient to generate a multi-dimensional pollution risk tracing map, which breaks the limitation of the single exceedance model judgment and realizes the dynamic evolution tracking and precise spatial positioning of hidden pollution sources in the urban micro-environment.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1This is a flowchart of a multi-source monitoring data analysis method for urban air quality according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the breakpoint reconstruction and gradient compensation effect of the continuous time-series concentration matrix of the present invention.

[0019] Figure 3 This is a schematic diagram of the dynamic calibration and feature mapping of the multi-source micro-sensor of the present invention.

[0020] Figure 4 This is a schematic diagram of the multidimensional pollution risk tracing map generated by extracting extreme nodes of the mutation gradient in this invention. Detailed Implementation

[0021] This application provides a method for analyzing multi-source monitoring data of urban air quality, which solves the problems of physical benchmark fragmentation, topological distortion caused by static repair of breakpoint data during multi-source monitoring data fusion, and the inability of risk assessment models based on a single threshold to accurately locate hidden and sudden pollution sources.

[0022] The overall approach of the scheme in this application embodiment is as follows: First, the pollutant identifiers and meteorological parameters of the acquired multi-source monitoring data stream are analyzed. Based on the pollutant identifiers, physical measurement benchmarks are determined, and the conversion mapping between reference state and actual state is performed according to categories. The normalized concentration matrix with unified physical state is then generated. Next, the time series of the normalized concentration matrix is ​​analyzed to construct a time series topology graph. The time series breakpoints are located and mapped as void nodes to be compensated. The first-order derivative of the spatiotemporal sequence of dense target sensing sequences in the spatial neighborhood is used to generate gradient compensation vectors to fill the void nodes, reconstructing a spatiotemporally continuous concentration matrix. Subsequently, the matrix is ​​split to extract the discrete deviation array of the first and second precision sequences. Combined with the environmental functional zoning attributes, spatial heterogeneity weighting is performed to obtain a dynamic calibration matrix. Feature mapping is performed on the continuous time series concentration matrix to output a global calibration concentration field. Finally, two sets of limit matrices, one for the transition stage and one for the basic items, are introduced to compare the global calibration concentration field. The step approximation difference vector of the spatial grid is calculated, and the extreme value nodes of the abrupt gradient are extracted and mapped to generate a multidimensional pollution risk tracing map.

[0023] Example 1; please refer to Figure 1This invention provides a technical solution: a method for analyzing multi-source monitoring data of urban air quality, comprising the following steps: S1, acquiring multi-source monitoring data streams and parsing pollutant identifiers and meteorological parameters; determining physical measurement benchmarks based on pollutant identifiers; converting the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to meteorological parameters; and mapping the original measured concentration vector of particulate matter to an actual state concentration vector; cascading the reference state concentration vector and the actual state concentration vector to generate a normalized concentration matrix; S2, parsing the timestamp sequence of the normalized concentration matrix to construct a time-series topology graph; extracting breakpoints in the time-series topology graph and mapping them to nodes to be compensated; extracting the target dense sensing sequence within the spatial neighborhood of the nodes to be compensated and performing spatiotemporal first derivative operations to generate a gradient compensation vector; injecting the gradient compensation vector into the nodes to be compensated. S3. The continuous time-series concentration matrix is ​​reconstructed from the hole node; S4. The continuous time-series concentration matrix is ​​decomposed into a first-precision standard station sequence and a second-precision micro-station sequence, and the discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window is calculated; the environmental air functional zoning category is introduced to perform spatial heterogeneity weighting on the discrete deviation array to generate a dynamic calibration matrix, and the dynamic calibration matrix is ​​used to perform feature mapping on the continuous time-series concentration matrix to reconstruct the output global calibration concentration field; S5. The transition stage limit matrix and the basic item limit matrix are introduced, and the global calibration concentration field is compared with the two sets of limit matrices. The step approximation difference vector of the spatial grid in the global calibration concentration field converges from the transition stage limit matrix to the basic item limit matrix is ​​calculated; the abrupt gradient extreme value node of the step approximation difference vector is extracted and mapped to generate a multi-dimensional pollution risk source map.

[0024] In this implementation scheme, step S1 functions to perform low-level alignment and unification of the accessed heterogeneous environmental monitoring data at the physical state level. The reference state concentration vector refers to the mathematical vector of gaseous pollutants calculated under standard atmospheric temperature of 298.15 K and standard atmospheric pressure of 1013.25 hPa; the actual state concentration vector refers to the particulate matter concentration vector retaining the actual temperature and pressure characteristics at the time of measurement; and the normalized concentration matrix is ​​a standardized two-dimensional data structure formed by stitching together various pollutant data after phase-state differentiation calculations according to a unified dimension. The technical role of this step is to strictly adhere to the state specifications for pollutants of different phases in the ambient air quality standards, thoroughly eliminating systematic deviations in physical measurement benchmarks from multi-source heterogeneous equipment at the source of data preprocessing, and preventing serious calculation errors caused by the mixing of dimensions and physical states in subsequent fusion calculations.

[0025] Step S2 involves reconstructing missing monitoring data by analyzing time-series features and utilizing spatial correlations. The time-series topology graph is a directed acyclic network structure built from the timestamps of monitoring data in chronological order, used by computers to quickly locate data interruptions. The missing nodes to be compensated are data nodes whose continuous effective sampling time did not reach the legal minimum due to power outages or equipment malfunctions. The spatiotemporal first derivative operation is a mathematical process of calculating the rate of change of high-frequency sensor data within the target spatial range along both the distance and time dimensions, used to obtain the physical diffusion direction and instantaneous concentration change rate of pollutants. The technical advantage of this step is that it abandons the traditional crude approach of directly removing incomplete data or using static means to fill in gaps. Instead, it utilizes the spatiotemporal correlation gradient of surrounding densely packed equipment to dynamically reconstruct the features of the missing grid, thereby preserving the transient abrupt changes in pollution distribution to the greatest extent possible while ensuring the statistical validity of environmental data.

