Enterprise ESG index determination method and system based on data fusion

Abstract

The invention relates to the technical field of enterprise evaluation, and discloses an enterprise ESG index determination method and system based on data fusion. According to the method, internal and external environment, social and governance data streams of an enterprise are captured in real time, and a standardized ESG data sequence is formed through noise filtering and format alignment processing; performing scale transformation on the sequence to eliminate unit difference, generating a dimensionless ESG data matrix, and further synthesizing the dimensionless ESG data matrix into a uniform ESG feature tensor; based on the feature tensor, calculating an enterprise ESG index reference value through dynamic weighted aggregation; performing matching analysis on the reference value and a pre-stored ESG index dynamic threshold interval, and if the reference value exceeds the interval, triggering an ESG deviation alarm; when an alarm is responded, the system collects the latest governance data sequence of the enterprise and calculates the fluctuation degree of the enterprise, the ESG index reference value is dynamically calibrated according to the fluctuation degree, and finally an accurate enterprise ESG index output value is generated. According to the method, scientificity, real-time performance and reliability of enterprise ESG index calculation are improved.

Description

Technical Field

[0001] This invention relates to the field of enterprise evaluation technology, specifically to a method and system for determining enterprise ESG indices based on data fusion. Background Technology

[0002] Current corporate ESG index assessments primarily employ periodic data collection and static weighted averaging methods. Existing technologies for processing ESG data largely rely on manual collection and organization, resulting in long data update cycles and poor timeliness. During the assessment process, various types of ESG data suffer from inconsistent dimensions and standards, making direct comprehensive calculation difficult. Index generation typically uses fixed-weight formulas, failing to adapt to the unique characteristics of different industries and companies. Static threshold settings cannot reflect changes in the market environment and policy requirements. The lack of index calibration mechanisms hinders rapid response and adjustment to anomalies. Existing methods need to address key technical issues such as data real-time performance, consistent dimensions, dynamic weight adjustment, and rapid response to anomalies.

[0003] Traditional ESG assessment methods have significant shortcomings in data integration and index calculation. Data is collected from multiple sources, with scattered data formats and high integration difficulty. Data preprocessing methods are simplistic, resulting in ineffective noise filtering and impacting data quality. Scaling transformation methods are limited and fail to effectively eliminate the dimensional influence between different indicators. Feature synthesis methods are crude, failing to fully consider the intrinsic correlations among the three ESG dimensions. Threshold settings are based on historical experience and lack dynamic adjustment mechanisms, leading to insufficient warning sensitivity. Calibration processes rely on manual judgment, resulting in high subjectivity and slow response times. Existing technologies require the establishment of a dynamic optimization scheme for the entire process from data acquisition to index generation. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for determining enterprise ESG indices based on data fusion, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for determining an enterprise ESG index based on data fusion, the method comprising: Real-time capture of environmental, social, and governance data streams from within and outside the enterprise; noise filtering and format alignment of the environmental, social, and governance data streams to form a standardized ESG data sequence; The standardized ESG data sequences are scaled to eliminate unit inconsistencies and generate dimensionless ESG data matrices; the dimensionless ESG data matrices are then synthesized into a unified ESG feature tensor. Based on the ESG feature tensor, the benchmark value of the enterprise's ESG index is calculated through dynamic weighted aggregation operation; the benchmark value of the enterprise's ESG index is matched and analyzed with the pre-stored dynamic threshold range of the ESG index; when the benchmark value of the enterprise's ESG index exceeds the dynamic threshold range, an ESG deviation alarm is generated. In response to ESG deviation alerts, the system collects the latest governance data sequence of enterprises and calculates the volatility of governance data. Based on the volatility of governance data, the system dynamically calibrates the benchmark value of the enterprise's ESG index and generates the enterprise's ESG index output value.

[0006] Preferably, the calculation of the enterprise ESG index benchmark value includes: decomposing the enterprise's operating structure into multiple functional modules, collecting environmental data streams, social data streams, and governance data streams for each functional module; assigning dynamic weight factors to the environmental data streams, social data streams, and governance data streams for each functional module, calculating the ESG local evaluation value for each functional module; and weighting and fusing the ESG local evaluation values ​​of all functional modules to obtain the enterprise ESG index benchmark value.

[0007] Preferably, the calculation of governance data volatility includes: obtaining the governance data temporal variability and the governance data spatial dispersion, and linearly superimposing the governance data temporal variability and the governance data spatial dispersion to obtain the governance data volatility.

[0008] Preferably, the acquisition of governance data time variability includes: identifying synchronous fluctuation periods between the governance data sequence and the ESG index benchmark value sequence within the evaluation time window; calculating the correlation ratio between the rate of change of governance data and the rate of change of ESG index benchmark value within the synchronous fluctuation period; screening highly synchronous periods based on the correlation ratio and recording them as key periods; measuring the proportion of the cumulative duration of the key periods to the total duration of the evaluation time window, and using this proportion as the governance data time variability.

[0009] Preferably, the process of identifying the synchronous fluctuation period between the governance data sequence and the ESG index benchmark value sequence includes: within the evaluation time window, constructing a time axis with equally spaced sampling points, and plotting curves showing the change of governance data values ​​over time and curves showing the change of ESG index benchmark values ​​over time; extracting the peaks and troughs of the governance data curves, calculating the slope between adjacent peaks and troughs to obtain the governance data change rate; extracting the peaks and troughs of the ESG index benchmark value curves, calculating the slope between adjacent peaks and troughs to obtain the ESG index benchmark value change rate; and comparing the direction and magnitude of the governance data change rate and the ESG index benchmark value change rate to determine the synchronous fluctuation period.

[0010] Preferably, obtaining the spatial dispersion of governance data includes: mapping the enterprise governance structure to a spatial grid, with each grid cell corresponding to a governance data sampling point; calculating the absolute difference between the measured and expected values ​​of the governance data for each grid cell, denoted as the grid governance deviation; comparing the grid governance deviation with the grid governance deviation tolerance: if the grid governance deviation is greater than or equal to the grid governance deviation tolerance, the grid is marked as a governance anomalous grid; simultaneously, calculating the absolute difference between the measured and expected values ​​of the ESG index benchmark for each grid cell, denoted as the grid ESG deviation; comparing the grid ESG deviation with the grid ESG deviation tolerance: if the grid ESG deviation is greater than or equal to the grid ESG deviation tolerance, the grid is marked as an ESG anomalous grid; counting the number of overlapping grids between governance anomalous grids and ESG anomalous grids; calculating the proportion of the total area of ​​overlapping grids to the total area of ​​the entire spatial grid, and using this proportion as the spatial dispersion of governance data.

[0011] Preferably, calculating the grid governance deviation includes: obtaining the real-time value of governance data for each grid cell through a sensor network or database interface; extracting the standard value of governance data for each grid cell from historical data; and calculating the absolute value of the difference between the real-time value of governance data and the standard value to obtain the grid governance deviation. The calculation of grid ESG bias includes: obtaining the real-time benchmark value of the ESG index for each grid cell through the monitoring system; obtaining the standard value of the ESG index for each grid cell from the benchmark model; and calculating the absolute value of the difference between the real-time benchmark value and the standard value of the ESG index to obtain the grid ESG bias.

[0012] Preferably, the calculation of the correlation ratio between the rate of change of governance data and the rate of change of the ESG index benchmark value during the synchronous fluctuation period includes: extracting numerical pairs of the rate of change of governance data and the rate of change of the ESG index benchmark value during the key period; calculating the ratio of the rate of change of governance data to the rate of change of the ESG index benchmark value for each numerical pair; and taking the arithmetic mean of the ratios of all numerical pairs to obtain the average correlation ratio, which is used as the correlation ratio. The method of selecting high synchronization periods based on correlation ratio includes: calculating the deviation of the correlation ratio of each key period from the ideal correlation ratio; calculating the standard deviation of the deviation of all key periods to obtain the synchronization stability index; and adjusting the selection threshold of key periods based on the synchronization stability index.

[0013] Preferably, the dynamic calibration of the corporate ESG index benchmark value includes: deriving an index correction factor from the volatility of governance data; multiplying the corporate ESG index benchmark value by the index correction factor to obtain the corporate ESG index output value; wherein the calculation of the index correction factor includes: extracting the synchronous stability index for all key periods; calculating the harmonic mean of the synchronous stability index as the index correction factor.

