Project authority management method, equipment and program product

Abstract

The invention provides a project permission management method and device and a program product. The method comprises the steps of obtaining user behavior characteristics of a target user, task characteristics of a target task and permission state characteristics related to the target user and the target task; and according to the user behavior characteristics, the task characteristics of the target task, and the permission state characteristics related to the target user and the target task, using a permission configuration model to output the optimal permission configuration of the target user in the target task and the confidence of the optimal permission configuration, the permission configuration comprises configuration of read-write permission, data access permission and / or model operation permission; the permission configuration model is trained by using a preset machine learning algorithm in advance. By means of the technical scheme, dynamic adjustment of authority management is achieved based on machine learning, and the efficiency and flexibility of project authority management are improved.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a method, device and permission management for project permissions. Background Art

[0002] Permission management refers to the fact that users can access and can only access the resources they are authorized to access according to the security rules or security policies set by the system. The current project permission management scheme generally relies on manual participation, and each business party controls it through permission approval. After the permissions are determined, they are relatively fixed, inconvenient to adjust, and have low management efficiency and are inflexible. Summary of the invention

[0003] The embodiments of the present invention provide a method, apparatus and device for project authority management, which realize dynamic adjustment of authority management based on machine learning and improve the efficiency and flexibility of project authority management.

[0004] In order to achieve the above object, on the one hand, a method for project authority management is provided, comprising:

[0005] Acquire user behavior characteristics of a target user, task characteristics of a target task, and permission status characteristics related to the target user and the target task, wherein the user behavior characteristics are used to characterize the target user's operation behavior, the frequency of access to project data, and / or the historical operation behavior trajectory; the task characteristics are used to characterize the context information of the task, and the context information of the task includes: the task type, the resources required for the task, the time limit of the task, the stage of the task, and / or the identity of the target user; the permission status characteristics related to the target user and the target task are used to characterize the current permission of the target user, the target user's use history of the current permission, and / or the permission allocation mode of similar tasks that meet a predetermined similarity condition with the target task;

[0006] Based on the user behavior characteristics, the task characteristics of the target task, and the permission status characteristics related to the target user and the target task, a permission configuration model is used to output the optimal permission configuration of the target user in the target task and the confidence of the optimal permission configuration, wherein the permission configuration includes the configuration of read and write permissions, data access permissions and / or model operation permissions; the permission configuration model is pre-trained using a preset machine learning algorithm.

[0007] Preferably, in the method, the permission configuration model is a Bayesian network model, and the method further comprises:

[0008] The permission configuration model dynamically adjusts the optimal permission of the target user in the target task according to the user behavior characteristics, the task characteristics of the target task and / or the permission status characteristics related to the target user and the target task;

[0009] The dynamically adjusted frequency and amplitude meet a preset frequency threshold and amplitude threshold.

[0010] Preferably, in the method, the user behavior characteristics further include: the user's operation log, permission change history, operation corresponding timestamp, current operation device information and / or current geographical location; the device information includes: the IP address of the device;

[0011] The method further comprises:

[0012] Embedded vectorization is performed on the user behavior feature to obtain an embedded vector for characterizing the user behavior;

[0013] The embedded vector is input into a pre-trained LSTM network to detect whether there is currently an abnormal permission operation, and in the case of an abnormal permission operation, a permission self-healing strategy is enabled through a pre-trained self-healing model, wherein the permission self-healing strategy includes: generating permission adjustment suggestions for the current user or directly adjusting the permissions of the current user; the permission adjustment includes: restricting specific operations of the current user, reducing the permission level of the current user and / or reminding the administrator to review the permissions of the current user; the machine learning algorithm used by the self-healing model includes: a reinforcement learning algorithm.

[0014] Preferably, the method, wherein

[0015] The abnormal authority operation includes: abnormal authority escalation, abnormal behavior patterns and / or abnormal high-frequency operations; and / or,

[0016] The permission self-healing strategy also includes: after adjusting the permissions of the current user, using a predetermined additional identity authentication method to perform additional identity authentication on the current user; if the verification passes, it is confirmed that the current user is a legitimate user and the operation is invalid, and the permissions of the current user are restored; otherwise, the permission adjustment of the current user is maintained.

[0017] Preferably, the method further comprises the step of implementing cross-project permission inheritance and migration from the source project to the current project for the target user; wherein implementing cross-project permission inheritance and migration comprises:

[0018] Obtain the target user's permission characteristics in the source project, the project characteristics of the current project, and the target user's operation behavior characteristics in the source project; the permission characteristics are used to characterize the target user's: permission allocation, permission operation history and / or identity information in the source project; the project characteristics are used to characterize the current project's: name, scale, project type and / or role distribution of participants; the target user's operation behavior characteristics in the source project are used to characterize the target user's frequency of use of permissions of different permissions in the source project and / or the risk level of the operation;

[0019] According to the obtained permission characteristics of the target user in the source project, the project characteristics of the current project, and the operation behavior characteristics of the target user in the source project, a pre-trained permission inheritance and migration model is used to output the permissions of the target user in the current project; wherein the permission inheritance and migration model includes: a dual-network migration structure and a dual-network migration structure.

[0020] Preferably, the method of outputting the target user's permissions in the current project according to the permission inheritance and migration model further comprises:

[0021] An anomaly detection model combining a graph network and LSTM is used to detect permission anomalies of the target user in the current project, and when permission anomalies exist, a reinforcement learning mechanism is used to perform permission self-healing operations.

[0022] Preferably, in the method, the user behavior characteristics further include: the user's historical task records and the user's mode of handling specific tasks; the method further includes one or more of the following:

[0023] When the target user has permissions for multiple tasks, the permissions for each of the multiple tasks are allocated using a K-Means clustering algorithm according to the user behavior characteristics;

[0024] When the target user has authority for multiple tasks, and when there is a task among the multiple tasks that meets a predetermined condition, selecting an agent for the task that meets the predetermined condition; the predetermined condition includes: the target user cannot complete the task;

[0025] A deep neural network is used to learn the user's mode of processing specific tasks, the specific tasks are set as tasks that can be processed automatically, and after the specific tasks occur, the task processing results are automatically generated; the specific tasks include: approval tasks.

[0026] Preferably, in the method, the user behavior characteristics further include: the user's historical permission usage records; the task characteristics further include: task requirements; the method further includes:

[0027] Using multi-agent reinforcement learning to optimize authority allocation according to the user behavior characteristics of the target user, the task characteristics of the target task, and the authority status characteristics related to the target user and the target task, the authority allocation optimization includes adjusting the authority of the target user to the target task; and

[0028] After the optimization of authority allocation is achieved, the behavior of the target user is monitored to obtain the behavior feedback of the target user, and the authority allocation is further optimized based on the obtained feedback, wherein the behavior feedback includes: positive feedback and negative feedback.

[0029] On the other hand, an electronic device is provided, including a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement any of the methods described above.

[0030] On the other hand, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the methods described above are implemented.

[0031] The above technical solution has the following technical effects:

[0032] The technical solution of the embodiment of the present invention can dynamically adjust the permissions of the target user by acquiring the user behavior characteristics of the target user, the task characteristics of the target task, and the permission status characteristics related to the target user and the target task, using a permission configuration model based on a preset machine learning algorithm, thereby improving the efficiency and flexibility of project permission management;

[0033] Furthermore, the technical solution of the embodiment of the present invention is applied to the data center of the steel plant, which can continuously adjust the permissions according to the actual needs of the project and user behavior, ensuring that each participant can obtain accurate and efficient permission control;

[0034] Furthermore, based on user behavior characteristics, task characteristics and related permission status characteristics, combined with the use of a variety of different machine learning algorithms, it is possible to achieve anomaly detection and permission self-healing of permission management, cross-project permission inheritance and migration, intelligent permission allocation and agency for user multi-tasks, and optimization of permission allocation, further improving the efficiency, flexibility, convenience and security of project permission management. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart of a method for project authority management according to an embodiment of the present invention;

[0036] Figure 2 A flowchart of a method for project authority management according to another embodiment of the present invention is shown. DETAILED DESCRIPTION

[0037] To further illustrate the various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, a person of ordinary skill in the art should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0038] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.

