An intelligent access management system based on dynamic resource mapping and multi-agent cooperation

Abstract

The application discloses an intelligent access management system based on dynamic resource mapping and multi-agent cooperation, belongs to the technical field of intelligent Internet of Things and distributed resource scheduling, and comprises the following: a physical resource layer composed of distributed intelligent access terminals; a virtual resource mapping layer used for dynamically mapping physical cabinet grids into virtual resource pools, and adopting a clustering algorithm to dynamically divide the cabinet grids into multiple virtual partitions, and supporting automatic adjustment of partition boundaries according to real-time loads; a behavior anchor point engine used for fusing multi-modal behavior data to predict user access intentions, and generating a dynamic permission anchor point curve; a multi-agent cooperative decision module containing a user agent, a cabinet grid agent and a management agent, and realizing resource negotiation and permission determination through a distributed consensus protocol; and a heterogeneous system fusion interface. The application realizes dynamic and flexible allocation of resources, intention-based permission deduction and decentralized autonomous cooperation, and significantly improves resource utilization and system robustness.

Description

Technical Field

[0001] This invention belongs to the field of intelligent Internet of Things and distributed resource scheduling technology, and particularly relates to an intelligent access management system based on dynamic resource mapping and multi-agent collaboration. Background Technology

[0002] Currently, with the rapid development of IoT and AI technologies, intelligent access control devices are widely used in multi-tenant scenarios such as campuses, enterprises, libraries, and public places. Existing intelligent access control systems generally employ technologies such as facial recognition, IC card authentication, and basic access control to achieve contactless access and basic data recording functions. In typical application scenarios, the system completes basic access operations by permanently binding or randomly assigning users to lockers, combined with static time-based or whitelisted permission rules. Simultaneously, some systems achieve preliminary cross-system collaboration by connecting directly with external business systems via APIs. These technical solutions, to a certain extent, improve the convenience and security of access management and meet basic business needs.

[0003] However, existing technologies still suffer from the following deep-seated technical shortcomings in practical applications. First, resource allocation is static, generally employing fixed "user-cabinet" binding or simple random allocation strategies. This fails to dynamically adjust based on actual usage frequency and time-of-day heat distribution, resulting in some cabinets remaining idle for extended periods while resources are insufficient during peak hours, limiting both resource utilization and user access experience. Second, access control mechanisms are passive and coarse-grained, often based on static rules such as fixed time periods or whitelists. They struggle to perceive user intent and contextual changes, failing to proactively predict and dynamically respond to abnormal behavior, and lacking effective support for time-sequential and contextualized access control. Third, inter-system collaboration mechanisms are rigid. Existing solutions often use point-to-point, tightly coupled API calls, resulting in rigid business processes. Adding new business systems or adjusting business rules requires modifying multiple interfaces, leading to poor system scalability and high maintenance costs. Fourth, decision-making mechanisms are centralized. Systems generally rely on centralized servers for unified decision-making. During network fluctuations or cloud failures, terminals cannot independently complete complex permission judgments, resulting in weak overall system reliability. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an intelligent access management system based on dynamic resource mapping and multi-agent collaboration, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, this invention provides an intelligent access management system based on dynamic resource mapping and multi-agent collaboration, comprising a physical resource layer composed of distributed intelligent access terminals, each terminal containing multiple independent cabinets, and each cabinet integrating an electronic lock, a status sensor, and an edge computing unit; characterized in that it further includes: The virtual resource mapping layer is used to dynamically map the physical cabinets of the physical resource layer into virtual resource pools, and to dynamically divide the cabinets into multiple virtual partitions based on the usage feature vectors of the cabinets using a clustering algorithm. The virtual partitions include high-frequency groups, low-frequency groups, dedicated groups, and shared groups. The dedicated groups are used for fixed users to bind for a long time, and the shared groups are used for dynamic temporary allocation. It also supports automatic adjustment of partition boundaries based on real-time load. The behavior anchor engine is used to integrate users' multimodal behavior data, predict users' access intentions based on temporal neural networks, and generate dynamic permission anchor curves with time as the horizontal axis and permission level as the vertical axis. The multi-agent collaborative decision-making module is deployed in the cloud and at the edge, and includes user agents, cabinet agents and management agents. It realizes resource negotiation and permission determination between user agents and cabinet agents through a distributed consensus protocol, and the management agent performs global policy arbitration and anomaly coordination. The heterogeneous system integration interface adopts an event-driven architecture and a semantic data bus to enable loosely coupled interfaces with external business systems and support dynamic orchestration of cross-system business processes.