[0026] Step S3 functions by using reliable data from high-precision stations to perform dynamic error calibration on a low-cost micro-sensor network. The first-precision standard station sequence represents a dataset of monitoring station data with high-precision calibration attributes at the national or provincial control level. The second-precision micro-station sequence represents a dataset of widely distributed but easily drift-prone IoT micro-sensors. Spatial heterogeneity weighting refers to the process where the computer, when allocating calibration weights, forcibly introduces the background concentration difference characteristics between Class I areas (such as nature reserves) and Class II areas (such as mixed residential and industrial areas) for nonlinear adjustment. The technical role of this step is to integrate the numerical stability of high-precision stations with the spatial high-resolution advantage of micro-stations at the electronic and digital levels. Through a heterogeneous calibration matrix, the systematic drift of low-cost sensors is smoothed out, reconstructing a global calibration concentration field that possesses both high reference accuracy and extremely high spatial resolution.

[0027] Step S4, based on a dual-track standard limit system, performs in-depth attribution and risk warning positioning of the calibrated concentration field. The transitional phase limit matrix and the basic project limit matrix represent the lenient implementation threshold set in the national standard and the final strict implementation threshold implemented nationwide, respectively. The step-approximation difference vector refers to the relative convergence rate matrix of the dynamic concentration values ​​within the spatial grid as they evolve from the current lenient standard to the future strict standard. The abrupt gradient extreme node refers to a specific coordinate point in the spatial grid where the concentration evolution curvature increases abnormally sharply and completely deviates from the surrounding environmental trend. The technical significance of this step lies in breaking away from the previous flat judgment logic that relied solely on whether a single absolute value exceeded the standard. By calculating the microscopic rate of concentration convergence towards the strict standard, it can keenly capture and accurately locate sudden pollution emission sources hidden in the complex urban microenvironment, providing a high-dimensional quantitative navigation basis for environmental law enforcement and pollution source tracing.

[0028] Specifically, the process of acquiring multi-source monitoring data streams and parsing pollutant identifiers and meteorological parameters, determining physical measurement benchmarks based on pollutant identifiers, converting the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to meteorological parameters, and mapping the original measured concentration vector of particulate matter into an actual state concentration vector is as follows: Parsing the multi-source monitoring data streams extracts the pollutant identifiers and environmental meteorological parameters encapsulated in the data packet header. The molecular phase attributes corresponding to the pollutant identifiers are identified. When the molecular phase attribute indicates gaseous pollutants, the atmospheric temperature and atmospheric pressure parameters within the environmental meteorological parameters are extracted. Preset standard reference temperature and pressure constants are introduced to establish a state transition equation. The original measured concentration vector of gaseous pollutants is substituted into the state transition equation to perform a physical benchmark alignment operation, outputting the reference state concentration vector. When the molecular phase attribute indicates particulate matter, the real-time monitoring temperature and real-time monitoring pressure parameters within the environmental meteorological parameters are locked to construct an actual state benchmark. The original measured concentration vector of particulate matter is projected in the same dimension according to the real-time monitoring temperature and real-time monitoring pressure parameters, outputting the actual state concentration vector.

[0029] In this implementation plan, the primary step in parsing multi-source monitoring data streams is to separate the identifiers encapsulated in the underlying communication data packet headers from the meteorological state data. Then, by identifying the molecular phase attributes corresponding to the identifiers, branch logic processing of the data stream is executed. Molecular phase identification is necessary because before performing underlying digital information fusion, the computer must rigorously eliminate systematic measurement errors caused by physical state misalignment in heterogeneous monitoring equipment. When the process enters the gaseous pollutant processing branch, the system invokes environmental meteorological parameters to construct a physical benchmark alignment calculation logic. The specific state transition equation is expressed as follows: ,in, This represents the x-th data component of the reference state concentration vector; The x-th data component represents the original measured concentration vector of gaseous pollutants; This represents the atmospheric temperature parameter extracted from environmental meteorological parameters and synchronized with the x-th data component; This represents the preset standard reference temperature constant. This indicates the preset standard reference pressure constant; This represents the atmospheric pressure parameter extracted from environmental meteorological parameters and synchronized with the x-th data component; x represents the linearly increasing sorting variable of the data component. Through the above nonlinear multiplication mapping equation, the computer uniformly converts the continuous gas measurement scalars under non-standard temperature and pressure environments to an absolute reference state. When the process enters the particulate matter processing branch, the system directly locks the real-time monitored temperature and pressure parameters to construct the actual state benchmark, and uses a direct projection operation of equal dimension to retain the original actual state concentration vector, thus perfectly aligning the underlying physical dimensions of multi-source heterogeneous environmental data with the measurement benchmark at the electronic digital processing level.

[0030] Specifically, the process of generating a normalized concentration matrix by cascading the reference state concentration vector and the actual state concentration vector is as follows: extract the sampling timestamp sequence and spatial geographic coordinate sequence carried by the reference state concentration vector and the actual state concentration vector; perform time granularity alignment operation on the sampling timestamp sequence and perform geographic grid discretization processing on the spatial geographic coordinate sequence to generate a spatiotemporal alignment feature index set; extract the reference state concentration vector and the actual state concentration vector, perform tensor concatenation operation according to the spatiotemporal alignment feature index set, and output a normalized concentration matrix that integrates the physical benchmark features of the multiphase states.

