Pollutant concentration background value calculation method based on reference sampling node and multi-source data

Abstract

The application relates to the field of big data analysis, and discloses a pollution concentration background value calculation method based on a benchmark sampling node and multi-source data, which comprises the following steps: obtaining target parameter time series data of the benchmark sampling node and a multi-source heterogeneous feature set; implementing dimension aggregation and normalization processing to construct a multi-dimensional environmental state tensor representing node dynamic evolution; extracting background fluctuation characteristics by using a feature subspace projection algorithm, calculating normalized interference weight coefficients of each time series slice in a high-dimensional topological space relative to a historical background baseline; and injecting the coefficients into a steady-state statistical smoothing operator to strip nonlinear disturbance characteristics through mean shift operation. The application quantifies local interference by using multi-dimensional feature coupling and topological distance measurement, eliminates the interference of micro-scale environmental factor fluctuation on background value extraction, and realizes accurate representation of a regional macro evolution trend sequence.

Description

Technical Field

[0001] This invention belongs to the field of big data analysis technology, and in particular relates to a method for calculating the background value of pollutant concentration based on benchmark sampling nodes and multi-source data. Background Technology

[0002] In the current data processing system of large-scale environmental monitoring networks, the core step in extracting background pollutant concentration sequences representing regional average conditions from continuous observation data streams to support the construction of atmospheric evolution models and the formulation of coordinated control strategies is to use big data analysis to remove short-term random disturbances caused by local pollution sources. The current mainstream practice in the industry mainly relies on the statistical distribution patterns based on single concentration time series, using percentile methods or robust statistical methods to set static or sliding calculation windows, thereby identifying and eliminating abnormally high-value points that deviate from the statistical envelope. However, when the benchmark sampling node is in a local static and stable environment such as low wind speed residence and strong boundary layer inversion, pollutants from near-field trace emission sources accumulate continuously. Such data exhibits low gradient and long duration characteristics, and its mathematical statistical distribution is strongly homomorphic to the baseline evolution of background air masses in large-scale regions.

[0003] Statistical algorithms based on independent scalars, lacking the physical constraints of multidimensional spatial dynamics, cannot distinguish between local enrichment and background concentration increases. This leads to local interference data penetrating the statistical model, causing inaccurate baseline drift in the background value sequence. Optimizing the physical location of sampling sites or adding vertical observation hardware can alleviate ground interference, but the cost of site deployment and limitations in equipment topology create bottlenecks for the analysis model in handling complex nonlinear coupled data. For example, Chinese invention patent application CN110348649A discloses a method for estimating PM2.5 background concentration based on meteorological elements and the vertical distribution characteristics of pollutants, analyzing the vertical gradient of meteorological and pollutant factors. To address the aforementioned issues, existing technologies attempt to introduce independent threshold blocking mechanisms for meteorological factors, using cutoff boundaries set for parameters such as wind speed and direction to assist in the screening. However, this approach presents new technical challenges when facing nonlinearly coupled environmental conditions. On the one hand, atmospheric diffusion processes exhibit nonlinear characteristics, and discrete thresholds for single factors cannot accurately characterize the local residence state under the synergistic effects of multiple variables. On the other hand, rigid blocking logic leads to large-scale data removal under specific meteorological conditions, causing logical breaks in the concentration sequence and impairing the temporal continuity of the background value sequence.

[0004] Therefore, the technical problem to be solved by this invention is how to break the inertia of the judgment of variables isolated from each other in the existing big data processing while maintaining the continuity of time series data, and to identify and eliminate penetrating noise by constructing a correlation logic with physical consistency. Summary of the Invention

[0005] This invention provides a method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data, comprising the following steps:

[0006] Step S101: Obtain the target parameter time series data of the reference sampling node, and obtain the multi-source heterogeneous feature set that is aligned with the acquisition phase of the target parameter time series data in the time dimension;

[0007] Step S102: The target parameter time series data and multi-source heterogeneous feature set are subjected to dimensional aggregation and variance normalization processing to construct a multi-dimensional environmental state tensor representing the dynamic evolution trajectory of the local microenvironment of the benchmark sampling node in a multi-dimensional feature coordinate system.

[0008] Step S103: The covariance matrix decomposition of the multidimensional environmental state tensor is performed using the feature subspace projection algorithm to extract the basis vector components that reflect the macro background fluctuation characteristics. The multidimensional environmental state tensor is mapped to a high-dimensional topological space. The deviation vector of each time slice in the high-dimensional topological space relative to the preset historical background baseline is calculated. Based on the negative correlation mapping function between the Euclidean distance magnitude of the deviation vector and the interference weight, the normalized interference weight coefficient of the corresponding time slice is determined.

[0009] Step S104: The normalized disturbance weight coefficient is injected into the preset steady-state statistical smoothing operator as the iterative learning weight. The steady-state statistical smoothing operator is used to perform weighted mean drift operation on the target parameter time series data to remove the nonlinear disturbance features induced by local micro-topography and meteorological dynamic field represented by the multi-source heterogeneous feature set. The pollutant concentration background value sequence reflecting the regional macro trend is output by the weighted mean drift operation.