[0039] Embodiment 1:

[0040] Figure 1 FIG. 1 is a flow chart of a method for project authority management according to an embodiment of the present invention. Figure 1 The project authority management method of this embodiment includes:

[0041] Obtain user behavior characteristics of the target user, task characteristics of the target task, and permission status characteristics related to the target user and the target task. The user behavior characteristics are used to characterize the target user's operation behavior, the frequency of access to project data, and / or the historical operation behavior trajectory; the task characteristics are used to characterize the context information of the task, and the context information of the task includes: the task type, the resources required for the task, the time limit of the task, the stage of the task, and / or the identity of the target user; the permission status characteristics related to the target user and the target task are used to characterize the current permission of the target user, the target user's use history of the current permission, and / or the permission allocation mode of similar tasks that meet the predetermined similarity conditions with the target task;

[0042] Based on user behavior characteristics, task characteristics of the target task, and permission status characteristics related to the target user and the target task, a permission configuration model is used to output the optimal permission configuration of the target user in the target task and the confidence of the optimal permission configuration, wherein the permission configuration includes configuration of read and write permissions, data access permissions and / or model operation permissions; the permission configuration model is pre-trained using a preset machine learning algorithm.

[0043] In other embodiments of the present invention, based on user behavior characteristics, task characteristics and related permission status characteristics, a variety of different machine learning algorithms are used in combination to achieve anomaly detection and permission self-healing of permission management, cross-project permission inheritance and migration, intelligent permission allocation and agency for user multi-tasking, and optimization of permission allocation.

[0044] Preferably, the permission configuration model is a Bayesian network model, and the method also includes: the permission configuration model dynamically adjusts the optimal permissions of the target user in the target task based on user behavior characteristics, task characteristics of the target task and / or permission status characteristics related to the target user and the target task; wherein the frequency and amplitude of the dynamic adjustment meet pre-set frequency thresholds and amplitude thresholds.

[0045] Preferably, the user behavior characteristics also include: the user's operation log, permission change history, timestamp corresponding to the operation, device information of the current operation and / or the current geographical location; the device information includes: the IP address of the device; the method also includes: embedding vector representation of the user behavior characteristics to obtain an embedded vector for characterizing the user behavior; inputting the embedded vector into a pre-trained LSTM network to detect whether there is currently any abnormal permission operation, and in the case of abnormal permission operation, enabling a permission self-healing strategy through a pre-trained self-healing model, the permission self-healing strategy includes: generating permission adjustment suggestions for the current user or directly adjusting the permissions of the current user; permission adjustment includes: restricting specific operations of the current user, reducing the permission level of the current user and / or reminding the administrator to review the permissions of the current user; the machine learning algorithm used by the self-healing model includes: reinforcement learning algorithm.

[0046] Preferably, abnormal authority operations include: abnormal authority escalation, abnormal behavior patterns and / or abnormal high-frequency operations; and / or,

[0047] The permission self-healing strategy also includes: after adjusting the permissions of the current user, using a predetermined additional identity authentication method to perform additional identity authentication on the current user; if the verification passes, it is confirmed that the current user is a legitimate user and the operation is unnecessary, and the permissions of the current user are restored; otherwise, the permission adjustment of the current user is maintained.

[0048] Preferably, the method further includes the step of implementing cross-project permission inheritance and migration from the source project to the current project for the target user; wherein implementing cross-project permission inheritance and migration includes:

[0049] Obtain the target user's permission characteristics in the source project, the project characteristics of the current project, and the target user's operation behavior characteristics in the source project; permission characteristics are used to characterize the target user's: permission allocation, permission operation history and / or identity information in the source project; project characteristics are used to characterize the current project's: name, scale, project type and / or role distribution of participants; the target user's operation behavior characteristics in the source project are used to characterize the target user's frequency of use of permissions of different permissions in the source project and / or the risk level of the operation;

[0050] According to the obtained permission characteristics of the target user in the source project, the project characteristics of the current project, and the operation behavior characteristics of the target user in the source project, a pre-trained permission inheritance and migration model is used to output the permissions of the target user in the current project; wherein the permission inheritance and migration model includes: a dual-network migration structure and a dual-network migration structure.

[0051] Preferably, outputting the target user's permissions in the current project for the permission inheritance and migration model also includes:

[0052] The anomaly detection model combining graph network and LSTM is used to detect permission anomalies of the target user in the current project, and when permission anomalies occur, the reinforcement learning mechanism is used to perform permission self-healing operations.

[0053] Preferably, the user behavior characteristics further include: the user's historical task records and the user's mode of handling specific tasks; the method further includes one or more of the following:

[0054] When the target user has permissions for multiple tasks, the K-Means clustering algorithm is used to allocate permissions for each of the multiple tasks based on the user's behavior characteristics;

[0055] When the target user has authority for multiple tasks, and when there is a task that meets a predetermined condition among the multiple tasks, an agent is selected for the task that meets the predetermined condition; the predetermined condition includes: the target user cannot complete the task;

[0056] Use deep neural networks to learn the patterns of users in handling specific tasks, set specific tasks as tasks that can be automated, and automatically generate task processing results after specific tasks occur; specific tasks include: approval tasks.

[0057] Preferably, the user behavior characteristics also include: the user's historical permission usage records; the task characteristics also include: task requirements; the method also includes: based on the user behavior characteristics of the target user, the task characteristics of the target task, and the permission status characteristics related to the target user and the target task, using multi-agent reinforcement learning to achieve permission allocation optimization, the permission allocation optimization including adjusting the target user's permissions to the target task; and after achieving the permission allocation optimization, monitoring the target user's behavior, obtaining the target user's behavior feedback, and further optimizing the permission allocation based on the feedback obtained, the behavior feedback including: positive feedback and negative feedback.

[0058] Embodiment 2:

[0059] Figure 2 FIG. 1 is a flow chart of a method for project authority management according to another embodiment of the present invention. Figure 2 The project authority management method of this embodiment includes:

[0060] After the user logs in to the corresponding system, the user role, i.e., identity information, is obtained;

[0061] Predict the user's permission requirements for the current task; in specific implementation, this prediction is achieved by using a preset machine learning algorithm and a trained related model;

[0062] Dynamically allocate permissions based on predicted demand;

[0063] Detect whether there is abnormal operation;

[0064] If so, perform exception handling, adjust user permissions, record the situation in the exception log, and end the process;

[0065] Otherwise, continue the operation and monitor the user's behavior in real time. At the same time, detect whether the user's behavior is abnormal. If an abnormality occurs, trigger an alarm and adjust permissions, such as tightening permissions. If no abnormality occurs, operate normally, adjust the log according to the permission adjustment record, and further perform feedback optimization according to the preset machine learning algorithm to output the optimized permission adjustment strategy.

[0066] Embodiment three:

[0067] The project authority management method of the embodiment of the present invention is used to implement authority changes based on user behavior. The specific implementation details of this embodiment are described below.

[0068] Technical implementation: Use Bayesian networks to predict user task requirements and automatically adjust permissions based on current tasks and contexts. Traditional Bayesian network models usually predict user behavior or needs by observing historical behaviors and tasks, calculating conditional probabilities between different events, and thus giving the permissions required for the current task.

[0069] 1.1 Input:

[0070] User behavior data: including user operation records, project data access frequency, types of tasks performed, and historical behavior trajectories. Task context information: current task type, resources required for the task, time limit, user's department and role. Permission status data: current user's permission information, permission usage history, and permission allocation patterns for other similar tasks.

[0071] 1.2 Output:

[0072] The output result is the optimal permission configuration of the current user in the task (such as read and write permissions, data access permissions, model operation permissions, etc.), and the corresponding credibility score (that is, the confidence of the permission configuration) is output.

[0073] 1.3 Processing flow:

[0074] Data preprocessing: Collect and clean user behavior data and context data, unify the format and perform feature extraction. Convert user behavior patterns into probabilistic input variables that can be used in Bayesian networks.

[0075] Probabilistic reasoning: Use Bayesian networks to build a probabilistic dependency model between user behavior and permission requirements, and predict the user's permission requirements in the current task based on past data. The model dynamically adjusts user permissions by learning the permission usage patterns of different users in different contexts.

[0076] Real-time feedback and adjustment: When a user has insufficient or too many permissions, the system automatically triggers the adjustment mechanism and updates the permission configuration. The weights of the Bayesian network are adjusted by feedback on user operation behavior to ensure more accurate permission configuration for future tasks.

[0077] 1.4 Innovation and improvement:

[0078] Context-aware optimization: The traditional Bayesian network model has a relatively simple processing of context. The innovation lies in the introduction of multi-dimensional context-aware technology (such as real-time data flow, task urgency, team structure), which makes the permission adjustment more refined. For example, when a task is urgent, the system can temporarily grant users more permissions to improve efficiency.

[0079] Adaptive Bayesian Network: Based on the traditional Bayesian network, an adaptive mechanism is added. By introducing reinforcement learning algorithms (such as Q-learning), the network will self-adjust according to the actual operation results of the user, so that the permission configuration can better adapt to the changes in user behavior.

[0080] Dynamically adjust thresholds: Different threshold control mechanisms are set to address the dynamic nature of permission configuration. The system analyzes historical data to determine the optimal permission adjustment frequency and thresholds to avoid frequent permission changes interfering with user workflows.

