Urban pollution reduction and carbon reduction collaborative evaluation method and system based on multi-source data fusion

Abstract

The invention discloses an urban pollution and carbon reduction collaborative evaluation method and system based on multi-source data fusion, and the method achieves the on-demand collection through a multi-source data collection rule set through a historical data 3 sigma principle and a dynamic collection mechanism triggered by a real-time event, remarkably reduces the redundancy of unnecessary data, saves the storage and transmission resources, and improves the efficiency. Meanwhile, it is ensured that key data such as sudden pollution events can be efficiently captured, and the value density of the data is improved; through space-time grid mapping and cross validation and calibration based on statistical characteristics, the problem that space-time references of multi-source data are not uniform is fundamentally solved, a high-precision and space-time consistent standardized data set is generated, and a reliable data basis is provided for subsequent analysis; the cooperative effect coefficient and the efficiency grade evaluation result calculated based on the high-quality data are high in accuracy and high in credibility, the cooperative conditions and weak links of pollution reduction and carbon reduction of different areas of the city are visually revealed, and finally the city is assisted to achieve the purpose of pollution reduction and carbon reduction cooperative interaction.

Description

Technical Field

[0001] This application relates to the technical field of environmental protection and carbon emission control, and in particular to a collaborative assessment method and system for urban pollution reduction and carbon reduction based on multi-source data fusion. Background Technology

[0002] Air pollutants and greenhouse gas emissions share the same origin and are highly synergistic. Therefore, scientifically and accurately assessing the synergistic effects of pollution reduction and carbon reduction at the city scale is crucial for formulating effective environmental policies.

[0003] Existing assessment methods mostly rely on single or limited data sources (such as data from ground monitoring stations) and employ traditional statistical analysis or model simulation techniques. These methods have significant limitations: First, at the data level, the acquisition patterns of multi-source heterogeneous data (such as ground sensors, remote sensing satellites, and mobile monitoring vehicles) are rigid, often involving fixed-frequency acquisition. This leads to data redundancy during non-critical periods, wasting storage and computing resources. Furthermore, insufficient acquisition frequency results in the loss of critical data when abnormal events occur. Simultaneously, these data are inconsistent in their spatiotemporal references, exhibiting issues such as sensor timestamp discrepancies and mismatched monitoring point locations. Without effective alignment and calibration during preprocessing, large fusion errors can occur, severely impacting data quality.

[0004] Secondly, at the data processing level, traditional outlier identification algorithms (such as isolated forest and 3σ principle) often ignore the spatiotemporal distribution characteristics of data. For example, there are significant differences in the background values ​​of pollutants between industrial areas and residential areas. Using a globally uniform threshold for anomaly detection results in a high misjudgment rate, which further contaminates the data foundation.

[0005] Finally, at the evaluation level, due to the existence of quality defects, gaps, and anomalies in the underlying data, the reliability and accuracy of the subsequent quantitative evaluation results of synergistic effects are difficult to guarantee, and cannot provide truly effective decision support for the refined and intelligent environmental management of cities. Summary of the Invention

[0006] This application provides a collaborative assessment method and system for urban pollution reduction and carbon reduction based on multi-source data fusion, in order to solve the problems of rigid patterns, low fusion efficiency and poor quality of multi-source datasets mentioned in the background technology.

[0007] To address the aforementioned technical problems, firstly, this application provides a method for coordinated assessment of urban pollution reduction and carbon reduction based on multi-source data fusion, comprising: Based on the 3σ principle of historical data, multi-source data acquisition rules and real-time event triggering are set to dynamically adjust the data acquisition frequency of various monitoring sensors. The multi-source data is processed using a spatiotemporal interpolation gridding method to uniformly map all the multi-source data onto a basic grid cell with a standard timestamp and a unified geographic coordinate system; and cross-validation and bias calibration are performed using the statistical characteristics of different data located in the same basic grid cell to generate a standardized dataset with consistent spatiotemporal reference. Based on the standardized dataset, an anomaly detection algorithm that integrates spatiotemporal context features is used to identify outliers; for the identified outliers, a spatiotemporal tensor completion algorithm based on Tucker decomposition is used to reconstruct the data and generate a high-quality fused dataset. Based on the high-quality fusion dataset, the synergistic effect coefficient of air pollutants and greenhouse gas emissions is calculated, the synergistic effectiveness level of pollution reduction and carbon reduction in different regions is evaluated, and a visual evaluation report is generated.

[0008] In one embodiment, the multi-source data acquisition rule based on the 3σ principle of historical data is as follows: Calculate the mean μ and standard deviation σ of the data collected by any monitoring sensor under the same historical spatiotemporal conditions, and set the data fluctuation range of the monitoring sensor at the normal acquisition frequency as (μ-3σ, μ+3σ); When the real-time collected data remains within the data fluctuation range, it is determined to be in a steady state and the first collection frequency is automatically maintained; when the real-time data exceeds the data fluctuation range, it immediately switches to the preset second collection frequency.

[0009] In one embodiment, the multi-source data is processed using a spatiotemporal interpolation gridding method. The method for uniformly mapping all the aforementioned multi-source data onto basic grid cells with standard timestamps and a unified geographic coordinate system is as follows: Based on the geographical scope and time evaluation period of the target city, a three-dimensional spatiotemporal basic grid framework is constructed. The X-axis and Y-axis of the basic grid framework are divided based on a unified geographic coordinate system to form a regular geographic planar grid. The Z-axis is the time axis, which is based on a standard timestamp and divided according to a preset time resolution to generate a series of equally spaced continuous time slices. Each raw data point from different data sources is associated with the corresponding basic grid cell in the basic grid framework based on its spatial coordinate information and timestamp information; for spatial coordinates, the data point is assigned to the geographic plane grid where the data point is located; for timestamps, the data point is assigned to the specific time slice to which the data point belongs. Traverse all basic grid cells, identify empty grid cells that do not contain any original data values, and generate a list of empty grid cells; For each cell in the list of missing grid cells, the spatiotemporal kriging interpolation algorithm is used to calculate the interpolation estimate of the missing grid cell and fill it in; Each basic grid cell contains a data value, generating a gridded dataset with a unified spatiotemporal reference. The data value includes the original data value or the interpolated estimate.

[0010] In one embodiment, the method for filling data using a spatiotemporal kriging interpolation algorithm for each cell in the list of missing grid cells is as follows: Based on the data in the non-empty grid cells, a spatiotemporal variogram model describing the spatial and temporal correlation of the data is calculated and fitted. Centered on the missing grid cell to be interpolated, all non-missing grid cells within a preset spatiotemporal neighborhood of the missing grid cell are selected as reference points; using the spatiotemporal variogram model, the weight of each reference point to be interpolated is calculated, where the weight represents the spatiotemporal correlation strength between the reference point and the point to be interpolated; The interpolation estimate of the missing grid cell is obtained by multiplying the data value of each reference point by its corresponding weight and then summing the results.