[0010] Preferably, the multi-source heterogeneous feature set includes scalar data characterizing the local topographic relief of the reference sampling node, vector data characterizing the real-time atmospheric dynamic field radius change, and time-varying data characterizing the emission intensity of known surrounding fixed emission sources; step S101 includes: applying nonlinear axial stretching to the target parameter time series data and the multi-source heterogeneous feature set through a dynamic time warping operator to eliminate the acquisition phase difference caused by sensor heterogeneity and generate a feature matrix under a unified time scale.

[0011] Preferably, the rule for establishing the normalized interference weight coefficient in step S103 follows the formula below: ,in, For the first Normalized interference weight coefficients for each time slice; The perturbation sensitivity factor is a preset value for the signal-to-noise ratio based on a multi-source heterogeneous feature set, and its value ranges from 0.5 to 1.5. For the first The Euclidean distance metric of each deviation vector relative to the historical background baseline in a high-dimensional topological space.

[0012] Preferably, when performing covariance matrix decomposition in step S103, the principal component basis vectors with a cumulative contribution rate greater than 85% are retained as feature basis vectors by calculating the eigenvalues ​​of the multidimensional environmental state tensor in each dimension; the deviation vector is defined as the residual component in the orthogonal complement of the feature subspace formed by the feature basis vectors of the current time slice, which is used to quantify the degree of influence of local disturbances on the benchmark sampling node.

[0013] Preferably, the steady-state statistical smoothing operator adopts a kernel density estimation algorithm with physical property boundary constraints; step S104 includes: injecting the normalized disturbance weight coefficient into the kernel function window of the kernel density estimation algorithm, and dynamically adjusting the contribution ratio of the corresponding time slice in the probability density evolution path so that the weight of the time slice disturbed by nonlinear disturbance characteristics in the calculation of the pollutant concentration background value sequence decreases monotonically with the increase of the magnitude of the deviation vector.

[0014] Preferably, after step S104, the method further includes the following repair steps: when a data gap is detected in the background value sequence of pollutant concentration due to long-term local pollution interference, the pressure gradient vector difference in the vector data of the missing time window boundary is extracted; continuous data segments that satisfy the same topological features are retrieved in the historical background database, and the matching data segments are mapped to the missing position of the sequence through a dynamic time warping algorithm to complete the data chain repair.

[0015] Preferably, the historical background baseline is established by the following calibration method: retrieve the historical observation records of the benchmark sampling node under calm wind conditions and when the surrounding fixed emission sources are in a zero-emission state, extract the static distribution centroid of the multidimensional environmental state tensor under the corresponding conditions, and define the static distribution centroid as the coordinate origin of the high-dimensional topological space as the benchmark reference point for measuring the real-time state shift.

[0016] Preferably, when performing the weighted mean drift operation in step S104, a preset convergence threshold is set. When the absolute deviation between the background values ​​generated by two adjacent iterations is less than the convergence threshold, the iterative learning process is stopped, and the value generated in the current iteration step is determined as the final pollutant concentration background value sequence, so as to ensure the numerical stability of the background value extraction logic when processing high-dimensional non-stationary data streams.

[0017] Preferably, step S104 includes: calculating the first-order time variability of the pollutant concentration background value sequence to identify the evolution cycle of large-scale weather systems; and adaptively adjusting the smoothing time window width of the steady-state statistical smoothing operator based on the duration of the evolution cycle to ensure that the background value calculation logic can effectively cover the variation law of the macro-meteorological field and suppress local micro-scale noise at different seasonal scales.

[0018] Preferably, the method further includes the following closed-loop calibration steps: performing residual subtraction analysis on the pollutant concentration background value sequence and the original target parameter time series data to obtain a difference sequence characterizing the contribution of local disturbances, and performing dynamic calibration on the value of the disturbance sensitivity factor in the formula based on the second moment fluctuation characteristics of the difference sequence to compensate for the weight allocation deviation caused by abrupt changes in environmental state.

[0019] Compared with existing technologies, the pollutant concentration background value calculation method of this invention based on benchmark sampling nodes and multi-source data has the following advantages:

[0020] 1. In the calculation of pollutant concentration background values ​​from multi-source data, a multi-dimensional slice tensor representing transient full-scale characteristics is constructed by performing structured splicing of pollutant concentration scalars and meteorological state vector flows in memory. This breaks through the linear fragmentation limitations caused by independent determination of variables in traditional environmental monitoring. Based on the manifold centroid vector and covariance inverse matrix generated from historical background datasets, the system can accurately capture local hidden disturbances that do not reach the limits of a single physical quantity but exhibit nonlinear combined distortions in multi-dimensional space by calculating the topological deviation distance of real-time data slices relative to the background constraint envelope model. This processing path, which transforms from one-dimensional scalar blocking to high-dimensional topological geometric determination, effectively removes local noise data with strong homomorphic characteristics to large-scale background evolution and avoids background baseline drift caused by nonlinear interference of physical variables.

[0021] 2. The calculation method introduces a time-scale inertial smoothing operator at the front end of matrix reconstruction. It performs a sliding window vector averaging operation on the input meteorological parameter vector flow. Based on the physical inertia of atmospheric advection, this mechanism filters out high-frequency vector jumps caused by local topographic turbulence, ensuring that the meteorological vectors sent to the projection screening unit represent the true macroscopic air mass movement trajectory. The feature projection screening unit combines the Mahalanobis distance calculation results with the preset tensor projection tolerance to convert the judgment results into normalized interference weight coefficients, and injects the coefficients inversely into the steady-state statistical smoothing operator. This dynamically adjusts the iterative learning weights of the corresponding time slices in the background value calculation. This soft-landing screening mechanism avoids the data flow logic discontinuity caused by the hard elimination of traditional Boolean logic, and enhances the logical stability of the computing system under complex and variable working conditions.