[0014] Preferably, the present invention also includes a data fusion-based enterprise ESG index determination system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described data fusion-based enterprise ESG index determination method.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Standardized environmental, social, and governance (ESG) data sequences are scaled to generate a dimensionless ESG data matrix. The scaling transformation employs range standardization or Z-score standardization to eliminate the influence of dimensions in the original data. Range standardization linearly transforms the data to a specific interval, while Z-score standardization ensures the data conforms to a standard normal distribution. Dimensionless processing makes ESG indicators of different properties and magnitudes comparable, facilitating subsequent comprehensive calculations. The generated dimensionless ESG data matrix preserves the original distribution characteristics and variability of each dimension. The dimensionless ESG data matrix is ​​then synthesized into a unified ESG feature tensor. Tensor synthesis integrates data from the environmental, social, and governance dimensions through a multi-dimensional array structure. The tensor dimensions correspond to time series, indicator categories, and data collection points, forming a complete data cube. The feature tensor preserves the temporal dynamics and indicator correlations of the ESG data, providing a structured data foundation for in-depth analysis. The unified tensor representation facilitates feature learning and pattern recognition using modern machine learning algorithms.

[0016] The system matches the company's ESG index benchmark value with a pre-stored dynamic threshold range, generating an ESG deviation alert when the benchmark value exceeds the threshold range. The dynamic threshold range is updated regularly based on industry characteristics, company size, and changes in the macro environment. Upper and lower boundaries are set for the threshold range to define a reasonable range of fluctuations in ESG performance. The matching analysis calculates the degree and direction of deviation between the benchmark value and the threshold boundaries, identifying anomaly types. Deviation alerts are triggered in stages based on the degree of deviation, ensuring the accuracy and timeliness of warnings. In response to ESG deviation alerts, the system collects the company's latest governance data sequence to calculate governance data volatility. The latest governance data includes key indicators such as board decision-making efficiency, internal control quality, and information disclosure transparency. Governance data volatility is quantified by calculating the standard deviation or coefficient of variation of the indicators, reflecting the stability of the governance status. Governance volatility assesses the company's inherent ability and responsiveness to ESG anomalies. The company's ESG index benchmark value is dynamically calibrated based on governance data volatility. The calibration process considers the magnitude and duration of governance volatility; the greater the volatility, the larger the calibration amplitude. The calibration algorithm uses a sliding window weighted average to smooth the impact of short-term fluctuations. Dynamic calibration makes the ESG index more accurately reflect the company's true level of sustainable development. Through the synergistic effects of scaling transformation, tensor synthesis, threshold matching, and volatility calibration, the scientific generation and real-time optimization of ESG indices are achieved. Multi-source data integration ensures comprehensive assessment, dynamic threshold mechanisms improve early warning sensitivity, and volatility calibration enhances index accuracy. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the enterprise ESG index determination method based on data fusion described in this invention. Figure 2 A flowchart illustrating the calculation method for the benchmark value of an enterprise's ESG index; Figure 3 A flowchart for methods of obtaining time variability of governance data; Figure 4 A graph showing the relationship between correlation ratio and synchronization stability index; Figure 5 This is a comparison chart of the synchronization stability index and the index correction factor during critical periods. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides a method and system for determining an enterprise's ESG index based on data fusion. The method includes: real-time capture of environmental, social, and governance data streams from within and outside the enterprise, sourced from various channels such as sensor networks, public databases, and internal enterprise systems. Noise filtering is applied to the environmental, social, and governance data streams, outliers are eliminated using a sliding window averaging method, and field alignment is achieved through data format conversion tools to form a standardized ESG data sequence. The standardized ESG data sequence undergoes scaling transformation, and a min-max normalization method is used to map data from different units to a unified interval, eliminating unit inconsistencies and generating a dimensionless ESG data matrix. The dimensionless ESG data matrix is ​​synthesized into a unified ESG feature tensor through tensor synthesis, which integrates multi-dimensional data features. Based on the ESG feature tensor, a benchmark value for the enterprise's ESG index is calculated through dynamic weighted aggregation, with the weighting factors dynamically adjusted according to data freshness and importance. The system matches and analyzes the company's ESG index benchmark value against a pre-stored dynamic threshold range for ESG indices. This dynamic threshold range is updated in real-time based on historical industry data. When the company's ESG index benchmark value exceeds the dynamic threshold range, the system automatically generates an ESG deviation alarm. In response to the ESG deviation alarm, the system collects the company's latest governance data sequence and calculates the governance data volatility, which quantifies the degree of variation in the governance data. Based on the governance data volatility, the system dynamically calibrates the company's ESG index benchmark value and generates the company's ESG index output value through a multiplicative correction factor, ensuring that the index reflects the latest governance status.

[0020] Example 1: See Figure 2In practice, the process of decomposing the enterprise's operational structure into multiple functional modules is based on the enterprise's organizational structure and business process logic. The criteria for dividing functional modules include departmental functional independence, data collectability, and the degree of impact on overall environmental, social, and governance performance. Typical functional modules include manufacturing, supply chain management, human resources, R&D, and quality management, each with its own independent data collection channel. When collecting environmental, social, and governance data streams for each functional module, environmental data streams primarily originate from the energy consumption monitoring system of the manufacturing module and logistics carbon emission records of the supply chain management module. Social data streams come from employee satisfaction surveys in the human resources module and customer complaint data from the quality management module. Governance data streams come from board decision records and compliance audit reports. In practice, dynamic weighting factors are assigned to the environmental, social, and governance data streams of each functional module. The calculation of these dynamic weighting factors is based on the module's strategic importance and data quality assessment results. Strategic importance is determined through expert scoring, while data quality assessment examines data completeness, timeliness, and accuracy indicators. For example, the environmental data flow weight factor for the manufacturing module might be set to 0.4, while the social data flow weight factor for the human resources module might be set to 0.3. These weight factors will be dynamically adjusted as the company's strategic priorities change. When calculating the local environmental, social, and governance evaluation values ​​for each functional module, a weighted average algorithm is used to aggregate multi-source data within the module. The environmental, social, and governance data flows also have sub-weights within the module, which are allocated based on the relevance of the data to the module's core functions.

[0021] In some embodiments, the calculation of the environmental, social, and governance (ESG) local evaluation values ​​of functional modules adopts a multi-layer weighted fusion method. First, the indicators within the environmental data stream are standardized. Then, environmental dimension scores are calculated according to sub-weights. The social and governance data streams are processed similarly. Finally, the environmental, social, and governance dimension scores are combined according to dynamic weighting factors. When weighting and fusing the ESG local evaluation values ​​of all functional modules, the weighting coefficients are determined based on the module's contribution to the overall enterprise operation. This contribution is quantified by the module's revenue share or resource input ratio. After obtaining the benchmark values ​​for the enterprise's ESG index, the governance data volatility calculation stage begins. This calculation requires obtaining two components: governance data temporal variability and governance data spatial dispersion. The calculation of governance data temporal variability is based on time series analysis methods, while governance data spatial dispersion relies on spatial statistical techniques. After the two components are calculated, they are linearly superimposed to generate the governance data volatility. The coefficients for linear superposition are preset according to the enterprise type and data characteristics. For manufacturing enterprises, a higher weight may be given to governance data spatial dispersion, while for service enterprises, more emphasis may be placed on governance data temporal variability.

[0022] In practical implementation, obtaining the temporal variability of governance data requires analyzing the fluctuation characteristics of the governance data sequence over time. The temporal variability reflects the drastic change of governance indicators over time. The calculation of the temporal variability of governance data employs a sliding window technique, calculating the variance and coefficient of variation of the governance data within a specified time interval, and combining this with trend analysis results to derive a quantitative value. Obtaining the spatial dispersion of governance data requires mapping the governance data to a spatial dimension. The spatial dispersion characterizes the degree of difference in governance levels across different geographical or organizational units. The calculation of the spatial dispersion of governance data uses spatial autocorrelation analysis, measuring the uneven distribution of governance data through the Moran index or Gini coefficient. When linearly superimposing the temporal variability and spatial dispersion of governance data, the formula is: Governance data volatility = α × Temporal variability of governance data + β × Spatial dispersion of governance data, where α and β are weighting coefficients, the values ​​of which are determined through historical data regression analysis.

