A business operation monitoring system based on AI data sharing

Abstract

This invention relates to the field of data monitoring technology, specifically to a business operation monitoring system based on AI data sharing, comprising: a federated learning module for aggregating model parameters of multiple business systems through a dynamic differential privacy mechanism to generate an anomaly detection model; and a knowledge graph module for constructing a dynamic service dependency graph based on logs, performance indicators, and user behavior data of the multiple business systems. In use, this invention combines a dynamic differential privacy mechanism with local training, achieving the goal of making user data available but invisible, meeting legal compliance requirements such as GDPR and data export restrictions. By automatically adjusting noise injection through a combination of information gain and differential privacy budget, the system improves model aggregation accuracy while avoiding privacy leaks. High-contribution nodes receive less noise injection, helping their high-quality local models to play a greater role in the global model. An automatic pause mechanism with an upper limit is set to prevent sensitive systems from exceeding acceptable privacy leakage limits due to excessive participation.

Description

Technical Field

[0001] This invention relates to the field of data monitoring technology, and more specifically to a business operation monitoring system based on AI data sharing. Background Technology

[0002] AI data sharing and business operation monitoring refers to the real-time or near-real-time collection, analysis, and early warning of multi-dimensional data such as core business processes, system services, user behavior, and resource usage through technical means, while ensuring data security, privacy protection, and compliance. It also enables the exchange and collaborative use of data resources required for AI training and operation across organizations, platforms, or systems, in order to ensure stable business operation, rapid response to anomalies, optimization of AI model performance, and comprehensive data support for intelligent operational decision-making.

[0003] Patent publication number CN113608969A describes in its specification that "This invention provides a business system operation monitoring system and method, applicable to the field of distributed technology. The system includes: a configuration module for receiving configuration information of the business system, wherein the configuration information is input by the user based on navigation steps, and the configuration information includes log collection configuration information and log processing configuration information; a probe installation module for installing a log collection probe on the business system based on the log collection configuration information, wherein the log collection probe is used to collect business system logs in real time after the business system starts and send them to Kafka; and a log processing module for generating log processing tasks based on log processing configuration information, wherein the log processing tasks are used to implement..." The system retrieves and processes business system logs from Kafka. This invention enables access to different types of business systems for real-time monitoring, offering simple operation and high timeliness. While the aforementioned technology replaces traditional code development with visual navigation configuration, combining real-time stream processing with batch storage to solve the adaptation problem of heterogeneous systems while maintaining timeliness and data accumulation advantages, large companies often have different business operations handled by different departments. The raw data collected by these different systems is subject to format and GDPR-like privacy regulations, preventing data sharing and creating data silos. Traditional differential privacy uses fixed noise injection, leading to a decrease in the utility of high-value data and insufficient privacy protection.

[0004] In conclusion, developing a business operation monitoring system based on AI data sharing remains a key issue that urgently needs to be addressed in the field of data monitoring technology. Summary of the Invention

[0005] The purpose of this invention is to address the problem in existing technologies where different departments within a large company are responsible for different business operations, resulting in distinct business operation systems. The raw data collected by these different systems is subject to format restrictions and privacy regulations such as GDPR, making data sharing impossible and leading to data silos. Traditional differential privacy methods employ fixed noise injection, which results in a decrease in the utility of high-value data and insufficient privacy protection.

[0006] To achieve the above objectives, the present invention provides a business operation monitoring system based on AI data sharing, comprising: a federated learning module, used to aggregate model parameters of multiple business systems through a dynamic differential privacy mechanism to generate an anomaly detection model; The knowledge graph module constructs a dynamic service dependency graph based on logs, performance metrics, and user behavior data from multiple business systems, and locates root cause nodes through graph neural networks. The dynamic trade-off module dynamically adjusts the noise injection amount and the model aggregation weight allocation based on the contribution of one of the business systems to the training of the anomaly detection model. The self-healing strategy module works in conjunction with the knowledge graph module to trigger repair actions, including scaling up and down Kubernetes containers and rolling back service versions, based on the root cause node location results.