[0031] In this implementation scheme, the process of generating the normalized concentration matrix aims to address the data structure fragmentation caused by the uneven sampling frequency of multi-source IoT devices and the fragmented spatial distribution of geographic coordinates. The system extracts the underlying time stamps and spatial coordinates, performs time granularity alignment on the sampling timestamp sequence, forcibly aligning irregular continuous floating-point time slices to preset discrete time anchors; simultaneously, it performs geographic grid discretization on the spatial geographic coordinate sequence, reducing the dimensionality of continuously distributed latitude and longitude positioning points to a regularly divided two-dimensional matrix coordinate grid, thereby generating a highly normalized spatiotemporal alignment feature index set. After obtaining this spatiotemporal alignment feature index set, the system performs multi-source tensor splicing operations according to the constructed three-dimensional grid coordinate system to reconstruct a normalized concentration matrix that integrates multi-phase physical benchmark features. The specific formula for generating the grid elements of the tensor splicing is as follows: ,in, Y represents the tensor element in the normalized concentration matrix that is located at the coordinates of time index u, latitude index v, and longitude index w; Y represents the total number of feature nodes contained in the spatiotemporal aligned feature index set; y represents the cyclic accumulation index of the feature node. This represents the concentration observation scalar corresponding to the y-th feature node in the reference state concentration vector or the actual state concentration vector; This represents a predefined discrete matching measure function; This represents the temporal discretization index of the y-th feature node after temporal granularity alignment. This represents the latitude discretization index output after the y-th feature node is processed by geographic grid discretization; This represents the longitude discretization index output after the y-th feature node is processed by geographic grid discretization; Represents the traversal coordinates of the time dimension of a multidimensional tensor; Represents the coordinates for traversing the dimensional dimensions of a multidimensional tensor; Traversal coordinates of the longitude dimension of a multidimensional tensor. For the key discrete matching measure function. One of its publicly disclosed methods for determining this is as follows: the system compares two discrete index parameters as input. If and only if the underlying digital formats of the two parameters are completely equivalent and consistent, the function outputs a scalar of one; otherwise, it outputs a scalar of zero. This ensures that only data nodes whose spatiotemporal features completely match the corresponding three-dimensional grid will be cascaded and superimposed into the multidimensional normalized concentration matrix. This provides underlying digital assets with completely aligned data and absolutely regular structure for subsequent temporal topology analysis and spatial correlation risk inference.

[0032] Specifically, the process of constructing a temporal topology graph by parsing the timestamp sequence of the normalized concentration matrix and extracting breakpoints in the temporal topology graph as nodes to be compensated is as follows: The timestamp sequence of the normalized concentration matrix is ​​divided according to a preset time granularity to generate a discrete time axis vector. A sliding verification time window is set along the discrete time axis vector, and the frequency of continuous valid samples within the sliding verification time window is counted. The frequency of continuous valid samples is compared with the preset statistical validity baseline condition. Time window intervals that do not meet the preset statistical validity baseline condition are extracted to construct a discrete time segment set. The discrete time segment set is projected onto the temporal network structure to generate a temporal topology graph. Breakpoint features that are interrupted in the temporal topology graph are located. Null value marking is performed on the breakpoint features to assign missing labels, and nodes to be compensated with additional missing labels are output.

[0033] In this implementation plan, the process of constructing a time-series topology graph by parsing the timestamp sequence of the normalized concentration matrix aims to map the fragmented time information with hidden faults at the bottom layer into a network graph structure that can be efficiently computed and traversed by a computer, thereby accurately locating non-compliant data faults from a graph theory perspective. The system first forcibly aligns the timestamp sequence of the normalized concentration matrix according to a preset time granularity to generate a discrete time axis vector. Then, a sliding verification time window is set along this discrete time axis vector, and the frequency of data packets carrying valid monitoring concentration scalars within the window is counted. Since ambient air quality monitoring has strict legal requirements for data validity, the system must compare the frequency of continuous valid sampling with the preset statistical validity baseline conditions. Therefore, the system constructs a fault network determination equation: ,in, This indicates the breakpoint status flag for the q-th sliding verification time window; Represents the step activation function; This indicates the preset statistical validity baseline condition; P represents the total number of discrete-time slices included in the sliding validation time window; p represents the traversal sequence number of the discrete-time slices. This represents the effective data indicator variable for the p-th discrete-time slice within the q-th sliding validation time window. Pre-set statistical validity baseline conditions are applied to key parameters. The system discloses a method for determining the type of average time of ambient air pollutants. If it is determined to be within the statistical range of one-hour average concentration, the system directly binds and sets the condition parameter to the data slice quantity limit corresponding to 45 minutes according to the legal standards for ambient air quality, thereby ensuring the absolute legality of the data statistics. When the output of the breakpoint status indicator is positive, it indicates that the monitoring data in that time window interval is severely missing. The system extracts and constructs a discrete time segment set, and projects the discrete time segment set onto the temporal network structure to generate a temporal topology graph. Through the scanning calculation of node degree in topology, the system directly locates the breakpoint features in the temporal topology graph that are characterized by disconnection and zero network degree. The breakpoint features are marked with null values ​​and given missing labels, and the missing node to be compensated is output with the missing label. This step visualizes the discrete missing phenomenon as spatial graph nodes, providing accurate repair targets for subsequent algorithm intervention.

