Dynamic Evaluation Method of Multi-Source Heterogeneous Time Series Data for Atmospheric Environmental Governance Process

By constructing a dynamic evaluation method for multi-source heterogeneous time-series data, the inaccuracy of multi-source data processing in atmospheric environmental governance is solved, enabling continuous characterization of the governance process and identification of key shortcomings, and generating effective intervention strategies.

CN123117540APending Publication Date: 2026-11-17XINJIANG UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611332636.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2026-08-20
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing atmospheric environmental governance assessment methods are difficult to adapt to multi-source heterogeneous data and cannot continuously characterize the time-delay coupling relationship in the governance process, resulting in inaccurate assessment results and an inability to identify key shortcomings and generate effective intervention strategies.

Method used

A dynamic evaluation method using multi-source heterogeneous time-series data is adopted. Through time alignment, missing data completion, standardization, and text intensity quantification, a time-varying coupled network of complex systems is constructed. Dimensional scores and dynamic evaluation values ​​are calculated to identify key weaknesses and generate intervention strategies.

Benefits of technology

It enables unified processing and time-varying relationship characterization of multi-source heterogeneous data, improves the accuracy and interpretability of evaluation results, and can identify key shortcomings and generate actionable intervention strategies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117540A_ABST
    Figure CN123117540A_ABST
Patent Text Reader

Abstract

This invention discloses a dynamic evaluation method for multi-source heterogeneous time-series data related to atmospheric environmental governance processes, specifically including the following steps: Step 1, acquiring multi-source heterogeneous time-series data related to atmospheric environmental governance in the target area within a preset time range; Step 2, performing time alignment and preprocessing on the multi-source heterogeneous time-series data; Step 3, mapping the processed data according to preset index mapping rules to form a multi-dimensional feature matrix under a continuous time window; Step 4, constructing a time-varying coupling network for a complex system based on the multi-dimensional feature matrix; Step 5, calculating dimensional scores and dynamic evaluation values ​​based on the time-varying coupling network and the multi-dimensional feature matrix; Step 6, outputting the evaluation results based on the difference between the target area and the benchmark area. This invention solves the problem of inaccurate evaluation results caused by existing evaluation methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of environmental governance data processing and decision support technology, and relates to a dynamic evaluation method for multi-source heterogeneous time-series data for atmospheric environmental governance processes. Background Technology

[0002] Atmospheric environmental governance involves multiple elements, including air quality monitoring, pollution control, financial investment, cross-departmental collaboration, implementation response, supervision and feedback, and policy document disclosure. It is characterized by diverse data sources, inconsistent time scales, time lags in the effectiveness of governance measures, and feedback coupling between different governance stages. With the intensification of the combined pollution problem of fine particulate matter and ozone, atmospheric environmental governance is no longer simply a matter of controlling single pollutant concentrations. Instead, it is a complex and dynamic process formed by the combined effects of environmental pressure, resource input, organizational coordination, implementation response, supervision and feedback, and governance results. Relying solely on the air quality index, pollutant concentrations, or periodic performance scores is insufficient to continuously characterize the operational status of the governance process, nor can it support the identification of key weaknesses and the generation of intervention strategies.

[0003] However, existing air quality assessment and atmospheric environmental governance performance evaluation schemes generally still focus on pollutant concentration assessment, air quality level determination, phased performance calculation, weight optimization, or influencing factor analysis. Their data processing objects are mostly single monitoring data, phased statistical data, or relatively static indicator systems, which are difficult to adapt to the application scenarios in actual governance processes where multiple sources of heterogeneous data coexist, such as monthly air quality data, annual governance investment data, publicly available policy text data, publicly available supervision and rectification data, monthly random inspection data, and annual administrative penalty data. At the same time, existing methods can usually only output evaluation values, levels, rankings, or weight results, lacking technical processes for time alignment, missing data completion, standardization, and governance dimension mapping of data from different sources under a unified time window. They also cannot depict the time-lag coupling relationship between governance pressure, governance investment, organizational coordination, execution response, supervision feedback, and governance results over time. Furthermore, they cannot further identify key shortcomings and conduct propagation simulation and comprehensive benefit ranking of candidate intervention strategies based on dynamic evaluation, leading to inaccurate final evaluation results. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, which solves the problem of inaccurate evaluation results caused by existing evaluation methods.

[0005] The technical solution adopted in this invention is a dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, which specifically includes the following steps: Step 1: Obtain multi-source heterogeneous time-series data related to atmospheric environmental governance in the target area within a preset time range; Step 2: Perform time alignment and preprocessing on the multi-source heterogeneous time series data; Step 3: Map the processed data according to the preset index mapping rules to form a multi-dimensional feature matrix under a continuous time window; Step 4: Construct a time-varying coupled network for complex systems based on multidimensional feature matrices; Step 5: Calculate the dimensional score and dynamic evaluation value based on the time-varying coupled network and multidimensional feature matrix of the complex system; Step 6: Output the evaluation results based on the differences between the target area and the benchmark area.