[0022] 3. The computational method is configured with a data void topology filling module. When continuous local pollution causes long-term voids in the concentration sequence, the system extracts the pressure gradient vector difference based on the void time window boundary and retrieves continuous data segments with the same topological characteristics from the historical clean background database. The matching data segments are mapped to the sequence voids through a dynamic time warping operator to complete the data link repair. This ensures that the reconstructed sequence maintains the continuity of temporal statistical characteristics while having clear physical causal constraints. This mechanism abandons mathematical linear interpolation without physical basis and avoids secondary calculation distortion introduced by the lack of spatial dynamic correlation, ensuring that the final output pollutant concentration background value sequence can present the macroscopic evolution trend of a large-scale region with high fidelity. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall process of the concentration background value calculation method of the present invention;

[0024] Figure 2 This is a diagram of the multi-source heterogeneous data fusion and environmental state tensor construction architecture of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0027] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0028] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0029] A method for calculating background pollutant concentrations based on baseline sampling nodes and multi-source data includes the following steps:

[0030] Step S101: Obtain the target parameter time series data of the reference sampling node, and obtain the multi-source heterogeneous feature set that is aligned with the acquisition phase of the target parameter time series data in the time dimension;

[0031] Step S102: The target parameter time series data and multi-source heterogeneous feature set are subjected to dimensional aggregation and variance normalization processing to construct a multi-dimensional environmental state tensor representing the dynamic evolution trajectory of the local microenvironment of the benchmark sampling node in a multi-dimensional feature coordinate system.

[0032] Step S103: The covariance matrix decomposition of the multidimensional environmental state tensor is performed using the feature subspace projection algorithm to extract the basis vector components that reflect the macro background fluctuation characteristics. The multidimensional environmental state tensor is mapped to a high-dimensional topological space. The deviation vector of each time slice in the high-dimensional topological space relative to the preset historical background baseline is calculated. Based on the negative correlation mapping function between the Euclidean distance magnitude of the deviation vector and the interference weight, the normalized interference weight coefficient of the corresponding time slice is determined.

[0033] Step S104: The normalized disturbance weight coefficient is injected into the preset steady-state statistical smoothing operator as the iterative learning weight. The steady-state statistical smoothing operator is used to perform weighted mean drift operation on the target parameter time series data to remove the nonlinear disturbance features induced by local micro-topography and meteorological dynamic field represented by the multi-source heterogeneous feature set. The pollutant concentration background value sequence reflecting the regional macro trend is output by the weighted mean drift operation.

[0034] Preferably, the multi-source heterogeneous feature set includes scalar data characterizing the local topographic relief of the reference sampling node, vector data characterizing the real-time atmospheric dynamic field radius change, and time-varying data characterizing the emission intensity of known surrounding fixed emission sources; step S101 includes: applying nonlinear axial stretching to the target parameter time series data and the multi-source heterogeneous feature set through a dynamic time warping operator to eliminate the acquisition phase difference caused by sensor heterogeneity and generate a feature matrix under a unified time scale.

[0035] Preferably, the rule for establishing the normalized interference weight coefficient in step S103 follows the formula below: ,in, For the first Normalized interference weight coefficients for each time slice; The perturbation sensitivity factor is a preset value for the signal-to-noise ratio based on a multi-source heterogeneous feature set, and its value ranges from 0.5 to 1.5. For the first The Euclidean distance metric of each deviation vector relative to the historical background baseline in a high-dimensional topological space.

[0036] Preferably, when performing covariance matrix decomposition in step S103, the principal component basis vectors with a cumulative contribution rate greater than 85% are retained as feature basis vectors by calculating the eigenvalues ​​of the multidimensional environmental state tensor in each dimension; the deviation vector is defined as the residual component in the orthogonal complement of the feature subspace formed by the feature basis vectors of the current time slice, which is used to quantify the degree of influence of local disturbances on the benchmark sampling node.

[0037] Preferably, the steady-state statistical smoothing operator adopts a kernel density estimation algorithm with physical property boundary constraints; step S104 includes: injecting the normalized disturbance weight coefficient into the kernel function window of the kernel density estimation algorithm, and dynamically adjusting the contribution ratio of the corresponding time slice in the probability density evolution path so that the weight of the time slice disturbed by nonlinear disturbance characteristics in the calculation of the pollutant concentration background value sequence decreases monotonically with the increase of the magnitude of the deviation vector.

[0038] Preferably, after step S104, the method further includes the following repair steps: when a data gap is detected in the background value sequence of pollutant concentration due to long-term local pollution interference, the pressure gradient vector difference in the vector data of the missing time window boundary is extracted; continuous data segments that satisfy the same topological features are retrieved in the historical background database, and the matching data segments are mapped to the missing position of the sequence through a dynamic time warping algorithm to complete the data chain repair.