[0011] In one embodiment, the method for generating a standardized dataset with consistent spatiotemporal reference by using the statistical characteristics of different data located within the same basic grid cell for cross-validation and bias calibration is as follows: The different data sources are defined as the baseline data source and the data source to be calibrated, respectively. For each base grid cell that has undergone spatiotemporal interpolation, extract the data values ​​from all data sources existing within the same spatiotemporal cell; the data values ​​from the data sources include values ​​from the reference data source and values ​​from at least one of the data sources to be calibrated. For each pair of data source combinations of the reference data source and the data source to be calibrated, statistical characteristics of their data values ​​within multiple spatiotemporal grid cells are calculated to quantify systematic bias; the statistical characteristics include: mean bias and linear regression parameters; The average deviation is calculated as the average difference between the data source to be calibrated and the reference data source across all matching grid cells. The linear regression parameters are obtained by performing a univariate linear regression with the data values ​​of the data source to be calibrated as the independent variable and the data values ​​of the benchmark data source as the dependent variable, resulting in two parameters: slope and intercept. Based on the statistical characteristics, a calibration model is constructed for each type of data source to be calibrated; the calibration model adopts the following formula: Calibrated value = (Original value to be calibrated - Intercept) / Slope; Or, the calibrated value = the original value to be calibrated - the average deviation; The calibration model is applied to all data points of the data source to be calibrated within the entire domain, and the data source to be calibrated is batch-corrected to eliminate the systematic deviation between the data source to be calibrated and the reference data source. The calibration data from all calibrated different data sources, as well as the baseline data from data sources that do not require calibration, are integrated into the spatiotemporal basic grid framework to generate a standardized dataset that eliminates inter-source systematic biases and is consistent with the spatiotemporal baseline.

[0012] In one embodiment, the method for identifying outliers based on the standardized dataset and employing an anomaly detection algorithm that incorporates spatiotemporal context features is as follows: Each basic grid cell in the standardized dataset is abstracted into a graph node, and a spatiotemporal correlation graph is constructed based on the spatial proximity and temporal correlation between grid cells; A feature vector is constructed for each graph node. The feature vector includes the target monitoring value of the grid cell at the current moment, the sequence of historical monitoring values ​​within a preset time window, and the characteristics of its spatial location. The feature vector to be detected is input into the trained spatiotemporal graph neural network model to obtain the predicted value of each grid cell data value; the reconstruction error between the actual value of each grid cell data value and the model prediction value is calculated, and the reconstruction error is standardized and used as the anomaly score of the grid cell. Grid cells with abnormal scores exceeding the dynamic threshold of the grid cells are marked as abnormal values, and the time, spatial location, and abnormal score of the abnormal values ​​are output; grid cells with abnormal scores less than the dynamic threshold of the grid cells are marked as valid values.

[0013] In one embodiment, the method for reconstructing the identified outliers using a spatiotemporal tensor completion algorithm based on Tucker decomposition to generate a high-quality fused dataset is as follows: Outliers in the standardized dataset are used to construct a third-order spatiotemporal data tensor X∈ R^(I×J×K) according to the three dimensions of longitude, latitude, and time; where I represents the number of grid cells in the longitude direction, J represents the number of grid cells in the latitude direction, and K represents the total number of time slices; the positions in the spatiotemporal data tensor X that have been marked as outliers are initialized as missing values; The original spatio-temporal data tensor X is Tucker decomposed into the product of a core tensor g ∈ R^(P×Q×R) and three factor matrices (A, B, C). The factor matrices A ∈ R^(I×P), B ∈ R^(J×Q), and C ∈ R^(K×R) correspond to the longitude, latitude, and time patterns respectively; the dimensions P, Q, R of the core tensor are automatically determined by the minimum description length criterion in the alternating least squares iterative process and satisfy P << I, Q << J, R << K to achieve high compression and feature extraction of the original data; The core tensor g and the factor matrices A, B, C are simultaneously solved by minimizing an objective function, thereby completing the filling of missing values; The alternating least squares method is used to solve the objective function. Fix two of the factor matrices and the core tensor, update the other factor matrix, and iterate alternately in turn until the change in the objective function value is less than a preset convergence threshold or the maximum number of iterations is reached.

[0014] Substitute the optimized optimal core tensor g and factor matrices A, B, C into the Tucker decomposition model, and calculate the complete reconstructed tensor X'_rec = g×A×B×C; all positions in the reconstructed tensor X'_rec are filled with estimated values reconstructed based on spatio-temporal global information; Map the data values in the reconstructed tensor X'_rec back to the original spatio-temporal base grid framework, replace the originally marked outliers, and jointly form the final high-quality fusion dataset with the valid values not marked as outliers.

[0015] In one embodiment, the specific calculation formula for calculating the synergy coefficient is: ; where S represents the synergy coefficient, ΔC and ΔP respectively represent the emission changes of greenhouse gases and air pollutants over a period of time, and C0 and P0 are the baseline emissions; when S > 0, it indicates synergy between pollution reduction and carbon emission reduction; when S < 0, it indicates a trade-off relationship.

[0016] In a second aspect, the present application also provides a system for collaborative assessment of urban pollution reduction and carbon emission reduction based on multi-source data fusion, which is applied to the system in the method for collaborative assessment of urban pollution reduction and carbon emission reduction based on multi-source data fusion. The system includes: A dynamic acquisition control module, which is used to communicate with the urban monitoring sensor network, generate instructions to dynamically adjust the data acquisition frequencies of various environmental and energy monitoring sensors based on the multi-source data acquisition rules set by the historical data 3σ principle and real-time event triggers; The spatiotemporal alignment and calibration engine module is connected to the dynamic acquisition and control module. It receives raw data from multiple sources and uses a spatiotemporal interpolation gridding method to uniformly map all data onto a basic grid cell with a standard timestamp and a unified geographic coordinate system. It also uses the statistical characteristics of different data located in the same basic grid cell to perform cross-validation and deviation calibration, generating a standardized dataset with consistent spatiotemporal reference. The data quality improvement module, connected to the spatiotemporal alignment and calibration engine module, includes a spatiotemporal context anomaly detection unit and a spatiotemporal tensor completion unit. The anomaly detection unit is configured to identify outliers based on the standardized dataset using an algorithm that incorporates spatiotemporal context features. The tensor completion unit is configured to reconstruct the identified outliers using a spatiotemporal tensor completion algorithm based on Tucker decomposition, thereby generating a high-quality fused dataset. The pollution reduction and carbon reduction synergistic assessment module is connected to the data quality improvement module. Based on the high-quality fusion dataset, it quantifies the synergistic effect coefficient of air pollutants and greenhouse gas emissions, assesses the synergistic effectiveness level of pollution reduction and carbon reduction in different regions, and generates a visual assessment report.