[0081] 1.5 Application Examples

[0082] Example scenario: Changes in permissions for the design department

[0083] Assume that in the design department of a steel plant, designer A is responsible for the design tasks of a certain project. Designer A usually uses specific permissions to complete his work, but as the project progresses or the work content changes, the permissions he needs may need to be dynamically adjusted.

[0084] 1.5.1. Initial permission settings

[0085] User identity: Designer A.

[0086] Initial permissions: Designer A is granted viewing permissions (Read) to access the design model in the permission management system, but does not have modification or deletion permissions (Write / Delete) to ensure the security of project data.

[0087] 1.5.2. Collection and analysis of behavioral data

[0088] The system monitors designer A's operation behavior over a period of time and collects the following behavior data:

[0089] Operation behavior: A frequently checks design drawings and project documents but does not make any modifications.

[0090] Task mode: In past design tasks, Designer A tended to only review models and documents at the beginning of the task, and only began to modify and edit the design drawings when the task entered the specific implementation stage.

[0091] Bayesian network model construction:

[0092] Input nodes: The system will build a Bayesian network based on behavioral data. The input nodes include designer A’s operational behavior (view / modify / delete), task progress (early / mid-term / late), and project requirements (design / modify / proofread, etc.).

[0093] Conditional Probability Table (CPT): The system calculates the probability relationship between operation behavior and task progress based on historical data. For example:

[0094] At the beginning of the task, the probability that designer A views the model is 0.8, and the probability that he modifies the model is 0.2.

[0095] In the later stage of the task, the probability of designer A modifying the model increases to 0.9, and the probability of viewing the model is 0.1.

[0096] 1.5.3. Dynamic Permission Adjustment

[0097] Based on the output of the Bayesian network model, the system can predict the authority requirements of designer A and make dynamic adjustments based on the current task stage.

[0098] Example 1: Early stage of the mission

[0099] Task progress: The project has just started, and A is mainly checking the design requirements and drawings without making any modifications.

[0100] Prediction results: The Bayesian network predicts that designer A only needs to view permissions (Read) most of the time in the early stage of the task, and the probability of modification is very low.

[0101] Permission adjustment: The system automatically maintains A's viewing permission (Read) but does not grant modification permission (Write) to ensure the integrity of project data.

[0102] Example 2: Mid-mission

[0103] Task progress: The project entered the implementation phase, and Designer A began to deal with specific design modifications.

[0104] Prediction results: Bayesian network analysis shows that designer A has entered the middle stage of the task, and the probability of modifying the design drawings has increased significantly.

[0105] Permission adjustment: The system dynamically grants modification permissions (Write) to designer A based on the prediction results, allowing him to edit the model. At this time, the system will also grant certain temporary permissions based on the task requirements to ensure that the modification operation can be completed smoothly.

[0106] Example 3: Abnormal Operation Detection

[0107] Behavioral analysis: If designer A suddenly tries to delete something at the beginning of a project, the system will detect that this behavior does not match the normal working pattern.

[0108] Processing method: The Bayesian network determines through probability calculation that A should not have the deletion permission, so it triggers an abnormal operation alarm and rejects A's deletion operation. At this time, the system will also record the operation behavior and consider further tightening A's permissions.

[0109] 1.5.4. Task completion and permission recovery

[0110] When the design task is completed, the system will automatically revoke Designer A's modification permission (Write) based on the task completion status and operation behavior records, while retaining the viewing permission (Read) to prevent data loss due to misoperation.

[0111] 1.5.5. Permission optimization and feedback

[0112] The system will conduct feedback learning based on designer A's operation records and task completion status. If A makes efficient use of modification permissions in the middle of a task, the system will be more inclined to grant A modification permissions in advance in future tasks; conversely, if A frequently makes operational errors or abnormal operations, the system will reduce the scope of his permissions.

[0113] Example extension: cross-project permission management

[0114] Assume that designer A is involved in multiple projects, and different projects have different permission requirements for him. Through the Bayesian network, the system can automatically infer A's permission requirements in multiple projects, and achieve cross-project permission inheritance and dynamic adjustment. For example:

[0115] Project A: Designer A is mainly responsible for reviewing and proofreading, and the Bayesian network will automatically maintain low-level review permissions for him.

[0116] Project B: Designer A participates in the design modification, and the system will dynamically grant him higher permissions during the project implementation phase.

[0117] Through this dynamic permission management mechanism, the system can automatically optimize permission configuration without increasing the management burden, thereby improving security and efficiency.

[0118] Embodiment 4:

[0119] The project rights management method of the embodiment of the present invention is used to implement anomaly detection and rights self-healing mechanism in rights management. The specific implementation details of this embodiment are described below.

[0120] 2.1. Time Series Analysis Using Long Short-Term Memory Network (LSTM)

[0121] LSTM networks are essentially suitable for processing sequence data. They can analyze user operation behaviors, monitor the time series of permission changes and operation records, and detect abnormal operations. The following are the specific implementation steps and improvements:

[0122] Input data processing:

[0123] Input: Input data can include user operation logs, permission change history, timestamps, device information, geographic location and other related context information. Each piece of data represents a complete information sequence of a user operation.

[0124] Data preprocessing: These log data need to be standardized, cleaned (noise data removed), normalized, and batch processed in the time dimension.

[0125] Innovation: Introducing embedded coding layers to embed vector representation of discrete permission types, user identities, and device information, improving the feature capture capability of LSTM.

[0126] LSTM network structure:

[0127] Model design: Based on the traditional LSTM, a multi-layer bidirectional LSTM structure can be designed to further improve the ability to capture complex operation behavior patterns.

[0128] Innovation: The introduction of the attention mechanism allows the model to focus on the part of the operation behavior sequence that has a greater impact on abnormal judgment, thereby improving the accuracy of anomaly detection.

[0129] Model output:

[0130] Output: The output of LSTM can be used to predict whether the user behavior at a certain moment deviates from the normal operation trajectory (abnormal operation). The output at each moment is a probability value used to evaluate whether the moment is abnormal.

[0131] Innovation: The time series score output by LSTM is combined with the Hidden Markov Model (HMM) to analyze the persistence and suddenness of abnormal operations, and make more intelligent judgments on possible malicious use of permissions or misoperations.

[0132] 2.2 Self-healing mechanism for abnormal permission operations

[0133] Once an abnormal permission operation is detected, the permissions are dynamically repaired and optimized through machine learning algorithms:

[0134] Input data processing:

[0135] Input: In addition to the log data in the anomaly detection phase, the permission self-healing mechanism also introduces more contextual information, such as the user's historical operation mode, system workload, and permission usage of other similar users.

[0136] Data aggregation: Aggregate user behavior data and system health status and input them into the self-healing model.

[0137] Self-healing strategy generation:

[0138] Model design: Combine reinforcement learning (RL) to develop a self-healing strategy. The "state" of the RL model is the user's current permissions and operation context, the "action" is to adjust or not adjust the permissions, and the "reward" mechanism is defined by the normality and security of the system operation after self-healing.

[0139] Innovation: Combining RL with transfer learning, by introducing permission operation modes of similar user groups, it can quickly make decisions on permission adjustments for new users and avoid the "cold start" problem.

[0140] Permission adjustment output:

[0141] Output: Generate specific permission adjustment suggestions or directly implement permission adjustment actions, such as restricting user-specific operations, reminding administrators to review permissions, etc.

[0142] Innovation: By combining LSTM with RL to form a two-way feedback loop, user behavior after permission adjustment is monitored in real time to further optimize the permission management strategy.

[0143] 2.3 Comparison between traditional implementation and innovative improvement:

[0144] Traditional implementation: Usually a static permission management system based on rules can only respond to abnormal permission behaviors through predefined rules, and does not have the ability of self-learning and self-optimization.

[0145] Innovation and improvement: Through the combination of LSTM and RL, dynamic and intelligent permission management is achieved. LSTM can autonomously learn user behavior patterns and detect complex abnormal operations, while RL can perform permission self-healing after detecting anomalies, realizing intelligent adjustment of permission management, greatly improving the flexibility and security of the system.

[0146] Through these innovative technical means, not only can we achieve fine-grained management in complex permission scenarios, but also dynamic self-healing of abnormal operations, greatly improving the stability, security and management efficiency of the system.

[0147] Example:

[0148] Assume that on a construction management platform, construction manager B is responsible for managing construction progress and materials, and usually has high authority, including project modification and approval authority, etc. However, when B's operation behavior is abnormal, the system needs to detect it in time and take authority self-healing measures.

[0149] 1. Normal permission behavior mode

[0150] User identity: Construction Manager B.