[0023] It is understandable that the granularity of functional module division affects the accuracy of environmental, social, and governance (ESG) index benchmark values. Overly coarse module division may mask internal differences, while overly fine division increases computational complexity. In implementation, a dynamic module division mechanism is adopted, automatically adjusting the number of modules based on enterprise size and organizational complexity. Large enterprises may be divided into more than ten functional modules, while small and medium-sized enterprises may only be divided into five to six core modules. The collection frequency of environmental, social, and governance data streams for each functional module also varies. Environmental data streams may be collected in real time, social data streams may be collected monthly, and governance data streams may be updated quarterly. The system automatically aligns data from different time granularities. When assigning dynamic weighting factors to the environmental, social, and governance data streams of each functional module, the dynamic weighting factor update cycle is synchronized with the enterprise's strategic assessment cycle, typically adjusted every six months or one year. When calculating the local ESG evaluation values ​​for functional modules, data quality verification is performed, outliers are removed, and missing data is supplemented to ensure the reliability of the evaluation values.

[0024] In some embodiments, the calculation of governance data temporal variability can employ various time series analysis models, such as autoregressive integral moving average models or seasonal decomposition models, with the model selection determined by the characteristics of the governance data. The calculation of governance data spatial dispersion can consider spatial interpolation techniques to estimate data for grid cells not directly sampled, improving spatial coverage integrity. The linear superposition coefficient of governance data temporal variability and governance data spatial dispersion can be optimized using machine learning algorithms, training a regression model with historical data to predict the optimal coefficient combination. The governance data volatility after linear superposition needs to be normalized so that its value falls between zero and one, facilitating the subsequent derivation of the index correction factor. The governance data volatility calculation module runs asynchronously with the enterprise environmental, social, and governance index benchmark value calculation module, exchanging data through a message queue to ensure system processing efficiency.

[0025] Optionally, the environmental data stream of the functional modules can be further subdivided into resource consumption data, emission data, and ecological impact data; the social data stream into employee rights data, community relations data, and product liability data; and the governance data stream into board structure data, shareholder rights data, and audit supervision data. Each subcategory has an independent data collection channel and quality control rules. After preprocessing, the subcategory data is aggregated into the environmental, social, and governance data streams of the functional modules. When assigning dynamic weighting factors to the environmental, social, and governance data streams of each functional module, the calculation of the dynamic weighting factors incorporates the entropy weighting method, automatically adjusting the weight allocation based on the amount of data information. When calculating the local environmental, social, and governance evaluation values ​​of the functional modules, fuzzy comprehensive evaluation methods can be used to process qualitative indicators, transforming verbal descriptions into numerical scores. The calculation of the time variability of governance data can incorporate multi-scale analysis, examining variation characteristics at different time granularities to avoid the limitations of single-scale analysis.

[0026] In practical implementation, the spatial dispersion calculation of governance data can employ a multi-level spatial grid. Different levels of grid units are set according to the enterprise's organizational structure, such as headquarters level, department level, and team level. Each level of grid unit corresponds to governance data sampling points of different granularities; higher-level grids use summary data, while lower-level grids use detailed data. The linear superposition of governance data temporal variability and spatial dispersion can be considered using non-linear combinations, such as introducing interaction terms or employing neural network fusion. However, linear superposition has the advantages of strong interpretability and computational simplicity. The results of governance data volatility calculation are stored in a historical database for analyzing volatility trends and setting volatility thresholds. When governance data volatility exceeds the threshold, a system warning is triggered. As a key input for the dynamic calibration of environmental, social, and governance (ESG) index benchmarks, the accuracy of governance data volatility calculation directly affects the quality of the final ESG index output values.

[0027] It is understandable that the decomposition of an enterprise's operational structure into functional modules should adhere to the principles of low coupling between modules and high cohesion within modules, ensuring that each functional module has relatively independent environmental, social, and data governance data flows. The functional module division scheme needs to be dynamically updated based on changes in the enterprise's business. When the enterprise undergoes organizational restructuring or business expansion, the functional module division needs to be adjusted accordingly. When collecting environmental, social, and governance data flows from functional modules, a data lineage tracing mechanism needs to be established to record data sources and processing procedures, meeting audit requirements. When assigning dynamic weighting factors to the environmental, social, and governance data flows of each functional module, the weighting rules need to be transparent and traceable to avoid subjective bias affecting the fairness of the environmental, social, and governance index benchmark values. The calculation of governance data temporal variability and governance data spatial dispersion requires sufficient historical data support. For newly established enterprises or those with insufficient data accumulation, industry benchmark values ​​can be used as a substitute, switching to enterprise-specific algorithms once sufficient data is available.

[0028] Optionally, the calculation of governance data temporal variability can incorporate volatility persistence analysis to examine whether governance data volatility exhibits long-term memory, thereby better predicting future volatility trends. The calculation of governance data spatial dispersion can consider spatial heterogeneity, assigning different importance weights to different regions, with deviations in governance data from important regions contributing more to the overall spatial dispersion. The linear superposition coefficient between governance data temporal variability and governance data spatial dispersion can be determined through sensitivity analysis, testing the impact of governance data volatility on the calibration effectiveness of environmental, social, and governance indices under different coefficient combinations. The governance data volatility calculation module can be deployed as a distributed microservice to improve the scalability and fault tolerance of large-scale enterprise data processing. The dynamic calibration relationship between governance data volatility and the benchmark values ​​of enterprise environmental, social, and governance indices can be modeled using feedback mechanisms from control theory to achieve more precise index adjustment.

[0029] Example 2: See Figure 3In practice, obtaining the time variability of governance data needs to be done within the assessment time window. The length of the assessment time window is set according to the company's reporting cycle and industry characteristics, with a common assessment time window being a fiscal quarter or half a fiscal year. Identifying the synchronous fluctuation periods between the governance data sequence and the benchmark values ​​of the environmental, social, and governance indices is a core step. Synchronous fluctuation periods refer to the time intervals during which the governance data sequence and the benchmark values ​​of the environmental, social, and governance indices show significant consistency in the direction and magnitude of change. The correlation ratio between the rate of change of governance data and the rate of change of the benchmark values ​​of the environmental and social governance indices within the synchronous fluctuation period is calculated. The correlation ratio is an important indicator for quantifying the coordination of changes in the two sequences. Highly synchronous periods are selected based on the correlation ratio and are designated as key periods. These key periods form the basis for subsequent calculations of the time variability of governance data. The cumulative duration of the key periods is measured as a proportion of the total assessment time window. This proportion is directly used as the numerical output of the time variability of governance data. A higher time variability indicates a stronger correlation between the volatility of governance data over time and the benchmark values ​​of the environmental, social, and governance indices.

[0030] In practical implementation, identifying the synchronous fluctuation periods between the governance data sequence and the benchmark value sequence of the environmental, social, and governance indices requires constructing a time axis using equally spaced sampling points. The interval length of the equally spaced sampling points is determined according to the data update frequency; for example, daily data uses daily sampling points, and weekly data uses weekly sampling points. Curves depicting the changes in governance data values ​​over time and the changes in the benchmark values ​​of the environmental, social, and governance indices over time are plotted separately. Line graphs or spline interpolation are used to smooth fluctuations. Local extremum detection algorithms are used to extract the peaks and troughs of the governance data curves. A peak is defined as a point where the value is greater than the values ​​of its immediate neighbors, and a trough is defined as a point where the value is less than the values ​​of its immediate neighbors. The slope between adjacent peaks and troughs is calculated using a two-point difference formula to obtain the rate of change of the governance data. A positive rate of change indicates an upward trend, and a negative rate of change indicates a downward trend. The same extremum detection algorithm is used to extract the peaks and troughs of the benchmark value curves of the environmental, social, and governance indices, and the slope between adjacent peaks and troughs is calculated to obtain the rate of change of the benchmark values ​​of the environmental, social, and governance indices. The direction and magnitude of the change rate of governance data and the change rate of the benchmark values ​​of environmental, social and governance indices are compared. The consistency of direction requires that the two change rates are both positive or both negative, and the similarity of magnitude requires that the ratio of the absolute values ​​of the two change rates is within a preset range. The period that meets both the direction and magnitude conditions is determined as the synchronous fluctuation period.

