Monitoring method and system for data governance process

Abstract

The invention relates to the technical field of data processing and governance, in particular to a monitoring method and system for a data governance process, and the method comprises the steps: constructing a quantum distribution model through extracting multi-modal features, so as to represent a complex relation between feature variables; analyzing the abnormal propagation path based on a causal chain model and a recursive reasoning algorithm, and generating basic abnormal feature data; performing parameter optimization on the governance rule by using a memetic evolutionary algorithm to generate optimized rule data; model parameters of the monitoring system are updated through multi-round iteration, and dynamic optimization of the system state is achieved. According to the method, the depth and precision of anomaly detection are remarkably improved, meanwhile, efficient iteration of a monitoring system is achieved, the treatment requirement under the complex multi-modal data environment is met, and the key problems that in the prior art, the analysis depth is insufficient, the real-time performance is poor, and the traceability is limited are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and governance, and in particular to a monitoring method and system for a data governance process. Background Art

[0002] Monitoring of the data governance process is a key link to ensure the success of the data governance system, and it runs through the entire life cycle of data collection, storage, processing, analysis and use. Through effective monitoring methods, we can not only ensure the quality, integrity and security of data, but also timely discover and handle abnormal data, reduce decision-making errors caused by data quality problems, and improve the efficiency of business processes.

[0003] The existing technology (Chinese invention patent, publication number: CN118394970B, name: Data supervision method for data governance process management) adopts a method based on vector discreteness analysis to identify malicious data by mapping data into vector space and detecting its discreteness. However, this method has the following defects: Relying on discreteness measurement, this method has weak analysis of deep causal relationships in data and has difficulty discovering non-intuitive abnormal propagation patterns; by gradually combining subsets and detecting discreteness, it has high computational complexity and low efficiency when processing large amounts of data; it is unable to accurately identify the source of abnormal data and its propagation path, limiting the optimization of subsequent governance rules; it mainly focuses on the numerical characteristics of data and lacks support for unstructured data (such as text, images, etc.) and multimodal features. Summary of the invention

[0004] In view of the many problems existing in the above-mentioned prior art, the present invention provides a monitoring method and system for the data governance process. The present invention constructs a causal chain model through adaptive multimodal feature extraction and quantum distribution modeling, combines a recursive reasoning algorithm to analyze the abnormal propagation path, and optimizes the governance rules through meme evolution. Finally, the status of the monitoring system is updated in real time, effectively improving the ability to identify, analyze and process abnormal data. The present invention has the characteristics of high efficiency, accuracy and scalability.

[0005] A method for monitoring a data governance process comprises the following steps: Collecting multimodal data from multiple data sources and preprocessing the collected multimodal data to generate adaptive multimodal feature data; A causal chain model is constructed based on adaptive multimodal feature data to generate causal chain graph data; the causal chain graph data is dynamically adjusted to reflect the real-time causal relationship between multimodal feature data, and the abnormal propagation path is analyzed through recursive causal reasoning to generate abnormal propagation path data; Based on the abnormal propagation path data, the quantum distribution model data is constructed using the distribution modeling method; based on the quantum distribution model data, the abnormal features outside the distribution are identified, and the fundamental abnormal feature data is generated through recursive abnormal location analysis; Generate preliminary governance rule data based on fundamental abnormal feature data, optimize the preliminary governance rule data through rule optimization method, and generate optimized governance rule data; use the optimized governance rule data to update the monitoring system status, and generate optimized monitoring system status data.

[0006] Preferably, the generation of the adaptive multimodal feature data includes: Structured features, unstructured features and time series features are extracted from multimodal preprocessed data through a feature extraction method; the structured features include statistical features of fields in a data table and the correlation between fields, the unstructured features include semantic vectors of text and pixel distribution features of images, and the time series features include trend values ​​and periodic features of time series; the structured features, unstructured features and time series features are fused and optimized using a dynamic semantic fusion mechanism to generate adaptive multimodal feature data.

[0007] Preferably, the construction of the causal chain model includes: Based on adaptive multimodal feature data, a causal reasoning algorithm is used to generate causal chain graph data; the nodes of the causal chain graph data represent the characteristic variables of the adaptive multimodal feature data, and the edges of the causal chain graph data represent the causal relationship between the characteristic variables in the adaptive multimodal feature data; the edge weights of the causal chain graph data are calculated by a neural network, and the calculation of the edge weights includes correlation analysis of the multimodal feature variables and causal determination based on conditional independence; the dynamic adjustment of the causal chain graph data includes updating the edge weights in combination with real-time data.

[0008] Preferably, the recursive causal reasoning includes: The abnormal propagation path is derived layer by layer from the causal chain graph data through a recursive causal reasoning algorithm, and the initial conditions of the recursive causal reasoning are the starting node of the abnormal propagation path and its initial abnormal value; the abnormal propagation probability of each node is calculated during the recursive causal reasoning process, and the abnormal propagation probability calculation includes a signal transmission formula based on the edge weights in the causal chain; the reasoning results of each abnormal propagation path record the impact range of the propagation and possible high-risk nodes to form abnormal propagation path data.

[0009] Preferably, the construction of the quantum distribution model data includes: Based on the abnormal propagation path data, a quantum state vector is generated through quantum state encoding; the quantum state vector is represented as the amplitude and phase distribution of a multimodal characteristic variable; a global quantum distribution model is generated through the quantum superposition effect, and the quantum superposition effect includes linear superposition of different characteristic quantum states to simulate the global distribution; the nonlinear correlation between the multimodal characteristic variables is simulated through the quantum entanglement effect to form quantum distribution model data.

[0010] Preferably, the identification of abnormal features outside the distribution includes: By performing feature measurement calculation on the quantum distribution model data, the feature measurement includes inner product calculation of quantum state vector to determine the distribution probability of each feature variable; the feature variable with distribution probability lower than a set threshold is marked as an out-of-distribution abnormal feature.

[0011] Preferably, the recursive anomaly location analysis includes: Based on the data of the out-of-distribution abnormal features, cluster analysis is performed on the out-of-distribution abnormal features through a recursive clustering algorithm; the cluster analysis includes a hierarchical clustering method based on feature distance; a local feature abnormal area is generated for each clustering result, and a reverse reasoning algorithm is used to trace the abnormal cause from the local feature abnormal area; in the tracing process, the temporal dependency of the feature variables and the abnormal propagation path of the causal chain diagram data are combined to form fundamental abnormal feature data.

[0012] Preferably, the generation of the optimized governance rule data includes: Based on the fundamental abnormal feature data, the preliminary governance rule data is optimized using a memetic evolution algorithm; the optimization process of the memetic evolution algorithm includes feature crossover and parameter variation of governance rule condition parameters, the feature crossover includes combining condition parameters of multiple governance rules, and the parameter variation includes randomly adjusting the upper and lower limits of the rule threshold; the governance rule with the highest fitness is selected through fitness evaluation to generate optimized governance rule data.

[0013] Preferably, the generation of the optimized monitoring system state includes: Based on the optimized governance rule data, the model parameters of the feature extraction, causal chain modeling and quantum distribution modeling of the monitoring system are updated respectively; the updated model parameters include the weight allocation rule for feature extraction, the causal relationship edge weight update rule and the phase adjustment rule for the quantum state vector; and the optimized monitoring system state is generated through multiple rounds of parameter iterative evaluation.