[0151] Initial permissions: B has advanced permissions such as "view, modify, approve, and upload files" in the system.

[0152] Normal behavior: B usually checks the construction plan regularly, approves the progress, and makes adjustments based on the actual construction situation (modifying the schedule, uploading the construction report, etc.).

[0153] 2. Collection and processing of behavioral data

[0154] Over a period of time, the system collects the following time series data by monitoring B's operation behavior:

[0155] Review operation: B reviews the construction schedule at the beginning of each day's work.

[0156] Modification operations: B makes modifications based on the issues in the schedule, generally 2-3 times a day.

[0157] Approval operation: B regularly approves project progress applications once a week.

[0158] 3. Abnormal operation detection

[0159] The LSTM model establishes a behavior pattern by learning the normal operation behavior sequence of construction manager B. This pattern can predict B's normal operation range within a specific time. Once abnormal behavior occurs, LSTM will automatically detect it and trigger the permission self-healing mechanism.

[0160] Abnormal situation 1: abnormal privilege escalation request

[0161] Detection behavior: Construction manager B suddenly tries to elevate his privileges during non-working hours, attempting to obtain system administrator privileges to modify the core configuration of the platform.

[0162] Model detection: LSTM found through time series analysis that B had never performed similar operations before, and the request for privilege escalation occurred late at night, far exceeding the confidence interval of B's ​​normal operating behavior.

[0163] Abnormal marking: The LSTM model compares the behavior pattern and finds that the behavior is seriously inconsistent with B's normal pattern, and marks it as an abnormal operation.

[0164] Abnormal situation 2: abnormal high-frequency operation

[0165] Detection behavior: B frequently modified the construction schedule in a short period of time, performing dozens of modification operations per minute, which clearly exceeded the normal working frequency.

[0166] Model detection: The LSTM model detects that B's behavior is abnormally high-frequency, which is a serious deviation from his previous modification behavior of 2-3 times a day and is an abnormal operation behavior.

[0167] Abnormal marking: The system marks it as a high-risk operation through timing analysis.

[0168] 4. Permission self-healing mechanism

[0169] When the LSTM model detects an abnormal operation, the system will automatically trigger the permission self-healing mechanism.

[0170] Permission self-healing steps:

[0171] Permission downgrade: Once an abnormality is detected, the system will automatically and temporarily downgrade B's high-level permissions to low-level permissions (such as viewing permissions only) to prevent him from continuing to perform dangerous operations.

[0172] Log records: The system will record detailed information of the abnormal operation, including time, operation type, user identity, etc., for future auditing.

[0173] Multi-factor verification: To prevent misjudgment, the system will send a verification request to construction manager B, requiring B to perform additional identity verification, such as SMS verification code or fingerprint recognition, to confirm whether the operation is legal.

[0174] Permission recovery: If B is verified to be a legitimate user and the operation is not malicious, the system will automatically restore his original permissions; if the verification fails, the system will maintain the permission downgrade and the administrator will receive a security alert.

[0175] 5. Self-healing feedback and learning

[0176] The system not only handles single abnormal operations, but also feeds these behaviors into the LSTM model to enhance the model's learning ability and enable it to better predict similar abnormal behaviors.

[0177] Feedback input: Abnormal behaviors and self-healing results are input into LSTM to enhance its ability to detect potential abnormal operations.

[0178] Model optimization: As more abnormal behavior data is collected, the LSTM model is continuously optimized and can gradually learn more abnormal patterns, improving detection accuracy and sensitivity.

[0179] Detailed explanation of input and output

[0180] enter

[0181] Behavioral data: time series data, including the user's operation type (view, modify, delete, etc.), operation frequency, time node, task context (such as whether it is during working hours), permission type, etc.

[0182] Environmental context: The environment in which the user is located (such as project type, operating device, login IP address, etc.), which can help LSTM better understand the user's operating context.

[0183] Output

[0184] Behavior pattern: LSTM generates a behavior pattern based on the input time series data to predict the range of normal user operation behavior.

[0185] Abnormal marking: The LSTM model outputs the probability of abnormal operation. Once it exceeds the set threshold, it triggers the abnormal operation alarm and starts the permission self-healing mechanism.

[0186] Permission adjustment instructions: The system outputs instructions for adjusting user permissions based on the model, including steps for permission downgrade, restoration, and request verification.

[0187] Innovative improvements to the LSTM model

[0188] To improve innovation, the LSTM model can introduce the following improvements:

[0189] Multi-task learning: By learning the operation behavior patterns of multiple users at the same time, the LSTM model can better detect anomalies by leveraging the correlation between different users. For example, if multiple users in the same project have abnormal operations, the system can improve the overall anomaly detection accuracy through collaborative detection.

[0190] Online update mechanism: The LSTM model can adopt an online learning mechanism to automatically update its internal weights after each new abnormal behavior occurs, so that it can be continuously optimized in continuous actual operation, improving the real-time and accuracy of detection.

[0191] Abnormal behavior reasoning: By combining with Bayesian networks, LSTM can not only detect anomalies, but also infer the potential causes of anomalies. For example, a request for privilege escalation may be related to an external attack or account theft, and a smarter security strategy can be provided through the reasoning process.

[0192] Embodiment five:

[0193] The project rights management method of the embodiment of the present invention is used to implement cross-project rights inheritance and migration. The specific implementation details of this embodiment are described below.

[0194] Technical implementation: Using transfer learning, when a user obtains valid permission configuration in a project, the system can migrate this configuration to similar projects. In the cross-project permission inheritance and migration function, the core of using transfer learning is that the permission allocation mode learned from one project or task can be migrated to another similar project, thus avoiding the manual configuration process of permissions from scratch. Users no longer need to repeatedly configure permissions in each project, saving time and resources. At the same time, if the permission requirements change with the special circumstances of the new project, the system will automatically adjust the permissions without affecting the overall process.

[0195] 3.1. Input and processing of permission inheritance and migration

[0196] enter:

[0197] Permission data: includes the user's permission allocation in the source project, permission operation history, and user identity information (role, position, etc.).

[0198] Project context: Enter relevant context information of the current project and the target project, such as project scale, project type, role distribution of participants, etc.

[0199] Operation behavior: Enter the user's actual operations on different permissions in the source project, including the frequency of permission use, the risk level of the operation, etc.

[0200] Data processing:

[0201] Data preprocessing: User permissions, project context, and operation behavior data are standardized and unified to generate feature vectors. Permission behavior data can be modeled through time series to extract the permission usage pattern of each user.

[0202] Innovation: Graph Embedding is used to represent the relationship between permissions. By building a graph structure of permission operations, the frequency of user usage and dependency of permissions are analyzed, which helps to better migrate permissions in the target project.

[0203] 3.2. Improvements to the algorithm for cross-project permission migration

[0204] In traditional transfer learning, the trained model parameters are usually applied to similar target domains. However, cross-project permission inheritance involves migration based on historical permission operation behaviors and differences in project environments, so new methods need to be introduced to improve the accuracy of transfer learning.

[0205] Model design for permission migration:

[0206] Innovation 1: Dual network migration structure

[0207] Design a dual network structure: a source network is responsible for learning the permission allocation model of the original project, and the other target network is responsible for applying the learned model to the new project. The source network and the target network transfer knowledge by sharing or partially sharing weights.

[0208] Through the adaptive adjustment mechanism, the target network can dynamically adjust the permission mapping in the source project to adapt to the permission structure in the new project.

[0209] Innovation 2: Permission transfer based on conditional generative adversarial networks (Conditional GANs)

[0210] The conditional generative adversarial network is used to generate appropriate permission mappings in the new project. The GAN generator generates a new permission configuration based on the permissions of the source project and the context of the new project, and the discriminator is used to evaluate whether the generated permission configuration is reasonable.

[0211] Through adversarial training, the model can automatically adapt to the permission migration requirements between different projects and improve the effectiveness of permission inheritance.

[0212] Processing of input data:

[0213] The permission usage patterns of the source and target projects are used as input to the dual network.

[0214] The input of the conditional GAN ​​is set as the source project permissions, the target project context, and the similarity of permission usage between projects (calculated by graph embedding).

[0215] After the input data is normalized, it is initially modeled through the source network, and the target network uses the shared weights to partially migrate the authority structure.

[0216] Output:

[0217] Output 1: Output the initial permission settings of each user in the target project and generate permission mapping results based on the source project.

[0218] Output 2: Generate permission adjustment suggestions to further optimize permission allocation during the project, such as adding new permissions, reducing permissions, or adjusting permissions based on operational behavior.

[0219] 3.3. Detection and improvement of abnormal permission operations

[0220] After permission migration, it is necessary to perform anomaly detection on permission operations to ensure that the migrated permission settings meet the user's actual operation needs and prevent risks caused by improper permission allocation.