[0006] Preferably, the virtual resource mapping layer includes a resource virtualization unit, a dynamic grouping algorithm unit, and an elastic partitioning management unit; The resource virtualization unit is used to abstract each physical cabinet into a virtual resource node with attributes such as capacity, location, ownership policy, and available time period. The dynamic grouping algorithm unit uses a clustering algorithm based on historical concurrent access data to dynamically divide the cabinet into high-frequency group, low-frequency group, dedicated group and shared group; The elastic partition management unit is used to automatically adjust the partition boundaries and capacity of each group according to the real-time load.

[0007] Preferably, in the virtual resource mapping layer, the K-means clustering algorithm is used to dynamically group the cabinets, and its objective function is: ; in, The feature vector of the cabinet includes usage frequency, average occupancy time, and peak heat value; For the first The cluster centers of the group, the grouping results for each Minutes based on sliding window size Historical data was recalculated.

[0008] Preferably, the behavior anchor engine includes a multimodal behavior modeling unit, an intent prediction model unit, and a dynamic permission anchor generation unit; The multimodal behavior modeling unit is used to integrate user identity features, time-series behavior, spatial trajectory data, and external business events to construct a behavior feature vector; The intent prediction model unit is used to perform temporal prediction of user access probability using a long short-term memory network. The dynamic permission anchor point generation unit is used to dynamically generate the dynamic permission anchor point curve based on the intent prediction result and external events.

[0009] Preferably, in the multi-agent collaborative decision-making module, the user agent represents the user to submit access intentions and constraints, the cabinet agent manages the status of the corresponding physical cabinet and participates in resource allocation negotiation, and the management agent is responsible for global strategy arbitration, anomaly coordination and cross-system task orchestration; the three types of agents achieve decentralized resource allocation and permission decision-making through a distributed consensus algorithm.

[0010] Preferably, the heterogeneous system fusion interface adopts an event-driven message passing mechanism, which unifies the business events generated by external business systems into the semantic data bus, and realizes dynamic configuration of business processes through visual orchestration tools without modifying the underlying interface.

[0011] Preferably, the virtual resource mapping layer is further used to calculate the real-time load factor: ; when At that time, the elastic buffer shifts to the high-frequency shared region with a step size Δ; when At that time, the backoff buffer capacity is adjusted. For high threshold, The threshold is low.

[0012] Preferably, in the dynamic permission anchor curve generated by the behavior anchor engine, the user... At any moment permission levels for: ; in, Basic time period permissions, , The influence coefficient is the intention. For the predicted access probability, Adjust the amount for external events.

[0013] Preferably, in the multi-agent collaborative decision-making module, when a user agent initiates an access request, it allocates resources with candidate cabinet agents according to a negotiation decision function, which is: ; in, Rate the spatial distance. To determine the degree of matching between the attribution policy and user permissions, This represents the current load factor of the cabinet.

[0014] Preferably, the system further includes a closed-loop self-optimization module for recording decision logs generated during system operation, the decision logs including system status, decision actions, and immediate rewards; The closed-loop self-optimization module optimizes the grouping strategy of the virtual resource mapping layer online based on the multi-armed gambling machine algorithm, and incrementally trains the temporal neural network of the behavior anchor engine based on the decision log to update the dynamic permission anchor curve.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention dynamically maps physical cabinets to virtual resource pools by setting up a virtual resource mapping layer. Based on a clustering algorithm, the cabinets are dynamically divided into high-frequency, low-frequency, dedicated, and shared groups, while also supporting automatic adjustment of partition boundaries according to real-time load. This technical solution decouples cabinet resources from static user binding, allowing for elastic grouping and dynamic allocation based on usage frequency, time-of-day activity, and real-time concurrent load. This effectively revitalizes idle resources, alleviates resource shortages during peak periods, and significantly improves cabinet utilization.