[0039] Preferably, the historical background baseline is established by the following calibration method: retrieve the historical observation records of the benchmark sampling node under calm wind conditions and when the surrounding fixed emission sources are in a zero-emission state, extract the static distribution centroid of the multidimensional environmental state tensor under the corresponding conditions, and define the static distribution centroid as the coordinate origin of the high-dimensional topological space as the benchmark reference point for measuring the real-time state shift.

[0040] Preferably, when performing the weighted mean drift operation in step S104, a preset convergence threshold is set. When the absolute deviation between the background values ​​generated by two adjacent iterations is less than the convergence threshold, the iterative learning process is stopped, and the value generated in the current iteration step is determined as the final pollutant concentration background value sequence, so as to ensure the numerical stability of the background value extraction logic when processing high-dimensional non-stationary data streams.

[0041] Preferably, step S104 includes: calculating the first-order time variability of the pollutant concentration background value sequence to identify the evolution cycle of large-scale weather systems; and adaptively adjusting the smoothing time window width of the steady-state statistical smoothing operator based on the duration of the evolution cycle to ensure that the background value calculation logic can effectively cover the variation law of the macro-meteorological field and suppress local micro-scale noise at different seasonal scales.

[0042] Preferably, the method further includes the following closed-loop calibration steps: performing residual subtraction analysis on the pollutant concentration background value sequence and the original target parameter time series data to obtain a difference sequence characterizing the contribution of local disturbances, and performing dynamic calibration on the value of the disturbance sensitivity factor in the formula based on the second moment fluctuation characteristics of the difference sequence to compensate for the weight allocation deviation caused by abrupt changes in environmental state.

[0043] Example 1: In the data aggregation and computing center of a regional atmospheric environment monitoring network, when the benchmark sampling node faces the condition of local enrichment of pollutants caused by topographic obstruction, the method of the present invention is used to extract background values. The processor receives the pollutant concentration time series and multi-source heterogeneous feature set with the same timestamp output by the benchmark sampling node. The multi-source heterogeneous feature set includes wind speed component, wind direction component and boundary layer height component. In the environment of topographic blockage and boundary layer height compression, pollutants generated by local emission sources accumulate in the region of the benchmark sampling node. At this time, the evolution trajectory of pollutant concentration values ​​in a single dimension shows characteristic homomorphism with the increase in regional background concentration. Traditional one-dimensional statistical extraction methods lack spatial dynamic constraints and are difficult to identify interference noise induced by local meteorological conditions at the concentration scalar level, resulting in non-true baseline drift in the data analysis results.

[0044] To eliminate the aforementioned noise, the processor extracts pollutant concentration values, along with corresponding wind speed, wind direction, and boundary layer height values ​​at the same time stamp, and constructs a multidimensional environmental state tensor representing the environmental state at that time stamp. The processor retrieves the historical context constraint envelope model from memory, which contains the centroid vectors of the high-dimensional manifold. and the inverse covariance matrix By aggregating multidimensional features, a physical relationship is established between concentration variability and atmospheric dynamic constraints. When the benchmark sampling node is in a combination state where the wind speed is below a preset threshold and the wind direction is located in a specific sector, even if each single-dimensional parameter does not touch the independent decision boundary, the deviation vector generated after its projection into the feature subspace will still deviate from the original value. The processor calculates the current multidimensional environment state tensor by defining the centroid of the distribution. Topological deviation distance relative to historical background baseline The specific calculation formula is as follows: In the formula, This represents the topological offset distance. For the first Multidimensional environmental state tensor at each sampling time Let the centroid vector of the historical background distribution be the centroid vector. is the inverse covariance matrix of the background distribution characteristics.

[0045] Based on the calculated topology deviation distance The processor calculates the Euclidean distance metric of the deviation vector in the topological space relative to the historical background baseline. And determine the normalized interference weight coefficients for the corresponding time slices. The normalized interference weight coefficient The rules for establishing this rule follow the formula below: ,in, For the first Normalized interference weight coefficients for each time slice, This is the disturbance sensitivity factor, with a value ranging from 0.5 to 1.5. For the first The processor normalizes the interference weight coefficients by taking the Euclidean distance metric of each deviation vector relative to the historical background baseline in the topological space. A steady-state statistical smoothing operator is injected to perform a weighted mean shift operation on the pollutant concentration time series. This weighting mechanism avoids time series gaps caused by hard data removal. When data loss due to long-term local pollution interference is detected, the system extracts the pressure gradient vector of the missing time period boundary and retrieves historical sequences with the same topological features from the database. The dynamic time warping algorithm is used to map the matching sequence to the missing position, completing the data chain repair and finally outputting a pollutant concentration background value sequence that represents the global trend of the region.

[0046] Example 2: At the atmospheric environment monitoring data computing center, a multi-source monitoring dataset was generated using an atmospheric boundary layer wind tunnel experimental platform. This platform adjusted the microscale terrain model and wind extraction system to simulate a monitoring environment of a closed canyon accompanied by temperature inversion. The experimental data included pollutant concentration time series from 3000 consecutive sampling points and a multi-source heterogeneous feature set with the same timestamp, with a sampling frequency set to 1 hour. To simulate sensor measurement bias, Gaussian white noise with a signal-to-noise ratio of 15 dB was superimposed on the concentration signal source. A perturbation sensitivity factor was also included. The setting is based on a balance between the strength of local short-term fluctuation suppression and the sensitivity of background evolution capture. When the value is below 0.5, the system's correction for concentration drift under weak wind speed combinations is insufficient. When the value is above 1.5, the background value extraction sequence exhibits a smooth state, losing the true large-scale concentration fluctuation characteristics of the region. For this typical canyon closure condition, [the following will be implemented / implemented]. Set to 1.0.