[0007] Furthermore, the operational process of the federated learning module includes: Each business system trains its local model using its private data. The dynamic differential privacy mechanism adds noise to the local model parameters before uploading them to protect user privacy, and then updates the local model parameters. of Norm clipping to the maximum threshold, where Indicates the first The difference in model parameter updates for each business system's local model during this iteration. Indicates the first The local model of the first business system Model parameters after training round Indicates the first The local model of the first business system After training the model parameters, Gaussian noise is then added to satisfy the desired parameters. Differential privacy, in which Indicates the first The level of information leakage allowed in each iteration round This represents the probability of differential privacy failure. The standard deviation of Gaussian noise is dynamically adjusted according to the model training epochs. Finally, the local model parameters uploaded to the business system are updated as follows: In the formula, Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. Indicates the first The updated model parameters are obtained from local training of the local model in each business system. This indicates that a random vector sampled from a normal distribution is used for differential privacy protection. Indicates the first Gaussian noise variance used in the next iteration Indicates the use of generating Dimensionally independent and identically distributed Gaussian noise.

[0008] Furthermore, the operational process of the federated learning module includes: The generated anomaly detection model aggregates parameter updates uploaded from local models of multiple business systems and employs a privacy-enhanced federated averaging algorithm, expressed as: In the formula, Indicates the global model at the 1st Parameters during round iteration, Indicates the first The global model parameters are updated after each iteration. Indicates the number of local models in the business system. Indicates the first Weight values ​​of the local model of each business system. This represents the sum of the weights of the local models of all business systems. Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. This represents the weighted average of all client updates, the aggregated global model. As an anomaly detection model.

[0009] Furthermore, the operation process of the knowledge graph module includes: The service name is extracted from the logs, the host IP is extracted from the performance metrics, and the API interface is extracted from the user behavior data. The dynamic service dependency graph is constructed as follows: In the formula Represents a dynamic service dependency graph. Denotes the set of edges. Represents a set of nodes. This represents the set of service names extracted from the logs. This represents the set of host IPs extracted from performance metrics. This represents the set of API interfaces extracted from user behavior data. The edge weights of the dynamic service dependency graph are updated every 0.5 seconds based on the real-time call chain. The expression is: In the formula, Indicates time Time Node To the node The weight of the dependent edge, This indicates the degree of influence of the call frequency metric on the total edge weight. Indicates time Time Node Call Node frequency, The weights representing response latency Represents a node Call Node In time The reciprocal of the response delay. Weights representing error rates Indicates time Time node Call Node Error rate.

[0010] Furthermore, the operation process of the knowledge graph module includes: The dynamic service relies on the node representation updates of the graph, and a multi-layer graph neural network is used to embed the nodes, expressed as: In the formula, Indicates the first The node representation matrix of the layer, Indicates the first The node representation matrix of the layer, Represents a non-linear activation function. This indicates that a self-loop has been added. yes The degree matrix, It is a symmetric normalized adjacency matrix. Indicates the first The trainable weight matrix of the layer.

[0011] Furthermore, the operation process of the knowledge graph module includes: The method for locating root cause nodes using a graph neural network employs an improved Personalized PageRank algorithm. The jump probability is calculated based on the fault history records of each node, expressed as: In the formula, Represents a node The probability of the final ranking. Represents a node Personalized redirect probability, This represents the probability that the random walk will continue to propagate in the graph. This represents all possible predecessor nodes. Summation, Indicates from node Jump to node The transition probability, Represents a node The probability of ranking Represents a node Personalized restart distribution.

[0012] Furthermore, the operational process of the dynamic trade-off module includes: No. The first round of federal learning The local model parameters uploaded by each business system are The set differential privacy budget cap is The contribution assessment, the first The information gain contribution of each business system is measured by the Kullback-Leibler divergence, which is used to measure the change in the model distribution. The expression is: In the formula, Indicates the first Wheel of Life Contribution of local model differences to each business system Indicates the Kullback-Leibler divergence. Indicates the first Parameter distribution of the wheel anomaly detection model. Indicates the first The business system in the first The parameter distribution of the local model trained in round 1. A vector representing the model parameters. Indicates the parameters of the anomaly detection model and local model parameters The logarithm of the probability ratios above, This represents the local model parameter values ​​for all possible business systems. Integrating, on the... Each business system has There are samples, and the average sample entropy is The information weight of the i-th business system is expressed as: In the formula, Indicates the first Wheel of Life Normalized contribution weights of each participant Indicates the first The number of samples used by each business system in this round of training Indicates the first Information entropy of the local model of each business system in this round, all The sum of the product of the data volume and information entropy of each business system is used for normalization.