[0016] This invention establishes a behavior anchor engine, integrates multimodal user behavior data, and employs a temporal neural network to predict user access intentions, thereby generating a dynamic permission anchor curve with time as the horizontal axis and permission level as the vertical axis. This technical solution enables permission control to move beyond static time periods or fixed whitelists, allowing it to dynamically and continuously change based on user behavior intentions, time characteristics, and external events. This enables refined and temporal permission determination for various scenarios such as "normal access," "temporary access requests," and "abnormal behavior blocking," significantly enhancing the proactiveness and adaptability of permission control.

[0017] This invention establishes user agents, cabinet agents, and management agents deployed at the cloud and edge, employing a distributed consensus protocol to achieve resource negotiation and permission determination among these agents. The management agent handles global policy arbitration and anomaly coordination. This technical solution eliminates the system's reliance on a single central server for unified decision-making. Even during network fluctuations or cloud node failures, the edge-based user agents and cabinet agents can still dynamically allocate and verify access resources through local negotiation, ensuring the continuous availability of core services and significantly improving the system's fault tolerance and robustness.

[0018] This invention achieves loosely coupled integration with external business systems by setting up a heterogeneous system fusion interface using an event-driven architecture and a semantic data bus, and supports dynamic orchestration of cross-system business processes. This technical solution eliminates the need to modify multiple underlying interfaces when adding new business systems or adjusting business rules; adaptation can be completed simply by accessing events through the semantic bus and re-orchestrating the process. This significantly reduces system integration complexity and maintenance costs, and enhances the system's scalability and adaptability in multi-tenant, multi-business scenarios.

[0019] In summary, this invention, through the synergistic effect of the virtual resource mapping layer, behavior anchor engine, multi-agent collaborative decision-making module, and heterogeneous system fusion interface, comprehensively improves the adaptive capability, resource utilization efficiency, operational reliability, and expansion flexibility of the intelligent access management system from four dimensions: resource scheduling, access control, decision-making mechanism, and system integration, demonstrating significant technological advancements. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the four-layer system architecture according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the multi-agent collaborative decision-making process according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the dynamic permission anchor point curve in an embodiment of the present invention; Figure 4 This is a schematic diagram of the dynamic grouping and elastic partitioning of virtual resources according to an embodiment of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Example 1 like Figure 1As shown, this embodiment provides an intelligent access management system based on dynamic resource mapping and multi-agent collaboration, including a physical resource layer composed of distributed intelligent access terminals. Each terminal contains multiple independent cabinets, and each cabinet integrates an electric lock, a status sensor, and an edge computing unit. The system is characterized by further comprising: The virtual resource mapping layer is used to dynamically map the physical cabinets of the physical resource layer into virtual resource pools, and to dynamically divide the cabinets into multiple virtual partitions based on the usage feature vectors of the cabinets using a clustering algorithm. The virtual partitions include high-frequency groups, low-frequency groups, dedicated groups, and shared groups. The dedicated groups are used for fixed users to bind for a long time, and the shared groups are used for dynamic temporary allocation. It also supports automatic adjustment of partition boundaries based on real-time load. Furthermore, the virtual resource mapping layer includes a resource virtualization unit, a dynamic grouping algorithm unit, and an elastic partitioning management unit; the resource virtualization unit is used to abstract each physical cabinet into a virtual resource node with attributes such as capacity, location, ownership strategy, and available time period; the dynamic grouping algorithm unit uses a clustering algorithm based on historical concurrent access data to dynamically divide the cabinets into high-frequency groups, low-frequency groups, dedicated groups, and shared groups; the elastic partitioning management unit is used to automatically adjust the partition boundaries and capacity of each group according to the real-time load.

[0024] Specifically, the resource virtualization model maps the physical cabinet set C={c1,c2,…,cN} to the virtual resource node set V={v1,v2,…,vN}, with each virtual node defining an attribute tuple: vi=id,loc,cap,policy,avail (1) Where loc is the spatial location coordinate; cap is the capacity (1 indicates available, 0 indicates occupied); policy∈{shared,dedicated,flexible} is the attribution policy; and available is the set of available time slots.

[0025] Dynamic grouping algorithm: Based on historical usage data, K-means clustering is used to divide the cabinets into K groups. The objective function is: (2) in, The feature vector of the cabinet includes usage frequency, average occupancy time, and peak heat value; For the first The cluster centers of the group, the grouping results for each Minutes based on sliding window size Historical data was recalculated.