[0047] In the validation of pollutant concentration gradients, the concentration contribution rates of local emission sources to the baseline sampling nodes were set to 10%, 25%, and 40%, respectively. The processor extracted pollutant concentration values, wind speed, wind direction, and boundary layer height values, and performed covariance matrix decomposition in the feature subspace, where the principal component basis vectors retained a cumulative contribution rate of no less than 85%. In the sample group with a local source strength contribution rate of 40%, the processor calculated the multidimensional environmental state tensor at the current moment. Euclidean distance metric relative to historical background baseline The value is 3.86. The normalized interference weight coefficient is calculated based on the weight mapping formula. The value was 0.206. As a control group, the calculation method based on a single threshold judgment was used in an environment with a wind speed of 1.35 m / s. Since the wind speed was not lower than the cutoff boundary of 1.0 m / s, the system fully included the concentration value affected by pollution in the calculation of the background baseline mean, and the background value deviation rate was 22.45%. The background extraction deviation rate of the experimental group using the method of this invention was 1.38% under the same gradient. This mechanism of using multidimensional topological distance to quantify the degree of interference makes the background baseline physically self-consistent under the critical wind speed condition.

[0048] Stress tests are performed on the parameter boundaries to adjust the perturbation sensitivity factor. Set them to 0.4 and 1.7 respectively; When the value is 0.4, the normalized interference weight coefficient Under the same pollution shock, the corresponding value increased to 0.395, and the background value series changed synchronously with fluctuations in local emission sources, with the baseline error increasing from 1.38% to 9.67%; when When the parameter value is 1.7, the system's response delay to regional large-scale background elevation increases from 3.2 hours to 11.5 hours, verifying that the parameter range of 0.5 to 1.5 is the working window to ensure the background value extraction model achieves synergistic optimization between signal-to-noise ratio and sensitivity; the processor calculates the normalized interference weight coefficients for each time slice. A steady-state statistical smoothing operator was injected, and the concentration time series was corrected through weighted mean drift operation. The correlation coefficient between the background value series of the experimental group under three pollution gradients and the preset large-scale background trajectory was greater than 0.985, while the correlation coefficient of the control group was between 0.72 and 0.81. A multidimensional environmental state tensor was established in the feature subspace. The topological metric relationship with the historical context-constrained envelope model is used to remove nonlinear homomorphic noise induced by local meteorological microenvironment fluctuations; the interference weight coefficients are normalized. The method achieves weight adjustment of interference components, avoiding the failure of one-dimensional truncation under multi-dimensional correlation interference; the decrease in background extraction deviation rate reflects the role of multi-dimensional feature aggregation in the fidelity of environmental big data analysis; under the condition of maintaining the sampling hardware configuration unchanged, the method solves the baseline drift of regional background value extraction under local enrichment conditions by multi-source data fusion, and meets the calculation needs of regional trend sequences in the field of environmental monitoring.

[0049] Example 3: In the background benchmark establishment phase of a newly established monitoring station, when the system faces the problem of missing historical background data and the need to build a background constraint envelope model in the initial stage of operation, the processor receives the multi-source heterogeneous feature set of the monitoring station after phase alignment processing and stores it in the buffer. The processor then processes the initial operating cycle... The multidimensional environment state tensor undergoes variance normalization, and the processor calculates the sample covariance matrix at the initial time according to the following formula. : ,in, Let covariance matrix be the variance matrix. The initial number of samples. For the first Multidimensional environmental state tensor at each sampling time Given the sample mean vector, the processor processes this covariance matrix. Perform singular value decomposition to obtain the eigenvalues The diagonal matrix is ​​constructed, and the cumulative contribution rate is calculated in descending order of eigenvalues. When the cumulative contribution rate reaches 85%, the corresponding set of eigenvectors is defined as the basis vector space reflecting the macroscopic background characteristics of the region. This 85% threshold corresponds to the consistent characteristic distribution within the 95% confidence interval during atmospheric evolution. Local random disturbance components are stripped to the orthogonal complement space.

[0050] After establishing the basis vector space, the processor generates the centroid vector of the manifold by calculating the center position of the projection of each sampling point into this space. And simultaneously calculate the covariance matrix. inverse matrix To construct a metric benchmark in a high-dimensional topological space, and addressing the lack of historical observation records under absolute zero emission conditions in the early stages of monitoring network deployment, an alternative calibration procedure based on air mass trajectory tracking was initiated. Historical meteorological data from the nighttime valley periods of the benchmark sampling nodes, where the atmospheric boundary layer height was greater than a preset turbulent mixing height threshold, were extracted. Backward air mass trajectories were generated using a regional upper-level wind field inversion model. Air mass trajectory samples passing through areas with known fixed emission source coordinates were removed. The mean values ​​of each dimension of the multidimensional environmental state tensor corresponding to the retained air mass samples were extracted and written into memory as the centroid vector of the alternative manifold. To address background drift caused by seasonal changes in the monitoring environment, a sliding window mechanism was used to adjust the covariance matrix. Perform the update, specifically, each time the processor receives... A valid background slice is used to analyze the covariance matrix using newly added sample vectors. Weighted correction is implemented to shift the generated topological distance judgment boundary with the evolution of regional atmospheric characteristics. This process provides the background extraction algorithm with the ability to perceive long-term environmental changes and suppresses the deviation of calculation results caused by the solidification of model parameters. When the system detects that the residual component magnitude of the sampling point in the orthogonal complement direction of the feature subspace exceeds the preset physical consistency threshold, the pollutant concentration value at that moment is identified as an interference term affected by local disturbance.