[0221] Input data:

[0222] The user's operation log after migration, including the use of migration permissions, operation frequency, operation content, etc.

[0223] Historical permission behavior data and project context to compare actual operations with expected operations.

[0224] Algorithm improvements:

[0225] Innovation: Anomaly detection model based on improved LSTM combined with graph network

[0226] The long short-term memory network (LSTM) is used to process the permission usage behavior in time series and extract the user's operation mode.

[0227] The dependency relationship between permission operations is modeled as a graph structure, and the chain reaction between permissions is analyzed in combination with Graph Neural Network (GNN).

[0228] Graph networks can help models discover potential abnormal patterns in permission operations, such as permissions that a user should not have or abnormally high-frequency operations.

[0229] Input processing:

[0230] LSTM receives time series data of user operation behaviors and extracts short-term and long-term operation features.

[0231] The graph network takes the relevance and hierarchical structure of permissions as input to help the model determine whether the use of permissions meets actual needs.

[0232] Output:

[0233] Output the detection results of abnormal operations, give risk levels (low, medium, high), and generate recommendations for automatic adjustment of permissions.

[0234] Innovation: The output includes not only the abnormal operation itself, but also the chain risk between permissions. For example, the incorrect assignment of a permission may lead to security vulnerabilities in the entire permission system, which can be automatically identified through the graph network.

[0235] 3.4. Permission self-healing and optimization mechanism

[0236] Once an abnormal operation is detected, a self-healing mechanism can be introduced to dynamically adjust permissions.

[0237] Implementation of self-healing mechanism:

[0238] Combination of Reinforcement Learning (RL) and Permission Self-Healing

[0239] The RL model is used to adjust the user's permission operation behavior in real time. The "state" of the RL model is the current configuration of the permission, the "action" is to increase, decrease or adjust the permission, and the "reward" is set based on whether the adjusted permission usage is more in line with the actual operation needs.

[0240] Innovation: Use Deep Q-Network (DQN) to enhance the model's ability to quickly self-heal complex permission systems. Through training, the system can intelligently optimize permission allocation.

[0241] Permission optimization steps:

[0242] Input: The abnormal permissions and risk levels output by the input anomaly detection module, as well as the user operation history.

[0243] Output: Generate permission optimization suggestions or automatically perform permission adjustment operations to reduce unnecessary permissions and increase permissions actually required.

[0244] 3.5. Improvement of technology:

[0245] Traditional implementation: In permission migration, permissions are usually inherited based on preset rules or manually configured experience, lacking an automated and intelligent tuning process.

[0246] Innovation and improvement:

[0247] Through conditional generative adversarial networks and dual network migration structures, permission migration is made more adaptive and intelligent. Using the anomaly detection model combining graph networks and LSTM, in-depth analysis of operational behaviors in complex permission systems is performed.

[0248] The introduction of reinforcement learning mechanism enables permission self-healing to be more dynamically adjusted, improving the security and efficiency of the system.

[0249] In this way, through these technical means, the system can automatically migrate, inherit and optimize permission management, while achieving efficient anomaly detection and permission self-healing, greatly improving the flexibility and security of permission management.

[0250] Example:

[0251] Assume that designer A is involved in two different construction projects: Project X and Project Y. In Project X, A is responsible for the architectural structure design and has relevant permissions; when A transfers to Project Y, the system automatically migrates his permission configuration in Project X and adjusts it according to the needs of Project Y.

[0252] 1. Initial permission configuration

[0253] Project X: A has the permissions of "view, modify building structure, submit design plan" in Project X, which is applicable to structural design tasks.

[0254] Project Y: A is assigned to Project Y and is responsible for similar architectural design tasks, but some permissions need to be adjusted due to different design standards of Project Y.

[0255] 2. Transfer learning and cross-project permission inheritance

[0256] The transfer learning model can analyze A's permission usage patterns in project X, learn its commonly used permission configurations, and intelligently migrate and adjust them in project Y.

[0257] Migration steps:

[0258] Permission feature extraction: The system extracts the permission usage features of A in project X (such as operation frequency, task type, usage time, etc.) through the transfer learning model and builds a user permission profile.

[0259] Project requirements comparison: The transfer learning model compares the requirements and standards (such as design process, approval rules, etc.) of project X and project Y to determine which permissions need to be adjusted or reallocated. For example, the design approval process of project Y may be more stringent, so the system needs to limit A's approval permissions when migrating permissions.

[0260] Dynamic permission allocation: Based on the comparison results, the model automatically assigns appropriate permissions to A in Project Y. For example, the system automatically retains A's permission to "view and modify building structures" in Project X, but may restrict his permission to submit design plans until he passes the approval process of the new Project Y.

[0261] 3. Automatic adaptation and optimization of permissions

[0262] In order to enable A to work more efficiently in project Y, the system further optimizes its permission configuration by combining reinforcement learning. Reinforcement learning adjusts permissions through real-time feedback and dynamically learns based on A's operational behavior.

[0263] Permission optimization steps:

[0264] Initial permission allocation: The system initially allocated permissions to A in project Y based on transfer learning, allowing him to view and modify the building structure design, but restricting his permission to submit design plans.

[0265] Operational behavior monitoring: The reinforcement learning model begins to monitor A's operational behavior in project Y in real time, such as whether A frequently accesses certain functions and whether its operational efficiency is consistent with its performance in project X.

[0266] Feedback mechanism: The system gives positive / negative feedback based on A's performance. If A is highly efficient in project Y and meets the project requirements, the system will gradually restore his permission to submit design plans; conversely, if it is found that he abuses his permissions or is inefficient, the system will further restrict certain operations.

[0267] 4. Innovation: Intelligent permission adjustment and adaptation to project differences

[0268] Compared with the traditional manual permission allocation method, the combination of transfer learning and reinforcement learning can realize the automatic inheritance and intelligent adjustment of permissions. In particular, the following innovations are made:

[0269] Automatic identification of permission differences: The system can automatically identify differences between projects and take these differences into account when migrating permissions without manual adjustments. For example, if some permissions are applicable in project X but not in project Y, the system will automatically restrict these permissions.

[0270] Permission learning and optimization: Through reinforcement learning, the system can adjust permissions based on the user's real-time operation behavior, which not only improves work efficiency but also reduces the risk of permission abuse.

[0271] Gradual permission unlocking: The system will not grant all permissions to users at once. Instead, it will gradually unlock more advanced permissions based on the user's actual performance through a reinforcement learning feedback mechanism, thereby avoiding excessive concentration of permissions.

[0272] Detailed explanation of input and output

[0273] enter

[0274] Historical permission data: including the user's permission configuration, usage frequency, operation habits, etc. in project X, as input for transfer learning. Project difference data: including the difference analysis between project X and project Y (such as design process, approval rules, etc.), to help the system make reasonable adjustments when permissions are transferred. Real-time operation data: in project Y, the user's actual operation behavior data (such as permission usage frequency, work efficiency, etc.) as input for reinforcement learning.

[0275] Output

[0276] Permission inheritance configuration: Based on the results of transfer learning, the system generates the user's initial permission configuration in the new project and automatically migrates common permissions.

[0277] Permission adjustment instructions: The reinforcement learning model outputs permission adjustment suggestions based on user performance, such as unlocking or restricting certain operation permissions.

[0278] Embodiment six:

[0279] The project rights management method of the embodiment of the present invention is used to implement intelligent rights allocation and agency in rights management. The specific implementation details of this embodiment are described below.

[0280] In the "Intelligent Permission Assignment and Agency" function, the traditional K-Means clustering algorithm is used to assign users or multiple tasks of users to different permission groups. However, this method may be too simple and lack flexibility when dealing with complex permission structures and multi-dimensional operation behaviors. In order to improve innovation, we can improve the clustering algorithm, introduce a hybrid model or enhance the intelligence of the clustering algorithm, and combine it with anomaly detection mechanisms to achieve more intelligent and adaptive permission allocation.

[0281] Innovative algorithm design: Intelligent permission allocation based on hybrid clustering and reinforcement learning

[0282] 4.1. Input and processing of intelligent permission allocation

[0283] enter:

[0284] User characteristic data: each user’s identity attributes (such as position, department, scope of responsibilities, etc.), historical authority allocation records, and the frequency and behavior of users’ use of authority.

[0285] Permission requirement data: The requirements of users or roles for different permissions in the current system, including the frequency of permission use, importance, operational risks, etc.

[0286] Operational behavior data: The user's actual operational behavior records, including the ratio of normal operations to abnormal operations, and the matching degree between historical behaviors and permission usage.