[0014] A system for implementing the monitoring method of the data governance process, the system comprising: A multimodal data acquisition module, used to collect multimodal data from multiple data sources and pre-process the multimodal data to generate adaptive multimodal feature data; A causal chain modeling module is used to construct a causal chain model based on the adaptive multimodal feature data and generate causal chain graph data; the causal chain modeling module is also used to dynamically adjust the causal chain graph data to reflect the real-time causal relationship between the multimodal feature data, and generate abnormal propagation path data through recursive causal reasoning analysis; A distribution modeling module, used to construct quantum distribution model data according to the abnormal propagation path data; the distribution modeling module is also used to identify abnormal features outside the distribution based on the quantum distribution model data, and generate fundamental abnormal feature data through recursive abnormal location analysis; A governance rule generation module, used to generate preliminary governance rule data based on the fundamental abnormal feature data, and optimize the preliminary governance rule data through a rule optimization method to generate optimized governance rule data; The monitoring system update module is used to update the monitoring system status based on the optimized governance rule data and generate optimized monitoring system status data.

[0015] Compared with the prior art, the advantages and beneficial effects of the present invention are: The present invention uses recursive causal reasoning technology to achieve accurate identification of abnormal propagation paths and real-time update of dynamic causal relationships, greatly improving the depth and accuracy of abnormal analysis; The present invention realizes unified modeling of structured and unstructured data through multi-modal characteristic quantum modeling technology, and enhances the adaptability of the system to complex data; The present invention optimizes governance rules through memetic evolution algorithm, realizes rapid iterative optimization of dynamic monitoring system, and significantly improves the real-time response capability of the system; The present invention realizes the accurate positioning of the root anomaly through recursive anomaly positioning and tracing technical means, and improves the efficiency and comprehensiveness of abnormal data management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 A schematic diagram of the relationship between causal reasoning and anomaly detection in the present invention; Figure 3 A schematic diagram of the relationship between rule optimization and system feedback in the present invention; Figure 4 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0017] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0018] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0019] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0020] like Figure 1 As shown, a method for monitoring a data governance process includes the following steps: Collecting multimodal data from multiple data sources and preprocessing the collected multimodal data to generate adaptive multimodal feature data; Preferably, the generation of the adaptive multimodal feature data includes: Structured features, unstructured features and time series features are extracted from multimodal preprocessed data through a feature extraction method; the structured features include statistical features of fields in a data table and the correlation between fields, the unstructured features include semantic vectors of text and pixel distribution features of images, and the time series features include trend values ​​and periodic features of time series; the structured features, unstructured features and time series features are fused and optimized using a dynamic semantic fusion mechanism to generate adaptive multimodal feature data.

[0021] Multimodal data refers to a collection of structured data, unstructured data, and time series data. These data have the characteristics of diverse sources and different characteristics. In order to achieve unified processing of these data, the present invention performs specific preprocessing operations on them to eliminate the heterogeneity between modalities and provide high-quality feature representation for subsequent steps. Specifically, structured data achieves consistency of field data through normalization and missing value completion, and extracts statistical features of fields (such as mean, variance) and correlations between fields (such as Pearson correlation coefficient). Unstructured data generates semantic vectors for text through natural language processing technology, such as using the BERT embedding method to extract contextual features of text; image data extracts texture information through convolutional neural networks (CNN) and generates fixed-dimensional pixel feature representations. The processing of time series data includes wavelet decomposition and fast Fourier transform (FFT) to extract periodicity, trend values, and short-term fluctuation characteristics from time series.

[0022] The generation of adaptive multimodal feature data is the core step of the present invention, which realizes the unified representation of multimodal features through feature extraction and dynamic semantic fusion mechanism. In the feature extraction stage, structured features, unstructured features and temporal features are obtained through the above method, and dynamic semantic fusion uses mutual information calculation to measure the correlation between each modal feature and the target feature, and dynamically adjusts the fusion weight of the modal feature based on the correlation. The calculation formula of mutual information is:

[0023] in, Representing modal characteristics With target features The mutual information between for and The joint probability distribution of and are their marginal probability distributions respectively. After the calculation is completed, the fusion weight of the modal features is allocated according to the mutual information value, and the weight calculation formula is:

[0024] in, Representing modal characteristics The weight of is the total number of modal features. By weighted summing each modal feature, a unified high-dimensional feature representation is generated, and its calculation formula is:

[0025] in, is the target feature representation after fusion, and Respectively modal features and their weights.

[0026] The present invention effectively solves the challenges of heterogeneity, feature expression consistency, and weight adaptability of multimodal data through the synergistic effect of feature extraction and dynamic semantic fusion. The generated adaptive multimodal feature data unifies the representation of multi-source data, improves the semantic consistency of features and the adaptability of the model, and provides strong feature support for subsequent causal modeling and anomaly detection. Taking customer data governance as an example, the collected structured data includes purchase records, the unstructured data is customer feedback text and product pictures, and the time series data is daily sales records. Through the feature extraction and dynamic semantic fusion methods of the present invention, a unified high-dimensional feature representation is generated. In practical applications, these features not only effectively improve the ability to analyze customer behavior patterns, but also significantly improve the accuracy and robustness of the model in anomaly detection tasks.

[0027] like Figure 2 As shown, a causal chain model is constructed based on adaptive multimodal feature data to generate causal chain graph data; the causal chain graph data is dynamically adjusted to reflect the real-time causal relationship between the multimodal feature data, and the abnormal propagation path is analyzed through recursive causal reasoning to generate abnormal propagation path data; The causal chain model mines the causal relationship between adaptive multimodal feature data through causal reasoning algorithms and represents it as causal chain graph data. The causal chain graph is a directed weighted graph, in which nodes represent the feature variables in the adaptive multimodal feature data, edges represent the causal relationship between variables, and the weight of the edge quantifies the strength and direction of the causal relationship.

[0028] In the specific implementation, the derivation of causal relationships is based on a method that combines conditional independence testing with correlation analysis. For conditional independence testing, it can be implemented in combination with a causal reasoning model based on partial least squares regression to further verify the rationality of the causal chain diagram structure.

[0029] The dynamic adjustment of the causal chain graph aims to dynamically update the edge weights in the causal chain model by combining new data flowing in in real time, so as to reflect the changes in the causal relationship between data in real time. Recursive causal reasoning derives the abnormal propagation path layer by layer through the directed structure of the causal chain graph to identify potential high-risk nodes and their propagation chains.

[0030] The causal chain model constructed by the causal reasoning algorithm can depict the causal relationship between multimodal feature data in a visual form, which not only reveals the complex interaction mechanism between data, but also provides a scientific basis for abnormal propagation analysis. The dynamic adjustment mechanism of the causal chain graph enables the model to respond to data changes in a timely manner, improving the dynamic adaptability and robustness of causal relationship modeling. Recursive causal reasoning analyzes the diffusion path of anomalies through layer-by-layer propagation. The generated abnormal propagation path data can intuitively display the diffusion chain and impact range of the anomaly, which helps to quickly locate the source of the anomaly and high-risk nodes.