[0051] The normalized interference weight coefficients for each time slice are calculated. The processor injects this into the kernel density estimation model within the steady-state statistical smoothing operator. This model uses a Gaussian kernel function to perform a locally weighted fit on the pollutant concentration time series, where the bandwidth of the kernel density is affected by the normalized disturbance weight coefficient. Dynamic adjustment, in During periods when the concentration value is below 0.3, the processor increases bandwidth to enhance the correction strength of the spatial physical constraints on the concentration value. During periods of near-pure background (approaching 1.0), the processor reduces bandwidth to preserve the fluctuation characteristics of large-scale background evolution. Based on the adaptive exponential scaling principle of the kernel density estimation algorithm, mathematical formulas are used... Real-time calculation of dynamic bandwidth for each time slice, where variables... The actual kernel function bandwidth value applicable to the current time slice; parameter The baseline time window is used as the initial value, determined based on the autocorrelation time scale of wind speed over the previous 30 natural days. The parameter k is the attenuation control coefficient. The root mean square error response curves under different pollution concentration levels in the historical dataset are extracted. The inflection point corresponding to the abrupt change in the slope of error growth is selected as the dimensionless coefficient setting value. The processor calculates the second derivative of the root mean square error response curve and locates the inflection point of error growth by identifying the extreme point where the sign of the second derivative reverses. This inflection point represents the critical state where the model transitions from stable fitting to divergent distortion. The corresponding ordinate value is determined as coefficient k after max-min normalization, ensuring that the kernel function bandwidth can generate sufficient nonlinear contraction when the pollution impact reaches the physical critical value. The system verifies the model state by comparing the background consistency deviation between the current output sequence and adjacent baseline sampling nodes. When the monitoring station experiences a continuous 48-hour period of light wind, the correlation coefficient between the reconstructed background value sequence and adjacent upper-level stations is not less than 0.96. The final generated background baseline sequence is sent to the environmental decision support platform. The processor utilizes the normalized interference weight coefficient when performing weighted mean drift calculation. By controlling the centroid movement step size of the sampling point in the feature space, a local feature circle is constructed with the pollutant concentration value at the current sampling time as the center. The normalized interference weight coefficients of all neighboring sampling points within the local feature circle are then calculated. The processor determines the steady-state center of the background distribution through iteration of the weighted mean partial vector under constraints, when the normalized disturbance weight coefficient at the corresponding sampling time is... When the interference threshold is lower than the preset threshold, the processor reduces the magnitude of the weighted mean bias vector to suppress the pulling effect of the sampling point on the centroid position at that moment. This weight-driven centroid search mechanism provides a closed-loop tracking of non-stationary background evolution in large-scale streaming data processing, avoiding background baseline drift caused by local high-concentration noise points. To obtain the baseline time window reference value for adaptive tuning, based on the principle of discrete-time signal frequency domain energy distribution, the first-order time rate of change, i.e., the discrete-time derivative, is obtained from the 30-day pollutant concentration background value sequence output by the weighted mean drift operation. The generated first-order time rate sequence is then subjected to a fast Fourier transform to generate a power spectral density distribution map. The power spectral density distribution map is scanned to extract the main frequency peak with the largest energy amplitude. The reciprocal of the frequency corresponding to the main frequency peak is determined as the actual evolution cycle duration of the large-scale weather system, and the duration value is written into the steady-state statistical smoothing operator configuration register to complete the baseline bandwidth update closed loop.

[0052] Example 4: During the network commissioning phase of the multi-sensor fusion atmospheric monitoring node, when the system faces the acquisition phase misalignment caused by the response delay of heterogeneous sensors, the processor corrects the time axis deviation between data according to the clock synchronization compensation operator, receives the original sampling sequences generated by the ultrasonic anemometer and the chemiluminescence pollutant concentration analyzer within the monitoring station, and retrieves the preset sensor physical hysteresis parameters from the register, including the transmission delay of the pollutant concentration analyzer's sample inlet pipeline. Based on the cross-sectional area of ​​the pipeline's inner diameter Pipeline length With real-time injection flow rate The determined parameters are calculated using the following formula: ,in, For the transmission delay of the sample injection line, The cross-sectional area of ​​the injection line is... This refers to the length of the injection line. To provide real-time sample flow rate, the processor adjusts the data based on the sample injection line transmission delay. The pollutant concentration time series is reverse-shifted to align the concentration data with the transient meteorological parameter vector on the same time reference to construct a multidimensional environmental state tensor.