[0017] In one embodiment, the spatiotemporal alignment and calibration engine module includes a spatiotemporal gridding unit and a data calibration unit; Spatiotemporal gridded units are used to establish a unified basic geographic grid and perform interpolation mapping on multi-source data; The data calibration unit is used to perform cross-validation and bias correction on multi-source data within the same grid.

[0018] Compared with the prior art, this application has at least the following beneficial effects: 1. By setting multi-source data acquisition rules based on the 3σ principle of historical data and a dynamic acquisition mechanism triggered by real-time events, on-demand data acquisition was achieved, significantly reducing the redundancy of unnecessary data, saving storage and transmission resources, and ensuring that key data can be efficiently captured during sudden pollution events, thereby increasing the value density of the data. 2. By using spatiotemporal mesh mapping and cross-validation and calibration based on statistical features, the problem of inconsistent spatiotemporal benchmarks for multi-source data is fundamentally solved, generating a high-precision, spatiotemporally consistent standardized dataset, which provides a reliable data foundation for subsequent analysis; 3. An anomaly detection algorithm that integrates spatiotemporal context features is adopted, which can dynamically adapt to the features of different functional areas and different times, effectively distinguishing normal activities from real abnormal events, and greatly reducing the false positive rate and false negative rate. 4. The spatiotemporal tensor completion algorithm based on Tucker decomposition can intelligently reconstruct outliers and missing values ​​by utilizing the global spatiotemporal correlation of data, generating a high-quality fusion dataset that is blank-free, outlier-free, smooth and continuous, thus completely solving the constraint of data quality issues on subsequent analysis. 5. The synergy effect coefficient and efficiency level assessment results based on data calculation are highly accurate and reliable. Combined with visualization reports, they can clearly and intuitively reveal the synergy status and weaknesses of pollution reduction and carbon reduction in different urban areas. This provides a strong scientific basis for managers to formulate precise and efficient synergy control measures, and ultimately helps cities achieve the goal of synergistic efficiency improvement in pollution reduction and carbon reduction. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the collaborative assessment method for urban pollution reduction and carbon reduction based on multi-source data fusion, as shown in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a collaborative assessment system for urban pollution reduction and carbon reduction based on multi-source data fusion, as shown in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein in the description of a method and system for coordinated assessment of urban pollution reduction and carbon reduction based on multi-source data fusion is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] This application provides a collaborative assessment method for urban pollution reduction and carbon reduction based on multi-source data fusion, including... S1. Based on the 3σ principle of historical data, multi-source data acquisition rules and real-time event triggering are set to dynamically adjust the data acquisition frequency of various monitoring sensors; The multi-source data acquisition rules based on the 3σ principle of historical data are as follows: S11. Calculate the mean μ and standard deviation σ of the data collected by any monitoring sensor under the same historical spatiotemporal conditions, and set the data fluctuation range of the monitoring sensor at the normal collection frequency as (μ-3σ, μ+3σ). S12. When the real-time collected data is continuously within the data fluctuation range, it is determined to be in a steady state and the first collection frequency is automatically maintained; when the real-time data exceeds the data fluctuation range, it is immediately switched to the preset second collection frequency.

[0023] Multi-source data includes environmental monitoring data, remote sensing and geospatial data, and emissions and energy consumption data; environmental monitoring data includes PM2.5, SO2, and NO collected by urban ambient air quality automatic monitoring stations and national / provincial monitoring points. x Concentrations of pollutants such as CO (unit: μg / m³ or mg / m³), meteorological parameters such as temperature, humidity, wind speed, and air pressure; also includes high spatiotemporal resolution pollutant concentrations (e.g., 1km grid / 1 minute sampling) and road dust emission data collected by mobile monitoring vehicles equipped with sensors and drone inspections; used to supplement blind spots of fixed stations and capture local pollution sources (e.g., construction sites, illegal discharges by enterprises). Remote sensing and geospatial data include satellite remote sensing data and geographic information data. Satellite remote sensing data, for example, includes atmospheric pollutant column concentrations (such as NO2 column concentration), surface temperature, vegetation cover, and land use types. It provides spatial distribution characteristics at a regional scale (such as 10km × 10km), compensating for the spatial limitations of ground-based monitoring. Geographic information data, such as functional zoning (industrial / residential / commercial) in 1km × 1km grid units, road network distribution, and pollution source locations (such as factory and power plant coordinates), provides spatial attribute labels for spatiotemporal interpolation and collaborative assessment.

[0024] Emissions and energy consumption data include industrial emissions data, such as SO2 and NO. x Emissions such as CO2, and production process parameters (such as fuel type and production capacity).

[0025] In a normal distribution, 99.73% of the data will fall within the ranges μ-3σ and μ+3σ (μ = mean, σ = standard deviation). The 3σ principle applies the statistical law of normal distribution to sensor acquisition and control. It determines whether the data is in a steady state or a non-steady state by whether the data is within this range, and dynamically adjusts the acquisition frequency to achieve on-demand acquisition.

[0026] For example, a PM2.5 sensor under the conditions of "winter weekday morning rush hour + industrial area + calm wind": Historical data statistics: μ=85μg / m³, σ=12μg / m³; fluctuation range=(85-3×12, 85+3×12)=(49μg / m³, 121μg / m³).

[0027] For example, in a steady-state situation: if real-time data remains within (μ-3σ, μ+3σ) for three consecutive periods with a change rate ≤ 5%, then the first sampling frequency is used, for example, once per hour. If the data exceeds this range, or the change rate is > 10%, for example, PM2.5 changes from 60 to 130 μg / m³, then the second sampling frequency is used, immediately switching to once per minute for 30 minutes. For real-time events: for example, if the predicted daily average PM2.5 is > 150 μg / m³ and lasts for more than 72 hours, or the hourly O3 concentration is > 200 μg / m³ and lasts for more than 8 hours, then the second sampling frequency is used, immediately switching to once per minute for 30 minutes.

[0028] S2. The multi-source data is mapped to a basic grid cell with a standard timestamp and a unified geographic coordinate system by using a spatiotemporal interpolation gridding method. The method of using spatiotemporal interpolation gridding to uniformly map all the multi-source data onto basic grid cells with standard timestamps and a unified geographic coordinate system is as follows: S21. Based on the geographical scope and time evaluation period of the target city, construct a three-dimensional spatiotemporal basic grid framework; wherein, the X-axis and Y-axis of the basic grid framework are divided based on a unified geographic coordinate system to form a regular geographic planar grid; the Z-axis is the time axis, which is based on a standard timestamp and divided according to a preset time resolution to generate a series of equally spaced continuous time slices. To address the issue of inconsistent spatiotemporal references for multi-source data, a standard, unified coordinate system is created, mapping all data from different sources into this coordinate system, thereby achieving data alignment and fusion.

[0029] For example, the geographical scope of the target city: We selected the main urban area of ​​Beijing (approximately 39.7°-40.1° North latitude, 116.1°-116.6° East longitude) as the study area.