[0006] The invention is further characterized by: In step 1, the target area is a prefecture-level city, a city cluster, a river basin, or a provincial-level administrative region; the preset time range is any one of the following: daily, weekly, monthly, quarterly, or annual scale. Air quality monitoring data includes one or more of the following: Air Quality Index (AQI), PM2.5 concentration, PM10 concentration, O3 concentration, NO2 concentration, SO2 concentration, CO concentration, percentage of days with good or excellent air quality, and percentage of days with heavy pollution. Meteorological environmental data include one or more of the following: temperature, humidity, wind speed, precipitation, boundary layer height, and diffusion condition index. The governance input data includes one or more of the following: environmental protection fiscal expenditure, monitoring resource input, law enforcement resource input, and digital governance input; Organizational collaboration data includes one or more of the following: frequency of joint meetings, number of joint law enforcement actions, completion rate of collaborative matters, and frequency of cross-regional collaboration. Execution response data includes one or more of the following: early warning response duration, emergency measure implementation rate, and rectification completion efficiency; Supervision and feedback data include one or more of the following: frequency of information disclosure, rate of handling public complaints, and rate of feedback on rectification and inspection. Policy text data includes one or more of the following: government work reports, special rectification plans, notification documents, and public governance documents.

[0007] In step 2, the time alignment method includes: mapping data from different sources to the same sampling period on a daily, weekly, monthly, quarterly, or annual scale; using linear interpolation or spline interpolation to fill in missing data with a missing rate not higher than the first threshold, and using weighted nearest neighbor estimation based on relevant indicators to fill in missing data with a missing rate higher than the first threshold. The preprocessing methods include: identifying outliers using an anomaly detection method based on the absolute deviation of the median, and correcting for anomalies using the truncated mean or neighborhood estimate within the corresponding time window.

[0008] In step 2, data from different sources are uniformly mapped to a monthly sampling period; for data with a missing rate of less than or equal to 20%, linear interpolation or spline interpolation is used for missing data, and for data with a missing rate of more than 20%, weighted nearest neighbor estimation based on relevant indicators is used for missing data; outliers are identified using an anomaly detection method based on the absolute deviation of the median, and anomaly correction is performed using the truncated mean or neighborhood estimate within the corresponding time window.

[0009] In step 2, for numerical indicators, standardization is performed; for positive indicators, the following formula is used: (1) For contrarian indicators, the following formula is used: (2) Where x_{j,t} represents the original observation value of the j-th indicator at time t; z_{j,t} represents the standardized indicator value; max(x_j) and min(x_j) represent the maximum and minimum values ​​of the j-th indicator within a preset time range or a rolling time window, respectively; when max(x_j) = min(x_j), the standardized value of the indicator within the corresponding time window is set to the preset neutral value; For text-based data, text intensity quantification includes: extracting environmental governance keywords, law enforcement keywords, collaborative governance keywords, and supervision feedback keywords from policy text data; and calculating text intensity feature values ​​based on the word frequency, inverse document frequency, and part-of-speech weight of each keyword within the current time window. Text-like features at time The strength value below satisfy: (3) in, Indicates the first Keywords in time The number of valid terms under, Indicates the first The word class weight of each keyword, Indicates the first Key words at any time The statistics below.

[0010] The specific process of step 3 is as follows: According to the preset indicator mapping rules, the processed data is mapped to the dimensions of governance pressure, governance input, organizational collaboration, execution response, supervision feedback, and governance results, forming a multidimensional feature matrix under a continuous time window. At time t, the multidimensional feature matrix is ​​represented as X_t=[z_{d,k,t}], where d=1,2,...,6, k=1,2,...,m_d; d represents the governance dimension, m_d represents the number of indicators under the d-th dimension, and z_{d,k,t} represents the standardized value of the k-th indicator under the d-th dimension at time t. The sequence of multidimensional feature matrices under a continuous time window is represented as {X_t | t=1,2,...,T}.

[0011] The specific process of step 4 is as follows: The network nodes are either indicator nodes or dimension nodes. The edges between nodes represent the coupling strength between different indicators within the current time window, using lengths of [missing information]. The rolling time window is used for window-by-window calculation, and any two index nodes in the time-varying coupled network of the complex system are calculated. and At any moment Coupled edge weights under satisfy: (4) in, Represents mutual information value, Indicates optimal time delay The correlation coefficient below, Indicates the degree of grey relational relationship. , and is a non-negative weighting coefficient, and: (5) Optimal time delay By searching within a preset time delay set, The largest Sure.

[0012] The specific process of step 5 is as follows: No. Each dimension at time Dimensional scores satisfy: (6) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below This represents the standardized index value; No. Each dimension at time Time-varying weights satisfy: (7) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below Indicates the first Information entropy in each dimension Indicates the first Normalized centrality of each dimension in a time-varying coupled network of a complex system at time t. and It is a non-negative adjustment coefficient; This represents the time-varying weight of the u-th dimension at time t; Target area at time Dynamic evaluation value under satisfy: (8) in, Represents the cooperative gain term. This indicates a penalty for risk imbalance. and It is a non-negative adjustment coefficient; Synergistic gain term satisfy: (9) Risk Imbalance Penalty satisfy: (10).