[0013] Furthermore, the operational process of the dynamic trade-off module includes: The dynamic adjustment of noise injection amount employs a Gaussian mechanism to add differential privacy noise to the anomaly detection model, where the noise variance is negatively correlated with the contribution. The model's aggregate weights are dynamically allocated, jointly adjusted based on the contribution and noise perception factor. The data is recorded after each update. The cumulative privacy loss of a business system, expressed as: In the formula, Indicates the first The business system in the first The differential privacy budget accumulated after rounds of iterations, Indicates the first The business system in the first Privacy budget after multiple iterations Indicates L2 sensitivity. This indicates the number of operations performed in this iteration. Indicates the first The business system in the first The Gaussian noise variance used in the wheel This represents the increase in privacy loss caused by a single-round noise mechanism. Then their participation in training will be suspended.

[0014] Furthermore, the operation process of the self-healing strategy module includes: The set of root cause nodes output by the graph neural network is: In the formula, Represents the set of root cause nodes. It is a set that contains multiple nodes. Represents the first element in the set. 1 node For a dynamic service dependency graph node set, extract the set of historical associated repair strategies for each root cause node, and jointly represent all candidate strategies as follows: In the formula, This represents the set of candidate repair strategies for all root cause nodes. This represents the set of all nodes belonging to the root cause. nodes , Represent each node The corresponding set of strategies, Indicates the first One repair strategy, This indicates the service node targeted by the repair strategy. Indicates the policy in the log record. The target node is equal to the node .

[0015] Furthermore, the operation process of the self-healing strategy module includes: The priority of the repair actions is ordered according to the following rules: for failures affecting more than 50% of nodes, Kubernetes container scaling is performed first; when the latest business system version is abnormal, the strategy of historical business system version repair success rate >90% is implemented to trigger version rollback; when automatic repair fails for 3 consecutive times, cross-cluster traffic switching is triggered and an anomaly report is generated to notify the operation and maintenance personnel via Webhook / API. Beneficial effects

[0016] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects: When in use, this invention combines a dynamic differential privacy mechanism with local training, achieving the goal of making user data available but not visible, thus meeting legal compliance requirements such as GDPR and data export restrictions. By combining information gain and differential privacy budget to automatically adjust noise injection, the model aggregation accuracy is improved while avoiding privacy leakage. High-contribution nodes are injected with less noise, which helps their high-quality local models to play a greater role in the global model. An automatic pause mechanism with an upper limit is set to prevent sensitive systems from exceeding the acceptable range of privacy leakage due to excessive participation. Attached Figure Description

[0017] Figure 1 This is a system diagram of a business operation monitoring system based on AI data sharing according to the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] The present invention will now be described in further detail with reference to the accompanying drawings: Example: Figure 1 As shown, the present invention provides a business operation monitoring system based on AI data sharing, including: a federated learning module, used to aggregate model parameters of multiple business systems through a dynamic differential privacy mechanism to generate an anomaly detection model;

[0021] Furthermore, the operational process of the federated learning module includes: Each business system trains its local model using its private data. The dynamic differential privacy mechanism adds noise to the local model parameters before uploading them to protect user privacy, and then updates the local model parameters. of Norm clipping to the maximum threshold, where Indicates the first The difference in model parameter updates for each business system's local model during this iteration. Indicates the first The local model of the first business system Model parameters after training round Indicates the first The local model of the first business system After training the model parameters, Gaussian noise is then added to satisfy the desired parameters. Differential privacy, in which Indicates the first The level of information leakage allowed in each iteration round This represents the probability of differential privacy failure. The standard deviation of Gaussian noise is dynamically adjusted according to the model training epochs. Finally, the local model parameters uploaded to the business system are updated as follows: In the formula, Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. Indicates the first The updated model parameters are obtained from local training of the local model in each business system. This indicates that a random vector sampled from a normal distribution is used for differential privacy protection. Indicates the first Gaussian noise variance used in the next iteration Indicates the use of generating Dimensionally independent and identically distributed Gaussian noise.