[0030] Time evaluation period: The selected time range is from 00:00:00 on September 1, 2023 to 23:00:00 on November 30, 2023.

[0031] Unified geographic coordinate system: CGCS2000 was selected as the unified geographic coordinate system, ensuring the consistency of the basis of all spatial data.

[0032] Spatial resolution (grid size): Set the size of each geographic grid to 2 km x 2 km. This is a commonly used resolution to balance accuracy and computational cost in urban-scale studies.

[0033] Time resolution: Set the smallest unit (time slice) on the time axis to 1 hour. This is a commonly used resolution for capturing the daily variation patterns of pollutants and carbon emissions.

[0034] Standard timestamp: UTC+8 (Beijing time) is used as the standard timestamp.

[0035] The east-west span of Beijing's main urban area is approximately (116.6 - 116.1) = 0.5 degrees of longitude, or about 55 kilometers. The north-south span is approximately (40.1 - 39.7) = 0.4 degrees of latitude, or about 44 kilometers.

[0036] Division: Divided into regular grids of 2km x 2km.

[0037] The X-axis (longitude direction) needs to be divided into: 55 km / 2 km = 27.5 > 28 grids. The Y-axis (latitude direction) needs to be divided into: 44 km / 2 km = 22 grids. This results in a planar grid covering the main urban area of ​​Beijing, consisting of 28 x 22 = 616 regular geographic grid cells. Each grid cell has a unique spatial index, such as (i, j), where i is the longitude number (1 to 28) and j is the latitude number (1 to 22).

[0038] The evaluation period is 3 months (30 days in September, 31 days in October, and 30 days in November), totaling 91 days.

[0039] Division: Divided into 1-hour time slices.

[0040] Total number of time slices: 91 days * 24 hours / day = 2184 hours. This results in a timeline consisting of 2184 consecutive time slices, starting from 00:00:00 on September 1, 2023, and ending at 23:00:00 on November 30, 2023. Each time slice has a unique time index k, corresponding to a specific timestamp. Combining the above 2D geographic grid and 1D timeline forms a three-dimensional, discretized spatiotemporal data cube. This cube consists of 28 (longitude) x 22 (latitude) x 2184 (time) = 1,345,344 basic spatiotemporal units. Each basic spatiotemporal unit can be located using a unique coordinate (i, j, k), representing a specific geographic location (such as a 2x2km grid) and a specific point in time (such as the start time of a certain hour).

[0041] S22. For each raw data point from different data sources, associate it with the corresponding basic grid cell in the basic grid framework according to its spatial coordinate information and timestamp information; for spatial coordinates, assign the data point to the geographic plane grid where the data point is located; for timestamps, assign the data point to the specific time slice to which the data point belongs.

[0042] For example, a three-dimensional spatiotemporal basic grid framework: Spatial grid: the main urban area of ​​Beijing, CGCS2000 coordinate system, 2 km x 2 km grid, a total of 28 (longitude) x 22 (latitude) grids.

[0043] Time slice: The time resolution is 1 hour, and the evaluation period is Fall 2023 (September-November).

[0044] We selected the time slice k = 1105, which corresponds to the hour from 13:00:00 on October 15, 2023 to 13:59:59 on October 15, 2023 in Beijing time.

[0045] Select the grid with index (i=15, j=10), assuming its geographical range is: longitude: 116.30°E ~ 116.32°E, latitude: 39.90°N ~ 39.92°N.

[0046] Table 1 Detection values ​​of the grid ; As shown in Table 1, through the above process, these three original data points from different sources and with different accuracies are associated with the same standardized basic grid cell. This achieves data normalization: discretizing and standardizing continuous, heterogeneous original data into a unified basic grid cell, enabling subsequent processing. It creates the prerequisite for "cross-validation and bias calibration." As shown in Table 1, there are three values ​​in the basic grid cell (15, 10, 1105). The system can calculate the average and variance of these three values, or use the data from the national control station as a benchmark to calibrate the values ​​from satellites and mobile monitoring vehicles.

[0047] S23. Traverse all basic grid cells, identify empty grid cells that do not contain any original data values, and generate a list of empty grid cells. For example, spatially, there are 28 (longitude) x 22 (latitude) = 616 grids; temporally, the total number of grid cells over 2184 hours is 28 x 22 x 2184 = 1,345,344. For each grid cell being accessed, the system performs a check: after the "data association" step, it checks whether any original data value has been assigned to this cell. If at least one data value is present, the cell is marked as "data present." If no data value is present, the cell is marked as "missing." This clearly identifies the location of the missing data, providing the data completion algorithm with a clear target list. It eliminates the need to perform calculations on the entire grid, saving significant computational resources and accelerating processing.

[0048] S24. For each cell in the list of missing grid cells, the method for calculating the interpolation estimate of the missing grid cell and filling it using the spatiotemporal kriging interpolation algorithm is as follows: S241. Based on the data in the non-empty grid cells, calculate and fit a spatiotemporal variogram model that describes the spatial and temporal correlation of the data. S242. Taking the missing grid cell to be interpolated as the center, select all non-missing grid cells within a preset spatiotemporal neighborhood of the missing grid cell as reference points; use the spatiotemporal variation function model to calculate the weight of each reference point to be interpolated, where the weight represents the spatiotemporal correlation strength between the reference point and the point to be interpolated. S243. Multiply the data values ​​of each reference point by their corresponding weights and sum them to obtain the interpolation estimate of the missing grid cell.

[0049] The spatiotemporal variogram model is a mathematical model that can describe how the data changes in space and time based on existing data. First, the known data points in the spatiotemporal range around the missing point are found, and the weight of each known point in estimating the value of the missing point is calculated using the above model. The values ​​of the surrounding known points are then weighted and averaged to obtain the final and optimal estimate.

[0050] The spatiotemporal variogram model is γ(h, Δt) = γ_Spatial(h) + γ_Temporal(Δt), where γ_Spatial(h) is the pure spatial variogram and γ_Temporal(Δt) is the pure temporal variogram.

[0051] The spatiotemporal distance (h, Δt) between each reference point and the point to be interpolated is calculated using a system of equations to determine the weight γi of each reference point.

[0052] γi: ; sum of weights Reference points that are spatially close and have short time intervals have higher weights. For example, a reference point that is 1km away and has a time difference of 1 hour has a weight of 0.4.

[0053] Interpolation estimate calculation formula: For example, a weighted summation of three reference points: =0.4*62 + 0.35*58 + 0.25*55 = 58.45 m3.

[0054] S25. Each basic grid cell contains a data value, generating a gridded dataset with a unified spatiotemporal reference. The data value includes the original data value or the interpolated estimate.