[0031] Preferably, the construction of the causal chain model includes: Based on adaptive multimodal feature data, a causal reasoning algorithm is used to generate causal chain graph data; the nodes of the causal chain graph data represent the characteristic variables of the adaptive multimodal feature data, and the edges of the causal chain graph data represent the causal relationship between the characteristic variables in the adaptive multimodal feature data; the edge weights of the causal chain graph data are calculated by a neural network, and the calculation of the edge weights includes correlation analysis of the multimodal feature variables and causal determination based on conditional independence; the dynamic adjustment of the causal chain graph data includes updating the edge weights in combination with real-time data.

[0032] The causal chain graph is a directed weighted graph, where nodes represent feature variables in adaptive multimodal feature data, edges represent causal relationships between feature variables, and edge weights quantify the strength of causal relationships. The generation of causal chain graph data relies on a causal inference algorithm, which combines correlation analysis and conditional independence testing to derive causal dependencies between multimodal feature variables. In the specific implementation: Each node Represents a feature variable in adaptive multimodal feature data, such as "sales", "inventory", and "customer sentiment score". The causal relationship between feature variables is quantified using a causal inference algorithm. The direction of the edge represents the direction of the causal relationship, and the weight represents the causal strength. The conditional independence test is used to verify whether the effect of a certain variable on the causal relationship can be fully explained by a third variable.

[0033] The calculation of edge weights is the core part of the causal chain graph data, and the neural network is used to perform comprehensive calculations on the correlation analysis and causal determination of feature variables. Specifically: Use correlation metrics (such as the Pearson correlation coefficient) to preliminarily measure the linear relationship between variables. Use statistical tests (such as causal inference models based on partial least squares regression) to assess whether there is conditional independence between variables, thereby determining whether there is a causal relationship. Based on a neural network, the correlation and conditional independence results are integrated to calculate the edge weights. , the formula is:

[0034] in, Feature variables right The causal weight of Indicates that the activation function (such as Sigmoid) is used to normalize the edge weights; express and correlation measures (e.g., Pearson correlation coefficient); represents the causality score of the conditional independence test; , represents the weight parameter used to balance the influence of relevance and causality scores.

[0035] The dynamic adjustment of the causal chain diagram updates the edge weights through real-time data to reflect the dynamic changes in the relationship between multimodal feature data. The specific implementation is: A new data stream is introduced to calculate the edge weight increment in the current causal chain graph .

[0036] Combine historical weights and real-time data increments to update edge weights:

[0037] in, Indicates Edge weights at the time of update; Indicates the weight increment calculated based on real-time data; Represents the smoothing coefficient, which is used to control the influence ratio of historical weight and real-time data (the range is [0,1]).

[0038] The causal chain diagram clearly depicts the causal dependency between feature variables through comprehensive correlation and causality analysis, providing a reliable basis for abnormal propagation and decision support. The dynamic adjustment mechanism ensures that the model can respond to data changes in real time and always maintain a high degree of consistency between the causal chain diagram and the data distribution. The calculation of edge weights combines the correlation and causality of multimodal features, improving the accuracy of the causal chain diagram in explaining complex data relationships.

[0039] Example: A company monitors the causal relationship between inventory, sales, and user feedback, and collects the following adaptive multimodal feature data: Node 1: Daily inventory changes (time series features).

[0040] Node 2: Daily sales (structured features).

[0041] Node 3: User feedback sentiment score (unstructured feature).

[0042] Causal chain diagram generation: Inventory changes, sales, and sentiment scores are used as nodes of the causal chain diagram.

[0043] Causal inference is used to calculate the causality between variables. The direction and weight of the edge represent the strength of the relationship, such as the edge weight of inventory change on sales. is 0.8.

[0044] The causal weight of inventory change caused by the addition of user feedback sentiment score has changed, and it is recalculated through the edge weight update formula .

[0045] The causal chain diagram ultimately shows that inventory changes are indirectly affected by user emotional feedback through sales.

[0046] Through the causal chain diagram model, companies can accurately identify the indirect impact of user feedback on inventory management and quickly adjust inventory strategies when sales fluctuate significantly, thereby optimizing supply chain management efficiency.

[0047] Preferably, the recursive causal reasoning includes: The abnormal propagation path is derived layer by layer from the causal chain graph data through a recursive causal reasoning algorithm, and the initial conditions of the recursive causal reasoning are the starting node of the abnormal propagation path and its initial abnormal value; the abnormal propagation probability of each node is calculated during the recursive causal reasoning process, and the abnormal propagation probability calculation includes a signal transmission formula based on the edge weights in the causal chain; the reasoning results of each abnormal propagation path record the impact range of the propagation and possible high-risk nodes to form abnormal propagation path data.

[0048] Recursive causal reasoning uses the starting node of the anomaly propagation path and its initial anomaly value as initial conditions. The starting node is the node where the anomaly first appears, and the initial anomaly value reflects the degree of anomaly of the node, which is usually represented by a standardized value provided by the detection module. For example, suppose the sales node is abnormal and the initial anomaly value is detected. =1.0, recursive reasoning will start propagation analysis from this node.

[0049] In the recursive reasoning process, the probability of abnormal propagation at each layer is determined by the edge weight in the causal chain graph and the abnormal propagation probability of the upper node. The propagation formula is as follows:

[0050] in, Indicates The probability of abnormal propagation of layer nodes; Representation Node The set of predecessor nodes; Represents a node in a causal chain graph For Node The causal edge weights of Representation Node In the The probability of abnormal propagation of the layer; Represents a nonlinear normalization function (such as Sigmoid), which is used to ensure that the value of the propagation probability is in the range of [0, 1].

[0051] This formula describes the mechanism by which anomalies diffuse layer by layer in a causal network. The propagation probability of each layer of nodes is obtained by accumulating the propagation probabilities of all its predecessor nodes and the corresponding edge weights, and is normalized by a nonlinear function to simulate the attenuation effect of anomaly diffusion in reality.

[0052] During the reasoning process, nodes with higher propagation probabilities (exceeding the set threshold) are marked and considered high-risk nodes. At the same time, the impact range of the anomaly is recorded based on the coverage of the propagation path. The impact range is represented by the set of all nodes with a propagation probability greater than zero. The identification of high-risk nodes and impact ranges provides a clear target area for anomaly tracing and governance.

[0053] After the recursive reasoning is completed, the complete anomaly propagation path data is constructed based on the propagation results of all levels. The data includes: the starting node; all nodes on the anomaly propagation path and their propagation probabilities; the list of high-risk nodes and their propagation probabilities; and the set of covered nodes in the anomaly impact range.

[0054] Through recursive causal reasoning, the present invention realizes dynamic analysis and visualization of the anomaly diffusion mechanism, with the following main effects: Through recursive propagation analysis of multi-layer causal relationships, the diffusion chain of anomalies is clearly displayed, providing a scientific basis for anomaly tracing. During the recursive reasoning process, high-risk nodes with high propagation probability are marked in real time, which facilitates the rapid locking of the core impact area of ​​the anomaly. Recording the coverage of anomaly diffusion helps to make the formulation of governance strategies more precise and regionalized.