[0053] When the system encounters deviations from the operating conditions in terms of topological criteria due to sensor zero-point drift or localized short-term sporadic interference, the processor corrects the model's metric boundary through a dynamic threshold calibration process for background consistency deviation. It retrieves historical datasets of regional clusters composed of adjacent reference sampling nodes from memory, calculates the residual component distribution density of the current monitoring station when the wind speed is between 2.0 m / s and 5.0 m / s and the vertical temperature difference is less than 0.2℃, and determines the 99th quantile of this distribution density as the physical consistency threshold. During the model iteration phase, the processor monitors the magnitude of the residual components at the sampling points relative to the physical consistency threshold. The deviation evolution trajectory is determined. When the second derivative of this deviation trajectory exceeds a preset stability factor for three consecutive sampling periods, the processor reconstructs the covariance matrix. The non-stationary change characteristics of the monitored environment are introduced into the background constraint envelope model.

[0054] Example 5: In the deployment phase of a regional monitoring network encompassing both plains and mountains, when the system faces the challenge of aligning the original feature sets due to differences in sampling frequencies from heterogeneous sensors, the processor invokes a resampling operator to reconstruct the time axis of the multi-source heterogeneous feature sets. It receives a wind speed vector sequence with a sampling period of 5 minutes and a pollutant concentration time series with a sampling period of 60 minutes. A linear interpolation algorithm is used to upsample the low-frequency sampling sequence to a 5-minute baseline. The processor then uses a dynamic time warping operator to calculate the minimum value of the path cost function within a 300-minute sliding time window to determine the time axis stretching mapping relationship. By Operating on multi-source heterogeneous feature sets, the processor executes a dimensional decoupling procedure based on the temporal dynamic attributes and spatial static attributes of the feature parameters. For time-varying vector data representing real-time atmospheric dynamic fields such as wind speed and direction components, and time-varying data on emission intensity from surrounding fixed emission sources, it applies a time-axis stretching mapping relationship f(t) to perform sequence phase alignment. For spatial static scalar data representing the local topographic relief of the benchmark sampling node, it bypasses the dynamic time warping operator and directly injects it as a spatial penalty weight coefficient into the corresponding feature dimension of the time-series feature matrix after time resolution unification. This avoids algorithm divergence caused by nonlinear stretching operations on physical constants without time evolution phases. The aforementioned spatial penalty... The specific transformation process of the weighting coefficients is as follows: The processor pre-extracts the topographic elevation variance within a 1km radius around the site as a scalar of undulation, and uses an S-shaped logistic regression function to map it to the [0,1] interval. When constructing the feature matrix, this coefficient serves as a fixed gain for that dimension of the feature. The higher the undulation, the larger the weighting coefficient is assigned, thereby strengthening the explanatory power of the topographic occlusion effect on concentration fluctuations in subsequent covariance decomposition. The processor generates a feature matrix with a uniform time resolution. To eliminate the physical difference in magnitude between the wind speed component and the pollutant concentration component, variance normalization is performed on the feature matrix, and the average value of each dimension of the feature over the past 30 days is calculated. with standard deviation and each original observation value Convert to dimensionless eigencomponents This process establishes the dimensional equivalence of the multidimensional environment state tensor when measured within the feature subspace.

[0055] When a monitoring network is being operated for the first time in a specific geographical environment and it is necessary to determine the physical consistency threshold. At that time, the processor uses offline statistical calibration to determine the starting benchmark of the topology criterion. The processor retrieves historical operation slices of the monitoring station under wind speed conditions above 3.0 m / s and calculates the multidimensional environmental state tensor corresponding to each slice. Relative to the center of the background distribution The processor calculates the topology deviation distance by performing a three-parameter gamma distribution fit on the set of topology deviation distances, and determines the distance value corresponding to the cumulative probability of 0.95 as the physical consistency threshold. During system operation, monitor the topological deviation distance of the multidimensional environmental state tensor at the current moment. Physical consistency threshold The processor maintains the current covariance matrix when the deviation ratio falls within the range of 0.8 to 1.2 for five consecutive sampling times. When the coefficient of variation of the deviation ratio increases to above 0.35 within 12 hours, the processor reconstructs the feature basis vector space. This process matches the background constraint envelope model with the atmospheric diffusion characteristics of a specific geographical location. Before performing dimensional aggregation, the processor receives real-time emission intensity values ​​from surrounding fixed emission sources and calculates the initial contribution component of the emission source to the monitoring node using a spatial distance attenuation model. The initial contribution component is then cross-multiplied with the real-time wind speed component and the boundary layer height component to generate a synthetic dimensional component characterizing the local diffusion potential. The processor then performs vector concatenation of the synthetic dimensional component with the topographic relief component and other meteorological dynamic components to construct a multidimensional environmental state tensor. This procedure establishes the collaborative expression of static geographical elements and dynamic source intensity behavior in the feature subspace, enabling the background constraint envelope model to identify the concentration homomorphic feedback caused by topographic masking and emission fluctuations, and establishing the physical correlation of multi-source data in topological spatial measurement.