[0031] In some embodiments, the evaluation time window can be divided using a sliding window mechanism, sliding the evaluation time window forward at fixed time intervals to achieve continuous calculation of the time variability of governance data. Identification of synchronous fluctuation periods can be achieved by introducing a dynamic time warping algorithm to eliminate small phase differences between the governance data sequence and the benchmark sequences of environmental, social, and governance indices on the time axis, improving the accuracy of time period matching. The correlation ratio can be calculated using a weighted average method, assigning different weights to pairs of values ​​at different time points, with higher weights for recent data and lower weights for older data. A correlation ratio threshold can be set for the selection of key periods; only synchronous fluctuation periods with a correlation ratio greater than the threshold are identified as key periods. The threshold is dynamically adjusted based on the distribution of historical data. The proportion of governance data time variability can be calculated using the area integration method, calculating the ratio of the area under the curve for key periods to the total area of ​​the evaluation time window, more accurately reflecting the impact of the fluctuation duration.

[0032] It is understandable that the density of equally spaced sampling points affects the accuracy of identifying synchronous fluctuation periods. Too sparse sampling points may miss brief fluctuations, while too dense sampling points increase the computational burden. In implementation, an adaptive sampling interval adjustment strategy is adopted, automatically increasing the number of sampling points during periods of severe data fluctuation and relaxing the sampling interval during periods of stable data. The curves showing the change of governance data values ​​over time and the curves showing the change of environmental, social, and governance index benchmark values ​​over time must use the same coordinate scale to ensure the comparability of the rates of change. The extraction of peaks and troughs requires setting minimum amplitude thresholds to avoid misjudging minor fluctuations as valid extreme points. The comparison between the rate of change of governance data and the rate of change of environmental, social, and governance index benchmark values ​​requires the introduction of a tolerance mechanism, allowing the direction to remain consistent after a short delay, and setting upper and lower buffer boundaries for the amplitude ratio range. The determination results of synchronous fluctuation periods need to undergo consistency verification, and multi-window cross-validation is used to reduce misjudgments.

[0033] Optionally, the acquisition of governance data sequences can include multi-dimensional indicators such as board meeting frequency, changes in executive compensation, and compliance audit results. The environmental, social, and governance (ESG) benchmark value sequences are derived from historical records of ESG benchmark values ​​calculated in real time. The starting point of the assessment time window can be selected as the date of a significant corporate governance event, such as a board reorganization or strategy release, making the analysis more targeted. Identification of periods of synchronized fluctuation can be achieved using event study methodology, focusing on the interconnected reactions between two sequences after a specific governance event. The calculation of correlation ratios can incorporate logarithmic ratio transformations to make the ratio distribution closer to a normal distribution, facilitating statistical analysis. The cumulative duration measurement of key periods can be calculated using timestamps accurate to the hour, improving the accuracy of calculating the time variability of governance data.

[0034] In practical implementation, the calculation of the rate of change in governance data can consider using instantaneous rate of change rather than piecewise average rate of change. The instantaneous rate of change for each sampling point can be obtained through differentiation, and then the ratio of the average rate of change within the synchronous period can be calculated. A similar method can be used to calculate the rate of change of the benchmark values ​​of environmental, social, and governance indices, improving the granularity of the rate of change calculation. The detection of peaks and troughs can incorporate multi-scale analysis to identify extreme points at different time scales, avoiding information loss at a single scale. The comparison of direction and amplitude can use the cosine of the vector angle as a similarity measure; the closer the cosine value is to 1, the better the synchronicity. The boundary determination of synchronous fluctuation periods can use fuzzy clustering algorithms to handle the fuzzy attribution issues at the time-period boundaries. The proportion of the temporal variability of governance data needs to be standardized, mapped to a standard range of zero to one, facilitating cross-enterprise comparisons and threshold setting.

[0035] In some embodiments, the synchronicity analysis of governance data sequences with benchmark sequences of environmental, social, and governance indices can incorporate Granger causality tests to verify whether changes in governance data statistically lead changes in benchmark values ​​of environmental and social governance indices, thereby determining the causal direction. The calculation of correlation ratios can be divided into rising-phase correlation ratios and falling-phase correlation ratios, examining the linkage characteristics of the two sequences during the rising and falling phases, respectively. The selection of key time periods can be combined with machine learning classification models, using historical data to train a classifier to automatically identify highly synchronized periods. The calculation of governance data time variability can incorporate time decay weights, with recent key periods having higher weights than distant key periods, making the governance data time variability more reflective of the current situation. The entire identification and calculation process can be encapsulated as an independent microservice, receiving input data and returning governance data time variability results through an application programming interface.

[0036] It is understandable that the selection of the evaluation time window needs to avoid the cyclical boundary effect; the window boundary should avoid periods of drastic data fluctuations to prevent truncation effects from affecting the analysis results. Preprocessing of the governance data sequence and the benchmark value sequence of the environmental and social governance indices needs to include missing value imputation and outlier smoothing to ensure data quality. The peak and trough detection algorithm needs to set minimum interval constraints to avoid detecting too many invalid extreme points in localized minor fluctuations. The comparison of the direction and magnitude of the rate of change needs to be tailored to different industries; asset-heavy industries may allow for larger tolerances for magnitude differences, while asset-light industries require stricter magnitude matching. The calculation of the proportion of governance data temporal variability needs to consider time factors such as leap years and changes in working days, using actual duration rather than calendar days for calculation. The identification results of synchronous fluctuation periods need to be visualized, providing interfaces for manual review and parameter adjustment to ensure algorithm transparency.

[0037] Optionally, principal component analysis can be used to construct the governance data sequence, reducing the dimensionality of multiple governance indicators to synthesize a comprehensive governance index sequence and minimizing noise impact. The calculation of the benchmark values ​​for environmental, social, and governance indices can incorporate moving average smoothing to eliminate random fluctuations and highlight trend changes. Identification of synchronous fluctuation periods can be achieved by combining wavelet coherence analysis to detect the resonance frequencies and resonance periods of two sequences in the time-frequency joint domain. The correlation ratio can be calculated using robust statistical methods, employing the median instead of the arithmetic mean to reduce the impact of outliers. The cumulative duration measurement of key periods can adopt the concept of effective duration, calculating only periods with fluctuation amplitudes exceeding a threshold and ignoring periods of minor fluctuations. The final output of the temporal variability of the governance data can include confidence intervals, and the statistical uncertainty of the variability estimate can be calculated using bootstrap resampling.

[0038] Example 3: In practical implementation, obtaining the spatial dispersion of governance data first requires mapping the enterprise governance structure to a spatial grid. The spatial grid is divided based on the enterprise's physical distribution and organizational hierarchy. Each grid cell corresponds to a governance data sampling point, and the grid size is adjusted according to data density and business needs. The absolute difference between the measured and expected values ​​of governance data for each grid cell is calculated. This absolute difference is calculated using absolute value operations and is denoted as the grid governance deviation. The grid governance deviation reflects the degree of deviation of governance data within a single grid cell. The grid governance deviation is compared with the grid governance deviation tolerance, which is a threshold set based on historical data statistics or industry standards. If the grid governance deviation is greater than or equal to the grid governance deviation tolerance, the grid is marked as a governance anomaly grid. Simultaneously, the absolute difference between the benchmark measured and expected values ​​of the environmental, social, and governance indices for each grid cell is calculated and denoted as the grid environmental, social, and governance index deviation. This deviation characterizes the abnormal level of the environmental, social, and governance indices at the grid level. The deviations of the grid's environmental, social, and governance indices are compared with the tolerance limits for these deviations. If the deviations are greater than or equal to the tolerance limits, the grid is marked as an anomalous grid for these indices. The number of overlapping grids with anomalous governance grids and anomalous environmental, social, and governance index grids is counted. Overlapping grids represent areas where both governance problems and anomalous environmental, social, and governance indices exist simultaneously. The proportion of the total area of ​​overlapping grids to the total area of ​​the entire spatial grid is calculated and used as the spatial dispersion of the governance data. A higher spatial dispersion value indicates a more significant spatial clustering of governance problems.

[0039] In practical implementation, calculating grid governance deviation requires obtaining real-time governance data values ​​for each grid cell through sensor networks or database interfaces. These real-time governance data values ​​include dynamic indicators such as board decision frequency and internal control scores. A standard value for the governance data of each grid cell is extracted from historical data. This standard value typically uses a moving average or seasonally adjusted value to reduce the impact of random fluctuations. The absolute value of the difference between the real-time governance data value and the standard value is calculated to obtain the grid governance deviation. The calculation of grid governance deviation follows the formula:

[0040] in: This indicates a deviation in grid management. This represents the real-time value of the governance data. This represents the standard value for governance data.