[0013] The specific process of step 6 is as follows: The benchmark region is a reference region with consistent administrative levels and similar governance tasks, or a benchmark sample constructed from the historical best-performing time window of the target region; for each indicator Calculate the difference between the target area and the reference area. And calculate the sensitivity of the corresponding indicators to the dynamic evaluation value. , Sensitivity Calculated using a single-index perturbation method, for the first... Each indicator increased and reduce Recalculated and ,but: (11) in, and They represent the first The number of indicators increased and reduce The subsequent dynamic evaluation value; The centrality eigenvalues ​​of the index in the time-varying coupled network of the complex system are calculated. By combining the difference value, sensitivity and centrality, the short-board priority index can be obtained. The key short-board ranking results of the six-dimensional features are output based on the short-board priority index, as shown in the following formula (12): (12) Among them, Q j This represents the bottleneck priority index of the j-th indicator at time t; for candidate intervention strategies First, update the target indicator vector within the current time window according to the preset intervention intensity. ; Update the target indicator vector within the current time window based on the intervention variables and intervention intensity corresponding to the candidate intervention strategies. According to the propagation iteration formula: (13) Execute preset number of steps The propagation calculation, in which, Indicates the propagation attenuation coefficient. This represents the coupling edge weight matrix under the current time window. Represents the benchmark index vector. This means that the calculation result is restricted to the interval [0,1]. Based on the incremental dynamic evaluation value after propagation Stability indicators and implementation cost indicators Calculate the comprehensive return value Candidate intervention strategies Overall return value satisfy: (14) in, It is a non-negative adjustment coefficient.

[0014] The beneficial effects of this invention are as follows: (1) By performing unified time alignment, missing data repair and text intensity quantization on multi-source heterogeneous time series data, the feasibility of different types of governance data entering the unified computing framework is improved; (2) By introducing time-delay correlation, mutual information value and grey relational degree, a time-varying coupled network of complex system is constructed, so that the time-varying relationship between governance pressure, governance input, organizational coordination, execution response, supervision feedback and governance results can be continuously characterized; (3) By combining the short-spot priority index with intervention propagation simulation, a closed-loop process from dynamic assessment, key short-spot identification to target intervention strategy generation was realized, which improved the interpretability and executability of the results output. Attached Figure Description

[0015] Figure 1 This is the overall flowchart of the dynamic evaluation method for multi-source heterogeneous time-series data in the atmospheric environmental governance process of this invention; Figure 2 This is a mapping diagram between multi-source data and a six-dimensional feature matrix in the multi-source heterogeneous time-series data dynamic evaluation method for atmospheric environmental governance processes according to the present invention; Figure 3 This is a graph showing the calculation results of dynamic evaluation values ​​under continuous time windows in the dynamic evaluation method for multi-source heterogeneous time-series data of the atmospheric environmental governance process of this invention; Figure 4 This invention relates to a dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, which uses a time-varying coupled network edge weight matrix diagram. Figure 5 This is a graph showing the calculation results of the key bottleneck priority index in the multi-source heterogeneous time-series data dynamic evaluation method for atmospheric environmental governance processes, as presented in this invention. Figure 6(a) shows the simulation results of candidate intervention strategy propagation in the dynamic evaluation method of multi-source heterogeneous time series data for atmospheric environmental governance process according to the present invention; Figure 6(b) is a graph showing the comprehensive benefit ranking results in the dynamic evaluation method of multi-source heterogeneous time series data for atmospheric environmental governance process according to the present invention. Detailed Implementation

[0016] The following detailed description is provided in conjunction with specific implementation methods.

[0017] Example 1 This invention provides a dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, such as... Figure 1As shown, the specific process is as follows: First, acquire multi-source heterogeneous time-series data related to atmospheric environmental governance in the target area within a preset time range. This multi-source heterogeneous time-series data includes at least air quality monitoring data, meteorological and environmental data, governance input data, organizational coordination data, execution response data, supervision feedback data, and policy text data. Second, align the multi-source heterogeneous time-series data according to a unified sampling period. Perform missing data completion, anomaly correction, unit unification, and standardization on numerical data, and perform word segmentation, stop word removal, keyword extraction, and text intensity quantization on text data to obtain a preprocessed governance dataset. Third, map the governance dataset to governance pressure, governance input, organizational coordination, execution response, supervision feedback, and governance result dimensions according to preset indicator mapping rules, constructing a multi-dimensional feature matrix of the target area within a continuous time window. Fourth, based on the multi-dimensional feature matrix, calculate the time-delay correlation, mutual information value, and grey relational degree between indicator nodes within each time window, determine the time-varying coupling edge weights between nodes, and construct a complex system time-varying coupling network. Fifth, based on the complex system time-varying coupling network… The time-varying weights of each dimension are determined by considering the centrality characteristics of the nodes, the dispersion of each dimension's indicators, and the magnitude of change within adjacent time windows. The dynamic evaluation values ​​of the target region under each time window are calculated by combining dimensional scores, synergistic gain terms, and risk imbalance penalty terms. The multidimensional feature matrices of the target region and the benchmark region under the same time window are compared. The weakness priority index of each indicator is calculated by combining the sensitivity of the indicators to the dynamic evaluation values, the centrality characteristics of the indicators in the time-varying coupled network of the complex system, and the indicator difference values, generating a ranking result of key weaknesses. For multiple candidate intervention strategies or combinations of candidate intervention strategies, the corresponding indicator values ​​are updated according to the preset intervention intensity. Multi-step propagation simulation is performed through the time-varying coupled network of the complex system to obtain the evaluation value increment, stability index, and implementation cost index corresponding to each candidate intervention strategy or combination of candidate intervention strategies. The comprehensive benefit value of each candidate intervention strategy or combination of candidate intervention strategies is calculated based on the evaluation value increment, stability index, and implementation cost index. The dynamic evaluation results of the atmospheric environmental governance process of the target region, the key weakness results, and the target intervention strategy generation results are output.