[0026] Elastic partition boundary adjustment: defining real-time load factors for: (3) when At that time, the elastic buffer zone shifts to the high-frequency shared area in steps Δ (typically ranging from 5% to 20% of the total number of cabinets); when At that time, the backoff buffer capacity is adjusted. For high threshold, The threshold is low.

[0027] In this embodiment, Figure 4 This diagram illustrates the dynamic grouping and elastic partitioning of virtual resources, specifically including the division and dynamic boundary adjustment process of high-frequency groups, dedicated groups, shared groups, and elastic buffers.

[0028] The behavior anchor engine is used to integrate users' multimodal behavior data, predict users' access intentions based on temporal neural networks, and generate dynamic permission anchor curves with time as the horizontal axis and permission level as the vertical axis. Furthermore, the behavior anchor engine includes a multimodal behavior modeling unit, an intent prediction model unit, and a dynamic permission anchor generation unit; the multimodal behavior modeling unit is used to integrate user identity features, time-series behavior, spatial trajectory data, and external business events to construct a behavior feature vector; the intent prediction model unit is used to perform time-series prediction of user access probabilities using a long short-term memory network; and the dynamic permission anchor generation unit is used to dynamically generate the dynamic permission anchor curve based on the intent prediction results and external events.

[0029] Specifically, multimodal user behavior modeling: constructing user... At any moment The behavioral feature vector is: (4) in: Time features (day of the week, time period coding, holiday marking); Spatial trajectory characteristics (nearest location, frequently visited area); External event characteristics (leave of absence, class, access control); This is a sequence of historical access behaviors.

[0030] Intent prediction model: A Long Short-Term Memory (LSTM) network is used to predict user access intentions in a temporal sequence. The input sequence is: The output is the access probability at the next time step. : (5) (6) (7) (8) (9) (10) Figure 3 This is a schematic diagram of the dynamic permission anchor point curve, showing the continuous curve after the basic permissions and incremental adjustments are superimposed. In the dynamic permission anchor curve, the user At any moment permission levels for: (11) in, Set the basic time period permissions (e.g., set the after-school period to 1 and the non-after-school period to 0). , The influence coefficient is the intention. For the predicted access probability, Adjustments for external events (increased during leave periods) Access control abnormally low ).

[0031] The multi-agent collaborative decision-making module is deployed in the cloud and at the edge, and includes user agents, cabinet agents and management agents. It realizes resource negotiation and permission determination between user agents and cabinet agents through a distributed consensus protocol, and the management agent performs global policy arbitration and anomaly coordination. Furthermore, in the multi-agent collaborative decision-making module, the user agent submits access intentions and constraints on behalf of the user, the cabinet agent manages the status of the corresponding physical cabinet and participates in resource allocation negotiation, and the management agent is responsible for global strategy arbitration, anomaly coordination, and cross-system task orchestration; the three types of agents achieve decentralized resource allocation and permission decision-making through a distributed consensus algorithm.

[0032] Specifically, the system defines three types of intelligent agents: User Intelligence : Representing users The status includes permission anchors. Historical behavior and constraints; Cabinet Smart Body : Represents physical cabinet The status includes occupancy status, ownership policy, and location; Management Intelligence Global coordinator, responsible for maintaining system policies and arbitration rules.

[0033] The distributed negotiation protocol employs a modified Paxos protocol for distributed negotiation. The negotiation process for a single access request is formalized as follows: (12) in, Rate the spatial distance. To determine the degree of matching between the attribution policy and user permissions, This represents the current load factor of the cabinet.

[0034] The negotiation shall be subject to the following constraints: (13) The consensus algorithm employs the Raft consensus protocol to ensure decision consistency among multiple agents. Leader election utilizes a random timeout mechanism, and log replication guarantees decision traceability. The system achieves a balance between partition tolerance (P), availability (A), and consistency (C), ultimately reaching eventual consistency.

[0035] In this embodiment, Figure 2 The flowchart for multi-agent collaborative decision-making includes the process of user agents initiating requests, cabinet agents broadcasting negotiations, and management agents arbitrating.

[0036] The heterogeneous system integration interface adopts an event-driven architecture and a semantic data bus to enable loosely coupled interfaces with external business systems and support dynamic orchestration of cross-system business processes.

[0037] Furthermore, the heterogeneous system integration interface adopts an event-driven message passing mechanism to uniformly access business events generated by external business systems into the semantic data bus, and realizes dynamic configuration of business processes through visual orchestration tools without modifying the underlying interface.