[0041] Calculating the deviation of the grid's environmental, social, and governance (ESG) indices requires obtaining the benchmark real-time values ​​of these indices for each grid cell through a monitoring system. These benchmark real-time values ​​are derived from a real-time computing engine. Standard values ​​for the ESG indices for each grid cell are obtained from a benchmark model, which is trained using a machine learning algorithm. The absolute value of the difference between the benchmark real-time values ​​and the standard values ​​is calculated to obtain the deviation of the grid's ESG indices. The calculation of the deviation employs a method similar to the absolute value difference approach.

[0042] In some embodiments, spatial grid mapping can employ an adaptive grid partitioning algorithm, dynamically adjusting the grid granularity based on the geographical density of enterprise branches, using finer grids in high-density areas and coarser grids in low-density areas. Acquiring measured governance data can integrate IoT sensors and business system interfaces to achieve automatic multi-source data collection. Determining governance data standard values ​​can incorporate a time decay factor, with higher weighting for recent historical data and lower weighting for older data, making the standard values ​​more reflective of current trends. Setting grid governance deviation tolerance can employ a percentile method, for example, using the 95th percentile of historical grid governance deviations as the tolerance threshold. Marking abnormal governance grids can be combined with spatial clustering analysis, merging adjacent abnormal grids into abnormal regions to improve the continuity of spatial analysis. Counting the number of overlapping grids can be achieved using spatial intersection operations, automatically calculating the grid overlap area using GIS tools.

[0043] It is understandable that the spatial grid division needs to ensure full coverage and no overlap, with each enterprise area belonging to a unique grid unit to avoid data duplication or omission. The placement of governance data sampling points should consider business importance, with higher-density sampling points set up in critical governance areas such as headquarters or data centers. The collection frequency of real-time governance data values ​​should be synchronized with the business rhythm, and data should be updated promptly after significant governance events. The calculation of governance data standard values ​​must exclude data from abnormal periods to prevent extreme values ​​from distorting the standard benchmark. Grid governance deviation tolerances should be reviewed and updated regularly to adapt to changes in the enterprise's governance level. The marking results of governance anomaly grids should be visualized to help managers identify problem hotspots. Overlapping grid statistics should handle grid boundary situations, using an area-weighted method to ensure accurate ratio calculations.

[0044] In some embodiments, the calculation of governance data spatial dispersion can introduce grid weight coefficients, with important grids having higher weights and minor grids having lower weights, making the governance data spatial dispersion more focused on key areas. The determination of abnormal governance grids can employ fuzzy logic methods, converting the comparison results of grid governance deviations and tolerances into membership degrees to handle boundary ambiguities. The labeling of abnormal grids in environmental and social governance indices can be combined with the severity classification of environmental and social governance index deviations, with different levels of abnormal grids contributing differently to overlap statistics. The calculation of overlapping grid area can use pixel counting or vector area calculation, selected based on the grid representation. The output of governance data spatial dispersion can be normalized to the range of zero to one for easy comparison across time. The entire governance data spatial dispersion calculation process can be encapsulated as a parallel processing module, supporting efficient computation of large-scale grid data.

[0045] Optionally, the spatial grid can be constructed using a hierarchical grid structure, with a coarse-grained top layer for macroscopic analysis and a fine-grained bottom layer for microscopic diagnosis. Acquiring measured values ​​of governance data can incorporate a data validation step, reducing errors by cross-checking multiple data sources. The generation of standard values ​​for governance data can incorporate regression prediction models, predicting current standard values ​​based on historical trends. Grid governance deviation tolerances can be set differently based on grid type, with wider tolerances for production grids and stricter tolerances for R&D grids. The marking of abnormal governance grids can be accompanied by confidence scores to reflect the reliability of the judgment results. The statistics of overlapping grids can be extended to multi-dimensional overlap, considering the combined overlap of governance anomalies with environmental and social anomalies. The calculation of the spatial dispersion of governance data can be combined with spatial autocorrelation indices to evaluate the spatial clustering patterns of abnormal grids.

[0046] In practical implementation, the spatial dispersion ratio of governance data is calculated using a geometric area ratio rather than a simple quantity ratio, which more accurately reflects the impact of spatial distribution. The spatial dispersion value of governance data is used in subsequent dynamic calibration; when the spatial dispersion is high, the adjustment range of the exponential correction factor increases. Mapping the spatial grid requires establishing a grid coding system, with each grid having a unique identifier for easy data tracking and management. Data collection at governance data sampling points uses standardized protocols to ensure comparability between data from different grids. Comparison of real-time governance data values ​​with standard values ​​considers data timeliness; expired data is automatically marked and excluded. The tolerance for grid governance deviations is set with reference to industry benchmarks, adjusted based on the company's own historical performance. Overlap analysis of abnormal governance grids with abnormal environmental and social governance index grids uses spatial database queries to optimize computational performance.

[0047] It is understandable that when mapping corporate governance structures to a spatial grid, the grid cell size needs to balance accuracy and computational cost. A grid that is too small leads to sparse data, while a grid that is too large masks local differences. The acquisition of measured governance data values ​​may be affected by network latency, necessitating the implementation of timeout mechanisms and failover strategies. The historical data window length of governance data standard values ​​affects the stability of these standard values; a window that is too long results in a sluggish response, while a window that is too short causes significant fluctuations. Dynamic adjustments to the grid governance deviation tolerance must be made cautiously to avoid frequent changes that could cause system oscillations. The labeling results of abnormal governance grids should record timestamps to support trend backtracking analysis. Overlapping grid statistics need to address the non-uniform shape of the grids, using area proportions rather than simple counting. The calculation results of the spatial dispersion ratio of governance data need to be supplemented with uncertainty estimates to reflect sampling errors.

[0048] In some embodiments, the calculation of spatial dispersion of governance data can be integrated with a real-time stream processing framework, automatically triggering dispersion recalculation when grid data is updated. The mapping of spatial grids can be combined with an enterprise organizational chart, mapping logical departments to virtual grid cells. The collection of measured governance data values ​​can adopt an edge computing model, performing preliminary processing at the data source before uploading. The determination of standard values ​​for governance data can employ robust statistics, such as using the median instead of the mean, to reduce the impact of outliers. Grid governance deviation tolerance can be set as a dynamic threshold, fluctuating with the overall governance level. The labeling of abnormal governance grids can incorporate a machine learning classifier to automatically learn abnormal patterns. The identification of overlapping grids can employ image processing techniques to identify the grids. Figure 2 After valueization, the overlapping region is calculated. The final value of the data spatial dispersion can be smoothed by using a moving average to eliminate random fluctuations.

[0049] It is understandable that the spatial dispersion of governance data, as a component of governance data volatility, directly impacts the calibration effectiveness of environmental and social governance indices. The spatial grid division scheme needs to be documented and version-managed to ensure reproducible results. Data quality monitoring of governance data sampling points needs to be routine, with failed sampling points promptly identified and excluded. The update cycle of governance data standard values ​​should be aligned with the business reporting cycle to avoid data inconsistencies. The absolute value difference function in the grid governance deviation calculation ensures that the deviation is always non-negative, conforming to the definition of dispersion. The statistical analysis of overlap between abnormal governance grids and abnormal grids in the environmental and social governance indices needs to consider temporal synchronization, with only abnormal grids from the same period participating in the calculation. The proportional calculation of the spatial dispersion of governance data uses double-precision floating-point arithmetic to prevent the accumulation of rounding errors.

[0050] Example 4: In specific implementation, calculating the correlation ratio between the rate of change of governance data and the rate of change of the benchmark values ​​of environmental, social, and governance indices during the synchronous fluctuation period needs to be performed within a key period. The key period is the time interval during which the governance data sequence and the benchmark value sequence of environmental, social, and governance indices are highly synchronized. Numerical pairs of the rate of change of governance data and the rate of change of the benchmark values ​​of environmental, social, and governance indices are extracted. These pairs are strictly aligned according to timestamps to ensure that each time point has a corresponding rate of change of governance data and the rate of change of the benchmark values ​​of environmental, social, and governance indices. For each numerical pair, the ratio of the rate of change of governance data to the rate of change of the benchmark values ​​of environmental and social governance indices is calculated, using the rate of change of governance data as the numerator and the rate of change of the benchmark values ​​of environmental, social, and governance indices as the denominator. The arithmetic mean of all ratios is then calculated by summing all ratios and dividing by the total number of numerical pairs to obtain the average correlation ratio. This average correlation ratio is the final correlation ratio output, reflecting the average magnitude of the change in governance data relative to the change in the benchmark values ​​of environmental and social governance indices.