[0022] Furthermore, the operational process of the federated learning module includes: The generated anomaly detection model aggregates parameter updates uploaded from local models of multiple business systems and employs a privacy-enhanced federated averaging algorithm, expressed as: In the formula, Indicates the global model at the 1st Parameters during round iteration, Indicates the first The global model parameters are updated after each iteration. Indicates the number of local models in the business system. Indicates the first Weight values ​​of the local model of each business system. This represents the sum of the weights of the local models of all business systems. Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. This represents the weighted average of all client updates, the aggregated global model. As an anomaly detection model; Specifically, this invention is used to collaboratively train anomaly detection models across multiple business systems under privacy protection. Based on a dynamic differential privacy mechanism, it improves the overall performance of the model while ensuring that data does not leave the local environment. It is particularly suitable for business systems involving sensitive data, such as those in finance, healthcare, and government. By combining the dynamic differential privacy mechanism with local training, the system achieves the goal of making user data available but not visible, thus meeting legal compliance requirements such as GDPR and data export restrictions.

[0023] The knowledge graph module constructs a dynamic service dependency graph based on logs, performance metrics, and user behavior data from multiple business systems, and locates root cause nodes through graph neural networks. Furthermore, the operation process of the knowledge graph module includes: The service name is extracted from the logs, the host IP is extracted from the performance metrics, and the API interface is extracted from the user behavior data. The dynamic service dependency graph is constructed as follows: In the formula Represents a dynamic service dependency graph. Denotes the set of edges. Represents a set of nodes. This represents the set of service names extracted from the logs. This represents the set of host IPs extracted from performance metrics. This represents the set of API interfaces extracted from user behavior data. The edge weights of the dynamic service dependency graph are updated every 0.5 seconds based on the real-time call chain. The expression is: In the formula, Indicates time Time Node To the node The weight of the dependent edge, This indicates the degree of influence of the call frequency metric on the total edge weight. Indicates time Time Node Call Node frequency, The weights representing response latency Represents a node Call Node In time The reciprocal of the response delay. Weights representing error rates Indicates time Time node Call Node Error rate.

[0024] Furthermore, the operation process of the knowledge graph module includes: The dynamic service relies on the node representation updates of the graph, and a multi-layer graph neural network is used to embed the nodes, expressed as: In the formula, Indicates the first The node representation matrix of the layer, Indicates the first The node representation matrix of the layer, Represents a non-linear activation function. This indicates that a self-loop has been added. yes The degree matrix, It is a symmetric normalized adjacency matrix. Indicates the first The trainable weight matrix of the layer.

[0025] Furthermore, the operation process of the knowledge graph module includes: The method for locating root cause nodes using a graph neural network employs an improved Personalized PageRank algorithm. The jump probability is calculated based on the fault history records of each node, expressed as: In the formula, Represents a node The probability of the final ranking. Represents a node Personalized redirect probability, This represents the probability that the random walk will continue to propagate in the graph. This represents all possible predecessor nodes. Summation, Indicates from node Jump to node The transition probability, Represents a node The probability of ranking Represents a node Personalized restart distribution; Specifically, this invention is used to achieve root cause analysis and node localization of faults in a multi-service system environment. The system automatically extracts key meta-information from logs, performance indicators, and user behavior data, and constructs a dynamically updated service dependency graph. It uses graph neural networks for node representation learning and root cause localization analysis to achieve a structured expression of complex system states, improves the edge weight adjustment mechanism at the 0.5-second level, is suitable for microservice architectures with high-frequency interactions, enhances system responsiveness, and achieves more accurate root cause localization by combining the multi-layer perception capability of graph neural networks with a personalized PageRank algorithm.