[0055] Example: The monitoring system of a retail enterprise records the following multimodal feature data: Node 1: Sales (structured feature).

[0056] Node 2: Inventory quantity (time series feature).

[0057] Node 3: Customer sentiment score (unstructured features).

[0058] Initial condition setting: Starting Node: Sales.

[0059] Initial outliers: =1.0.

[0060] Recursive reasoning process: First-level propagation: According to the edge weight of "sales → inventory" in the causal chain graph , calculate the propagation probability of the inventory node:

[0061] Layer 2 propagation: According to the edge weight of "inventory quantity → sentiment score" , calculate the propagation probability of the sentiment score node:

[0062] According to the propagation probability threshold , the inventory node is marked as a high-risk node.

[0063] Anomaly propagation path data generation: Starting Node: Sales.

[0064] Communication path: sales → inventory → sentiment score.

[0065] High-risk node: inventory level.

[0066] Impact: Sales, inventory levels, sentiment scores.

[0067] Through recursive causal reasoning, companies can clearly identify the chain impact of sales anomalies on inventory management and customer emotional feedback, and quickly identify inventory management as high-risk areas, providing an accurate basis for formulating optimized inventory strategies.

[0068] Based on the abnormal propagation path data, the quantum distribution model data is constructed using the distribution modeling method; based on the quantum distribution model data, the abnormal features outside the distribution are identified, and the fundamental abnormal feature data is generated through recursive abnormal location analysis; The quantum distribution model simulates the global distribution and complex correlation of various features in the abnormal propagation path by introducing technologies such as quantum state superposition and quantum entanglement. Recursive anomaly positioning further traces the propagation path of abnormal features and their root causes, providing efficient and accurate support for anomaly analysis and governance decisions in the data governance process.

[0069] The quantum distribution model is a method that combines the characteristics of quantum states to efficiently model data distribution. It uses quantum superposition and quantum entanglement technology to simulate the nonlinear correlation and global distribution characteristics between characteristic variables. The specific implementation includes the following steps: 1. Quantum state encoding: Encode each characteristic variable in the abnormal propagation path data into a quantum state vector. The mathematical representation of the quantum state is:

[0070] in, represents the quantum state of the characteristic variable, is the amplitude of the characteristic variable, indicating its contribution to the global distribution, is the ground state of the characteristic variable, is the number of feature variables.

[0071] 2. Quantum superposition state generation: The quantum state of all characteristic variables is combined into a global distribution model using the quantum superposition effect, which is expressed as:

[0072] in, is the global quantum distribution model, is the weight coefficient, indicating the importance of the feature variable, is the number of features in the anomaly propagation path.

[0073] 3. Introduction of quantum entanglement: Model the nonlinear correlation between characteristic variables and generate entangled states to describe the complex dependencies between variables. The mathematical form of the entangled state is:

[0074] Entangled states enhance the ability to model implicit relationships between characteristic variables.

[0075] 4. Distribution expression: Integrate superposition state and entangled state to construct a complete quantum distribution model data, which is used to efficiently describe the joint distribution of characteristic variables in abnormal propagation path data.

[0076] Recursive anomaly location analysis determines the root cause of the anomaly by tracing the out-of-distribution anomaly features layer by layer. The specific process is as follows: Hierarchical clustering is performed on out-of-distribution features to mark groups of feature variables with similar distribution behaviors.

[0077] Combined with the causal chain diagram in the abnormal propagation path data, the abnormal propagation chain is traced from the clustering results, and the possible fundamental abnormal characteristics are located by tracing back layer by layer.

[0078] The located fundamental anomaly features and their associated propagation weights are used to form fundamental anomaly feature data for use by subsequent governance modules.

[0079] The quantum distribution model can efficiently express the complex correlation and global distribution characteristics of characteristic variables in the abnormal propagation path, significantly improving the accuracy of anomaly detection. Through quantum state measurement operations and distribution probability calculations, it accurately identifies abnormal characteristics outside the distribution and ensures the reliability of anomaly detection results. Recursive anomaly location analysis combines causal chain deduction with cluster analysis to quickly identify the key causes of anomalies and optimize anomaly governance strategies.

[0080] Preferably, the construction of the quantum distribution model data includes: Based on the abnormal propagation path data, a quantum state vector is generated through quantum state encoding; the quantum state vector is represented as the amplitude and phase distribution of a multimodal characteristic variable; a global quantum distribution model is generated through the quantum superposition effect, and the quantum superposition effect includes linear superposition of different characteristic quantum states to simulate the global distribution; the nonlinear correlation between the multimodal characteristic variables is simulated through the quantum entanglement effect to form quantum distribution model data.

[0081] Quantum state encoding is the first step, which captures the importance and characteristics of each characteristic variable in the abnormal propagation path data by converting it into a quantum state. The quantum state is represented by the amplitude and phase of the characteristic variable, the amplitude represents the importance of the variable, and the phase represents the distribution pattern of the variable. The quantum states of all characteristic variables form a quantum state vector, which provides a basis for subsequent model construction.

[0082] The quantum superposition effect is used to generate a global quantum distribution model, which linearly combines all characteristic quantum states in the form of weights to uniformly represent the distribution characteristics of multimodal characteristic variables. This superposition process retains the individual information of the characteristic variables and effectively integrates them into the overall distribution, ensuring that the model can describe the global data distribution.

[0083] The quantum entanglement effect simulates the nonlinear correlation between characteristic variables and captures the complex dependencies between variables. This process expresses the relationship between characteristic variables by constructing entangled states, enhancing the model's ability to describe the implicit correlation between multimodal features.

[0084] Quantum state superposition integrates the global distribution of multimodal feature variables, improving the accuracy of data modeling. Quantum entanglement technology effectively captures the nonlinear correlation between feature variables, bringing higher expressive power to distribution modeling. Based on the data support of quantum distribution model, the efficiency and accuracy of anomaly detection and anomaly root cause location are significantly improved.

[0085] Example: The abnormal propagation path data of a certain retail platform includes: Feature 1: sales; Feature 2: inventory level; Feature 3: customer feedback score.

[0086] Implementation process: Quantum state encoding: Sales, inventory levels, and customer feedback are encoded as quantum state vectors, where the amplitude reflects the importance of the feature variable and the phase expresses its distribution pattern.

[0087] Quantum superposition generates a global model: combining the quantum states of all feature variables to generate a global quantum distribution model that represents the joint distribution of multimodal features.

[0088] Quantum entanglement modeling nonlinear correlations: Modeling implicit relationships between feature variables, such as the indirect impact of customer feedback on sales, and expressing this complex correlation through entangled states.

[0089] Through the construction of a quantum distribution model, the impact of abnormal customer feedback on inventory management and sales is accurately captured, providing data support for enterprises to optimize inventory strategies and customer management.

[0090] Preferably, the identification of abnormal features outside the distribution includes: By performing feature measurement calculation on the quantum distribution model data, the feature measurement includes inner product calculation of quantum state vector to determine the distribution probability of each feature variable; the feature variable with distribution probability lower than a set threshold is marked as an out-of-distribution abnormal feature.