[0051] Screening for high-synchronization periods based on correlation ratios requires calculating the deviation of the correlation ratio for each key period from the ideal correlation ratio. The ideal correlation ratio is the theoretical ratio under a preset ideal synchronization state. The standard deviation of the deviations for all key periods is calculated using the sample standard deviation formula to determine the dispersion of the deviations, yielding a synchronization stability index. The selection threshold for key periods is adjusted based on this synchronization stability index. A lower index value indicates high stability for key periods, allowing for a wider correlation ratio threshold to include more periods. Conversely, a higher index value indicates poor stability for key periods, requiring an increase in the correlation ratio threshold to screen for more reliable periods.

[0052] In practical implementation, when extracting numerical pairs of governance data change rates and environmental, social, and governance index benchmark change rates, the extraction of numerical pairs needs to cover all valid sampling points within the key time period. Invalid sampling points, such as those with missing data or outliers, need to be removed beforehand. The numerical pairs of governance data change rates and environmental, social, and governance index benchmark change rates are stored in a two-dimensional array. The first column represents the governance data change rate, and the second column represents the environmental and social governance index benchmark change rate. When calculating the ratio for each numerical pair, it is necessary to check whether the environmental and social governance index benchmark change rate is zero. If it is zero, a minimum value is used to avoid division by zero errors. The arithmetic mean is calculated using a simple average method, but a truncated average can be introduced to handle extreme ratios. The ideal correlation ratio is usually set to 1, indicating that the change in governance data perfectly matches the change in the environmental and social governance index benchmark values. The deviation is calculated using an absolute difference or relative difference formula to measure the degree of difference between the actual correlation ratio and the ideal correlation ratio. The standard deviation is calculated using an unbiased estimation formula to accurately reflect the fluctuation of the deviation. The adjustment of the threshold during critical periods uses a linear or nonlinear mapping function to convert the synchronization stability index into the threshold adjustment amount.

[0053] As can be understood (referring to Table 1), the alignment accuracy of numerical pairs directly affects the accuracy of correlation ratio calculation. Timestamps need to be standardized to the same granularity, such as millisecond or second-level timestamps. Ratio calculation must consider sign control; a negative rate of change indicates a negative sign, reflecting an inverse relationship. Arithmetic averaging assumes equal weight for all numerical pairs, but can be weighted according to time density. The ideal correlation ratio can be set based on historical data optimization; different industries can set different ideal values. The deviation calculation method affects sensitivity to differences; absolute deviation focuses on absolute differences, while relative deviation focuses on proportional differences. Standard deviation calculation requires a sufficient sample size; when the critical period is too short, other discrete metrics should be used. Threshold adjustments should be smooth and gradual to avoid drastic fluctuations in the number of critical periods caused by threshold jumps.

[0054] Table 1: Parameters for Correlation Ratio Calculation

[0055] In some embodiments, the correlation ratio can be calculated using the geometric mean instead of the arithmetic mean to reduce the impact of extreme values ​​on the average correlation ratio. Interpolation methods can be introduced to extract numerical pairs and resample non-uniformly sampled rate-of-change data to ensure time point alignment. Logarithmic transformation can be added to the ratio calculation to make the ratio distribution more symmetrical and facilitate statistical analysis. The ideal correlation ratio can be dynamically adjusted, automatically updated based on the median of the correlation ratio in recent key periods. Deviation can be calculated using the squared difference form, increasing the weight of large deviations. Standard deviation calculation can be combined with moving window calculation to update the synchronization stability index in real time. The adjustment of the threshold for selecting key periods can employ fuzzy control rules, intelligently adjusting according to the magnitude and trend of the synchronization stability index.

[0056] Optionally, the numerical pairs of governance data change rates and environmental and social governance index benchmark change rates can be stored in a circular buffer, supporting real-time streaming processing. Ratio calculations can incorporate range restrictions; ratios exceeding a reasonable range are considered invalid and discarded. Arithmetic mean calculations can use weighted arithmetic mean, with higher weights for recent values. Ideal correlation ratios can be set for different time periods, using different benchmarks for daytime and nighttime. Deviation calculations can incorporate normalization to make deviations comparable across different key time periods. Standard deviation calculations can use robust standard deviation estimation, replacing standard deviation with median absolute deviation. Threshold adjustments can set upper and lower limits to prevent thresholds from being too high, resulting in the absence of key time periods, or too low, leading to the inclusion of low-quality time periods.

[0057] In practical implementation, the correlation ratio calculation module can be deployed as an independent service, receiving data streams for key time periods and outputting correlation ratio sequences in real time. Numerical pairs are managed using a time-series database, supporting efficient querying and aggregation calculations. Detailed logs are recorded during the ratio calculation process, facilitating anomaly investigation and algorithm optimization. The average correlation ratio result is cached for a certain period to reduce redundant calculation overhead. Deviation calculation considers the uncertainty of the correlation ratio, estimating the deviation error through error propagation theory. Standard deviation calculation employs an incremental algorithm, adaptable to streaming data scenarios. The threshold adjustment mechanism provides a manual overriding interface, allowing expert experience to intervene in adjustments.

[0058] It is understandable that the length of the critical period affects the numerical value relative to the quantity; an excessively short critical period may lead to unreliable correlation ratio statistics. The calculation methods for the rate of change in governance data and the benchmark rate of change in the environmental and social governance indices need to be consistent, both using instantaneous rates of change or interval average rates of change. When the denominator in the ratio calculation is close to zero, even using substitute values ​​may result in unstable results, requiring labeling as low-reliability. Arithmetic means are sensitive to outliers and require data cleaning procedures. Setting the ideal correlation ratio requires industry knowledge; unreasonable ideal values ​​can lead to distorted deviations. The standard deviation of the deviation is affected by the number of critical periods; too few periods result in a larger standard deviation estimation error. The threshold adjustment function needs thorough testing to avoid oscillations or divergent behavior.

[0059] Optionally, quality indicators can be added to the extraction of numerical pairs, excluding low-quality data with low rates of change from the correlation ratio calculation. Ratio calculation can employ robust estimation methods, such as using the Huber function to reduce the impact of outliers. Arithmetic mean calculation can be performed by grouping and merging data, first calculating the sub-interval averages and then the overall average. The ideal correlation ratio can be set in interval form, with the correlation ratio falling within the interval considered as zero deviation. Deviation calculation can incorporate confidence intervals, considering the impact of sampling errors. Standard deviation calculation can use stratified calculation, first calculating the within-group standard deviations and then merging them. Threshold adjustment can incorporate adaptive control algorithms, such as PID controllers, to achieve smooth regulation. The correlation ratio calculation results can be validated using cross-validation, backtesting historical data to test the algorithm's effectiveness.

[0060] In practical implementation, the correlation ratio calculation process needs to be calibrated regularly to ensure that the calculation parameters match the actual data characteristics. Numerical pairs are stored in a columnar storage format to improve batch calculation efficiency. The anomaly handling mechanism for ratio calculation needs to be improved, with clear strategies for handling exceptions such as division by zero and overflow. The output of the average correlation ratio includes a quality score, reflecting the reliability of the calculation process. Deviation calculation supports multiple measurement methods, allowing for flexible selection based on application scenarios. The results of standard deviation calculation are smoothed and filtered to eliminate random fluctuations. The threshold adjustment logic requires version management, recording the parameters and effects of each adjustment. The entire correlation ratio calculation and key period screening system needs to integrate monitoring and alarm functions to detect and handle anomalies in real time.

[0061] See Figure 4 This chart is a core visualization result of the correlation ratio and synchronization stability index analysis. The horizontal axis, correlation ratio, quantifies the average magnitude correlation between the rate of change of governance data and the rate of change of the ESG index benchmark value, while the vertical axis, synchronization stability index, measures the stability of the synchronization relationship between the governance data sequence and the ESG index benchmark value sequence during key periods. The colored scatter points in the chart correspond to multiple sets of measured data pairs, and the red trend line visually presents the positive correlation between the two: the higher the correlation ratio, the stronger the synchronization stability index is usually. This perfectly aligns with the technical logic of using the correlation ratio to screen for key periods of high synchronization, and the synchronization stability index reflecting the strength of synchronization during periods. This chart provides crucial support for calculating the time variability of governance data: by analyzing the linkage pattern between the two, key periods can be accurately identified, thereby quantifying the time variability of governance data, ultimately providing a core basis for deriving the index correction factor in the dynamic calibration of the ESG index.