[0034] Please see Figure 2 Specifically, the process of extracting the target dense sensing sequence within the spatial neighborhood of the void node to be compensated, performing spatiotemporal first derivative operations to generate a gradient compensation vector, and injecting the gradient compensation vector into the void node to be compensated to reconstruct and output a continuous temporal concentration matrix is ​​as follows: A spatial buffer neighborhood is established by reading the latitude and longitude coordinate features bound to the void node to be compensated; the target dense sensing sequence output by the micro-sensor nodes deployed within the spatial buffer neighborhood is captured; partial derivative operations are simultaneously performed on the target dense sensing sequence along the spatial distance evolution dimension and the temporal evolution dimension to extract the concentration change rate and physical diffusion direction features in the local space, generating a multi-source temporal feature tensor; a nonlinear feature vector synthesis operation is performed on the multi-source temporal feature tensor to output a gradient compensation vector; the gradient compensation vector is written into the void node to be compensated according to the timestamp index sequence to perform a missing state overwrite operation, reconstructing and generating a continuous temporal concentration matrix without discontinuities.

[0035] In this implementation scheme, the execution logic of performing spatiotemporal first-order derivative operations on the target dense sensing sequence within the spatial neighborhood of the void node to be compensated abandons the defect of traditional static mean interpolation operations that easily smooth out the transient extreme values ​​of pollutants, and instead utilizes a high-resolution sensor array to capture the dynamic physical diffusion trend of pollutants. In this process, the system adopts a typical edge-cloud collaborative architecture, using the micro-sensor nodes deployed at the bottom layer as the edge computing end of the Internet of Things. The system reads the latitude and longitude coordinate features bound to the void node to be compensated to establish a spatial buffer neighborhood, captures the target dense sensing sequence output by the micro-sensor nodes deployed within the neighborhood, and directly performs partial derivative calculations on the target dense sensing sequence along the spatial distance evolution dimension and the time evolution dimension locally at the edge, extracting the concentration change rate and physical diffusion direction features in the local space to generate a multi-source temporal feature tensor. This edge-side (end-end) pre-computation processing effectively reduces the network bandwidth pressure and latency caused by directly uploading massive amounts of high-frequency raw data to the cloud. Subsequently, the system (cloud) performs nonlinear feature vector synthesis operations on the multi-source temporal feature tensor to output a gradient compensation vector. The system then performs a nonlinear eigenvector synthesis operation on the multi-source time-series feature tensor to output a gradient compensation vector. The vector synthesis calculation formula is expressed as follows: ,in, This represents the gradient compensation vector generated during the reconstruction of the d-th hole node to be compensated; denoted as the nearest effective concentration benchmark scalar preceding the d-th hole node to be compensated in the time-series topology graph; F represents the total number of micro-sensor nodes recalled within the spatial buffer neighborhood; f represents the traversal and accumulation index of the micro-sensor nodes. This represents the spatial distance attenuation weight of the f-th microsensor node; Represents the spatial diffusion gradient mapping coefficients; This indicates that the f-th micro-sensor node is in the corresponding The target dense sensing sequence values ​​output at any time; This represents the relative spatial orientation distance between the miniature sensing node and the void node to be compensated; The gradient mapping coefficients represent the time evolution; t represents the continuous-time independent variable at the bottom layer of the algorithm. This represents the temporal span step occupied by the d-th hole node to be compensated. For the key parameter, spatial distance decay weight... A determination method is disclosed: the system obtains the Euclidean spatial distance between each micro-sensor node and the void node to be compensated, calculates the inverse distance using an inverse distance spatial weighting model, and performs normalized division on the inverse distances of all sensor nodes within the spatial buffer neighborhood of the void node to be compensated to obtain the spatial distance attenuation weight, ensuring that the contribution of sensor networks with closer physical distances to the convergence of gradient compensation vectors increases exponentially. Finally, the system accurately writes the gradient compensation vector into the void node to be compensated based on the timestamp index sequence to perform a missing state overwrite operation, reconstructing and generating a continuous temporal concentration matrix without discontinuities. While strictly meeting the requirements for the continuity of environmental data time statistics, it faithfully reproduces the spatiotemporal abrupt evolution characteristics of air pollutants propagating with wind direction and over time.

[0036] Specifically, the process of decomposing the continuous time-series concentration matrix into a first-precision standard station sequence and a second-precision micro-station sequence, and calculating the discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window is as follows: Parse the device attribute tags bound to the continuous time-series concentration matrix and perform a classification and decomposition operation; extract data nodes with national control benchmark precision attributes to construct the first-precision standard station sequence, and extract data nodes with spatially dense deployment attributes to construct the second-precision micro-station sequence; align the timestamp features carried by the first-precision standard station sequence and the second-precision micro-station sequence according to a preset synchronization time span, and establish a time slice alignment vector; perform discrete deviation calculation operations on the benchmark concentration nodes associated with the first-precision standard station sequence and the drift concentration nodes associated with the second-precision micro-station sequence that are in the same spatial buffer neighborhood within the time slice alignment vector, and summarize the difference feature component sets output by the discrete deviation calculation operation to generate a discrete deviation array.