[0026] The dynamic trade-off module dynamically adjusts the noise injection amount and the model aggregation weight allocation based on the contribution of one of the business systems to the training of the anomaly detection model. Furthermore, the operational process of the dynamic trade-off module includes: No. The first round of federal learning The local model parameters uploaded by each business system are The set differential privacy budget cap is The contribution assessment, the first The information gain contribution of each business system is measured by the Kullback-Leibler divergence, which is used to measure the change in the model distribution. The expression is: In the formula, Indicates the first Wheel of Life Contribution of local model differences to each business system Indicates the Kullback-Leibler divergence. Indicates the first Parameter distribution of the wheel anomaly detection model. Indicates the first The business system in the first The parameter distribution of the local model trained in round 1. A vector representing the model parameters. Indicates the parameters of the anomaly detection model and local model parameters The logarithm of the probability ratios above, This represents the local model parameter values ​​for all possible business systems. Integrating, on the... Each business system has There are samples, and the average sample entropy is The information weight of the i-th business system is expressed as: In the formula, Indicates the first Wheel of Life Normalized contribution weights of each participant Indicates the first The number of samples used by each business system in this round of training Indicates the first Information entropy of the local model of each business system in this round, all The sum of the product of the data volume and information entropy of each business system is used for normalization.

[0027] Furthermore, the operational process of the dynamic trade-off module includes: The dynamic adjustment of noise injection amount employs a Gaussian mechanism to add differential privacy noise to the anomaly detection model, where the noise variance is negatively correlated with the contribution. The model's aggregate weights are dynamically allocated, jointly adjusted based on the contribution and noise perception factor. The data is recorded after each update. The cumulative privacy loss of a business system, expressed as: In the formula, Indicates the first The business system in the first The differential privacy budget accumulated after rounds of iterations, Indicates the first The business system in the first Privacy budget after multiple iterations Indicates L2 sensitivity. This indicates the number of operations performed in this iteration. Indicates the first The business system in the first The Gaussian noise variance used in the wheel This represents the increase in privacy loss caused by a single-round noise mechanism. Then their participation in training will be suspended; Specifically, this invention is used in federated learning to dynamically adjust the differential privacy noise injection amount and model aggregation weight allocation strategy based on the contribution of each business system to the training of the anomaly detection model, so as to improve the balance between the accuracy of the global model and privacy protection. By combining information gain and differential privacy budget to automatically adjust noise injection, the model aggregation accuracy is improved while avoiding privacy leakage. High-contribution nodes are injected with less noise, which helps their high-quality local models to play a greater role in the global model. An upper limit automatic pause mechanism is set to prevent sensitive systems from exceeding the acceptable privacy leakage range due to excessive participation.

[0028] The self-healing strategy module works in conjunction with the knowledge graph module to trigger repair actions, including scaling up and down Kubernetes containers and rolling back service versions, based on the root cause node location results. Furthermore, the operation process of the self-healing strategy module includes: The set of root cause nodes output by the graph neural network is: In the formula, Represents the set of root cause nodes. It is a set that contains multiple nodes. Represents the first element in the set. 1 node For a dynamic service dependency graph node set, extract the set of historical associated repair strategies for each root cause node, and jointly represent all candidate strategies as follows: , In the formula, This represents the set of candidate repair strategies for all root cause nodes. This represents the set of all nodes belonging to the root cause. nodes , Represent each node The corresponding set of strategies, Indicates the first One repair strategy, This indicates the service node targeted by the repair strategy. Indicates the policy in the log record. The target node is equal to the node .

[0029] Furthermore, the operation process of the self-healing strategy module includes: The priority of the repair actions is ordered according to the following rules: for failures affecting more than 50% of nodes, Kubernetes container scaling is prioritized; when the latest business system version is abnormal, the strategy of historical business system version repair success rate >90% is implemented to trigger version rollback; when automatic repair fails for 3 consecutive times, cross-cluster traffic switching is triggered and an anomaly report is generated to notify the operation and maintenance personnel via Webhook / API. Specifically, this invention can deeply integrate with the knowledge graph module to achieve automated repair based on root cause localization results. By accurately locating root cause nodes in the service dependency graph through graph neural networks, the self-healing strategy module can trigger the most suitable repair strategy in a targeted manner, which helps to shorten the average repair time. By using indicators such as historical repair success rate and impact range to build a strategy priority ranking mechanism, it ensures that the strategy execution has a high success rate and business continuity guarantee. When automatic repair fails, it can seamlessly trigger cross-cluster redundant resource switching to minimize the risk of single point of failure.