[0038] The closed-loop self-optimization module is used to record the decision logs generated during system operation. The decision logs include system status, decision actions, and immediate rewards. The closed-loop self-optimization module optimizes the grouping strategy of the virtual resource mapping layer online based on the multi-armed gambling machine algorithm, and incrementally trains the temporal neural network of the behavior anchor engine based on the decision log to update the dynamic permission anchor curve.

[0039] Specifically, the system generates decision logs during operation. ,in: This represents the system status (resource distribution, user queues, permission curves). For decision-making actions (cabinet allocation, permission adjustment); For immediate rewards (success rate of access, negative waiting time).

[0040] The multi-armed bandit algorithm is used to optimize the resource grouping strategy online. Each partition... Maintenance revenue estimation The strategy selection follows the UCB principle: (14) in For partitioning The number of times it was selected This represents the total number of decisions made.

[0041] Strategy iteration update: every The period is used to incrementally train the LSTM intent prediction model based on accumulated logs, with the loss function being cross-entropy. (15) in For actual access to tags, To predict probabilities, the dynamic permission anchor curve is updated synchronously after the model is updated, enabling the adaptive evolution of the policy.

[0042] Beneficial effects of this embodiment: This embodiment breaks away from the traditional direct binding model of "user-cabinet" and constructs a four-layer architecture of "physical resource layer - virtual resource mapping layer - behavior anchor engine - multi-agent collaborative decision-making module". The physical cabinet is abstracted into resource nodes with dynamic attributes through virtualization technology. The behavior anchor engine generates dynamic permission curves based on user behavior characteristics, realizing the decoupling and dynamic matching of resources and permissions.

[0043] This embodiment uses a clustering algorithm to dynamically group cabinets according to usage heat, spatial distribution, and ownership strategy. It supports various virtual partitions such as high-frequency groups, dedicated groups, and shared groups, and automatically adjusts the group boundaries and partition capacity according to real-time load to achieve efficient resource reuse and elastic supply.

[0044] This embodiment constructs a multimodal user behavior model, integrating temporal features, spatial trajectories, and business events. It uses a temporal neural network to predict user access intentions and generate a dynamic "time-permission" anchor curve. Permissions are no longer static switches, but dynamic functions that change continuously with time and context.

[0045] This embodiment introduces a distributed multi-agent architecture, deploying three types of intelligent agents: user agents, cabinet agents, and management agents. Access requests are initiated by the user agent, cabinet agents autonomously participate in resource negotiation, and the management agent is responsible for policy arbitration and anomaly coordination. Decentralized decision-making is achieved through a distributed consensus algorithm, significantly improving system robustness and scalability.

[0046] This embodiment adopts an event-driven architecture, unifying external system events (such as leave approval and access control) into a semantic data bus to achieve loosely coupled integration of heterogeneous systems. Business processes can be dynamically adjusted through visual orchestration without modifying the underlying interfaces.

[0047] The entire decision-making process of the system is fully recorded, forming a decision log dataset. The behavior anchor engine and resource grouping algorithm are based on this dataset and perform incremental learning on a regular basis to continuously optimize the permission model and resource allocation strategy.

[0048] Example 2 Based on the inventive concept of Embodiment 1, this embodiment is applied to the mobile phone access management scenario of a middle school.

[0049] (a) Dynamic allocation of access resources through multi-agent collaboration; The system is deployed at the school gate, with a total of 6 sets of access terminals, each set with 24 cabinets, for a total of 144 physical cabinets.

[0050] During system initialization, the virtual resource mapping layer uses K-means clustering algorithm to divide the cabinets into three virtual partitions based on two weeks of historical access data: High-frequency shared area (60 cabinets): for use by day students during peak hours, dynamically allocated; Dedicated binding area (48 cabinets): Permanently allocated to boarding students for long-term use; Flexible buffer zone (36 compartments): Dynamically add high-frequency or dedicated areas based on real-time load.

[0051] When student A enters school in the morning, their user agent, based on the behavior anchor engine, predicts that the student is a day student, with historical access times concentrated between 7:20 and 7:40, and a predicted current access probability of 92%. The user agent broadcasts an access request to the cabinet agent in the high-frequency shared area, carrying the permission anchor (valid until 21:00 that day). The cabinet agent selects an available cabinet based on its own resource status and ownership policy and provides confirmation. The management agent supervises the negotiation process and issues the execution command after confirming there are no conflicts. The entire process is completed within 0.8 seconds, without requiring real-time decision-making from a central server.