[0091] Feature measurement determines the probability value of each feature variable in the global distribution by calculating the inner product of the quantum state vector. The quantum state vector represents the state of each feature variable, while the global quantum distribution model represents the joint distribution of all feature variables. Feature measurement quantifies the distribution probability of feature variables by measuring the relationship between the quantum state and the global distribution model to evaluate the normality of feature variables.

[0092] In the calculation, the distribution probability of the characteristic variable is determined by the inner product of its quantum state and the global quantum distribution model. The size of the distribution probability reflects the consistency of the characteristic variable in the overall distribution. A lower distribution probability indicates that the characteristic variable may be abnormal. In this way, the scheme can quickly screen out characteristic variables outside the distribution.

[0093] After calculating the distribution probability of each feature variable, these probability values ​​are compared with the preset abnormal threshold. The threshold is set based on field experience and historical data, reflecting the lowest acceptable probability value of the feature variable in the normal distribution. When the distribution probability of a feature variable is lower than the threshold, the feature variable is marked as an out-of-distribution abnormal feature.

[0094] By quantitatively analyzing the distribution probability of each feature variable, this method can accurately screen out the out-of-distribution abnormal features, which is especially suitable for complex multimodal data. The inner product measurement has high computational efficiency and can quickly complete the distribution probability calculation of all feature variables even in the face of large-scale data. The marked out-of-distribution abnormal features provide a clear goal for subsequent anomaly governance and root cause location, significantly improving the efficiency of anomaly handling.

[0095] Example: The monitoring system of a logistics company collects the following multimodal feature data: Node 1: logistics transportation time; Node 2: daily order volume; Node 3: customer satisfaction score.

[0096] Implementation process: Feature measurement: Based on the quantum distribution model data, the logistics transportation time, daily order volume and customer satisfaction score are measured respectively, and their distribution probabilities are calculated.

[0097] For example, the distribution probability of logistics transportation time is low, indicating that this variable may be outside the normal range.

[0098] Abnormal marking: The distribution probability threshold is set to 0.05. The measurement results show that the distribution probability of logistics transportation time is 0.03, which is lower than the threshold, so it is marked as an abnormal feature outside the distribution.

[0099] By identifying anomalies in logistics transportation time, companies quickly discovered the existence of transportation delays and located the core causes of the delays, providing data support for optimizing logistics strategies.

[0100] Through feature measurement based on quantum distribution model and out-of-distribution anomaly marking, the present invention achieves accurate screening of abnormal features, which not only improves the accuracy of anomaly detection, but also lays a solid foundation for the subsequent governance process. The efficiency and scalability of the solution make it suitable for a variety of complex scenarios, providing innovative support for the monitoring process of data governance.

[0101] Preferably, the recursive anomaly location analysis includes: Based on the data of the out-of-distribution abnormal features, cluster analysis is performed on the out-of-distribution abnormal features through a recursive clustering algorithm; the cluster analysis includes a hierarchical clustering method based on feature distance; a local feature abnormal area is generated for each clustering result, and a reverse reasoning algorithm is used to trace the abnormal cause from the local feature abnormal area; in the tracing process, the temporal dependency of the feature variables and the abnormal propagation path of the causal chain diagram data are combined to form fundamental abnormal feature data.

[0102] Recursive anomaly location analysis is a process of clustering out-of-distribution anomaly features through a recursive clustering algorithm, and tracing the source by combining the reverse reasoning algorithm and causal chain graph data, and finally forming the fundamental anomaly feature data. This process first applies a recursive clustering algorithm to the out-of-distribution anomaly feature data based on feature distance calculation, and gradually merges the local feature anomaly areas through feature similarity.

[0103] The clustering method uses hierarchical clustering technology. The calculation of feature distance can select appropriate metrics based on domain requirements, such as Euclidean distance or cosine similarity, to quantify the correlation between feature variables. After forming a local feature anomaly area, the recursive anomaly location analysis uses a reverse reasoning algorithm to combine the abnormal propagation path of the temporal dependency and causal chain graph data to trace the cause of the anomaly layer by layer. In the tracing process, the time series characteristics of the feature variables are analyzed through the temporal dependency relationship to determine the propagation order of the anomaly, for example, whether the anomaly of a certain variable causes a change in its successor variable; at the same time, combined with the causal relationship between the variables in the causal chain graph and the edge weight information, the cause of the anomaly is deduced forward, and the most likely fundamental abnormal feature is continuously focused on through recursive reasoning.

[0104] Ultimately, the fundamental anomaly feature data records the key feature variables that cause the anomaly and its propagation path, providing a clear target for subsequent anomaly governance. The advantage of recursive anomaly location analysis is that it can accurately locate the root cause of the anomaly. By combining multiple analysis methods, it avoids the ambiguity or misjudgment of the source of the anomaly by traditional methods, while effectively improving the efficiency and accuracy of anomaly handling, providing strong support for solving complex anomaly problems in the data governance process. During the implementation process, combined with the out-of-distribution anomaly feature data of the supply chain scenario, the system discovered abnormal areas of inventory volume, customer satisfaction score, and order completion time through recursive clustering analysis, and combined with the causal chain diagram to trace the source to locate inventory anomalies as the main cause, providing a reliable basis for enterprises to formulate inventory optimization strategies.

[0105] like Figure 3 As shown, preliminary governance rule data is generated based on fundamental abnormal feature data, and the preliminary governance rule data is optimized through a rule optimization method to generate optimized governance rule data; the optimized governance rule data is used to update the monitoring system status to generate optimized monitoring system status data.

[0106] Using the fundamental abnormal feature data as input, the rule generation algorithm extracts governance suggestions and initially forms governance rule data, and then adjusts and enhances the preliminary rules in combination with the rule optimization method to ensure the applicability and accuracy of the rules. Finally, the optimized governance rule data is used to update the monitoring system status, realize the dynamic adjustment and optimization of the system, and thus generate optimized monitoring system status data to ensure the efficient operation of the system after abnormal governance.

[0107] The preliminary governance rule data is the first step output generated based on the fundamental anomaly feature data. The fundamental anomaly feature data contains abnormal variables and their associated paths, and is the basis for the generation of governance rules. Through the rule generation algorithm, the fundamental anomaly feature data is mapped to specific governance operation suggestions. For example, if the fundamental anomaly feature is "insufficient inventory", the rule generation algorithm will generate the following preliminary governance rules based on the relationship between inventory and sales: "Increase the frequency of inventory replenishment" or "Adjust the inventory safety threshold". The rule generation process usually combines the experience of domain experts and historical data analysis, extracts governance suggestions through rule templates or rule mining methods, and forms preliminary governance rule data.

[0108] The rule optimization method aims to improve the applicability and effectiveness of the preliminary governance rules. Its core is to evaluate and adjust the rules through a multi-objective optimization algorithm, including the optimization of key indicators such as rule coverage, execution cost and adaptability. The specific process is as follows: Analyze the applicable scenarios and target variables of the preliminary rules to ensure the wide applicability of the rules in different abnormal situations. Combine the cost model to evaluate the resource requirements and costs of rule execution, and remove rules with high costs and limited effects. Verify the effectiveness of the rules in actual applications through simulation or experiments, and adjust the rule parameters based on feedback to make them more in line with actual needs. For example, for the rule of "increasing the frequency of inventory replenishment", the specific value of the replenishment frequency can be optimized to reduce logistics costs and delays by analyzing the timeliness data of each link in the supply chain.