[0030] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A business operation monitoring system based on AI data sharing, characterized in that, include: The federated learning module is used to aggregate model parameters from multiple business systems through a dynamic differential privacy mechanism to generate an anomaly detection model. The knowledge graph module constructs a dynamic service dependency graph based on logs, performance metrics, and user behavior data from multiple business systems, and locates root cause nodes through graph neural networks. The dynamic trade-off module dynamically adjusts the noise injection amount and the model aggregation weight allocation based on the contribution of one of the business systems to the training of the anomaly detection model. The self-healing strategy module works in conjunction with the knowledge graph module to trigger repair actions, including scaling up and down Kubernetes containers and rolling back service versions, based on the root cause node location results.

2. The business operation monitoring system based on AI data sharing according to claim 1, characterized in that, The operation process of the federated learning module includes: Each business system trains its local model using its private data. The dynamic differential privacy mechanism adds noise to the local model parameters before uploading them to protect user privacy, and then updates the local model parameters. of Norm clipping to the maximum threshold, where Indicates the first The difference in model parameter updates for each business system's local model during this iteration. Indicates the first The local model of the first business system Model parameters after training round Indicates the first The local model of the first business system After training the model parameters, Gaussian noise is then added to satisfy the desired parameters. Differential privacy, in which Indicates the first The level of information leakage allowed in each iteration round This represents the probability of differential privacy failure. The standard deviation of Gaussian noise is dynamically adjusted according to the model training epochs. Finally, the local model parameters uploaded to the business system are updated as follows: In the formula, Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. Indicates the first The updated model parameters are obtained from local training of the local model in each business system. This indicates that a random vector sampled from a normal distribution is used for differential privacy protection. Indicates the first Gaussian noise variance used in the next iteration Indicates the use of generating Dimensionally independent and identically distributed Gaussian noise.

3. The business operation monitoring system based on AI data sharing according to claim 2, characterized in that, The operation process of the federated learning module includes: The generated anomaly detection model aggregates parameter updates uploaded from local models of multiple business systems and employs a privacy-enhanced federated averaging algorithm, expressed as: In the formula, Indicates the global model at the 1st Parameters during round iteration, Indicates the first The global model parameters are updated after each iteration. Indicates the number of local models in the business system. Indicates the first Weight values ​​of the local model of each business system. This represents the sum of the weights of the local models of all business systems. Indicates the first The updated model parameters of the local model in each business system after adding differential privacy noise. This represents the weighted average of all client updates, the aggregated global model. As an anomaly detection model.

4. The business operation monitoring system based on AI data sharing according to claim 3, characterized in that, The operation process of the knowledge graph module includes: The service name is extracted from the logs, the host IP is extracted from the performance metrics, and the API interface is extracted from the user behavior data. The dynamic service dependency graph is constructed as follows: In the formula Represents a dynamic service dependency graph. Denotes the set of edges. Represents a set of nodes. This represents the set of service names extracted from the logs. This represents the set of host IPs extracted from performance metrics. This represents the set of API interfaces extracted from user behavior data. The edge weights of the dynamic service dependency graph are updated every 0.5 seconds based on the real-time call chain. The expression is: In the formula, Indicates time Time Node To the node The weight of the dependent edge, This indicates the degree of influence of the call frequency metric on the total edge weight. Indicates time Time Node Call Node frequency, The weights representing response latency Represents a node Call Node In time The reciprocal of the response delay. Weights representing error rates Indicates time Time node Call Node Error rate.

5. A business operation monitoring system based on AI data sharing according to claim 4, characterized in that, The operation process of the knowledge graph module includes: The dynamic service relies on the node representation updates of the graph, and a multi-layer graph neural network is used to embed the nodes, expressed as: In the formula, Indicates the first The node representation matrix of the layer, Indicates the first The node representation matrix of the layer, Represents a non-linear activation function. This indicates that a self-loop has been added. yes The degree matrix, It is a symmetric normalized adjacency matrix. Indicates the first The trainable weight matrix of the layer.