[0055] If a grid cell already has an original data value after data association, that original value is retained and used directly. If a grid cell is marked as missing, the interpolated estimate calculated using the spatiotemporal kriging interpolation algorithm is used as its data value. For example, for cell (15, 10, 1105), if it is associated with data from national monitoring stations, satellites, and monitoring vehicles (as in the previous example), its value is the result of aggregating these original values ​​within the grid (e.g., averaging), for example, (68+72+80) / 3 ≈ 73.3 μg / m³, which is an original data value. If it is identified as missing and its value is calculated to be 68.7 μg / m³ using kriging, then its value is 68.7 μg / m³, which is an interpolated estimate.

[0056] S26. The method for generating a standardized dataset with consistent spatiotemporal reference by using the statistical characteristics of different data located within the same basic grid cell for cross-validation and bias calibration is as follows: S261. Define the different data sources as the reference data source and the data source to be calibrated, respectively. During system initialization, the level of the data source is predefined based on factors such as monitoring principles, equipment accuracy, and authority. For example, the national control station is defined as the benchmark data source, and the micro-sensor is defined as the data source to be calibrated.

[0057] S262. For each base grid cell that has undergone spatiotemporal interpolation, extract the data values ​​of all data sources existing within the same spatiotemporal cell; the data values ​​of the data sources include values ​​from the reference data source and values ​​from at least one of the data sources to be calibrated. The system traverses each basic grid cell. For a specific cell (i=15, j=10, k=1105), the system checks which data sources it contains. It finds that this cell contains both a national control station value B_value = 70 μg / m³ and a microsensor value T_value = 85 μg / m³, providing a sample pair for the next step of statistical comparison.

[0058] S263. For each pair of data source combinations of the reference data source and the data source to be calibrated, calculate the statistical characteristics of their data values ​​in multiple spatiotemporal grid cells to quantify systematic bias; the statistical characteristics include: average bias and linear regression parameters. The system will not use only one pair of data for calibration, but will search for all grid cells in the entire domain that simultaneously have both types of data sources, perform statistical analysis, and quantify systematic bias.

[0059] S264. The average deviation: Calculate the average difference between the data source to be calibrated and the reference data source across all matching grid cells; for example, assuming 1000 matching cells are found across the entire range, calculate the difference between these 1000 (microsensor values ​​- national control station values), and then calculate the average of these differences.

[0060] S265. The linear regression parameters are as follows: using the data values ​​from the data source to be calibrated as the independent variable and the data values ​​from the benchmark data source as the dependent variable, a univariate linear regression is performed to obtain two parameters: slope and intercept. Using 1000 microsensor values ​​as the X-axis and 1000 corresponding national control station values ​​as the Y-axis, a univariate linear regression Y = a * X + b is performed to obtain the slope a and intercept b. For example, suppose the fitted equation is Y = 0.8 * X + 5.

[0061] S266. Based on the statistical characteristics, construct a calibration model for each type of data source to be calibrated; the calibration model adopts the following formula: The calibrated value = (original value to be calibrated - intercept) / slope, based on a linear regression model; Alternatively, the calibrated value = the original value to be calibrated - the average deviation, based on the model of the average deviation; S267. Apply the calibration model to all data points of the data source to be calibrated corresponding to the calibration model in the entire domain, perform batch calibration of the data source to be calibrated, and eliminate the systematic deviation between the data source to be calibrated and the reference data source. The calibration model (e.g., a linear regression model) is applied to all data points of the microsensor network across the entire city and over the entire time period, not just the 1000 matching points.

[0062] For example, for another unit with only microsensor data (i=16, j=11, k=1200), its raw value T_value = 90 μg / m³. Apply the calibration model: The calibrated value = (90 - 5) / 0.8 = 85 / 0.8 = 106.25 μg / m³. This value, which may have had a systematic bias, has been corrected to a level that is closer to the true value.

[0063] S268. Integrate all calibrated data from different data sources, as well as reference data from data sources that do not require calibration, into the spatiotemporal grid framework to generate a standardized dataset that eliminates inter-source system bias and is consistent with the spatiotemporal reference.

[0064] All data values ​​from microsensors (already calibrated) and national control stations (not requiring calibration) were integrated into the initial spatiotemporal basic grid framework.

[0065] S3. Based on the standardized dataset, an anomaly detection algorithm that integrates spatiotemporal context features is used to identify outliers; S31. The method for identifying outliers based on the standardized dataset and using an anomaly detection algorithm that integrates spatiotemporal context features is as follows: S311. Abstract each basic grid cell in the standardized dataset into a graph node; construct a spatiotemporal correlation graph based on the spatial proximity and temporal correlation between grid cells; abstract each basic grid cell (e.g., (15,10, 1105)) into a graph node. Spatial proximity: calculate the spatial distance between nodes. For example, connect node (15,10, 1105) with its four directly adjacent grid cells (e.g., (14,10,1105), (16,10,1105), (15,9,1105), (15,11,1105)) using edges. This indicates that they are spatially adjacent, and pollutant concentrations will influence each other. Calculate the temporal correlation of the nodes themselves. Connecting node (15, 10, 1105) with its nodes in the previous few time slices (such as (15,10,1104), (15,10,1103)) with edges indicates that the historical data at this position has an impact on the current value, resulting in a graph structure that can characterize the spatiotemporal dependencies of the data.

[0066] S312. Construct a feature vector for each graph node; the feature vector includes the target monitoring value of the grid cell at the current moment, the historical monitoring value sequence within a preset time window, and the characteristics of its spatial location; construct a digital feature vector for each node (i.e., each grid cell) as input to the model. This vector includes: the target monitoring value at the current moment: for example, the PM2.5 value of node (15,10,1105) is 73.3 μg / m³; the historical monitoring value sequence: for example, the values ​​of this node in the previous 6 hours [62, 58, 55, 68, 70, 73.3] (i.e., the values ​​from k=1099 to 1105); spatial location characteristics: for example, the functional area type of the grid where the node is located, such as industrial area code 1, residential area code 2, altitude, and other static characteristics.

[0067] S313. Input the feature vector to be detected into the trained spatiotemporal graph neural network model to obtain the predicted value of each grid cell data value; calculate the reconstruction error between the actual value of each grid cell data value and the model prediction value; and use the reconstruction error as the anomaly score of the grid cell after standardization. Spatiotemporal graph neural network models are trained using data from historical periods of normalcy, such as data from the past year without pollution events. Through learning, these models acquire knowledge of normal variation patterns within specific spatiotemporal contexts, such as the normal variation patterns of PM2.5 concentrations in industrial areas during the morning.

[0068] Input the data to be detected (such as feature vectors) into the trained model. The model will predict the predicted value of the current node based on the spatiotemporal pattern it has learned, and calculate the reconstruction error: Error = |True value A_value - Predicted value P_value|; Standardization: The reconstruction error is converted into a Z-score to obtain the final anomaly score. The higher the score, the greater the deviation of the data point from the expected spatiotemporal pattern, and the more likely it is to be an anomaly.

[0069] S314. Mark grid cells with abnormal scores exceeding the grid cell dynamic threshold as abnormal values, and output the time, spatial location, and abnormal score of the abnormal values; mark grid cells with abnormal scores less than the grid cell dynamic threshold as valid values.