[0109] The optimized governance rule data is the result of rule optimization, with higher adaptability and execution effect. For example, an optimized governance rule may clearly state: "Adjust the frequency of inventory replenishment from once a week to once every two days, focusing on Class A products", which is significantly more accurate than the preliminary rule.

[0110] Using optimized governance rule data to update the monitoring system status is a key step in achieving a closed-loop governance. Rule data is loaded into the system for real-time monitoring and dynamic adjustment, thereby generating optimized monitoring system status data. Specifically, system status updates include: Rule deployment: Convert optimized rules into system executable logic.

[0111] Real-time monitoring adjustment: Dynamically adjust the monitoring strategy based on the effect of real-time data verification rules.

[0112] Establishment of feedback loop: Through updated monitoring system status data, governance effects are continuously collected to provide support for further optimization.

[0113] Rule generation and optimization ensure that governance rules can cover a variety of scenarios for abnormal problems, improving the applicability and breadth of rules. The rule optimization method reduces execution costs and improves the feasibility and economy of governance operations. The dynamic update of the monitoring system status ensures that the system can adapt to changes in the status after governance, providing a solid foundation for continuous optimization.

[0114] Example: A retail enterprise monitoring system generates the following fundamental anomaly feature data through recursive anomaly location analysis: Abnormal feature 1: abnormal inventory; Abnormal feature 2: Order delay.

[0115] Implementation process: Initial governance rule generation: Based on the abnormal inventory level, a preliminary rule is generated: "Increase the frequency of inventory replenishment."

[0116] Based on order delays, generate preliminary rules: “Optimize order delivery process.”

[0117] Rule optimization: Regarding the inventory replenishment rules, the replenishment cycle was adjusted from every week to every three days by optimizing the specific values ​​of the replenishment frequency.

[0118] Based on order delay rules, we analyze historical data to optimize delivery routes and give priority to warehouses with strong delivery capabilities.

[0119] Monitor system status updates: Deploy optimized rules to load replenishment frequency adjustment and delivery optimization into the monitoring system.

[0120] Adjust replenishment and distribution strategies in real time and monitor the effectiveness of rule execution to ensure the stability of optimization results.

[0121] Through optimized governance rule data, the company successfully reduced the occurrence of inventory out-of-stock problems, while reducing order delay rates, and overall improved supply chain efficiency and customer satisfaction.

[0122] Preferably, the generation of the optimized governance rule data includes: Based on the fundamental abnormal feature data, the preliminary governance rule data is optimized using a memetic evolution algorithm; the optimization process of the memetic evolution algorithm includes feature crossover and parameter variation of governance rule condition parameters, the feature crossover includes combining condition parameters of multiple governance rules, and the parameter variation includes randomly adjusting the upper and lower limits of the rule threshold; the governance rule with the highest fitness is selected through fitness evaluation to generate optimized governance rule data.

[0123] The memetic evolution algorithm is an optimization method based on genetic evolution. It simulates the natural selection and evolution process of rules through feature crossover and parameter variation, and finally selects the optimal rule set. The present invention is suitable for the optimization of preliminary governance rule data, and can quickly find the rule with the highest fitness among a large number of possible rule parameter combinations.

[0124] The optimization process includes: 1. Construction of the initial rule population: The initial rule population is a combination of multiple rules generated based on the preliminary governance rule data. Each rule contains governance condition parameters, such as the triggering conditions and execution thresholds of the rule. Each individual in the population represents a possible governance rule.

[0125] 2. Feature Crossover: Feature crossover is a key step in the memetic evolution algorithm, which is used to generate new rule combinations. New governance rules are generated by combining the conditional parameters of two rules. For example, the conditional parameter of rule A is "inventory replenishment threshold = 50%", and the conditional parameter of rule B is "order delay time = 24 hours". Through feature crossover, a new rule C can be generated, and its conditional parameters may be "inventory replenishment threshold = 50%, order delay time = 24 hours".

[0126] 3. Parameter variation: After generating new rules through feature crossover, the diversity of rules is increased through parameter variation to avoid falling into local optimality. Parameter variation includes randomly adjusting the upper and lower limits of the rule thresholds. For example, the inventory replenishment threshold is adjusted from 50% to 45%-55%.

[0127] 4. Fitness evaluation: Fitness evaluation is the core of the optimization process and is used to measure the effect and applicability of each rule. The fitness function is calculated comprehensively based on the coverage, execution cost and governance effect of the governance rules. The higher the fitness value of the rule, the better its optimization effect is, and it is more in line with the needs of abnormal governance.

[0128] 5. Optimal rule selection: Based on the fitness evaluation results, select the governance rule with the highest fitness to generate optimized governance rule data, and remove the rules with lower fitness to simplify the rule set.

[0129] By optimizing the parameters of rule conditions through feature crossover and parameter mutation, the memetic evolution algorithm can generate rules that better meet governance needs. Combining multiple rule conditions to generate new rule combinations increases the flexibility and applicability of governance rules. Inefficient rules are eliminated through fitness evaluation to ensure that the optimized rules meet governance needs while being cost-effective.

[0130] Example: In a logistics system, the preliminary governance rule data includes the following rules: Rule A: When the inventory replenishment rate is less than 50%, execute the replenishment plan.

[0131] Rule B: When the order delay exceeds 24 hours, priority will be given to delivery.

[0132] Implementation process: Rule population construction: Build an initial population that includes Rule A and Rule B, as well as different parameter combinations. For example, "inventory replenishment rate = 45%-55%" or "order delay time = 18-30 hours".

[0133] Feature intersection: Perform feature intersection on rule A and rule B to generate new rule C, whose condition parameters are "inventory replenishment rate = 50%, order delay time = 24 hours".

[0134] Parameter mutation: mutate the newly generated rules, for example, adjust the inventory replenishment rate to 45%-55%, and generate multiple mutated rules.

[0135] Fitness evaluation: The execution effect of the rules is simulated based on the historical data of the logistics system, and each rule is evaluated based on the coverage, execution cost and governance effect. Rules with high fitness values ​​will be retained.

[0136] Generate optimized rules: After multiple rounds of evolution and evaluation, optimized governance rule data is finally generated, for example: Rule 1: When the inventory replenishment rate is lower than 48%-52%, execute the replenishment plan; Rule 2: When the order delay exceeds 22-26 hours, priority delivery will be given.

[0137] Through the optimization of the memetic evolution algorithm, the logistics system successfully adjusted the governance rules to a more accurate and efficient version. The optimized rule data significantly improved the effectiveness and efficiency of abnormal governance. For example, the execution success rate of replenishment plans increased by 20%, the response time of order delay issues was shortened by 15%, and the overall logistics operation costs were reduced. The optimized rules showed stronger adaptability and flexibility in practical applications, providing a solid foundation for the dynamic adjustment of the monitoring system.