[0052] (ii) Dynamic access control driven by behavior anchors: Student B, who lives on campus, is normally only allowed to retrieve his phone after school on Fridays. The behavior anchor engine combines his historical behavior, class schedule, and dormitory check-in records to construct a permission anchor curve: permission level 0 (prohibited) from Monday to Thursday, permission level 1 (allowed) from 15:30 to 18:00 on Friday, and permission level dynamically increases to 2 (allowed to retrieve at any time) during leave.

[0053] One Wednesday afternoon, the student's parent submitted a temporary leave application for them through the leave application system. The heterogeneous system fusion interface captured the leave approval event and pushed it to the behavior anchor engine via the semantic data bus. The engine immediately updated the student's permission anchor curve, raising the permission level to 2 during the leave period. When Student B left school on leave, the user's AI initiated a request to retrieve an item, carrying the updated permission anchor. The cabinet's AI verified the permission level and allowed the item retrieval. The system simultaneously pushed the item retrieval record and permission change log to the management AI for record-keeping.

[0054] (iii) Fault-tolerant mechanisms for distributed collaboration; Based on dynamic access control driven by behavior anchors, assuming a cloud network failure temporarily renders the central management AI unavailable, when Student C initiates an access request, their user AI and the cabinet AI within the target area can still negotiate resources and determine permissions through a distributed consensus protocol based on locally cached policies and historical data. The cabinet AI stores the decision records locally, synchronizing them with the cloud once the network is restored. During this period, the system's access service remains uninterrupted, significantly improving system reliability.

[0055] (iv) Closed-loop self-optimization and strategy iteration.

[0056] After running for a semester, the system accumulated a large amount of access records, permission decisions, and resource allocation logs. The management system periodically performed offline analysis on the dataset and found that: There was a brief 15-minute congestion in the high-frequency shared area on Monday morning; The activation threshold for the elastic buffer is set too high, causing resource response delays. Some boarding students had a high concentration of students picking up their belongings on Fridays, resulting in localized queues.

[0057] Based on the above analysis, the system will automatically adjust: (1) Temporarily increase the virtual capacity of the high-frequency shared area by 20% on Monday morning; (2) Lower the activation threshold of the elastic buffer; (3) On Friday afternoons, the designated binding area was subdivided to guide boarding students to retrieve their items at off-peak times. The new strategy was automatically implemented after simulation verification, and the average waiting time for accessing items in the next semester decreased by 27%.

[0058] The beneficial effects of this embodiment: Improved resource utilization: Through dynamic virtualization and flexible grouping, the utilization rate of cabinet resources is increased by more than 30% compared with the static allocation mode.

[0059] Fine-grained access control: Dynamic permission deduction driven by behavior anchors enables a leap from "static switches" to "dynamic functions," supporting more complex and user-friendly access rules.

[0060] System collaboration flexibility: Event-driven architecture and semantic bus enable system integration to shift from "hard-coded" to "programmable", reducing the integration time of new business systems by 80%.

[0061] Enhanced reliability: The multi-agent distributed decision-making mechanism enables the system to maintain core services even when the central node fails, significantly improving availability.

[0062] Self-optimization capability: The closed-loop optimization mechanism enables the system to adaptively adjust to changes in usage patterns and continuously maintain optimal operating status.

[0063] In summary, by introducing core technologies such as dynamic resource virtualization, behavior anchoring, and multi-agent collaboration, this invention constructs an innovative technology system for intelligent access scenarios, providing a more forward-looking, robust, and adaptive solution for item access management in multi-tenant environments.