[0056] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data, characterized in that, Includes the following steps: Step S101: Obtain the target parameter time series data of the reference sampling node, and obtain the multi-source heterogeneous feature set that is aligned with the acquisition phase of the target parameter time series data in the time dimension; Step S102: The target parameter time series data and multi-source heterogeneous feature set are subjected to dimensional aggregation and variance normalization processing to construct a multi-dimensional environmental state tensor representing the dynamic evolution trajectory of the local microenvironment of the benchmark sampling node in a multi-dimensional feature coordinate system. Step S103: The covariance matrix decomposition of the multidimensional environmental state tensor is performed using the feature subspace projection algorithm to extract the basis vector components that reflect the macro background fluctuation characteristics. The multidimensional environmental state tensor is mapped to a high-dimensional topological space. The deviation vector of each time slice in the high-dimensional topological space relative to the preset historical background baseline is calculated. Based on the negative correlation mapping function between the Euclidean distance magnitude of the deviation vector and the interference weight, the normalized interference weight coefficient of the corresponding time slice is determined. Step S104: The normalized disturbance weight coefficient is injected into the preset steady-state statistical smoothing operator as the iterative learning weight. The steady-state statistical smoothing operator is used to perform weighted mean drift operation on the target parameter time series data to remove the nonlinear disturbance features induced by local micro-topography and meteorological dynamic field represented by the multi-source heterogeneous feature set. The pollutant concentration background value sequence reflecting the regional macro trend is output by the weighted mean drift operation.

2. The method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, The multi-source heterogeneous feature set includes scalar data characterizing the local topographic relief of the reference sampling node, vector data characterizing the real-time atmospheric dynamic field radius change, and time-varying data characterizing the emission intensity of known surrounding fixed emission sources; step S101 includes: applying nonlinear axial stretching to the target parameter time series data and the multi-source heterogeneous feature set through a dynamic time warping operator to eliminate the acquisition phase difference caused by sensor heterogeneity and generate a feature matrix under a unified time scale.

3. The method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, The rule for establishing the normalized interference weight coefficient in step S103 follows the formula below: ,in, For the first Normalized interference weight coefficients for each time slice; The perturbation sensitivity factor is a preset value for the signal-to-noise ratio based on a multi-source heterogeneous feature set, and its value ranges from 0.5 to 1.

5. For the first The Euclidean distance metric of each deviation vector relative to the historical background baseline in a high-dimensional topological space.

4. The method for calculating background pollutant concentration based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, In step S103, when performing covariance matrix decomposition, the principal component basis vectors with a cumulative contribution rate greater than 85% are retained as feature basis vectors by calculating the eigenvalues ​​of the multidimensional environmental state tensor in each dimension. The deviation vector is defined as the residual component of the current time slice in the direction of the orthogonal complement of the feature subspace formed by the feature basis vectors, and is used to quantify the degree of influence of local disturbances on the reference sampling node.

5. The method for calculating background pollutant concentration based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, The steady-state statistical smoothing operator adopts a kernel density estimation algorithm with physical property boundary constraints; step S104 includes: injecting the normalized disturbance weight coefficient into the kernel function window of the kernel density estimation algorithm, and dynamically adjusting the contribution ratio of the corresponding time slice in the probability density evolution path so that the weight of the time slice disturbed by nonlinear disturbance characteristics in the calculation of pollutant concentration background value sequence monotonically decreases as the magnitude of the deviation vector increases.

6. The method for calculating background pollutant concentration based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, After step S104, the method further includes the following repair steps: when a data gap is detected in the background value sequence of pollutant concentration due to long-term local pollution interference, the pressure gradient vector difference in the vector data of the missing time window boundary is extracted; continuous data segments that satisfy the same topological features are retrieved in the historical background database, and the matching data segments are mapped to the missing position of the sequence through a dynamic time warping algorithm to complete the data chain repair.

7. The method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, The historical background baseline was established by retrieving historical observation records of the benchmark sampling node under calm wind conditions and with surrounding fixed emission sources in a zero-emission state. The static distribution centroid of the multidimensional environmental state tensor under the corresponding conditions was extracted and defined as the origin of the coordinates in the high-dimensional topological space, serving as a benchmark reference point for measuring the shift in real-time state.

8. The method for calculating background pollutant concentration based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, In step S104, when performing the weighted mean drift operation, a preset convergence threshold is set. When the absolute deviation between the background values ​​generated by two adjacent iterations is less than the convergence threshold, the iterative learning process is stopped, and the value generated in the current iteration step is determined as the final pollutant concentration background value sequence to ensure the numerical stability of the background value extraction logic when processing high-dimensional non-stationary data streams.

9. The method for calculating background pollutant concentrations based on benchmark sampling nodes and multi-source data according to claim 1, characterized in that, Step S104 includes: calculating the first-order time variability of the pollutant concentration background value sequence to identify the evolution cycle of large-scale weather systems; and adaptively adjusting the smoothing time window width of the steady-state statistical smoothing operator based on the duration of the evolution cycle to ensure that the background value calculation logic can effectively cover the variation law of the macro-meteorological field and suppress local micro-scale noise at different seasonal scales.

10. The method for calculating background pollutant concentration based on benchmark sampling nodes and multi-source data according to claim 3, characterized in that, The method also includes the following closed-loop calibration steps: performing residual subtraction analysis on the pollutant concentration background value sequence and the original target parameter time series data to obtain the difference sequence characterizing the contribution of local disturbances, and performing dynamic calibration on the value of the disturbance sensitivity factor in the formula based on the second moment fluctuation characteristics of the difference sequence to compensate for the weight allocation deviation caused by abrupt changes in environmental state.

Images