[0138] Through the application of memetic evolution algorithm, the present invention realizes efficient conversion from preliminary rules to optimal rules in terms of rule optimization, ensures the accuracy and practicality of governance rules, and ensures that the optimized rules can meet actual needs to the greatest extent through fitness evaluation and multiple rounds of evolution. This process greatly improves the efficiency and effect of handling abnormal problems in the data governance process.

[0139] Preferably, the generation of the optimized monitoring system state includes: Based on the optimized governance rule data, the model parameters of the feature extraction, causal chain modeling and quantum distribution modeling of the monitoring system are updated respectively; the updated model parameters include the weight allocation rule for feature extraction, the causal relationship edge weight update rule and the phase adjustment rule for the quantum state vector; and the optimized monitoring system state is generated through multiple rounds of parameter iterative evaluation.

[0140] The parameter update of the feature extraction module adjusts the weight distribution rules of multimodal data features based on the optimized governance rule data. The weight update is intended to reflect the priority adjustment of feature variables by the optimized governance rules. For example, for key feature variables in anomaly governance, such as inventory or order delay time, their weights will be increased to gain higher attention in subsequent anomaly detection. The formula for weight distribution update can be expressed as:

[0141] in, Represents feature variables The updated weight of represents the initial weight before optimization; Represents the weight adjustment value calculated based on the optimized governance rule data.

[0142] This updating method ensures that key feature variables are given higher detection priority during the monitoring process, enhancing the system's feature extraction capabilities.

[0143] The causal chain modeling module updates the weight rules of the causal relationship edges to reflect the adjustment of the optimized governance rule data to the variable dependency strength in the causal path. The optimized governance rule data may identify that the correlation strength between certain variables needs to be increased or decreased, thereby adjusting the edge weights in the causal relationship graph. The edge weight update formula can be expressed as:

[0144] in, Representation variables and Update the edge weights between ; represents the initial edge weight before optimization; Represents the weight adjustment value calculated based on the optimized governance rule data.

[0145] Through multiple rounds of iterative updates, the causal chain modeling module can more accurately reflect the dependencies between variables, improve the reliability of causal chain analysis and the accuracy of abnormal propagation paths.

[0146] The quantum distribution modeling module optimizes the representation of characteristic variables in the global distribution by adjusting the phase of the quantum state vector. The optimized governance rule data may identify that the distribution pattern of certain characteristic variables needs to be fine-tuned to better meet governance needs. The quantum state vector phase adjustment formula is:

[0147] in, Feature variables The updated phase of represents the initial phase before optimization; Represents the phase adjustment value calculated based on the optimized governance rule data.

[0148] The phase adjustment ensures that the quantum distribution modeling module reflects the abnormal patterns of characteristic variables more accurately in the global distribution, thus improving the accuracy of out-of-distribution anomaly identification.

[0149] After all module parameters are updated, the system enters a multi-round parameter iteration evaluation phase, where the optimized model parameters are actually tested through the monitoring system, and the parameters are further adjusted according to the test results until the optimal state is reached. Each round of evaluation includes the following steps: Load the currently updated parameters into each module; test the monitoring effect of the system by simulating abnormal scenarios and record performance indicators; adjust the model parameters according to the test results and enter the next round of iteration.

[0150] By updating feature extraction weights, causal chain edge weights, and quantum state phases, the optimized monitoring system can more accurately capture abnormal features and their propagation paths. Multiple rounds of parameter iteration evaluation ensure consistent performance of the system in different abnormal scenarios, improving the system's dynamic adaptability. The resulting optimized monitoring system state comprehensively improves monitoring efficiency and anomaly detection capabilities in the data governance process.

[0151] Example: The abnormal data collected by a logistics company’s monitoring system contains the following characteristics: Abnormal inventory levels; Abnormal order delay time; Customer satisfaction ratings declined.

[0152] Implementation process: Feature extraction module update: For inventory abnormalities, the weight is increased from 0.3 to 0.5 to reflect its priority in abnormality management.

[0153] Update of the causal chain modeling module: Based on the optimized governance rule data, the edge weight between order delay time and inventory is adjusted from 0.4 to 0.6.

[0154] Quantum distribution modeling module update: Adjust the quantum state phase of the inventory variable, increase the initial phase value by 10 degrees to enhance the ability to capture abnormal patterns.

[0155] Multiple rounds of parameter iteration evaluation: Conduct abnormal scenario tests in a simulated environment, record the monitoring results after each round of updates, and ultimately determine the optimal parameter combination.

[0156] Through parameter updates and multiple rounds of iterative evaluation, the optimized monitoring system status can capture the changing trends of abnormal inventory levels in real time, while more accurately identifying the abnormal propagation paths, thereby improving the execution effect of governance rules and the efficiency of exception handling.

[0157] By updating the monitoring system status through optimized governance rule data, the present invention realizes the optimization of the entire process of the monitoring system from feature extraction to causal analysis to anomaly detection. This process ensures the performance improvement of the system in anomaly governance through multiple rounds of parameter adjustment and evaluation, providing effective guarantee for the continuous optimization of data governance.

[0158] like Figure 4 As shown, a system for implementing the monitoring method of the data governance process, the system comprising: The multimodal data acquisition module is used to collect multimodal data from multiple data sources and preprocess the multimodal data to generate adaptive multimodal feature data; the multimodal data acquisition module is responsible for acquiring structured data (such as database fields), unstructured data (such as text content and image pixels), and time series data (such as sensor readings) from multiple data sources. The collected data is preprocessed by cleaning, denoising, and standardization to generate adaptive multimodal feature data, and the dynamic semantic fusion method is used to map different types of data to a unified feature space, providing high-quality, unified feature representation for subsequent analysis.

[0159] The causal chain modeling module is used to build a causal chain model based on adaptive multimodal feature data and generate causal chain graph data; the causal chain modeling module is also used to dynamically adjust the causal chain graph data to reflect the real-time causal relationship between multimodal feature data, and generate abnormal propagation path data through recursive causal reasoning analysis; the causal chain modeling module builds a causal chain model based on adaptive multimodal feature data, and generates causal chain graph data through causal reasoning algorithm, where nodes represent variables and edges represent causal relationships and their weights between variables. The module dynamically adjusts edge weights to reflect real-time data changes, and uses recursive causal reasoning algorithm to analyze abnormal propagation paths, accurately locates the starting point and propagation path of the abnormality, and generates abnormal propagation path data.

[0160] The distribution modeling module is used to construct quantum distribution model data based on abnormal propagation path data; the distribution modeling module is also used to identify abnormal features outside the distribution based on the quantum distribution model data, and generate fundamental abnormal feature data through recursive abnormal location analysis; the distribution modeling module constructs a quantum distribution model based on abnormal propagation path data, uses quantum state vectors to describe the amplitude and phase of characteristic variables, and simulates the complex nonlinear relationship between variables through quantum superposition and entanglement effects. The module calculates the probability of characteristics outside the distribution through characteristic measurement, identifies abnormal characteristics, and generates fundamental abnormal feature data in combination with recursive location analysis, providing a traceability basis for abnormal governance.