[0064] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent access management system based on dynamic resource mapping and multi-agent collaboration, comprising a physical resource layer consisting of distributed intelligent access terminals, each terminal containing multiple independent cabinets, each cabinet integrating an electronic lock, a status sensor, and an edge computing unit; characterized in that, Also includes: The virtual resource mapping layer is used to dynamically map the physical cabinets of the physical resource layer into virtual resource pools, and to dynamically divide the cabinets into multiple virtual partitions based on the usage feature vectors of the cabinets using a clustering algorithm. The virtual partitions include high-frequency groups, low-frequency groups, dedicated groups, and shared groups. The dedicated groups are used for fixed users to bind for a long time, and the shared groups are used for dynamic temporary allocation. It also supports automatic adjustment of partition boundaries based on real-time load. The behavior anchor engine is used to integrate users' multimodal behavior data, predict users' access intentions based on temporal neural networks, and generate dynamic permission anchor curves with time as the horizontal axis and permission level as the vertical axis. The multi-agent collaborative decision-making module is deployed in the cloud and at the edge, and includes user agents, cabinet agents and management agents. It realizes resource negotiation and permission determination between user agents and cabinet agents through a distributed consensus protocol, and the management agent performs global policy arbitration and anomaly coordination. The heterogeneous system integration interface adopts an event-driven architecture and a semantic data bus to enable loosely coupled interfaces with external business systems and support dynamic orchestration of cross-system business processes.

2. The system according to claim 1, characterized in that, The virtual resource mapping layer includes a resource virtualization unit, a dynamic grouping algorithm unit, and an elastic partitioning management unit; The resource virtualization unit is used to abstract each physical cabinet into a virtual resource node with attributes such as capacity, location, ownership policy, and available time period. The dynamic grouping algorithm unit uses a clustering algorithm based on historical concurrent access data to dynamically divide the cabinet into high-frequency group, low-frequency group, dedicated group and shared group; The elastic partition management unit is used to automatically adjust the partition boundaries and capacity of each group according to the real-time load.

3. The system according to claim 1, characterized in that, In the virtual resource mapping layer, the K-means clustering algorithm is used to dynamically group the cabinets, and its objective function is: ； in, The feature vector of the cabinet includes usage frequency, average occupancy time, and peak heat value; For the first The cluster centers of the group, the grouping results for each Minutes based on sliding window size Historical data was recalculated.

4. The system according to claim 1, characterized in that, The behavior anchor engine includes a multimodal behavior modeling unit, an intent prediction model unit, and a dynamic permission anchor generation unit; The multimodal behavior modeling unit is used to integrate user identity features, time-series behavior, spatial trajectory data, and external business events to construct a behavior feature vector; The intent prediction model unit is used to perform temporal prediction of user access probability using a long short-term memory network. The dynamic permission anchor point generation unit is used to dynamically generate the dynamic permission anchor point curve based on the intent prediction result and external events.

5. The system according to claim 1, characterized in that, In the multi-agent collaborative decision-making module, the user agent represents the user to submit access intentions and constraints, the cabinet agent manages the status of the corresponding physical cabinet and participates in resource allocation negotiation, and the management agent is responsible for global strategy arbitration, anomaly coordination and cross-system task orchestration; the three types of agents achieve decentralized resource allocation and permission decision-making through a distributed consensus algorithm.

6. The system according to claim 1, characterized in that, The heterogeneous system integration interface adopts an event-driven message passing mechanism, which unifies the business events generated by external business systems into the semantic data bus, and realizes dynamic configuration of business processes through visual orchestration tools without modifying the underlying interface.

7. The system according to claim 1, characterized in that, The virtual resource mapping layer is also used to calculate the real-time load factor: ； when At that time, the elastic buffer shifts to the high-frequency shared region with a step size Δ; when At that time, the backoff buffer capacity is adjusted. For high threshold, The threshold is low.

8. The system according to claim 1, characterized in that, In the dynamic permission anchor curve generated by the behavior anchor engine, the user At any moment permission levels for: ； in, Basic time period permissions, , The influence coefficient is the intention. For the predicted access probability, Adjust the amount for external events.

9. The system according to claim 1, characterized in that, In the multi-agent collaborative decision-making module, when a user agent initiates an access request, it allocates resources with candidate cabinet agents according to a negotiation decision function, which is: ； in, Rate the spatial distance. To determine the degree of matching between the attribution policy and user permissions, This represents the current load factor of the cabinet.

10. The system according to claim 1, characterized in that, It also includes a closed-loop self-optimization module, which is used to record decision logs generated during system operation. The decision logs include system status, decision actions, and immediate rewards. The closed-loop self-optimization module optimizes the grouping strategy of the virtual resource mapping layer online based on the multi-armed gambling machine algorithm, and incrementally trains the temporal neural network of the behavior anchor engine based on the decision log to update the dynamic permission anchor curve.

Images