[0287] Data processing:

[0288] Data preprocessing: Standardize and vectorize user feature data and permission requirement data to generate multi-dimensional feature vectors. Permission usage behavior can be modeled as a time series to extract operation behavior patterns.

[0289] Innovation: Through the graph neural network (GNN), the collaborative relationship between users is further modeled and the permission dependencies between users are extracted. For example, some permission operations may require the collaboration of multiple people. By analyzing the operation correlation between users, collaborative permissions can be dynamically allocated.

[0290] 4.2. Improvement of permission allocation algorithm

[0291] The traditional K-Means clustering algorithm usually assigns users to a fixed number of cluster groups, but this method is insufficient when dealing with high-dimensional and dynamically changing permission requirements. To this end, reinforcement learning (RL) and hybrid clustering (such as GMM, Gaussian mixture model) can be combined to achieve more flexible permission allocation.

[0292] Improved algorithm design:

[0293] Innovation 1: Dynamic clustering based on Gaussian mixture model (GMM)

[0294] Compared with K-Means, GMM can handle more complex permission allocation scenarios, especially when the permission requirements of different users are unevenly distributed. GMM generates a multi-dimensional Gaussian distribution to estimate the probability of each user's permission requirements, thereby more accurately determining the permission allocation group.

[0295] GMM can also adapt to dynamic changes in permission requirements. For example, a user may have different permission requirements in different projects or tasks. GMM can adapt to such changes by continuously updating weights.

[0296] Innovation 2: Reinforcement learning-driven permission adjustment mechanism

[0297] The introduction of reinforcement learning allows the system to optimize in real time based on the user's actual operation behavior after assigning permissions. The "state" of the reinforcement learning model is the user's current permissions, the "action" is to increase, decrease or modify permissions, and the "reward" is based on whether the permissions meet user needs and whether there is any abuse of permissions or abnormal operations.

[0298] This mechanism allows the system to dynamically learn the user's behavior patterns after permissions are allocated, and adjust permission allocation in a timely manner to avoid over-allocation or under-allocation of permissions.

[0299] Steps of the permission allocation algorithm:

[0300] Step 1: Initial clustering assignment: Use the Gaussian mixture model (GMM) to perform preliminary clustering on user feature vectors, and assign users to initial permission groups based on their historical operation behaviors, roles, and permission requirements.

[0301] Step 2: Reinforcement learning optimization: Use the reinforcement learning model to monitor the user's operation behavior in real time and dynamically adjust permissions. If the user's permissions are used less frequently or often exceed the scope of permissions, the system will automatically increase or decrease permissions.

[0302] Innovation: The "reward" function design of reinforcement learning can be more intelligent. For example, risk scoring is performed on user operation behaviors. If certain operations are high-risk permissions, the system will automatically adjust the permission level or require secondary authorization.

[0303] 4.3. Input, Output and Data Processing

[0304] Input data:

[0305] User permission requirements: Based on the user's historical operations and roles, enter the user's current permission requirement distribution.

[0306] Permission allocation history: Permission allocation records in the system, including the actual usage of permissions after users obtain them.

[0307] Anomaly detection data: Real-time monitoring of user permission usage behavior, and whether the input user has performed abnormal operations (can be combined with anomaly detection mechanism).

[0308] Data processing:

[0309] Gaussian mixture model: Based on the multi-dimensional feature distribution of users, the probability of each user belonging to a cluster is calculated, and the user is assigned to the most appropriate permission group.

[0310] Reinforcement learning model: Use Q-Learning or Deep Q Network (DQN) to score user operation behaviors and adjust permissions based on operation feedback.

[0311] Innovation: Combined with the LSTM network to process the time series data of user operations, the reinforcement learning model is used to learn long-term permission allocation strategies, thereby reducing short-term errors.

[0312] Output:

[0313] Output 1: Each user's initial permission assignment group, and permission adjustment suggestions based on permission usage frequency.

[0314] Output 2: The automatically adjusted permission allocation list updates the user permission groups in real time to ensure that the allocated permissions match the actual needs.

[0315] 4.4. Abnormal Operation Detection and Adaptive Agent

[0316] The intelligent permission allocation system also needs to be combined with anomaly detection mechanisms to ensure that the allocation and use of permissions do not exceed the normal range.

[0317] Improvements in anomaly detection:

[0318] Innovation: Anomaly detection mechanism based on LSTM and adaptive network

[0319] Use the LSTM network to process the time series data of user operation behavior, combine the operation history to perform anomaly detection, and determine whether there is an abnormal pattern in the user's permission usage behavior.

[0320] Adaptive networks are introduced to repair abnormal permission operations detected in real time to prevent excessive allocation or abuse of permissions. The adaptive network temporarily authorizes or cancels some abnormal operations by adjusting the user's proxy permissions.

[0321] Authority proxy function:

[0322] K-NN-based permission proxy allocation: For temporary permission requirements, the K-Nearest Neighbors (K-NN) algorithm is used to automatically assign appropriate temporary permissions to the current user based on the permission proxy records of historical users with the most similar characteristics to the current user.

[0323] Innovation: Based on the traditional K-NN, a time-weighted mechanism is introduced to give higher weights to newer permission allocation records, thereby improving the timeliness and rationality of permission delegation.

[0324] Output:

[0325] Output 1: Permission anomaly detection results, including a list of high-risk operations and permission usage deviations.

[0326] Output 2: Real-time permission adjustment or permission proxy suggestion.

[0327] 4.5. Summary of technical improvements and innovations:

[0328] Traditional implementation: usually uses simple rules or static clustering algorithms to allocate permissions, lacking dynamic adjustment and intelligent optimization mechanisms.

[0329] Innovation and improvement: Through the combination of Gaussian mixture model and reinforcement learning, permission allocation is more flexible and can adapt to multi-dimensional data and dynamic needs.

[0330] LSTM combined with adaptive networks realizes anomaly detection and self-healing of complex permission behaviors, greatly improving the security and intelligence of permission allocation.

[0331] The K-NN-based authority proxy allocation adds a time-weighted mechanism to improve the rationality and timeliness of proxy authority.

[0332] Through these innovative technical designs, intelligent permission allocation and proxy functions can more accurately meet the permission requirements in complex projects and ensure the security and rationality of permission allocation.

[0333] Example:

[0334] Assume that project manager B is responsible for multiple construction projects, involving numerous tasks and complex authority allocation. For example, B needs to review design documents, check financial reports, and coordinate construction progress, but due to time constraints, he cannot handle all tasks in real time. The system needs to use intelligent analysis to assign appropriate authority to B, or assign a subordinate to perform specific tasks when necessary.

[0335] 1. Initial Permission Analysis

[0336] When entering a new project, the system first configures permissions based on B's past operation records (such as approval frequency, project participation, etc.).

[0337] Applications of K-Means Clustering:

[0338] The system uses the K-Means clustering algorithm to divide users into different permission groups. For example, B's tasks are divided into "approval group", "finance group" and "construction group".

[0339] Clustering process:

[0340] The system takes B's past behaviors (such as login frequency, accessed functions, time and efficiency of processing tasks) as input and assigns it to the most appropriate permission group through clustering.

[0341] For example, B frequently approves design documents but rarely processes financial statements. Based on this pattern, the system assigns B to the "Approval Group" and grants him higher approval authority, while financial-related authority may be relatively less.

[0342] 2. Real-time task allocation and agent selection

[0343] When B is unable to complete certain tasks due to time or authority constraints, the system needs to intelligently select an agent for him to ensure that the task is completed smoothly.

[0344] Applications of Deep Neural Networks:

[0345] The system uses a deep neural network model to analyze the skills, historical task performance, and availability of project team members and select appropriate agents from them.

[0346] Agent selection process:

[0347] Input features: The system analyzes B’s tasks, current status, agent availability, skill matching and other information as input to the deep neural network.

[0348] Output agent recommendation: The deep neural network analyzes these input features and outputs an agent selection recommendation. For example, if the approval task that B is handling belongs to structural design, and designer C in the project has relevant authority and rich experience, then the system selects C as the agent to approve on behalf of B.

[0349] 3.Task automation and permission optimization

[0350] To ensure efficient operation of the project, the system also has the ability to automatically perform certain repetitive tasks and reduce manual intervention.

[0351] Task automation steps:

[0352] The system learns B’s common task patterns, such as its approval process and frequency, through historical task data, and automatically handles relatively simple approval tasks.

[0353] If B often makes similar decisions on certain tasks, the system will gradually learn and automate these operations, and with B's authorization, the system can automatically perform these tasks.

