Multi-access point network collaborative optimization system based on multiple agents

Abstract

The invention belongs to the crossing field of artificial intelligence and wireless communication, and discloses a multi-agent-based multi-AP network collaborative optimization architecture, the architecture is divided into a distributed system and a centralized system, the distributed system comprises AP equipment, a user agent and a terminal STA connected with the AP, the centralized system MAP Coordinator is deployed on a main AP, and the user agent is connected with the terminal STA. The system comprises an agent coordinator, an agent pool and an arbitration agent. The agents of the two systems communicate with each other through an Http service, and the agents of the centralized system communicate with each other through a shared memory. According to the method, user requirements are sensed in a distributed manner, and a universal decision framework suitable for various collaborative technologies can be designed based on a multi-agent system. And the method has relatively high practicability and effectiveness.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the interdisciplinary field of artificial intelligence and wireless communication, and particularly relates to a multi-agent multi-access point network collaborative optimization system. Background Technology

[0002] As the cornerstone of Wireless Local Area Networks (WLANs), Wi-Fi technology has evolved around three core objectives: increasing speed, reducing latency, and increasing connection density. From Wi-Fi 5 (802.11ac) to Wi-Fi 6 / 6E (802.11ax), and now to Wi-Fi 7 (802.11be), each generation of technology has addressed the explosive growth in wireless data demand by introducing more advanced modulation techniques (such as 1024-QAM to 4096-QAM), wider spectrum channels (from 80MHz to 320MHz), and multi-user access mechanisms (such as OFDMA and MU-MIMO).

[0003] However, traditional Wi-Fi network architecture is essentially "single AP autonomy." Each access point (AP) communicates independently with clients (STAs) and makes decisions based on local information (such as the AP's load and interference). This model exposes serious drawbacks in high-density deployment scenarios (such as enterprise offices, large venues, and smart homes): multiple APs cannot coordinate resources, and competition between them may lead to resource conflicts, such as co-channel interference and hidden terminals; furthermore, APs are unaware of each other's service information, resulting in unreasonable resource allocation. In modern large-scale cluster access scenarios, it is impossible to guarantee the communication needs of high-priority traffic.

[0004] To this end, the IEEE 802.11 standards organization has initiated the development of the next-generation standard—Wi-Fi 8 (IEEE 802.11bn, also known as 802.11 Ultra High Reliability). Its core paradigm shifts from "performance extraction from a single AP" to "cooperative networking of multiple APs," aiming to build multiple distributed APs into an organically unified network, thereby maximizing system-level performance. Multi-AP Coordination technology has thus become the most revolutionary feature and key research area of ​​Wi-Fi 8.

[0005] To address the complex interference environment faced by Wi-Fi 8 multi-AP coordination (MAPC), a centralized MAS architecture based on a central controller offers a promising solution for achieving efficient coordination. Compared to traditional rule-based or mathematically optimized centralized scheduling, the MAS residing in the central controller fully leverages global observability, using the network's channel state and traffic demands as joint inputs. By decomposing the complex network-level coordination problem into collaborative tasks for multiple agents, this mechanism not only effectively avoids the computational complexity explosion problem of traditional optimization algorithms when handling high-dimensional resource allocation, but also accurately achieves joint beamforming and coordinated transmission through zero-latency information sharing and collaboration among agents. This centralized AI decision-making paradigm significantly improves the system's response speed and scheduling accuracy to dynamic interference environments while ensuring optimal global performance of Wi-Fi 8.

[0006] Currently, multi-AP collaboration technology, proposed by WiFi 8, aims to improve wireless network performance through cooperation between multiple access points. However, in actual deployment, it does face challenges from various aspects, including technical implementation, protocol standards, and network deployment and maintenance.

[0007] First, existing collaborative frameworks lack uniformity and scalability. Existing multi-AP collaborative schemes exhibit significant heterogeneity in resource domains (time, space, power), required information, and optimization objectives, making it difficult to build a unified framework to dynamically select and jointly apply multiple collaborative schemes. Furthermore, the existing architecture is ill-suited to adapting to new collaborative mechanisms (such as Co-EDCA and Co-Roaming) introduced by future standards (such as Wi-Fi 8), resulting in insufficient versatility.