[0070] The grid cell dynamic threshold does not use a globally uniform threshold. The system dynamically adjusts the judgment threshold according to the functional area type and time period of each grid cell.

[0071] Industrial areas, transportation hubs, morning and evening rush hours: Use a higher threshold (e.g., alarm when the anomaly score > 3.0) because the normal values in these places are high and fluctuate greatly, to avoid false alarms.

[0072] Residential areas, parks, at night: Use a lower threshold (e.g., alarm when the anomaly score > 2.0) because the background values in these places are low, and once a high value appears, it is very likely to be a real anomaly.

[0073] Result output: Mark the units with anomaly scores exceeding their dynamic thresholds as anomaly values, record their spatio-temporal coordinates and scores, and mark the rest as valid values. The dynamic threshold mechanism enables the system to automatically adapt to the characteristics of different regions and different time periods in the city, without the need for manual repeated parameter adjustment.

[0074] S32. For the identified anomaly values, use the spatio-temporal tensor completion algorithm based on Tucker decomposition to reconstruct the data and generate a high-quality fusion dataset; the specific method is as follows: S321. For the anomaly values in the standardized dataset, construct a third-order spatio-temporal data tensor X ∈ R^(I×J×K) according to the three dimensions of longitude, latitude, and time; where, I represents the number of grid cells in the longitude direction, J represents the number of grid cells in the latitude direction, and K represents the total number of time slices; the positions in the spatio-temporal data tensor X that have been marked as anomaly values are initialized as missing values; Regard the structure of the entire standardized dataset as a third-order tensor X, dimensions: I (longitude, e.g., 28 grids) × J (latitude, e.g., 22 grids) × K (time, e.g., 2184 hours).

[0075] Assignment: For each position (i, j, k) in the tensor: If the data at this position is marked as a valid value, then X(i,j,k) = the original valid value; if the data at this position is marked as an anomaly value, then X(i,j,k) = NaN (missing value).

[0076] S322. Perform Tucker decomposition on the original spatio-temporal data tensor X into the product of a core tensor g ∈ R^(P×Q×R) and three factor matrices (A, B, C). The factor matrices A ∈ R^(I×P), B ∈ R^(J×Q), and C ∈ R^(K×R) correspond to the longitude, latitude, and time modes respectively; the dimensions P, Q, R of the core tensor are automatically determined by the minimum description length criterion in the alternating least squares iterative process, and satisfy P << I, Q << J, R << K to achieve high compression and feature extraction of the original data; A (Size I×P): Spatial (longitude) mode, which can be understood as extracting P main spatial distribution patterns (such as "uniform distribution across the city", "southeast-northwest gradient distribution", "high in the center and low around"). B (Size J×Q): Spatial (latitude) mode, combined with A, jointly depicting a complex two-dimensional spatial field. C (Size K×R): Temporal mode, which can be understood as extracting R main temporal change patterns (such as "daily change cycle", "weekly change cycle", "persistent pollution mode under stable weather"). Dimension determination: P, Q, R are much smaller than I, J, K (i.e., P<<I, Q<<J, R<<K), achieving high compression.

[0077] S323. Simultaneously solve the core tensor g and factor matrices A, B, C by minimizing an objective function to complete the filling of missing values. S324. Use the alternating least squares method to solve the objective function; fix two of the factor matrices and the core tensor, update the other factor matrix, and iterate alternately in sequence until the change in the objective function value is less than a preset convergence threshold or the maximum number of iterations is reached. Find an optimal g, A, B, C such that their product is as close as possible to all known, non-missing data points in the original tensor X.

[0078] Alternating least squares (ALS): It is very difficult to directly solve all unknowns simultaneously. ALS adopts a "divide and conquer" strategy: fix B, C, g, and update A; fix A, C, g, and update B; Fix A, B, g, and update C; fix A, B, C, and update g. Repeat the above steps until the result converges.

[0079] S325. Substitute the optimized optimal core tensor g and factor matrices A, B, C into the Tucker decomposition model, and calculate the complete reconstructed tensor X'_rec = g×A ×B ×C; all positions in the reconstructed tensor X'_rec are filled with estimated values reconstructed based on spatio-temporal global information. When the ALS algorithm converges, the optimal g, A, B, C are obtained. Substitute them into the Tucker decomposition formula: X'_rec = g×A ×B ×C. X'_rec is a complete three-dimensional tensor without any missing values.

[0080] S326. Map the data values in the reconstructed tensor X'_rec back to the original spatio-temporal basic grid framework, replace the originally marked outliers, and jointly form the final high-quality fusion dataset with the valid values not marked as outliers.

[0081] The values ​​in the reconstructed tensor X'_rec are "reverted" to the original spatiotemporal base grid framework. For positions that were originally valid values, their original values ​​are retained; for positions that were originally outliers / missing values, the calculated reconstructed estimates are used to fill them.

[0082] S4. Based on the aforementioned high-quality fused dataset, calculate the synergistic effect coefficient of air pollutants and greenhouse gas emissions, assess the synergistic effectiveness level of pollution reduction and carbon reduction in different regions, and generate a visual assessment report. The specific formula for calculating the synergistic effect coefficient is as follows: Where S represents the synergistic effect coefficient, and ΔC and ΔP represent the synergistic effect coefficient over a period of time. The changes in emissions of greenhouse gases and air pollutants, with C0 and P0 as baseline emissions; when S>0, it indicates that pollution reduction and carbon reduction are synergistic; when S<0, it indicates that there is a trade-off relationship.

[0083] ΔC represents the change in greenhouse gas emissions, which is the difference between emissions in the base year and the policy year. ΔC = C_Base Year - C_Policy Year. For example, if the city's total CO2 emissions from September to November 2022 were 50 million tons, and the same period in 2023 were 48 million tons, then ΔC = 50 million - 48 million = +2 million tons (a decrease is positive). C0 represents the base year greenhouse gas emissions, i.e., emissions of 50 million tons in the base year. ΔP represents the change in air pollutant emissions, calculated as ΔP = P_Base Year - P_P_Policy Year. For example: The city's total PM2.5 emissions in the same period of 2022 were 50,000 tons, and in 2023 they were 42,000 tons. Therefore, ΔP = 50,000 - 42,000 = +8,000 tons (a positive decrease). P0 represents the baseline emissions of air pollutants, i.e., 50,000 tons. Then S = (200 / 5000) / (0.8 / 5) = 0.04 / 0.16 = 0.25. This value specifically means that for every 1% reduction in air pollutants, greenhouse gases decrease by 0.25% in the same period, quantifying the strength of the synergistic effect. The above calculation is for the entire city. This application can perform the same calculation for each district, or even each 2km × 2km grid, to obtain the synergistic effect coefficient S for each region. For example, it can be divided into four levels: Level 1 (High Synergy and High Efficiency): Regions with the highest S-value and the largest absolute amount of emission reduction; these are the benchmark regions with the most significant effects.