[0037] In this implementation scheme, the process of disassembling the continuous time-series concentration matrix and calculating the discrete deviation array aims to solve the systematic numerical drift problem caused by environmental temperature and humidity and device aging in the long-term operation of massive low-cost IoT micro-sensors. The system first parses the device attribute tags in the underlying data structure, classifying data nodes that have been periodically calibrated by national or provincial control authorities, have absolutely reliable values, but are sparsely distributed spatially into a first-precision standard station sequence. Simultaneously, it classifies densely distributed nodes with error fluctuations into a second-precision micro-station sequence. To ensure the rigor of error comparison, the system forcibly aligns the timestamps of both nodes according to a preset synchronization time span to construct a time slice alignment vector. Then, it performs discrete deviation calculations on the benchmark and drift nodes within the same spatial buffer neighborhood. The calculation equation for the generated discrete deviation array is expressed as: ,in Let g represent the deviation characteristic component of the g-th second-precision microstation sequence in the discrete deviation array within the h-th synchronization time window; g represents the linear traversal index of the second-precision microstation sequence; and h represents the auto-incrementing index of the synchronization time window. This represents the total number of time slice alignment vectors contained within a single synchronization time window; c represents the internal step index of the time slice alignment vector. The node scalar represents the drift concentration of the second-precision microstation sequence in the corresponding time slice; This represents the first-precision standard station sequence mapping index that is in the same spatial buffer neighborhood as the g-th microstation; This represents the reference concentration node scalar corresponding to the first-precision standard station sequence; This represents the spatial relative distance weighting coefficient. For the key parameter, the spatial relative distance weighting coefficient... A method for determining the spatial distance between a microstation and a standard station is disclosed: the system obtains the physical distance between the microstation and the standard station, calculates the reciprocal of this physical distance, and then normalizes it by dividing the reciprocal by the sum of the reciprocals of the distances of all microstations in the buffer neighborhood. Through the above calculation, the system accurately extracts the dynamic drift characteristics of each microstation node relative to the absolute benchmark, providing a solid and reliable data array foundation for subsequent implementation of global spatial concentration compensation.

[0038] Please see Figure 3 Specifically, the process of introducing ambient air functional zoning categories to perform spatial heterogeneity weighting on discrete deviation arrays to generate dynamic calibration matrices, and then using the dynamic calibration matrix to perform feature mapping reconstruction on continuous time-series concentration matrices to output a global calibration concentration field is as follows: The ambient air functional zoning category parameters associated with the monitoring nodes are retrieved to divide the spatial grid into Class I and Class II feature vectors. The discrete deviation array is mapped to the Class I and Class II feature vectors, and spatial heterogeneity weighting is performed to generate region-specific offset weight parameters. The corresponding time evolution parameters within the continuous time-series concentration matrix are extracted, and the region-specific offset weight parameters are tensor-extended along the time evolution parameters to construct the dynamic calibration matrix. The continuous time-series concentration matrix is ​​retrieved, and the continuous time-series concentration matrix and the dynamic calibration matrix are subjected to tensor inner product operation for feature mapping reconstruction to output a global calibration concentration field.

[0039] This implementation plan introduces a process of generating a dynamic calibration matrix and reconstructing the concentration field based on ambient air functional zoning categories. This aims to overcome the algorithmic flaws of traditional one-size-fits-all error compensation schemes that ignore urban micro-geographical differences. This process primarily relies on an IoT cloud server. Due to the significant spatial heterogeneity in background pollutant concentrations and diffusion patterns between Class I zones (such as nature reserves) and Class II zones (mixed industrial and residential areas), the system (cloud) strictly divides the spatial grid into Class I and Class II feature vectors based on the zoning parameters associated with the monitoring nodes. Subsequently, the system maps the previously acquired discrete deviation array onto the aforementioned feature vectors to perform spatial heterogeneity weighting and extracts the time evolution parameters within the continuous time-series concentration matrix for tensor expansion, constructing a calibration matrix that dynamically evolves over time. Finally, feature mapping reconstruction is achieved through tensor inner product operations. Through an end-cloud collaborative mechanism combining feature extraction from edge nodes (end) and global matrix inner product calculation from the central server (cloud), the reconstruction equation for the global calibration concentration field is expressed as: ,in, This represents the calibration reconstruction concentration scalar corresponding to the k-th spatial grid at the l-th time evolution node within the global calibration concentration field; A two-dimensional matrix traversal index representing a spatial grid; A discrete, incremental index representing the time evolution parameter; This represents the original observation tensor input to the corresponding spatiotemporal node in the continuous temporal concentration matrix; Represents the tensor inner product multiplication operator; This represents the static mean of the characteristic components mapped to the k-th spatial grid in the discrete bias array; This represents the environmental air functional zoning heterogeneity weighting coefficient associated with the k-th spatial grid; This represents the decay compensation spread function that indicates the dynamic change of parameters over time. It also includes the weighting coefficient for the heterogeneity of key parameters in the environmental air functional zoning. A determination method is disclosed: the system compares the zoning attributes of the spatial grid. If it is determined to be a Class I area, a strict low-tolerance penalty weight constant is assigned due to its extremely low background concentration and sensitivity to pollution fluctuations; if it is determined to be a Class II area, a conventional tolerance weight constant is assigned. This reconstruction step, by integrating spatial attributes and temporal evolution characteristics, elevates the error correction of isolated microstations to a global heterogeneous grid smoothing operation, completely eliminating the local distortion interference of low-cost sensors, and outputting a large-scale continuous concentration field matrix with both high spatial resolution and extremely high numerical reliability.

[0040] Specifically, the process of introducing the transition phase limit matrix and the basic project limit matrix, comparing the global calibration concentration field with the two sets of limit matrices, and calculating the step approximation difference vector of the spatial grid in the global calibration concentration field converging from the transition phase limit matrix to the basic project limit matrix is ​​as follows: The time buffer between the implementation of the transition phase limit matrix association and the regular limit application interval for the implementation of the basic project limit matrix association are read from the preset standard switching time node. The real-time dynamic concentration scalar carried by a single spatial grid is extracted by analyzing the global calibration concentration field. The initial environmental redundancy scalar is calculated by comparing the real-time dynamic concentration scalar with the first-level constraint threshold contained in the transition phase limit matrix. The ultimate bearing approximation scalar is calculated by comparing the real-time dynamic concentration scalar with the second-level constraint threshold contained in the basic project limit matrix. Spatiotemporal evolution dimensionality reduction analysis is performed on the initial environmental redundancy scalar and the ultimate bearing approximation scalar to calculate the convergence rate of the spatial grid evolving towards the regular limit application interval. The convergence rate is integrated to output the step approximation difference vector.