[0018] Example 2 This invention provides a dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, such as... Figure 1 As shown, the specific process is as follows: including seven steps: data acquisition and preprocessing, six-dimensional index mapping, time-varying coupling modeling of complex systems, dynamic evaluation, key weakness identification, intervention propagation simulation, and strategy generation.

[0019] Step 1: Obtain multi-source heterogeneous time-series data related to atmospheric environmental governance in the target area within a preset time range.

[0020] Step 2: Perform time alignment and preprocessing on the multi-source heterogeneous time series data; Step 3: Map the processed data to the dimensions of governance pressure, governance input, organizational collaboration, execution response, supervision feedback, and governance results according to the preset indicator mapping rules, forming a multi-dimensional feature matrix under a continuous time window; Step 4: Construct a time-varying coupled network for complex systems based on multidimensional feature matrices; Step 5: Calculate the dimensional score and dynamic evaluation value based on the time-varying coupled network and multidimensional feature matrix of the complex system; Step 6: Output the evaluation results based on the differences between the target area and the benchmark area.

[0021] Example 3 The specific process of step 1 is as follows: The target area can be a prefecture-level city, a city cluster, a river basin, or a provincial-level administrative region; the preset time range can be any of the following scales: daily, weekly, monthly, quarterly, or annual.

[0022] Taking publicly available data from a certain prefecture-level city from 2019 to 2025 as an example: Air quality monitoring data includes one or more of the following: Air Quality Index (AQI), PM2.5 concentration, PM10 concentration, O3 concentration, NO2 concentration, SO2 concentration, CO concentration, percentage of days with good or excellent air quality, and percentage of days with heavy pollution. Meteorological environmental data includes one or more of the following: temperature, humidity, wind speed, precipitation, boundary layer height, and diffusion condition index; governance input data includes one or more of the following: environmental protection fiscal expenditure, monitoring resource input, law enforcement resource input, and digital governance input. Organizational collaboration data includes one or more of the following: frequency of joint meetings, number of joint law enforcement actions, completion rate of collaborative matters, and frequency of cross-regional collaboration. Execution response data includes one or more of the following: early warning response duration, emergency measure implementation rate, and rectification completion efficiency; Supervision and feedback data include one or more of the following: frequency of information disclosure, rate of handling public complaints, and rate of feedback on rectification and inspection. Policy text data includes one or more of the following: government work reports, special rectification plans, notification documents, and public governance documents.

[0023] Example 4 The specific process of step 2 is as follows: Time alignment methods include: mapping data from different sources to the same sampling period on a daily, weekly, monthly, quarterly, or annual scale; using linear interpolation or spline interpolation to fill in missing data with a missing rate not higher than the first threshold, and using weighted nearest neighbor estimation based on relevant indicators to fill in missing data with a missing rate higher than the first threshold. The preprocessing methods include: identifying outliers using an anomaly detection method based on the absolute deviation of the median, and correcting for anomalies using the truncated mean or neighborhood estimate within the corresponding time window.

[0024] For example: map data from different sources to a monthly sampling period; use linear interpolation or spline interpolation to impute missing data with a missing rate of no more than 20%, and use weighted nearest neighbor estimation based on relevant indicators to impute missing data with a missing rate of more than 20%; use anomaly detection based on median absolute deviation to identify outliers, and use the truncated mean or neighborhood estimate within the corresponding time window to correct for anomalies.

[0025] For numerical indicators, standardization is applied. For positive indicators, the following formula is used: (1) For contrarian indicators, the following formula is used: (2) Where x_{j,t} represents the original observation value of the j-th indicator at time t; z_{j,t} represents the standardized indicator value; max(x_j) and min(x_j) represent the maximum and minimum values ​​of the j-th indicator within a preset time range or rolling time window, respectively. Positive indicators are those whose larger values ​​indicate better governance, while negative indicators are those whose larger values ​​indicate worse governance. When max(x_j) = min(x_j), the standardized value of the indicator within the corresponding time window can be set to a preset neutral value.

[0026] Through the above processing, indicators with different dimensions are converted into a unified scale range.

[0027] For text-based data, text intensity quantification includes: extracting environmental governance keywords, law enforcement keywords, collaborative governance keywords, and supervision feedback keywords from policy text data; and calculating text intensity feature values ​​based on the word frequency, inverse document frequency, and part-of-speech weight of each keyword within the current time window. Text-like features at time The strength value below satisfy: (3) in, Indicates the first Keywords in time The number of valid terms under, Indicates the first The word class weight of each keyword, Indicates the first Key words at any time The statistics below.