[0084] Level 2 (High Synergy, Low Efficiency): Regions with high S-values ​​but small absolute emission reductions have good synergy, but may have a small emission base.

[0085] Level 3 (Low Synergy, High Efficiency): Regions with low S-values ​​but large absolute emission reductions have achieved significant emission reductions, but have failed to achieve good synergy and require optimization measures.

[0086] Level 4 (Low Synergy and Low Efficiency): Regions with low S values ​​and small absolute emission reductions are weak links that need to be focused on in the future. It provides the key quantitative indicator of synergy effect coefficient S, enabling decision-makers to go beyond qualitative feelings and accurately grasp the degree of synergy in pollution reduction and carbon reduction from quantitative data.

[0087] Secondly, this application also provides a collaborative assessment system for urban pollution reduction and carbon reduction based on multi-source data fusion, applied in a collaborative assessment method for urban pollution reduction and carbon reduction based on multi-source data fusion, the system comprising: The dynamic acquisition and control module 1 is used to communicate with the urban monitoring sensor network. Based on the multi-source data acquisition rules and real-time event triggering set by the 3σ principle of historical data, it generates instructions to dynamically adjust the data acquisition frequency of various environmental and energy monitoring sensors. The spatiotemporal alignment and calibration engine module 2 is connected to the dynamic acquisition and control module. It receives raw data from multiple sources and uses a spatiotemporal interpolation gridding method to uniformly map all data onto a basic grid cell with a standard timestamp and a unified geographic coordinate system. It also uses the statistical characteristics of different data located in the same basic grid cell to perform cross-validation and deviation calibration to generate a standardized dataset with consistent spatiotemporal reference. The data quality improvement module 3, connected to the spatiotemporal alignment and calibration engine module, includes a spatiotemporal context anomaly detection unit and a spatiotemporal tensor completion unit. The anomaly detection unit is configured to identify outliers based on the standardized dataset using an algorithm that incorporates spatiotemporal context features. The tensor completion unit is configured to reconstruct the identified outliers using a spatiotemporal tensor completion algorithm based on Tucker decomposition to generate a high-quality fused dataset. The pollution reduction and carbon reduction synergistic assessment module 4 is connected to the data quality improvement module. Based on the high-quality fusion dataset, it quantifies the synergistic effect coefficient of air pollutants and greenhouse gas emissions, assesses the synergistic effectiveness level of pollution reduction and carbon reduction in different regions, and generates a visual assessment report.

[0088] In one embodiment, the spatiotemporal alignment and calibration engine module includes a spatiotemporal gridding unit and a data calibration unit; Spatiotemporal gridded units are used to establish a unified basic geographic grid and perform interpolation mapping on multi-source data; The data calibration unit is used to perform cross-validation and bias correction on multi-source data within the same grid.

[0089] The options described in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application's embodiments can be found in the above system embodiments, and will not be repeated in this embodiment.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A collaborative assessment method for urban pollution reduction and carbon reduction based on multi-source data fusion, characterized in that, include: Based on the 3σ principle of historical data, multi-source data acquisition rules and real-time event triggering are set to dynamically adjust the data acquisition frequency of various monitoring sensors. The multi-source data is processed using a spatiotemporal interpolation gridding method to uniformly map all the multi-source data onto a basic grid cell with a standard timestamp and a unified geographic coordinate system. Furthermore, cross-validation and bias calibration are performed using the statistical characteristics of different data located within the same basic grid cell to generate a standardized dataset with consistent spatiotemporal reference. Based on the standardized dataset, an anomaly detection algorithm that integrates spatiotemporal context features is used to identify outliers; for the identified outliers, a spatiotemporal tensor completion algorithm based on Tucker decomposition is used to reconstruct the data and generate a high-quality fused dataset. Based on the high-quality fusion dataset, the synergistic effect coefficient of air pollutants and greenhouse gas emissions is calculated, the synergistic effectiveness level of pollution reduction and carbon reduction in different regions is evaluated, and a visual evaluation report is generated.

2. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 1, characterized in that, The multi-source data acquisition rules based on the 3σ principle of historical data are as follows: Calculate the mean μ and standard deviation σ of the data collected by any monitoring sensor under the same historical spatiotemporal conditions, and set the data fluctuation range of the monitoring sensor at the normal acquisition frequency as (μ-3σ, μ+3σ); When the real-time collected data remains within the data fluctuation range, it is determined to be in a steady state and the first collection frequency is automatically maintained; when the real-time data exceeds the data fluctuation range, it immediately switches to the preset second collection frequency.

3. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 1, characterized in that, The method of using spatiotemporal interpolation gridding to uniformly map all the multi-source data onto basic grid cells with standard timestamps and a unified geographic coordinate system is as follows: Based on the geographical scope and time evaluation period of the target city, a three-dimensional spatiotemporal basic grid framework is constructed. The X-axis and Y-axis of the basic grid framework are divided based on a unified geographic coordinate system to form a regular geographic planar grid. The Z-axis is the time axis, which is based on a standard timestamp and divided according to a preset time resolution to generate a series of equally spaced continuous time slices. Each raw data point from different data sources is associated with the corresponding basic grid cell in the basic grid framework based on its spatial coordinate information and timestamp information; for spatial coordinates, the data point is assigned to the geographic plane grid where the data point is located; for timestamps, the data point is assigned to the specific time slice to which the data point belongs. Traverse all basic grid cells, identify empty grid cells that do not contain any original data values, and generate a list of empty grid cells; For each cell in the list of missing grid cells, the spatiotemporal kriging interpolation algorithm is used to calculate the interpolation estimate of the missing grid cell and fill it in; Each basic grid cell contains a data value, generating a gridded dataset with a unified spatiotemporal reference. The data value includes the original data value or the interpolated estimate.

4. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 3, characterized in that, The method for filling data for each cell in the list of missing grid cells using the spatiotemporal kriging interpolation algorithm is as follows: Based on the data in the non-empty grid cells, a spatiotemporal variogram model describing the spatial and temporal correlation of the data is calculated and fitted. Centered on the missing grid cell to be interpolated, all non-missing grid cells within a preset spatiotemporal neighborhood of the missing grid cell are selected as reference points; using the spatiotemporal variogram model, the weight of each reference point to be interpolated is calculated, where the weight represents the spatiotemporal correlation strength between the reference point and the point to be interpolated; The interpolation estimate of the missing grid cell is obtained by multiplying the data value of each reference point by its corresponding weight and then summing the results.

5. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 3, characterized in that, The method for generating a standardized dataset with consistent spatiotemporal reference by utilizing the statistical characteristics of different data located within the same basic grid cell for cross-validation and bias calibration is as follows: The different data sources are defined as the baseline data source and the data source to be calibrated, respectively. For each base grid cell that has undergone spatiotemporal interpolation, extract the data values ​​from all data sources existing within the same spatiotemporal cell; the data values ​​from the data sources include values ​​from the reference data source and values ​​from at least one of the data sources to be calibrated. For each pair of data source combinations of the reference data source and the data source to be calibrated, the statistical characteristics of their data values ​​in multiple space-time grid cells are calculated to quantify systematic bias. The statistical characteristics include: mean deviation and linear regression parameters; The average deviation is calculated as the average difference between the data source to be calibrated and the reference data source across all matching grid cells. The linear regression parameters are obtained by performing a univariate linear regression with the data values ​​of the data source to be calibrated as the independent variable and the data values ​​of the benchmark data source as the dependent variable, resulting in two parameters: slope and intercept. Based on the statistical characteristics, a calibration model is constructed for each type of data source to be calibrated; the calibration model adopts the following formula: Calibrated value = (Original value to be calibrated - Intercept) / Slope; Alternatively, the calibrated value = the original value to be calibrated - the average deviation; The calibration model is applied to all data points of the data source to be calibrated within the entire domain, and the data source to be calibrated is batch-corrected to eliminate the systematic deviation between the data source to be calibrated and the reference data source. The calibration data from all calibrated different data sources, as well as the baseline data from data sources that do not require calibration, are integrated into the spatiotemporal basic grid framework to generate a standardized dataset that eliminates inter-source systematic biases and is consistent with the spatiotemporal baseline.

6. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 3, characterized in that, The method for identifying outliers based on the standardized dataset and using an anomaly detection algorithm that incorporates spatiotemporal context features is as follows: Each basic grid cell in the standardized dataset is abstracted into a graph node; a spatiotemporal correlation graph is constructed based on the spatial proximity and temporal correlation between grid cells. Construct a feature vector for each graph node; The feature vector includes the target monitoring value of the grid cell at the current moment, the historical monitoring value sequence within a preset time window, and the characteristics of its spatial location; The feature vector to be detected is input into the trained spatiotemporal graph neural network model to obtain the predicted value of the data value of each grid cell; Calculate the reconstruction error between the actual value and the model prediction value of the data value of each grid cell; after normalizing the reconstruction error, use it as the anomaly score of the grid cell. Mark the grid cells with anomaly scores exceeding the dynamic threshold of the grid cell as outliers, and output the time, spatial location, and anomaly scores of the outliers; mark the grid cells with anomaly scores less than the dynamic threshold of the grid cell as valid values.

7. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 6, characterized in that, For the identified outliers, the method of using the spatio-temporal tensor completion algorithm based on Tucker decomposition to perform data reconstruction and generate a high-quality fusion dataset is as follows: Construct a third-order spatio-temporal data tensor X ∈ R^(I×J×K) from the outliers in the standardized dataset according to the three dimensions of longitude, latitude, and time; where I represents the number of grid cells in the longitude direction, J represents the number of grid cells in the latitude direction, and K represents the total number of time slices; the positions in the spatio-temporal data tensor X that have been marked as outliers are initialized as missing values. Perform Tucker decomposition on the original spatio-temporal data tensor X into the product of a core tensor g ∈ R^(P×Q×R) and three factor matrices (A, B, C). The factor matrices A ∈ R^(I×P), B ∈ R^(J×Q), and C ∈ R^(K×R) correspond to the longitude, latitude, and time patterns respectively; the dimensions P, Q, R of the core tensor are automatically determined by the minimum description length criterion in the alternating least squares iteration process, and satisfy P << I, Q << J, R << K to achieve high compression and feature extraction of the original data. Simultaneously solve the core tensor g and the factor matrices A, B, C by minimizing an objective function to complete the missing values. Use the alternating least squares method to solve the objective function; fix two of the factor matrices and the core tensor, and update the other factor matrix, and alternately iterate in a loop until the change in the objective function value is less than the preset convergence threshold or the maximum number of iterations is reached. Substitute the optimized optimal core tensor g and factor matrices A, B, C into the Tucker decomposition model, and calculate the complete reconstruction tensor X'_rec = g × A × B × C; all positions in the reconstruction tensor X'_rec are filled with estimated values reconstructed based on spatio-temporal global information. Map the data values in the reconstruction tensor X'_rec back to the original spatio-temporal base grid framework, replace the originally marked outliers, and jointly form the final fusion dataset with the valid values that have not been marked as outliers.

8. The urban pollution reduction and carbon reduction collaborative assessment method based on multi-source data fusion as described in claim 1, characterized in that, The specific calculation formula of the synergy coefficient is as follows: Where S represents the synergistic effect coefficient, and ΔC and ΔP represent the synergistic effect coefficient over a period of time. The change in emissions of greenhouse gases and air pollutants inside, C0 and P0 are the baseline emissions respectively; when S > 0, it indicates the synergy of pollution reduction and carbon emission reduction; when S < 0, it indicates a trade-off relationship.

9. A collaborative assessment system for urban pollution reduction and carbon reduction based on multi-source data fusion, characterized in that, A system applied to the method for collaborative assessment of urban pollution reduction and carbon emission reduction based on multi-source data fusion according to any one of claims 1 to 8, the system includes: The dynamic acquisition and control module is used to communicate with the urban monitoring sensor network. Based on the 3σ principle of historical data, it sets multi-source data acquisition rules and real-time event triggers, and generates instructions to dynamically adjust the data acquisition frequency of various environmental and energy monitoring sensors. The spatiotemporal alignment and calibration engine module is connected to the dynamic acquisition and control module. It receives raw data from multiple sources and uses a spatiotemporal interpolation gridding method to uniformly map all data onto a basic grid cell with a standard timestamp and a unified geographic coordinate system. It also uses the statistical characteristics of different data located in the same basic grid cell to perform cross-validation and deviation calibration, generating a standardized dataset with consistent spatiotemporal reference. The data quality improvement module, connected to the spatiotemporal alignment and calibration engine module, includes a spatiotemporal context anomaly detection unit and a spatiotemporal tensor completion unit. The anomaly detection unit is configured to identify outliers based on the standardized dataset using an algorithm that incorporates spatiotemporal context features. The tensor completion unit is configured to reconstruct the identified outliers using a spatiotemporal tensor completion algorithm based on Tucker decomposition, thereby generating a high-quality fused dataset. The pollution reduction and carbon reduction synergistic assessment module is connected to the data quality improvement module. Based on the high-quality fusion dataset, it quantifies the synergistic effect coefficient of air pollutants and greenhouse gas emissions, assesses the synergistic effectiveness level of pollution reduction and carbon reduction in different regions, and generates a visual assessment report.

10. The dynamic popularity evaluation system that integrates time decay and user behavior weights as described in claim 9, characterized in that, The spatiotemporal alignment and calibration engine module includes a spatiotemporal gridding unit and a data calibration unit; Spatiotemporal gridded units are used to establish a unified basic geographic grid and perform interpolation mapping on multi-source data; The data calibration unit is used to perform cross-validation and bias correction on multi-source data within the same grid.

Images