[0354] 4. Innovation: Dynamic permission adjustment and intelligent proxy mechanism

[0355] Different from the traditional permission allocation system, this system has the following innovations:

[0356] Dynamic adjustment of permissions based on behavior: The K-Means clustering algorithm dynamically adjusts permission allocation based on the user's behavior patterns, rather than static permission configuration. For example, the system can automatically increase or decrease certain permissions of users based on their current task requirements and past behavior.

[0357] Intelligent agent mechanism: By analyzing agents through deep neural networks, the system can automatically assign the most suitable agent when the user is unable to complete the task, ensuring the continuity and efficiency of the project.

[0358] Automated task processing: The system can learn the user's task patterns and automatically process certain common tasks to reduce manual intervention. It is particularly suitable for repetitive approval or report generation tasks.

[0359] Detailed explanation of input and output

[0360] enter

[0361] User behavior data: including B's login frequency, operation behavior, task processing history, etc., as input features for K-Means clustering and deep neural networks. Project team data: including team members' skills, task processing capabilities, current status (such as availability), past task records, etc., for agent selection and authority optimization. Task data: specific task requirements and complexity in the current project, as a reference for system authority allocation and proxy mechanism.

[0362] Output

[0363] Intelligent permission configuration: The system analyzes B's behavior, dynamically adjusts his permissions in the current project, and outputs a suitable permission configuration plan. Agent selection suggestions: When B is unable to perform a task, the system outputs the most suitable agent and automatically notifies him to take over the task. Automated task processing: After the system identifies tasks that can be automated, it automatically generates the corresponding task processing results.

[0364] Embodiment seven:

[0365] The method of project authority management in the embodiment of the present invention is used to implement authority optimization and feedback learning. The specific implementation details of this embodiment are described below. In the "authority optimization and feedback learning" function, the use of reinforcement learning to optimize authority management is a common idea. Traditional reinforcement learning methods can gradually improve authority allocation through trial and error and reward mechanisms. However, to achieve innovative improvements, innovative designs can be made from aspects such as algorithm architecture, data processing, and strategy optimization to enhance the system's adaptability and accuracy in complex scenarios. Innovative improvements: Authority optimization based on multi-agent reinforcement learning and adaptive reward mechanisms.

[0366] 5.1. Limitations of Traditional Methods

[0367] In traditional reinforcement learning, a single agent usually takes actions based on the state of the environment and receives rewards. This approach is relatively simple for complex permission scenarios (multi-user, multi-permission, multi-task). It cannot handle the collaborative relationship between multiple users and the dynamic adjustment of cross-project permissions.

[0368] Innovative improvement ideas:

[0369] Multi-Agent Reinforcement Learning (MARL): Multiple agents jointly participate in permission decision-making. Each agent can represent a user or a type of permission. They learn collaboratively with each other and dynamically optimize the entire permission system.

[0370] Adaptive reward mechanism: Introduce a context-based reward function to dynamically adjust the reward value according to the urgency of the task, the importance of the user role, and the risk level of the operation to ensure optimal permission allocation and adjustment.

[0371] 5.2. Improved implementation of permission optimization and feedback learning

[0372] 5.2.1 Data input and processing

[0373] enter:

[0374] Permission allocation history data: permission usage records, historical permission adjustments, task completion status, and abnormal operation records for each user. Task context: current task urgency, user operation role level, required permission type (such as read, write, execute, etc.), and task risk level. User behavior data: sequence data of user operation behavior, including operation time, operation frequency, operation type, and matching degree with task objectives.

[0375] Data processing:

[0376] Behavioral feature extraction: Use convolutional neural networks (CNN) or LSTM to extract features from time series data of user operations and capture behavioral patterns of permission usage.

[0377] Innovation: Based on the urgency and risk level of the task, the attention mechanism is used to weight the behavior patterns to ensure that the system can give priority to high-risk or high-priority permission allocation requests.

[0378] Collaborative learning data generation: In a multi-agent learning scenario, each agent (such as different users or permission groups) needs to share some permission usage data and task information to generate a collaborative learning state space. This state space can include the operational relevance between users and the collaborative use of different permissions by users.

[0379] 5.2.2 Permission Optimization Algorithm

[0380] Multi-Agent Reinforcement Learning (MARL)

[0381] Agent definition: Each agent represents a user or permission role. By continuously learning the user's operating behavior and task requirements, agents can collaborate on permission allocation by sharing information.

[0382] Innovation: The Policy Gradient method is introduced to optimize the operation strategy of each agent. The interaction between different agents is learned through policy optimization to achieve the optimal authority allocation.

[0383] Status and Actions:

[0384] Status: the current user's authority allocation, task urgency, operation behavior history, user's authority interaction records with other intelligent entities, etc.

[0385] Action: Increase, decrease, or adjust the user's current permissions, or temporarily grant certain permissions (such as read and write permissions), and determine whether the permissions need to be revoked.

[0386] Reward mechanism: Basic reward: If the user effectively utilizes the assigned authority in the task and successfully completes the task, a positive reward will be given.

[0387] Innovation: Adaptive reward function: The reward value is adjusted based on task risk, role priority, and permission sensitivity. For example, when operating a high-risk task, incorrect use of permissions will result in a larger negative reward, while a successful task will result in a larger positive reward. Through this mechanism, the system can allocate sensitive permissions more carefully.

[0388] Permission optimization process:

[0389] Initialize permission allocation: Based on historical data, the agent preliminarily allocates permissions and performs basic permission settings based on the user's historical permission usage records and current task requirements.

[0390] Operation behavior monitoring and learning: The LSTM network performs time-series analysis on user operation behaviors to determine whether the user's permission usage is consistent with expected operations. For infrequently used or sensitive permissions, the agent reduces permissions by adjusting strategies.

[0391] Multi-agent collaborative optimization: Multiple agents share operation data and conduct collaborative learning. For example, some tasks require cross-departmental collaboration permissions, and the agents dynamically adjust the granting and withdrawal of cross-departmental permissions through policy gradient optimization.

[0392] Task completion and feedback learning: After the task is completed, the system adjusts the agent's operation strategy based on the consistency between the task completion and the user's operation behavior. Agents that use permissions efficiently will receive positive rewards, while abuse of permissions will trigger a penalty mechanism and gradually adjust the permission allocation strategy.

[0393] 5.2.3 Abnormal operation detection and adaptive adjustment

[0394] In order to prevent users from abusing their rights or making mistakes in assigning rights, the system combines LSTM with reinforcement learning to detect abnormal operations. LSTM network: The time series of user operation behaviors is input into the LSTM network to analyze in real time whether the user operation is abnormal. For example, the system can learn the user's behavior patterns in normal task operations, and trigger an abnormal alarm once an operation behavior that exceeds the normal mode (such as abuse of write permissions or unauthorized sensitive operations) occurs.

[0395] Adaptive adjustment mechanism:

[0396] Innovation: Through reinforcement learning, the system automatically fixes abnormal permission allocation. For example, if a user frequently abuses permissions during tasks, the system will automatically reduce their permission level or require secondary authorization.

[0397] Reward adjustment for unusual actions: unusual actions trigger negative rewards, and the agent learns how to avoid these negative actions and allocate permissions more accurately in future tasks.

[0398] 5.2.4 Output and Feedback

[0399] Output:

[0400] Permission allocation suggestions: The system outputs dynamic permission adjustment suggestions based on the user's task requirements and operating behaviors, and generates a personalized permission list.

[0401] Abnormal operation report: Outputs abnormal behavior records in user operations and recommends appropriate permission recovery or authorization adjustment.

[0402] Feedback Learning:

[0403] Reward feedback: The success or failure of user permission operations will affect the agent's future permission allocation strategy. Through continuous feedback learning, the system can gradually achieve optimal permission allocation.

[0404] 5.3. Innovation Improvement

[0405] Traditional implementation:

[0406] General permission management systems rely on static rules or simple single-agent reinforcement learning mechanisms, which are difficult to adapt to complex tasks and multi-dimensional user permission scenarios.

[0407] Innovative improvements:

[0408] Multi-agent reinforcement learning: allows multiple agents to collaboratively optimize permission allocation, solving the problem of collaboration among multiple users and roles in complex permission scenarios. Adaptive reward mechanism: a dynamic adjustment mechanism based on task risk and user operation behavior improves the flexibility and security of permission allocation. LSTM combined with anomaly detection and adaptive repair: through time series analysis, real-time detection and repair of permission anomalies ensure the rational use of permissions and reduce the risk of permission abuse. Policy gradient optimization: the policy gradient method is used to optimize the collaboration ability between agents, improving the accuracy of permission allocation across tasks and roles. This innovative solution can dynamically and intelligently allocate and adjust permissions based on the user's task requirements and behavior patterns, greatly improving the security and efficiency of permission management in complex scenarios.