[0062] Example 5: In practical implementation, the dynamic calibration of the environmental, social, and governance (ESG) index benchmark values ​​is a systematic process. Its core is deriving the index correction factor from the volatility of governance data, which is a comprehensive quantitative indicator of the degree of variation in governance data. The enterprise's ESG benchmark value is multiplied by the index correction factor; this multiplication mathematically achieves scalar scaling, yielding the enterprise's ESG output value. This output value is the final evaluation result after adjustment for governance data volatility. The calculation of the index correction factor includes extracting synchronization stability indicators for all key time periods. Synchronization stability indicators are parameters that measure the stability of the synchronization relationship between the governance data sequence and the ESG benchmark value sequence within key time periods. The harmonic mean of the synchronization stability indicators is calculated. This calculation requires taking the arithmetic mean of the reciprocals of all synchronization stability indicators and then taking the reciprocal. The resulting harmonic mean is used as the index correction factor. The characteristic of the harmonic mean is that it tends to have smaller values, thus avoiding excessive influence from a few abnormally high synchronization stability indicators on the index correction factor.

[0063] In practice, extracting synchronization stability indices for all key time periods requires first identifying all time intervals marked as key time periods within the evaluation time window. Each key time period corresponds to a pre-calculated synchronization stability index value. The synchronization stability index extraction operation retrieves metadata for key time periods from the time series database by timestamp and reads the synchronization stability index calculation results stored in the metadata. When calculating the harmonic mean of the synchronization stability indices, it is necessary to check whether the set of synchronization stability indices contains zero or negative values, because the harmonic mean requires all values ​​to be positive. The calculation process of the harmonic mean follows the standard mathematical definition, that is, the harmonic mean of the synchronization stability indices is equal to the reciprocal of the total number of synchronization stability indices multiplied by the reciprocal of the sum of the reciprocals of each synchronization stability index value. The assignment operation of the exponential correction factor directly uses the calculated harmonic mean of the synchronization stability indices as the factor value. The value range of the exponential correction factor is usually between zero and two, ideally close to one, indicating that no significant adjustment is needed. The multiplication of the benchmark value of the corporate environmental, social and governance index with the index correction factor is performed in real time in the calculation engine. The multiplication result is rounded to a specified number of decimal places to generate the corporate environmental, social and governance index output value.

[0064] It is understandable that governance data volatility serves as the triggering condition and input basis for the dynamic calibration process, and the accuracy of its calculation directly affects the rationality of the index correction factor. As a proportionality coefficient, a value greater than the index correction factor will raise the benchmark values ​​of the environmental and social governance indices, while a value less than it will lower them. The use of the harmonic mean, a synchronous stability indicator, makes the index correction factor insensitive to outliers. When there are individual instances of extremely low synchronous stability indicators, the harmonic mean will significantly reduce the index correction factor, reflecting a robust calibration approach. The output values ​​of the enterprise's environmental and social governance indices need to be compared again with the dynamic threshold range of the environmental and social governance indices to ensure that the calibrated values ​​are within a reasonable range. The entire dynamic calibration process can be iterative; when new governance data volatility occurs, a new round of calibration can be initiated to generate updated output values ​​for the enterprise's environmental and social governance indices.

[0065] In some embodiments, deriving the index correction factor from governance data volatility can employ a nonlinear mapping function, rather than solely relying on the harmonic mean of the synchronization stability index. The relationship between governance data volatility and the index correction factor can be established as a lookup table, directly mapping a predefined index correction factor based on the interval to which the governance data volatility belongs. The calculation of the harmonic mean of the synchronization stability index can introduce weighting coefficients, assigning higher weights to synchronization stability indicators in recent key periods, making the index correction factor more reflective of the current synchronization status. The multiplication operation between the benchmark value of the enterprise environmental and social governance index and the index correction factor can be extended to matrix multiplication, achieving dimension-by-dimensional calibration when the benchmark value of the environmental and social governance index is a multi-dimensional vector. The generation of the output values ​​of the enterprise environmental and social governance index can incorporate boundary checking functionality, automatically pruning to boundary values ​​when output values ​​exceed a reasonable range.

[0066] In practical implementation, a specific example of the data volatility derivation index correction factor can be illustrated as follows: Suppose a company identified three key periods within its most recent assessment period, each corresponding to a different synchronization stability index value. The synchronization stability index for the first key period is 0.85, for the second key period it is 0.92, and for the third key period it is 0.78. The calculation process for the harmonic mean of the synchronization stability index is as follows: First, calculate the reciprocal of each synchronization stability index, i.e. Then calculate the arithmetic mean of these reciprocals. Finally, take the reciprocal of this arithmetic mean. Therefore, the index adjustment factor is approximately 0.846. If the benchmark value for the company's environmental and social governance index is 75.6, then the output value of the company's environmental and social governance index will be calculated as follows: This example demonstrates how the harmonic mean reduces the exponential correction factor, as the relatively low synchronization stability index of 0.78 in the third key period has a significant impact on the harmonic mean.

[0067] Optionally, the calculation of the index correction factor can incorporate smoothing by weighting the currently calculated index correction factor with historical index correction factors to reduce factor jumps. The calculation of the harmonic mean of the synchronous stability index can exclude extreme values; when a synchronous stability index deviates significantly from the median, it is considered an outlier and excluded from the harmonic mean calculation. The multiplication operation between the benchmark value and the index correction factor of the corporate environmental and social governance index can use high-precision decimal data type to avoid precision loss caused by floating-point operations. Uncertainty estimates can be appended to the output values ​​of the corporate environmental and social governance indexes, and the impact of the uncertainty of the index correction factor on the output values ​​of the environmental and social governance indexes can be calculated using the propagation law.

[0068] In practical implementation, the triggering conditions for the dynamic calibration process can be diversified. Besides addressing data volatility exceeding thresholds, it can also be triggered periodically based on time rules. The calculation of the index correction factor can cache intermediate results; when only some key periods of the synchronization stability index are updated, only the changed portion needs to be recalculated instead of the entire index. The calculation of the harmonic mean of the synchronization stability index can be optimized into an incremental algorithm, supporting real-time updates in streaming data scenarios. The multiplication operation between the benchmark value and the index correction factor of the enterprise environment and social governance index can be encapsulated as a microservice, receiving input parameters and returning calibration results via an API interface. The publication of the output values ​​of the enterprise environment and social governance index can be integrated into the enterprise reporting system, automatically generating calibration documentation.

[0069] It is understandable that the mapping relationship between governance data volatility and the derived index correction factor needs to be optimized according to industry characteristics, as the impact of governance data volatility on the environmental and social governance indices may differ across industries. The calculation of the harmonic mean of the synchronization stability index requires that the synchronization stability index be positive for all key periods; a negative synchronization stability index is unreasonable from a business perspective and requires data verification before calculation. The value of the index correction factor needs to be limited to a reasonable range to avoid distortion of the environmental and social governance index output values ​​due to calculation anomalies causing the factor to be too large or too small. The enterprise's environmental and social governance index output values ​​should retain calibration records, documenting the baseline values ​​of the environmental and social governance indices and the index correction factors used before each calibration, to meet audit requirements.

[0070] In some embodiments, deriving the index correction factor from governance data volatility can employ a machine learning model, using governance data volatility and related characteristics as input to train a regression model to predict the optimal index correction factor. The calculation of the harmonic mean of the synchronous stability index can be combined with the geometric mean to form a mixed average index, balancing the impact of magnitude. The multiplication operation between the benchmark value of the corporate environmental and social governance index and the index correction factor can be extended to a multivariate function, introducing other moderating variables such as industry prosperity indices and macroeconomic indicators. The generation of the output value of the corporate environmental and social governance index can incorporate confidence interval estimation, providing both point and interval estimates as output formats.

[0071] Optionally, the calculation of the index correction factor can incorporate seasonal adjustments, setting different benchmark index correction factors for different seasons, and then fine-tuning based on the harmonic mean of the synchronous stability index. The harmonic mean of the synchronous stability index can be calculated using a truncated average method, removing the highest and lowest synchronous stability indices before calculating the harmonic mean, further improving robustness. The multiplication operation between the benchmark value and the index correction factor of the corporate environmental and social governance index can consider a non-linear transformation, first performing a logarithmic transformation on the benchmark values ​​of the environmental and social governance indexes, multiplying by the index correction factor, and then performing an exponential transformation. The release of the corporate environmental and social governance index output values ​​can be differential, only releasing calibration results whose changes relative to the previous period exceed a certain threshold, reducing unnecessary frequent updates.