[0161] The governance rule generation module is used to generate preliminary governance rule data based on fundamental abnormal feature data, and optimize the preliminary governance rule data through rule optimization methods to generate optimized governance rule data; the governance rule generation module generates preliminary governance rules based on fundamental abnormal feature data, optimizes the parameter configuration of the rules through the memetic evolution algorithm, and improves the adaptability and execution efficiency of the rules. The optimized governance rule data covers high-priority governance strategies and provides an accurate rule set for dynamic adjustment and exception handling of the system.

[0162] The monitoring system update module is used to update the monitoring system status based on the optimized governance rule data and generate optimized monitoring system status data. The monitoring system update module uses the optimized governance rule data to dynamically adjust system parameters, including key parameters such as feature extraction weights, causal chain edge weights, and quantum state phases. The module generates optimized monitoring system status data through multiple rounds of iterative evaluation to ensure that the system can adapt to environmental changes in real time and continuously and efficiently complete abnormal monitoring and governance tasks.

[0163] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0164] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for monitoring a data governance process, characterized in that: The following steps are involved: Collecting multimodal data from multiple data sources and preprocessing the collected multimodal data to generate adaptive multimodal feature data; A causal chain model is constructed based on adaptive multimodal feature data to generate causal chain graph data. The causal chain graph data is dynamically adjusted to reflect the real-time causal relationship between multimodal feature data, and anomaly propagation paths are analyzed through recursive causal reasoning to generate anomaly propagation path data. Based on the abnormal propagation path data, the quantum distribution model data is constructed using the distribution modeling method; based on the quantum distribution model data, the abnormal characteristics outside the distribution are identified, and the fundamental abnormal characteristic data is generated through recursive abnormal location analysis; Generate preliminary governance rule data based on fundamental abnormal feature data, and optimize the preliminary governance rule data through rule optimization methods to generate optimized governance rule data; use the optimized governance rule data to update the monitoring system status to generate optimized monitoring system status data.

2. The data governance process monitoring according to claim 1, characterized in that: The generation of the adaptive multimodal feature data includes: Structured features, unstructured features and time series features are extracted from multimodal preprocessed data through a feature extraction method; the structured features include statistical features of fields in a data table and the correlation between fields, the unstructured features include semantic vectors of text and pixel distribution features of images, and the time series features include trend values ​​and periodic features of time series; the structured features, unstructured features and time series features are fused and optimized using a dynamic semantic fusion mechanism to generate adaptive multimodal feature data.

3. The data governance process monitoring according to claim 1, characterized in that: The construction of the causal chain model includes: Based on adaptive multimodal feature data, a causal reasoning algorithm is used to generate causal chain graph data; the nodes of the causal chain graph data represent the characteristic variables of the adaptive multimodal feature data, and the edges of the causal chain graph data represent the causal relationship between the characteristic variables in the adaptive multimodal feature data; the edge weights of the causal chain graph data are calculated by a neural network, and the calculation of the edge weights includes correlation analysis of the multimodal feature variables and causal determination based on conditional independence; the dynamic adjustment of the causal chain graph data includes updating the edge weights in combination with real-time data.

4. The data governance process monitoring according to claim 1, characterized in that: The recursive causal reasoning includes: The abnormal propagation path is deduced layer by layer from the causal chain graph data through a recursive causal reasoning algorithm, and the initial conditions of the recursive causal reasoning are the starting node of the abnormal propagation path and its initial abnormal value; the abnormal propagation probability of each node is calculated during the recursive causal reasoning process, and the abnormal propagation probability calculation includes a signal transmission formula based on the edge weights in the causal chain; the reasoning results of each abnormal propagation path record the impact range of the propagation and possible high-risk nodes to form abnormal propagation path data.

5. The data governance process monitoring according to claim 1, characterized in that: The construction of the quantum distribution model data includes: Based on the abnormal propagation path data, a quantum state vector is generated through quantum state encoding; the quantum state vector is represented as the amplitude and phase distribution of multimodal characteristic variables; a global quantum distribution model is generated through the quantum superposition effect, and the quantum superposition effect includes linear superposition of different characteristic quantum states to simulate the global distribution; the nonlinear correlation between the multimodal characteristic variables is simulated through the quantum entanglement effect to form quantum distribution model data.

6. The data governance process monitoring according to claim 1, characterized in that: The identification of abnormal features outside the distribution includes: By performing feature measurement calculation on the quantum distribution model data, the feature measurement includes inner product calculation of the quantum state vector to determine the distribution probability of each feature variable; the feature variable with a distribution probability lower than a set threshold is marked as an out-of-distribution abnormal feature.

7. The data governance process monitoring according to claim 1, characterized in that: The recursive exception location analysis includes: Based on the data of the out-of-distribution abnormal features, a cluster analysis is performed on the out-of-distribution abnormal features through a recursive clustering algorithm; the cluster analysis includes a hierarchical clustering method based on feature distance; a local feature abnormal area is generated for each clustering result, and a reverse reasoning algorithm is used to trace the cause of the abnormality from the local feature abnormal area; in the tracing process, the temporal dependency of the feature variables and the abnormal propagation path of the causal chain diagram data are combined to form the fundamental abnormal feature data.

8. The data governance process monitoring according to claim 1, characterized in that: The generation of the optimized governance rule data includes: Based on the fundamental abnormal feature data, the preliminary governance rule data is optimized using a memetic evolution algorithm; the optimization process of the memetic evolution algorithm includes feature intersection and parameter variation of the conditional parameters of the governance rules, the feature intersection includes combining the conditional parameters of multiple governance rules, and the parameter variation includes randomly adjusting the upper and lower limits of the rule threshold; the governance rule with the highest fitness is selected through fitness evaluation to generate optimized governance rule data.

9. The data governance process monitoring according to claim 1, characterized in that: The generation of the optimized monitoring system state includes: Based on the optimized governance rule data, the model parameters of the monitoring system's feature extraction, causal chain modeling, and quantum distribution modeling are updated respectively; the updated model parameters include the weight allocation rule for feature extraction, the causal relationship edge weight update rule, and the phase adjustment rule for the quantum state vector; and the optimized monitoring system state is generated through multiple rounds of parameter iterative evaluation.

10. A system for implementing the data governance process monitoring method according to any one of claims 1 to 9, characterized in that: The system includes: A multimodal data acquisition module, configured to acquire multimodal data from multiple data sources and pre-process the multimodal data to generate adaptive multimodal feature data; A causal chain modeling module is used to construct a causal chain model based on adaptive multimodal feature data and generate causal chain graph data. The causal chain modeling module is also used to dynamically adjust the causal chain graph data to reflect the real-time causal relationship between the multimodal feature data and generate abnormal propagation path data through recursive causal reasoning analysis; A distribution modeling module is used to construct quantum distribution model data based on the abnormal propagation path data; the distribution modeling module is also used to identify abnormal features outside the distribution based on the quantum distribution model data, and generate fundamental abnormal feature data through recursive abnormal location analysis; A governance rule generation module is used to generate preliminary governance rule data based on fundamental abnormality feature data, and optimize the preliminary governance rule data through a rule optimization method to generate optimized governance rule data; The monitoring system update module is used to update the monitoring system status based on the optimized governance rule data and generate optimized monitoring system status data.

Images