[0041] This implementation plan introduces a process of calculating the difference vector between a transitional phase limit matrix and a basic project limit matrix to address the lag in risk assessment caused by the dual-track implementation standards during the transition period between old and new environmental management policies. Since urban air quality standards will implement stricter basic project limits after a specific time point, the system needs not only to determine whether current standards are exceeded but also to quantify the dynamic risk of approaching future stricter standards. The system first reads the preset standard transition time point to segment the timeline into a time buffer zone associated with the implementation of the transitional phase limit matrix and the application interval of the regular limits. Then, it analyzes the global calibration concentration field to extract the real-time dynamic concentration scalar carried by a single spatial grid. The system uses an algorithm to compare the real-time dynamic concentration scalar with the first-level constraint threshold to obtain the initial environmental redundancy scalar, and compares it with the second-level constraint threshold to obtain the ultimate carrying capacity approximation scalar. To achieve dimensionality reduction extraction of high-dimensional features, the system performs spatiotemporal evolution dimensionality reduction analysis on the initial environmental redundancy scalar and the ultimate bearing capacity approximation scalar, generating a convergence rate characterizing the drastic evolution of the spatial grid. The specific formula for generating this step-approximation difference vector is as follows: ,in, Indicates the first The spatial grid in the ... Step-approximation difference vectors for each evolution time node; The linear retrieval ordinal number representing the spatial grid; An index representing the arrangement of evolution time points; This represents the preset stage limit sensitivity weight coefficient; This represents the real-time dynamic concentration scalar extracted from the global calibration concentration field. This represents the first-level constraint threshold of the corresponding spatial grid, i.e., the transition phase limit. This represents the second-level constraint threshold of the corresponding spatial grid, i.e., the basic project limit. The system continuous time variable represents the continuous operation of the system. This represents the preset policy time decay constant; This indicates the remaining time step from the current time point to the application interval of the standard limit. It also includes the preset sensitivity weighting coefficient for the stage limit. The system discloses a method for determining the weighting coefficient: the system obtains the number of days remaining until the strict standard is officially implemented, divides the remaining number of days by the total number of days designed for the transition phase to obtain a dynamic ratio value, and uses it as the weighting coefficient. This ensures that as the switching node approaches, the algorithm model will automatically and smoothly shift the focus of risk assessment from static exceedance judgment to the monitoring of concentration deterioration rate, enabling the system to have a proactive algorithmic early warning capability for future strict standards.

[0042] Please see Figure 4 Specifically, the process of extracting the extreme value nodes of the abrupt gradient from the step approximation difference vector to generate a multidimensional pollution risk source map is as follows: The step approximation difference vector is divided into segments according to a preset step size to construct a local spatial evolution feature matrix. Second-order partial derivatives are calculated on the local spatial evolution feature matrix to obtain local curvature peak attributes. Data nodes whose spatial gradient changes exceed a preset variation threshold and deviate from the overall background concentration evolution trend of the region are selected as extreme value nodes of the abrupt gradient. Spatial density clustering is performed on the extreme value nodes of the abrupt gradient to generate a cluster of abrupt target activity coordinates. The cluster of abrupt target activity coordinates is then combined with the geographic information system coordinate system to perform a multidimensional spatial feature projection operation, reconstructing and outputting a multidimensional pollution risk source map.

[0043] In this implementation scheme, the process of extracting extreme nodes of the abrupt gradient and mapping them to generate a source map aims to accurately extract microscopic and hidden abnormal pollution sources from the macroscopic concentration difference vector field using computer graphics topological feature analysis technology. The system performs spatial domain segmentation and recombination of the step-approximation difference vector according to a preset step size to construct a local spatial evolution feature matrix. By performing second-order partial derivative calculations on the local spatial evolution feature matrix, the local curvature peak attribute, representing the intensity of concentration gradient abrupt changes, is obtained. Since conventional meteorological diffusion also causes concentration fluctuations, the system must combine local background trends for filtering and masking. The determination equation for its local curvature peak attribute is expressed as: ,in, The longitude direction index in the two-dimensional spatial grid coordinate system is... And the latitude direction index is The local curvature peak attribute of the data nodes; Discrete coordinate variables representing the longitude direction of the spatial grid; Discrete coordinate variables representing the latitudinal direction of the spatial grid; This indicates the preset background trend penalty coefficient; This indicates the total number of grid cells within the local region to which the data node belongs; This represents the cumulative traversal of the grid within a local region; Indicates the first [unit] within a local region The system provides a step-approximation difference vector scalar corresponding to each grid. A method for determining the preset mutation threshold for parameters is disclosed: the system extracts the statistical distribution set of local curvature peak attributes from areas with no abnormal records over the past three months, calculates the discrete standard deviation of this set, and sets the value corresponding to three times the positive standard deviation as the preset mutation threshold. The system compares and determines that when the calculated local curvature peak attribute exceeds this preset mutation threshold, the coordinate data node is identified as deviating from the overall background concentration evolution trend of the area and is established as a mutation gradient extreme value node. Finally, the system schedules a spatial density clustering algorithm to perform unsupervised analytical aggregation on the discrete mutation gradient extreme value nodes to generate a mutation target activity coordinate cluster. This coordinate cluster is then combined with the geographic information system coordinate system to perform multi-dimensional spatial feature projection operations, rendering a high-resolution multi-dimensional pollution risk tracing map. Through the above data denoising processing based on Laplace second-order spatial differentiation and density clustering, a substantial technological leap has been achieved from massive concentration monitoring reports to directly outputting the specific latitude and longitude of hidden pollution sources.