[0409] Example:

[0410] Assume that construction manager D is responsible for multiple construction projects, and his authority involves construction progress management, personnel allocation, task approval, etc. As the project progresses, the system needs to continuously optimize D's authority settings to ensure that he can quickly handle high-priority tasks while avoiding security risks caused by excessive openness of authority.

[0411] 1. Initial authority allocation and operation behavior analysis

[0412] The system assigns permissions to D in the initial stage. Based on D's role and task requirements, the system sets permissions for him such as approving construction progress and viewing personnel allocation.

[0413] Initial configuration of permissions:

[0414] According to D's position and task responsibilities, the system assigns construction-related permissions, such as progress approval, task assignment, etc. For example, D can manage the task assignment of construction personnel, approve daily construction plans, and view construction progress reports.

[0415] Behavioral data analysis:

[0416] Input: The system records D's daily operational behavior, such as approval frequency, which reports are viewed, which team members are interacted with, and other behavioral data.

[0417] Output: These operation behaviors are input into the reinforcement learning model as the basis for the system to learn how to optimize the permission configuration.

[0418] 2. Reinforcement Learning Permission Optimization Process

[0419] The core elements of reinforcement learning:

[0420] State: The current user's permission configuration and operating environment. The system uses D's current permission status and behavior as the input for reinforcement learning. Action: The system can make adjustments to D's permissions, such as adding approval permissions, restricting certain operating permissions, etc. Reward: The system evaluates the rationality of the current permission configuration based on D's feedback and operating efficiency, and calculates the corresponding "reward". For example, when D completes certain tasks efficiently, the system records it as a positive reward; if D's permissions are not sufficient to complete certain important tasks, it is recorded as a negative reward.

[0421] Permission optimization process:

[0422] Input features: The system obtains D's daily operation behavior, task processing time, number of erroneous operations, and other behavioral data. These behavioral data are used as input by the reinforcement learning model to evaluate the effectiveness of the current permission configuration.

[0423] Execute action:

[0424] The system selects the optimal permission adjustment action based on the reinforcement learning algorithm. For example, if the system finds that D frequently applies for temporary permissions to view personnel allocation, the reinforcement learning model will recommend that the system increase D's permissions in this regard to avoid frequent temporary applications. Conversely, if D does not use certain permissions, the system will automatically reduce his infrequently used permissions to improve the security of overall permission management.

[0425] Feedback Mechanism:

[0426] After each permission adjustment, the system monitors D's behavioral feedback. If the new permission configuration makes D's task processing smoother, the system will record the positive feedback and further optimize the permission configuration. If the permission adjustment reduces D's work efficiency, the system will record the negative feedback and adjust the learning strategy to avoid making similar wrong adjustments again.

[0427] The project authority management method of the embodiment of the present invention can dynamically manage and optimize the complex authority structure of the steel plant data center through machine learning algorithms and deep behavioral analysis; it can not only adjust the authority according to the actual user behavior, but also predict the potential authority requirements, and has high-end functions such as authority self-healing, intelligent agent, and ecological evolution; this comprehensive authority management not only improves the collaborative efficiency of the project, but also greatly enhances the security of the data and the accuracy of authority allocation; this futuristic technical architecture brings a new idea to authority management and is a revolutionary breakthrough in the traditional authority system.

[0428] Embodiment eight:

[0429] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement any of the methods described above.

[0430] Embodiment nine:

[0431] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.

[0432] Embodiment ten:

[0433] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the method described above are implemented.

[0434] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A method for project authority management, characterized in that: include: Acquire user behavior characteristics of a target user, task characteristics of a target task, and permission status characteristics related to the target user and the target task, wherein the user behavior characteristics are used to characterize the target user's operation behavior, access frequency to project data, and / or historical operation behavior trajectory; The task feature is used to characterize the context information of the task, and the context information of the task includes: the type of task, the resources required for the task, the time limit of the task, the stage of the task and / or the identity of the target user; the permission status feature related to the target user and the target task is used to characterize the current permission of the target user, the use history of the current permission by the target user and / or the permission allocation mode of similar tasks that meet the predetermined similarity condition with the target task; Based on the user behavior characteristics, the task characteristics of the target task, and the permission status characteristics related to the target user and the target task, a permission configuration model is used to output the optimal permission configuration of the target user in the target task and the confidence of the optimal permission configuration, wherein the permission configuration includes the configuration of read and write permissions, data access permissions and / or model operation permissions; the permission configuration model is pre-trained using a preset machine learning algorithm.

2. The method according to claim 1, characterized in that The permission configuration model is a Bayesian network model, and the method further includes: The permission configuration model dynamically adjusts the optimal permission of the target user in the target task according to the user behavior characteristics, the task characteristics of the target task and / or the permission status characteristics related to the target user and the target task; The dynamically adjusted frequency and amplitude meet a preset frequency threshold and amplitude threshold.

3. The method according to claim 1, characterized in that: The user behavior characteristics also include: the user's operation log, permission change history, operation corresponding timestamp, current operation device information and / or current geographical location; the device information includes: the IP address of the device; The method further comprises: Embedded vectorization is performed on the user behavior feature to obtain an embedded vector for characterizing the user behavior; The embedded vector is input into a pre-trained LSTM network to detect whether there is currently an abnormal permission operation, and in the case of an abnormal permission operation, a permission self-healing strategy is enabled through a pre-trained self-healing model, wherein the permission self-healing strategy includes: generating permission adjustment suggestions for the current user or directly adjusting the permissions of the current user; the permission adjustment includes: restricting specific operations of the current user, reducing the permission level of the current user and / or reminding the administrator to review the permissions of the current user; the machine learning algorithm used by the self-healing model includes: a reinforcement learning algorithm.

4. The method according to claim 3, characterized in that The abnormal authority operation includes: abnormal authority escalation, abnormal behavior patterns and / or abnormal high-frequency operations; and / or, The permission self-healing strategy also includes: after adjusting the permissions of the current user, using a predetermined additional identity authentication method to perform additional identity authentication on the current user; if the verification passes, it is confirmed that the current user is a legitimate user and the operation is invalid, and the permissions of the current user are restored; otherwise, the permission adjustment of the current user is maintained.

5. The method according to claim 1, characterized in that The method also includes the steps of implementing the cross-project permission inheritance and migration from the source project to the current project for the target user; wherein the implementation of the cross-project permission inheritance and migration includes: Obtain the target user's permission characteristics in the source project, the project characteristics of the current project, and the target user's operation behavior characteristics in the source project; the permission characteristics are used to characterize the target user's: permission allocation, permission operation history and / or identity information in the source project; the project characteristics are used to characterize the current project's: name, scale, project type and / or role distribution of participants; the target user's operation behavior characteristics in the source project are used to characterize the target user's frequency of use of permissions of different permissions in the source project and / or the risk level of the operation; According to the obtained permission characteristics of the target user in the source project, the project characteristics of the current project, and the operation behavior characteristics of the target user in the source project, a pre-trained permission inheritance and migration model is used to output the permissions of the target user in the current project; wherein the permission inheritance and migration model includes: a dual-network migration structure and a dual-network migration structure.

6. The method according to claim 1, characterized in that The permissions of the target user in the current project output for the permission inheritance and migration model also include: An anomaly detection model combining a graph network and LSTM is used to detect permission anomalies of the target user in the current project, and when permission anomalies exist, a reinforcement learning mechanism is used to perform permission self-healing operations.

7. The method according to claim 1, characterized in that The user behavior characteristics also include: the user's historical task records and the user's mode of handling specific tasks; the method also includes one or more of the following: When the target user has permissions for multiple tasks, the permissions for each of the multiple tasks are allocated using a K-Means clustering algorithm according to the user behavior characteristics; When the target user has authority for multiple tasks, and when there is a task among the multiple tasks that meets a predetermined condition, selecting an agent for the task that meets the predetermined condition; the predetermined condition includes: the target user cannot complete the task; A deep neural network is used to learn the user's mode of processing specific tasks, the specific tasks are set as tasks that can be processed automatically, and after the specific tasks occur, the task processing results are automatically generated; the specific tasks include: approval tasks.

8. The method according to claim 1, characterized in that The user behavior characteristics also include: the user's historical permission usage records; the task characteristics also include: task requirements; the method also includes: Using multi-agent reinforcement learning to optimize authority allocation according to the user behavior characteristics of the target user, the task characteristics of the target task, and the authority status characteristics related to the target user and the target task, the authority allocation optimization includes adjusting the authority of the target user to the target task; and After the optimization of authority allocation is achieved, the behavior of the target user is monitored to obtain the behavior feedback of the target user, and the authority allocation is further optimized based on the obtained feedback, wherein the behavior feedback includes: positive feedback and negative feedback.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement the method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images