[0072] In practical implementation, performance optimization of the dynamic calibration system can consider parallel computing techniques. When there are numerous critical periods, the harmonic mean of synchronization stability indices can be calculated in parallel groups before merging. An early termination mechanism can be introduced for the calculation of the exponential correction factor; if the exponential correction factor is found to be significantly outside the reasonable range during calculation, the calculation is terminated and an exception handling process is triggered. Numerical stability processing can be added to the calculation of the harmonic mean of synchronization stability indices, with special handling for synchronization stability indices approaching zero to avoid floating-point underflow. The multiplication of the benchmark value and the exponential correction factor of the enterprise environmental and social governance index can use fixed-point arithmetic to improve computational efficiency while ensuring accuracy. The system for generating the output values ​​of the enterprise environmental and social governance index needs to be highly available, deploying redundant nodes and failover mechanisms to ensure continuous availability of calibration services.

[0073] Optionally, the process of deriving the index correction factor based on governance data volatility can include a manual review step. When the index correction factor deviates from the historical average by more than a certain threshold, it should be submitted to management for confirmation before application. The calculation of the harmonic mean of the synchronization stability index can dynamically adjust the weights of key periods, allocating different weights based on the length of the key period and data quality. The multiplication operation between the benchmark value of the enterprise environment and social governance index and the index correction factor can be performed within a transaction to ensure data consistency. The generation of the output values ​​of the enterprise environment and social governance index can support multiple output formats to meet the integration needs of different downstream systems. The entire dynamic calibration process requires the establishment of a complete monitoring system to track the changing trend of the index correction factor and business indicators of the calibration effect.

[0074] See Figure 5 This chart is the core visualization result of the dynamic calibration process for the ESG index. The horizontal axis represents key time periods, indicating the periods of highly synchronous governance data selected within the assessment time window; the vertical axis represents numerical values, showing the quantitative results of the synchronization stability index and the index correction factor. From a technical perspective, the chart intuitively presents the correlation between poorer synchronization stability and smaller correction factors, reflecting the dynamic calibration idea that weaker governance data synchronization leads to a larger calibration magnitude for the ESG benchmark value. This step is a crucial visualization support for deriving the index correction factor from the volatility of governance data, ultimately generating accurate ESG index output values. It ensures that the index calibration reflects both the characteristics of governance data and possesses statistical robustness, providing a quantitative basis for the dynamic assessment of corporate ESG performance.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating enterprise ESG indices based on dynamic integration of multi-source data, characterized in that, The method includes the following steps: Real-time capture of environmental, social, and governance data streams from within and outside the enterprise; noise filtering and format alignment of the environmental, social, and governance data streams to form a standardized ESG data sequence; The standardized ESG data sequences are scaled to eliminate unit inconsistencies and generate dimensionless ESG data matrices; the dimensionless ESG data matrices are then synthesized into a unified ESG feature tensor. Based on the ESG feature tensor, the benchmark value of the enterprise's ESG index is calculated through dynamic weighted aggregation operation; the benchmark value of the enterprise's ESG index is matched and analyzed with the pre-stored dynamic threshold range of the ESG index; when the benchmark value of the enterprise's ESG index exceeds the dynamic threshold range, an ESG deviation alarm is generated. In response to ESG deviation alerts, the system collects the latest governance data sequence of enterprises and calculates the volatility of governance data. Based on the volatility of governance data, the system dynamically calibrates the benchmark value of the enterprise's ESG index and generates the enterprise's ESG index output value.

2. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 1, characterized in that, The calculation of the enterprise ESG index benchmark value includes: decomposing the enterprise's operating structure into multiple functional modules, collecting environmental data streams, social data streams, and governance data streams for each functional module; assigning dynamic weight factors to the environmental data streams, social data streams, and governance data streams for each functional module, calculating the ESG local evaluation value for each functional module; and weighting and fusing the ESG local evaluation values ​​of all functional modules to obtain the enterprise ESG index benchmark value.

3. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 1, characterized in that, The calculation of governance data volatility includes: obtaining the time variability and spatial dispersion of governance data, and linearly superimposing the time variability and spatial dispersion of governance data to obtain the governance data volatility.

4. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 3, characterized in that, The acquisition of governance data time variability includes: identifying synchronous fluctuation periods between the governance data sequence and the ESG index benchmark value sequence within the assessment time window; calculating the correlation ratio between the rate of change of governance data and the rate of change of ESG index benchmark value within the synchronous fluctuation period; screening highly synchronous periods based on the correlation ratio and recording them as key periods; measuring the proportion of the cumulative duration of key periods to the total duration of the assessment time window, and using this proportion as the governance data time variability.

5. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 4, characterized in that, The process of identifying the synchronous fluctuation periods of the governance data sequence and the ESG index benchmark value sequence includes: within the evaluation time window, constructing a time axis with equally spaced sampling points, and plotting curves showing the changes in governance data values ​​and ESG index benchmark values ​​over time; extracting the peaks and troughs of the governance data curves, calculating the slope between adjacent peaks and troughs to obtain the governance data change rate; extracting the peaks and troughs of the ESG index benchmark value curves, calculating the slope between adjacent peaks and troughs to obtain the ESG index benchmark value change rate; and comparing the direction and magnitude of the governance data change rate and the ESG index benchmark value change rate to determine the synchronous fluctuation period.

6. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 3, characterized in that, Obtaining the spatial dispersion of governance data includes: mapping the enterprise governance structure to a spatial grid, with each grid cell corresponding to a governance data sampling point; calculating the absolute difference between the measured and expected values ​​of the governance data for each grid cell, denoted as the grid governance deviation; comparing the grid governance deviation with the grid governance deviation tolerance: if the grid governance deviation is greater than or equal to the grid governance deviation tolerance, the grid is marked as a governance anomalous grid; simultaneously, calculating the absolute difference between the measured and expected values ​​of the ESG index benchmark for each grid cell, denoted as the grid ESG deviation; comparing the grid ESG deviation with the grid ESG deviation tolerance: if the grid ESG deviation is greater than or equal to the grid ESG deviation tolerance, the grid is marked as an ESG anomalous grid; counting the number of overlapping grids between governance anomalous grids and ESG anomalous grids; calculating the proportion of the total area of ​​overlapping grids to the total area of ​​the entire spatial grid, and using this proportion as the spatial dispersion of the governance data.

7. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 6, characterized in that, Calculating grid governance deviation includes: obtaining real-time governance data values ​​for each grid cell through sensor networks or database interfaces; extracting standard governance data values ​​for each grid cell from historical data; and calculating the absolute value of the difference between the real-time governance data values ​​and the standard values ​​to obtain the grid governance deviation. The calculation of grid ESG bias includes: obtaining the real-time benchmark value of the ESG index for each grid cell through the monitoring system; obtaining the standard value of the ESG index for each grid cell from the benchmark model; and calculating the absolute value of the difference between the real-time benchmark value and the standard value of the ESG index to obtain the grid ESG bias.

8. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 4, characterized in that, The calculation of the correlation ratio between the rate of change of governance data and the rate of change of the ESG index benchmark value during the synchronous fluctuation period includes: extracting numerical pairs of the rate of change of governance data and the rate of change of the ESG index benchmark value during the key period; calculating the ratio of the rate of change of governance data to the rate of change of the ESG index benchmark value for each numerical pair; and taking the arithmetic mean of the ratios of all numerical pairs to obtain the average correlation ratio, which is used as the correlation ratio. The method of selecting high synchronization periods based on correlation ratio includes: calculating the deviation of the correlation ratio of each key period from the ideal correlation ratio; calculating the standard deviation of the deviation of all key periods to obtain the synchronization stability index; and adjusting the selection threshold of key periods based on the synchronization stability index.

9. The method for generating enterprise ESG indices based on dynamic integration of multi-source data as described in claim 1, characterized in that, The dynamic calibration of the corporate ESG index benchmark value includes: deriving an index correction factor from the volatility of governance data; multiplying the corporate ESG index benchmark value by the index correction factor to obtain the corporate ESG index output value; wherein the calculation of the index correction factor includes: extracting the synchronous stability index for all key periods; calculating the harmonic mean of the synchronous stability index as the index correction factor.

10. A system for determining an enterprise ESG index based on data fusion, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the enterprise ESG index determination method based on data fusion as described in any one of claims 1 to 9.

Images