[0044] In summary, this application has at least the following effects: A multi-source monitoring data analysis method for urban air quality eliminates systematic benchmark biases caused by measurement misalignment between gaseous pollutants and particulate matter by performing classification physical benchmark alignment on multi-source heterogeneous data streams at the underlying data structure. For temporal topological breakpoints caused by equipment malfunctions, gradient compensation is performed using the spatiotemporal first derivative of densely packed sensor sequences in the spatial neighborhood, preserving the transient abrupt changes in pollution evolution without loss while strictly meeting the legal statistical validity baseline. Simultaneously, relying on high-precision standard station sequences and introducing heterogeneity weights from environmental air functional zoning, dynamic calibration and feature mapping are performed on the micro-sensor network, effectively mitigating systematic drift errors of sensor nodes and reconstructing a highly refined and realistic global calibration concentration field. Finally, based on this calibration concentration field and a dual-track threshold matrix, a step-approximation difference vector is calculated, overcoming the limitations of traditional single static threshold exceedance judgment. By extracting extreme value nodes of abrupt gradients deviating from background trends, dynamic convergence early warning and high-resolution spatial precise positioning of hidden and sudden pollution sources in the urban microenvironment are achieved.

[0045] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for analyzing multi-source monitoring data of urban air quality, characterized in that, Includes the following steps: S1. Acquire multi-source monitoring data streams and parse pollutant identifiers and meteorological parameters. Determine the physical measurement benchmark based on the pollutant identifiers. Convert the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to the meteorological parameters. Map the original measured concentration vector of particulate matter to the actual state concentration vector. Cascade the reference state concentration vector and the actual state concentration vector to generate a normalized concentration matrix. S2. Analyze the timestamp sequence of the normalized concentration matrix to construct a time-series topology graph, extract the breakpoints in the time-series topology graph and map them as nodes to be compensated; extract the target dense sensing sequence in the spatial neighborhood of the nodes to be compensated and perform spatiotemporal first derivative operation to generate gradient compensation vector; inject the gradient compensation vector into the nodes to be compensated to reconstruct and output a continuous time-series concentration matrix. S3. Decompose the continuous time-series concentration matrix into a first-precision standard station sequence and a second-precision micro-station sequence, and calculate the discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window; introduce the environmental air functional zoning category to perform spatial heterogeneity weighting on the discrete deviation array to generate a dynamic calibration matrix, and use the dynamic calibration matrix to perform feature mapping on the continuous time-series concentration matrix to reconstruct the output global calibration concentration field. S4. Introduce the transition phase limit matrix and the basic project limit matrix, compare the global calibration concentration field with the two sets of limit matrices, and calculate the step approximation difference vector of the spatial grid in the global calibration concentration field that converges from the transition phase limit matrix to the basic project limit matrix; extract the abrupt gradient extreme value node mapping of the step approximation difference vector to generate a multidimensional pollution risk source map. 2.The urban air quality multi-source monitoring data analysis method according to claim 1, characterized in that: The specific process of acquiring multi-source monitoring data streams and parsing pollutant identifiers and meteorological parameters, determining physical measurement benchmarks based on pollutant identifiers, converting the original measured concentration vector of gaseous pollutants into a reference state concentration vector according to meteorological parameters, and mapping the original measured concentration vector of particulate matter to the actual state concentration vector is as follows: The pollutant identifier and environmental meteorological parameters encapsulated in the data packet header of the multi-source monitoring data stream are extracted by parsing. The molecular phase attributes corresponding to the pollutant identifier are identified. When the molecular phase attribute indicates gaseous pollutants, the atmospheric temperature parameter and atmospheric pressure parameter in the environmental meteorological parameters are extracted. A state transition equation is established by introducing a preset standard reference temperature constant and a preset standard reference pressure constant. The original measured concentration vector of gaseous pollutants is substituted into the state transition equation to perform physical reference alignment operation and output the reference state concentration vector. When the molecular phase properties indicate particulate matter, the real-time monitoring temperature parameter and the real-time monitoring pressure parameter within the environmental meteorological parameters are locked to construct the actual state benchmark. The original measured concentration vector of particulate matter is then projected in the same dimension according to the real-time monitoring temperature parameter and the real-time monitoring pressure parameter to output the actual state concentration vector.

3. The urban air quality multi-source monitoring data analysis method according to claim 2, characterized in that: The specific process of generating a normalized concentration matrix from the cascaded reference state concentration vector and the actual state concentration vector is as follows: Extract the sampling timestamp sequence and spatial geographic coordinate sequence carried by the reference state concentration vector and the actual state concentration vector. Perform time granularity alignment operation on the sampling timestamp sequence and perform geographic grid discretization processing on the spatial geographic coordinate sequence to generate a spatiotemporal alignment feature index set. Extract the reference state concentration vector and the actual state concentration vector, perform tensor concatenation operation according to the spatiotemporal alignment feature index set, and output a normalized concentration matrix that integrates the physical benchmark features of the multiphase state.

4. The urban air quality multi-source monitoring data analysis method according to claim 1, characterized in that: The specific process of constructing a time-series topology graph by parsing the timestamp sequence of the normalized concentration matrix and extracting breakpoints in the time-series topology graph as nodes with holes to be compensated is as follows: A discrete time axis vector is generated by dividing the timestamp sequence of the normalized concentration matrix according to the preset time granularity. A sliding verification time window is set along the discrete time axis vector, and the frequency of continuous effective sampling within the sliding verification time window is counted. By comparing the frequency of continuous effective sampling with the preset statistical validity baseline conditions, the time window intervals that do not meet the preset statistical validity baseline conditions are extracted to construct a discrete time segment set. The discrete time segment set is projected onto the temporal network structure to generate a temporal topology graph. The breakpoint features that are interrupted in the temporal topology graph are located. The breakpoint features are marked with null values ​​and given missing labels. The missing label is added to the hole nodes to be compensated.