6. A business operation monitoring system based on AI data sharing according to claim 5, characterized in that, The operation process of the knowledge graph module includes: The method for locating root cause nodes using a graph neural network employs an improved Personalized PageRank algorithm. The jump probability is calculated based on the fault history records of each node, expressed as: In the formula, Represents a node The probability of the final ranking. Represents a node Personalized redirect probability, This represents the probability that the random walk will continue to propagate in the graph. This represents all possible predecessor nodes. Summation, Indicates from node Jump to node The transition probability, Represents a node The probability of ranking Represents a node Personalized restart distribution.

7. A business operation monitoring system based on AI data sharing according to claim 6, characterized in that, The operation process of the dynamic trade-off module includes: No. The first round of federal learning The local model parameters uploaded by each business system are The set differential privacy budget cap is The contribution assessment, the first The information gain contribution of each business system is measured by the Kullback-Leibler divergence, which is used to measure the change in the model distribution. The expression is: In the formula, Indicates the first Wheel of Life Contribution of local model differences to each business system Indicates the Kullback-Leibler divergence. Indicates the first Parameter distribution of the wheel anomaly detection model. Indicates the first The business system in the first The parameter distribution of the local model trained in round 1. A vector representing the model parameters. Indicates the parameters of the anomaly detection model and local model parameters The logarithm of the probability ratios above, This represents the local model parameter values ​​for all possible business systems. Integrating, on the... Each business system has There are samples, and the average sample entropy is The information weight of the i-th business system is expressed as: In the formula, Indicates the first Wheel of Life Normalized contribution weights of each participant Indicates the first The number of samples used by each business system in this round of training Indicates the first Information entropy of the local model of each business system in this round, all The sum of the product of the data volume and information entropy of each business system is used for normalization.

8. A business operation monitoring system based on AI data sharing according to claim 7, characterized in that, The operation process of the dynamic trade-off module includes: The dynamic adjustment of noise injection amount employs a Gaussian mechanism to add differential privacy noise to the anomaly detection model, where the noise variance is negatively correlated with the contribution. The model's aggregate weights are dynamically allocated, jointly adjusted based on the contribution and noise perception factor. The data is recorded after each update. The cumulative privacy loss of a business system, expressed as: In the formula, Indicates the first The business system in the first The differential privacy budget accumulated after rounds of iterations, Indicates the first The business system in the first Privacy budget after multiple iterations Indicates L2 sensitivity. This indicates the number of operations performed in this iteration. Indicates the first The business system in the first The Gaussian noise variance used in the wheel This represents the increase in privacy loss caused by a single-round noise mechanism. Then their participation in training will be suspended.

9. A business operation monitoring system based on AI data sharing according to claim 8, characterized in that, The operation process of the self-healing strategy module includes: The set of root cause nodes output by the graph neural network is: In the formula, Represents the set of root cause nodes. It is a set that contains multiple nodes. Represents the first element in the set. 1 node For a dynamic service dependency graph node set, extract the set of historical associated repair strategies for each root cause node, and jointly represent all candidate strategies as follows: In the formula, This represents the set of candidate repair strategies for all root cause nodes. This represents the set of all nodes belonging to the root cause. nodes , Represent each node The corresponding set of strategies, Indicates the first One repair strategy, This indicates the service node targeted by the repair strategy. Indicates the policy in the log record. The target node is equal to the node .

10. A business operation monitoring system based on AI data sharing according to claim 9, characterized in that, The operation process of the self-healing strategy module includes: The priority of the repair actions is ordered according to the following rules: for failures affecting more than 50% of nodes, Kubernetes container scaling is performed first; when the latest business system version is abnormal, the strategy of historical business system version repair success rate >90% is implemented to trigger version rollback; when automatic repair fails for 3 consecutive times, cross-cluster traffic switching is triggered and an anomaly report is generated to notify the operation and maintenance personnel via Webhook / API.

Images