[0028] Example 5 Step 3 specifically involves: The preset indicator mapping rules map the indicators processed in step 2 to six governance dimensions according to field semantics, data source attributes, and governance stage attributes. Specifically: pollutant concentration, unfavorable diffusion conditions, and industrial pressure fields are mapped to governance pressure dimension D1; fiscal environmental protection expenditure, monitoring resources, and law enforcement resources input fields are mapped to governance input dimension D2; joint meetings, joint law enforcement, and collaborative matter completion fields are mapped to organizational collaboration dimension D3; early warning response, emergency measure implementation, and rectification efficiency fields are mapped to execution response dimension D4; information disclosure, complaint handling, and supervision and rectification feedback fields are mapped to supervision feedback dimension D5; and the number of days with good air quality, heavily polluted days, and pollutant concentration improvement rate fields are mapped to governance results dimension D6.

[0029] The governance pressure dimension includes pollution load level, unfavorable diffusion conditions index, and industrial structure pressure index; the governance input dimension includes the intensity of fiscal environmental protection expenditure, the level of investment in monitoring facilities, and the intensity of investment in law enforcement resources. Organizational collaboration dimensions include the frequency of joint meetings, the number of joint law enforcement actions, and the completion rate of regional collaborative matters; The execution response dimensions include early warning response duration, emergency measure implementation rate, and rectification completion efficiency; The dimensions of supervision and feedback include the frequency of information disclosure, the rate of handling public complaints, and the rate of feedback on rectification and inspection. The dimensions of governance outcomes include the proportion of days with good air quality, the proportion of days with heavy pollution, and the improvement rate of major pollutant concentrations.

[0030] At time t, the multidimensional feature matrix is ​​represented as X_t=[z_{d,k,t}], where d=1,2,...,6, k=1,2,...,m_d; d represents the governance dimension, m_d represents the number of indicators in the d-th dimension, and z_{d,k,t} represents the standardized value of the k-th indicator in the d-th dimension at time t. The sequence of multidimensional feature matrices under continuous time windows is represented as {X_t | t=1,2,...,T}.

[0031] Monthly air quality data, annual data on governance investment, publicly available policy documents, publicly available data on inspection and rectification, monthly random inspection data, and annual administrative penalty data can be mapped to a six-dimensional feature matrix. This mapping relationship is as follows: Figure 2 As shown.

[0032] Figure 2The monthly air quality data, annual governance investment data, publicly available policy text data, publicly available inspection and rectification data, monthly random inspection data, and annual administrative penalty data are all governance data obtained in Step 1 and processed in Step 2 through time alignment, missing data completion, anomaly correction, unit unification, standardization, and text intensity quantification. The meteorological environment data in Step 1 can be represented as the unfavorable diffusion conditions index in the governance pressure dimension; organizational coordination data can be represented by fields such as joint meetings, joint enforcement, and policy coordination disclosure; execution response data can be represented by fields such as random inspections, administrative penalties, rectification efficiency, and implementation of emergency measures; and supervision feedback data can be represented by fields such as publicly available inspection and rectification data, information disclosure frequency, handling and feedback rate of public complaints.

[0033] Example 6 Step 4 is as follows: The nodes of the network are indicator nodes or dimension nodes, and the edges between nodes are used to represent the coupling strength between different indicators within the current time window.

[0034] Use a length of The rolling time window is used for window-by-window calculation to reflect the changes in the relationship structure of the governance process at different times.

[0035] Any two index nodes in a time-varying coupled network of a complex system and At any moment Coupled edge weights under satisfy: (4) in, Represents mutual information value, Indicates optimal time delay The correlation coefficient below, Indicates the degree of grey relational relationship. , and is a non-negative weighting coefficient, and: (5) Optimal time delay By searching within a preset time delay set, The largest Sure.

[0036] To mitigate the impact of differences in the range of values ​​for various statistical measures, the mutual information value, optimal time-delay correlation coefficient, and grey relational degree can be normalized before weighted fusion. For node pairs with edge weights below the second threshold, the corresponding edges can be removed from the coupled network to reduce the interference of weakly correlated edges on subsequent propagation simulations. For example, the normalized coupling edge weights between nodes in each dimension can be calculated based on the six-dimensional feature matrix under continuous time windows, forming a time-varying coupled network edge weight matrix, such as... Figure 4 As shown.

[0037] Example 7 The specific process of step 5 is as follows: No. Each dimension at time Dimensional scores satisfy: (6) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below This represents the standardized index value.

[0038] No. Each dimension at time Time-varying weights satisfy: (7) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below Indicates the first Information entropy in each dimension Indicates the first Normalized centrality of each dimension in a time-varying coupled network of a complex system at time t. and It is a non-negative adjustment coefficient. This represents the time-varying weight of the u-th dimension at time t, used to characterize the relative contribution of this dimension in the dynamic evaluation of the current governance process. This time-varying weight is jointly determined by factors such as dimensional information entropy, network centrality, and the magnitude of changes in adjacent time windows.

[0039] Target area at time Dynamic evaluation value under satisfy: (8) in, Represents the cooperative gain term. This indicates a penalty for risk imbalance. and It is a non-negative adjustment coefficient; Synergistic gain term Used to characterize the positive linkage effect between governance input, organizational coordination, execution response, and supervision feedback; risk imbalance penalty item. It is used to characterize the degree of mismatch between governance pressure and governance outcomes.