5. The urban air quality multi-source monitoring data analysis method according to claim 1, characterized in that: The specific process of extracting the dense sensing sequence of the target within the spatial neighborhood of the hole node to be compensated, performing spatiotemporal first derivative operations to generate a gradient compensation vector, and injecting the gradient compensation vector into the hole node to be compensated to reconstruct the output continuous temporal concentration matrix is ​​as follows: A spatial buffer neighborhood is established by reading the latitude and longitude coordinate features bound to the void node to be compensated. The target dense sensing sequence output by the micro-sensor nodes deployed in the spatial buffer neighborhood is captured. The partial derivative calculation is performed on the target dense sensing sequence along the spatial distance evolution dimension and the time evolution dimension simultaneously. The concentration change rate and physical diffusion direction features in the local space are extracted to generate a multi-source time series feature tensor. A nonlinear feature vector synthesis operation is performed on the multi-source temporal feature tensor to output a gradient compensation vector. Based on the timestamp index sequence, the gradient compensation vector is written into the missing state overwrite operation of the hole node to be compensated, and a continuous temporal concentration matrix without discontinuities is reconstructed.

6. The method for analyzing multi-source monitoring data of urban air quality according to claim 1, characterized in that: The continuous time-series concentration matrix is ​​decomposed into a first-precision standard station sequence and a second-precision micro-station sequence. The specific process for calculating the discrete deviation array of the first-precision standard station sequence and the second-precision micro-station sequence within the synchronization time window is as follows: The device attribute tags bound to the continuous time-series concentration matrix are parsed and classified. Data nodes with national control benchmark accuracy are extracted to construct a first-precision standard station sequence, and data nodes with spatially dense deployment are extracted to construct a second-precision micro-station sequence. Align the timestamp features carried by the first-precision standard station sequence and the second-precision micro station sequence according to the preset synchronization time span, and establish a time slice alignment vector; Discrete bias calculation is performed on the reference concentration nodes associated with the first-precision standard station sequence and the drift concentration nodes associated with the second-precision micro-station sequence that are in the same spatial buffer neighborhood within the time slice alignment vector. The set of difference feature components output by the discrete bias calculation is summarized to generate a discrete bias array.

7. The method for analyzing multi-source monitoring data of urban air quality according to claim 1, characterized in that: The specific process of introducing ambient air functional zoning categories to perform spatial heterogeneity weighting on discrete bias arrays to generate dynamic calibration matrices, and then using the dynamic calibration matrix to perform eigenmap reconstruction on continuous time-series concentration matrices to output a global calibration concentration field is as follows: The environmental air functional zoning category parameters associated with the monitoring nodes are retrieved to divide the spatial grid into Class I and Class II feature vectors. The discrete deviation array is mapped to the Class I and Class II feature vectors, and spatial heterogeneity weighted calculation is performed to generate region-specific offset weight parameters. Extract the corresponding time evolution parameters in the continuous time-series concentration matrix, and construct a dynamic calibration matrix by tensor expansion of the region-specific offset weight parameters along the time evolution parameters; Retrieve the continuous time-series concentration matrix, perform tensor inner product operation on the continuous time-series concentration matrix and the dynamic calibration matrix to reconstruct the feature mapping, and output the global calibration concentration field.

8. The method for analyzing multi-source monitoring data of urban air quality according to claim 1, characterized in that: Introducing the transition phase limit matrix and the basic item limit matrix, and comparing the global calibration concentration field with the two sets of limit matrices, the specific process of calculating the step-approximation difference vector of the spatial grid within the global calibration concentration field converging from the transition phase limit matrix to the basic item limit matrix is ​​as follows: Read the time buffer between the preset standard switching time node and the time buffer between the implementation transition phase limit matrix and the regular limit application interval between the implementation basic project limit matrix, and analyze the global calibration concentration field to extract the real-time dynamic concentration scalar carried by a single spatial grid. The initial environmental margin scalar is calculated by comparing the real-time dynamic concentration scalar with the first-level constraint threshold contained in the transition phase limit matrix, and the ultimate bearing capacity approximation scalar is calculated by comparing the real-time dynamic concentration scalar with the second-level constraint threshold contained in the basic project limit matrix. Spatiotemporal evolution dimensionality reduction analysis is performed on the initial environmental redundancy scalar and the ultimate bearing capacity approximation scalar to calculate the convergence rate of the spatial grid evolving into the conventional limit application interval, and the convergence rate is integrated to output the step approximation difference vector.

9. The method for analyzing multi-source monitoring data of urban air quality according to claim 1, characterized in that: The specific process of extracting the abrupt gradient extreme value node mapping of the step approximation difference vector to generate a multidimensional pollution risk source map is as follows: The local spatial evolution feature matrix is ​​constructed by dividing the step approximation difference vector according to the preset step size, and the local curvature peak attribute is obtained by performing second-order partial derivative calculation on the local spatial evolution feature matrix. Data nodes whose spatial gradient changes exceed a preset mutation threshold and deviate from the overall background concentration evolution trend of the region are selected as extreme value nodes of the mutation gradient by combining the local curvature peak attribute. Spatial density clustering is performed on the extreme nodes of the mutation gradient to generate the activity coordinate cluster of the mutation target. The activity coordinate cluster of the mutation target is combined with the geographic information system coordinate system to perform multi-dimensional spatial feature projection operation, and a multi-dimensional pollution risk source tracing map is reconstructed and output.

Images