[0040] Synergistic gain term satisfy: (9) Risk Imbalance Penalty satisfy: (10) The above design enables the linkage level of the input-coordination-response-feedback chain and the pressure-outcome imbalance to jointly participate in the dynamic evaluation value calculation. A dynamic evaluation value sequence is obtained after calculation and normalization using continuous monthly time windows, such as... Figure 3 As shown.

[0041] Example 8 Step 6 is as follows: The benchmark region can be a reference region with consistent administrative levels and similar governance tasks, or it can be a benchmark sample constructed from the historical best-performing time window of the target region; for each indicator Calculate the difference between the target area and the reference area. And calculate the sensitivity of the corresponding indicators to the dynamic evaluation value. .

[0042] Sensitivity Calculated using a single-index perturbation method, for the first... Each indicator increased and reduce Recalculated and ,but: (11) in, and They represent the first The number of indicators increased and reduce The subsequent dynamic evaluation value.

[0043] Simultaneously, the centrality eigenvalues ​​of the index in the time-varying coupled network of the complex system are calculated. Combining the difference value, sensitivity, and centrality, a short-board priority index is obtained; based on the short-board priority index, the key short-board ranking results with six dimensions are output, such as... Figure 5As shown: (12) Q j This represents the priority index of the j-th indicator at time t. A propagation simulation is performed on candidate intervention strategies. Candidate intervention strategies may include one or more of the following: increasing governance investment, enhancing organizational coordination, reducing implementation response time lag, strengthening supervision and feedback, controlling key pollution sources, and policy combination linkage. Regarding the candidate intervention strategies... First, update the target indicator vector within the current time window according to the preset intervention intensity. .

[0044] Multi-step propagation simulation includes: updating the target indicator vector within the current time window based on the intervention variables and intervention intensity corresponding to the candidate intervention strategies. According to the propagation iteration formula: (13) Execute preset number of steps The propagation calculation, in which, Indicates the propagation attenuation coefficient. This represents the coupling edge weight matrix under the current time window. Represents the benchmark index vector. This means that the calculation results are restricted to the interval [0,1]. Through the above propagation calculation, it is possible to simulate the indirect impact of changes in direct intervention variables on related indicators and scores of each dimension.

[0045] Based on the multidimensional feature matrix after propagation, the dynamic evaluation value is recalculated to obtain the candidate intervention strategy. Corresponding dynamic evaluation value increment .

[0046] Stability Indicators The stability index can be determined by the degree of fluctuation of the dynamic evaluation value sequence within a continuous time window after the intervention. The smaller the degree of fluctuation, the higher the stability index. Implementation cost indicators It can be determined by one or more of the following: capital investment, execution costs, and organizational coordination costs.

[0047] Based on the incremental dynamic evaluation value after propagation Stability indicators and implementation cost indicators Calculate the comprehensive return value Candidate intervention strategies Overall return value satisfy: (14) in, This is a non-negative adjustment coefficient. Based on the ranking of the comprehensive benefit values ​​corresponding to different candidate intervention strategies or combinations of candidate intervention strategies, the target intervention strategy is determined. Multi-step propagation simulations are performed on multiple candidate intervention strategies, and the comprehensive benefit values ​​are calculated to obtain the dynamic evaluation value change path and comprehensive benefit ranking results for each strategy.

[0048] Candidate intervention strategies are set as one or more of the following: strategy to increase governance input, strategy to strengthen execution response, strategy to strengthen supervision and feedback, strategy to strengthen organizational coordination and supervision linkage, and strategy to address weaknesses. After propagation simulation, the dynamic evaluation value change path, stability results, cost changes and comprehensive benefit value of each candidate strategy can be obtained. Based on this, the candidate intervention strategy or combination of candidate intervention strategies with the highest comprehensive benefit value can be selected.

[0049] The final output can include dynamic assessment level, list of key weaknesses, priority ranking of weaknesses, target intervention strategies, expected improvement range, stability results and cost information, and can be output in the form of a page-based interface, analysis report or early warning prompt.

[0050] This invention constructs a dynamic evaluation and intervention strategy generation mechanism for multi-source heterogeneous time-series data oriented towards atmospheric environmental governance. It realizes a complete closed loop from multi-source governance data acquisition, unified preprocessing, time-varying coupled modeling, dynamic evaluation, key weakness identification to intervention propagation simulation and strategy generation, which can provide more technically feasible process evaluation and result output for regional atmospheric environmental governance.

[0051] Example 9 Taking a continuous monthly window from September 2024 to December 2025 as an example, this paper illustrates how six types of data sources, namely MAQD, ADGE, ODPD, PADIR, RSMD, and ADAP, are mapped to a six-dimensional feature matrix after preprocessing. like Figure 2 As shown, MAQD maps 5 fields to the D1 governance pressure dimension and 3 fields to the D6 governance outcome dimension; ADGE maps 4 fields to the D2 governance input dimension; ODPD and PADIR map relevant fields to the D3 organizational collaboration dimension and the D5 supervision feedback dimension, respectively; RSMD and ADAP mainly map fields to the D4 execution response dimension, while ADAP also maps fields to the D5 supervision feedback dimension. Figure 2 In this table, D1 represents the governance pressure dimension, D2 represents the governance input dimension, D3 represents the organizational coordination dimension, D4 represents the implementation response dimension, D5 represents the supervision and feedback dimension, and D6 represents the governance outcome dimension. MAQD represents monthly air quality data, ADGE represents annual governance input data, ODPD represents publicly available policy text data, PADIR represents publicly available data on inspection and rectification, RSMD represents monthly random inspection data, and ADAP represents annual administrative penalty data.

[0052] Based on the aforementioned six-dimensional feature matrix, the system calculates the dynamic evaluation value sequence from September 2024 to December 2025. For example... Figure 3 As shown, the normalized dynamic assessment values ​​for each month are 1.00, 0.88, 0.56, 0.88, 0.65, 0.30, 0.00, 0.32, 0.15, 0.11, 0.52, 0.49, 0.71, 0.68, 0.19, and 0.75, respectively. This series is used to reflect the continuous changes in the status of atmospheric environmental governance in the target area.

[0053] The system calculates the coupling edge weight matrix between six-dimensional nodes based on time-delay correlation, mutual information value, and grey relational degree. For example... Figure 4 As shown, node pairs such as D3 and D5, D4 and D6, and D1 and D6 have high coupling edge weights, indicating that there is a strong dynamic correlation between organizational collaboration, execution response, supervision feedback, governance pressure, and governance results.

[0054] In the critical weakness identification step, the system calculates the weakness priority index Q by comprehensively considering the difference value, sensitivity, and network centrality. j .like Figure 5 As shown, D4 performs Q-axis response. j The value is 0.86, Q of the D1 governance pressure dimension. j The value is 0.70, the Q value of the D3 organizational synergy dimension. j The value is 0.48, the Q value of the D5 supervised feedback dimension. j The value is 0.47, Q of the D6 governance outcome dimension. j The value is 0.30, representing the Q value in the D2 governance input dimension. j The value is 0.25. Therefore, the key weaknesses are ranked as D4, D1, D3, D5, D6, and D2, with candidate intervention strategies prioritized for D4 and D1.

[0055] In the intervention propagation simulation steps, five candidate intervention strategies, P1 to P5, are set up. P1 represents the strategy of increasing governance input, P2 represents the strategy of strengthening execution response, P3 represents the strategy of strengthening supervision and feedback, P4 represents the strategy of organizational coordination and supervision linkage, and P5 represents the intervention strategy of addressing weaknesses. As shown in Figure 6(a), all five candidate intervention strategies can improve the dynamic evaluation value after three steps of propagation, with P5 and P4 showing the highest improvement. As shown in Figure 6(b), P5 has the highest overall benefit value, followed by P4. Therefore, the system determines the P5 intervention strategy of addressing weaknesses as the priority target intervention strategy. When it is necessary to reduce the complexity of organizational coordination, the P4 strategy of organizational coordination and supervision linkage can be used as an alternative target intervention strategy.

Claims

1. A dynamic evaluation method for multi-source heterogeneous time-series data in atmospheric environmental governance processes, characterized by: Specifically, the steps include the following: Step 1: Obtain multi-source heterogeneous time-series data related to atmospheric environmental governance in the target area within a preset time range; Step 2: Perform time alignment and preprocessing on the multi-source heterogeneous time series data; Step 3: Map the processed data according to the preset index mapping rules to form a multi-dimensional feature matrix under a continuous time window; Step 4: Construct a time-varying coupled network for complex systems based on multidimensional feature matrices; Step 5: Calculate the dimensional score and dynamic evaluation value based on the time-varying coupled network and multidimensional feature matrix of the complex system; Step 6: Output the evaluation results based on the differences between the target area and the benchmark area.

2. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 1, characterized in that: In step 1, the target area is a prefecture-level city, a city cluster, a river basin, or a provincial-level administrative region; the preset time range is any one of daily, weekly, monthly, quarterly, or annual scales. Air quality monitoring data includes one or more of the following: Air Quality Index (AQI), PM2.5 concentration, PM10 concentration, O3 concentration, NO2 concentration, SO2 concentration, CO concentration, percentage of days with good or excellent air quality, and percentage of days with heavy pollution. Meteorological environmental data include one or more of the following: temperature, humidity, wind speed, precipitation, boundary layer height, and diffusion condition index. The governance input data includes one or more of the following: environmental protection fiscal expenditure, monitoring resource input, law enforcement resource input, and digital governance input; Organizational collaboration data includes one or more of the following: frequency of joint meetings, number of joint law enforcement actions, completion rate of collaborative matters, and frequency of cross-regional collaboration. Execution response data includes one or more of the following: early warning response duration, emergency measure implementation rate, and rectification completion efficiency; Supervision and feedback data include one or more of the following: frequency of information disclosure, rate of handling public complaints, and rate of feedback on rectification and inspection. Policy text data includes one or more of the following: government work reports, special rectification plans, notification documents, and public governance documents.

3. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 2, characterized in that: In step 2, the time alignment method includes: mapping data from different sources to the same sampling period in the daily, weekly, monthly, quarterly, or annual scale; using linear interpolation or spline interpolation to fill in missing data with a missing rate not higher than the first threshold, and using weighted nearest neighbor estimation based on relevant indicators to fill in missing data with a missing rate higher than the first threshold. The preprocessing methods include: identifying outliers using an anomaly detection method based on the absolute deviation of the median, and correcting for anomalies using the truncated mean or neighborhood estimate within the corresponding time window.

4. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 3, characterized in that: In step 2, data from different sources are uniformly mapped to a monthly sampling period; for data with a missing rate of less than or equal to 20%, linear interpolation or spline interpolation is used for missing data, and for data with a missing rate of more than 20%, weighted nearest neighbor estimation based on relevant indicators is used for missing data; outliers are identified using an anomaly detection method based on the absolute deviation of the median, and anomaly correction is performed using the truncated mean or neighborhood estimate within the corresponding time window.

5. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 4, characterized in that: In step 2, for numerical indicators, standardization is performed; for positive indicators, the following formula is used: (1) For contrarian indicators, the following formula is used: (2) Where x_{j,t} represents the original observation value of the j-th indicator at time t; z_{j,t} represents the standardized indicator value; max(x_j) and min(x_j) represent the maximum and minimum values ​​of the j-th indicator within a preset time range or a rolling time window, respectively; when max(x_j) = min(x_j), the standardized value of the indicator within the corresponding time window is set to the preset neutral value; For text-based data, text intensity quantification includes: extracting environmental governance keywords, law enforcement keywords, collaborative governance keywords, and supervision feedback keywords from policy text data; and calculating text intensity feature values ​​based on the word frequency, inverse document frequency, and part-of-speech weight of each keyword within the current time window. Text-like features at time The strength value below satisfy: (3) in, Indicates the first Keywords in time The number of valid terms under, Indicates the first The word class weight of each keyword, Indicates the first Key words at any time The statistics below.

6. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 5, characterized in that: The specific process of step 3 is as follows: According to the preset indicator mapping rules, the processed data is mapped to the dimensions of governance pressure, governance input, organizational collaboration, execution response, supervision feedback, and governance results, forming a multidimensional feature matrix under a continuous time window. At time t, the multidimensional feature matrix is ​​represented as X_t=[z_{d,k,t}], where d=1,2,...,6, k=1,2,...,m_d; d represents the governance dimension, m_d represents the number of indicators under the d-th dimension, and z_{d,k,t} represents the standardized value of the k-th indicator under the d-th dimension at time t. The sequence of multidimensional feature matrices under a continuous time window is represented as {X_t | t=1,2,...,T}.

7. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 6, characterized in that: The specific process of step 4 is as follows: The network nodes are either indicator nodes or dimension nodes. The edges between nodes represent the coupling strength between different indicators within the current time window, using lengths of [missing information]. The rolling time window is used for window-by-window calculation, and any two index nodes in the time-varying coupled network of the complex system are calculated. and At any moment Coupled edge weights under satisfy: (4) in, Represents mutual information value, Indicates optimal time delay The correlation coefficient below, Indicates the degree of grey relational relationship. , and is a non-negative weighting coefficient, and: (5) Optimal time delay By searching within a preset time delay set, The largest Sure.

8. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 7, characterized in that: The specific process of step 5 is as follows: No. Each dimension at time Dimensional scores satisfy: (6) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below This represents the standardized index value; No. Each dimension at time Time-varying weights satisfy: (7) in, Indicates the first The first dimension Each indicator at time The weights of the indicators below Indicates the first Information entropy in each dimension Indicates the first Normalized centrality of each dimension in a time-varying coupled network of a complex system at time t. and It is a non-negative adjustment coefficient; This represents the time-varying weight of the u-th dimension at time t; Target area at time Dynamic evaluation value under satisfy: (8) in, Represents the cooperative gain term. This indicates a penalty for risk imbalance. and It is a non-negative adjustment coefficient; Synergistic gain term satisfy: (9) Risk Imbalance Penalty satisfy: (10)。 9. The method for dynamic evaluation of multi-source heterogeneous time-series data for atmospheric environmental governance processes according to claim 6, characterized in that: The specific process of step 6 is as follows: The benchmark region is a reference region with consistent administrative levels and similar governance tasks, or a benchmark sample constructed from the historical best-performing time window of the target region; for each indicator Calculate the difference between the target area and the reference area. And calculate the sensitivity of the corresponding indicators to the dynamic evaluation value. , Sensitivity Calculated using a single-index perturbation method, for the first... Each indicator increased and reduce Recalculated and ,but: (11) in, and They represent the first The number of indicators increased and reduce The subsequent dynamic evaluation value; The centrality eigenvalues ​​of the index in the time-varying coupled network of the complex system are calculated. By combining the difference value, sensitivity and centrality, the short-board priority index can be obtained. The key short-board ranking results of the six-dimensional features are output based on the short-board priority index, as shown in the following formula (12): (12) Among them, Q j This represents the bottleneck priority index of the j-th indicator at time t; for candidate intervention strategies First, update the target indicator vector within the current time window according to the preset intervention intensity. ; Update the target indicator vector within the current time window based on the intervention variables and intervention intensity corresponding to the candidate intervention strategies. According to the propagation iteration formula: (13) Execute preset number of steps The propagation calculation, in which, Indicates the propagation attenuation coefficient. This represents the coupling edge weight matrix under the current time window. Represents the benchmark index vector. This means that the calculation result is restricted to the interval [0,1]. Based on the incremental dynamic evaluation value after propagation Stability indicators and implementation cost indicators Calculate the comprehensive return value Candidate intervention strategies Overall return value satisfy: (14) in, It is a non-negative adjustment coefficient.