[0008] Second, the coupling of multi-dimensional resource parameters makes optimization difficult and complex. The parameter configurations between different collaborative mechanisms (such as Co-SR and Co-BF) are strongly coupled, and simple independent or sequential adjustments cannot achieve optimal performance. Furthermore, the computational complexity of joint optimization for power, time slots, and spatial orientation is extremely high, making it difficult to implement in real-time systems.

[0009] Third, they cannot adapt to diverse service quality requirements. Existing technologies mostly use fixed or general optimization models, which are difficult to simultaneously meet the vastly different QoS requirements in modern networks (such as the coexistence of high throughput and low latency) as well as personalized QoE requirements based on user perception and application context, and lack adaptive capabilities under multi-AP collaboration. Summary of the Invention

[0010] To address the problems existing in the prior art, this invention provides a multi-agent, multi-AP network collaborative optimization architecture.

[0011] This invention is implemented as follows: a multi-agent, multi-AP network collaborative optimization architecture, characterized in that the architecture consists of two systems: a distributed system and a centralized system. The distributed system includes AP devices, user agents, and terminal units (STAs) connected to the APs. The centralized system, MAP Coordinator, is deployed on the main AP and includes an agent coordinator, an agent pool, and an arbitration agent. The agents in the two systems communicate with each other via HTTP service, while the agents in the centralized system communicate with each other via shared memory.

[0012] Furthermore, the user intelligent agent is deployed on each AP participating in collaborative work. Every certain period, it collects data from each layer of the AP through the MCP protocol or other similar means, analyzes user needs based on the current situation, and transmits them to the centralized system using the HTTP-based A2A (Agent-to-Agent) communication protocol. User needs should follow the JSON format and should include whether each layer meets all business needs in the current period and the resource allocation required for the next period.

[0013] The collected data should include: application layer traffic changes, service priorities, latency requirements, etc.; MAC layer queue status, current cycle throughput and latency changes, current cycle rate changes, etc.; PHY layer power and RSSI changes, interference time changes, etc.

[0014] Furthermore, the intelligent agent controller, which is deployed on the main AP, is the first link in the centralized system. After receiving information from all user intelligent agents, it begins to identify suitable multi-AP collaborative tasks for this round and assigns them to the identified sub-intelligent agents.

[0015] Since each sub-agent carries its own agent profile, which includes the agent's capabilities, input / output requirements, and the multi-AP collaborative schemes it supports, the agent controller should scan all sub-agents in the agent pool before making a decision. This allows it to understand the multi-AP collaborative schemes it can use, select the scheme most likely to meet the requirements of all APs, and then call the sub-agents in the order specified by the scheme. This process includes two stages: task identification and task allocation. The former is the process of selecting a scheme, and the latter is the process of calling the sub-agents according to the input requirements of each agent.

[0016] Furthermore, the agent pool, a collection of agents deployed on the main AP and exposing call interfaces, constitutes the second ring of the centralized system; each sub-agent should adhere to the following standards when defined:

[0017] (1) Provide an Agent Card to demonstrate your capabilities and scope of application, serving as a standard pattern that the intelligent agent controller can scan;

[0018] (2) Provide input requirements and formats, as well as output formats, so that the intelligent agent can call them according to the requirements;

[0019] (3) Provide a list of all supported multi-AP collaborative schemes; We can predefine the calling order of a set of agents that can cooperate with each other, and assign them labels and applicable scenarios, which is called a multi-AP collaborative scheme; This mechanism can enhance the robustness of the agent controller and reduce the computational complexity of the agent controller.

[0020] (4) Part of the output can be saved into shared memory named after itself, making it convenient for other intelligent agents to read; it can also read the shared memory and read the data saved by other intelligent agents' protocols.

[0021] (5) It should be responsible for a specific task and should not have obvious functional overlap with other intelligent agents.

[0022] Furthermore, the arbitration agent is deployed on the main AP, receiving and evaluating the results of the multi-agent decisions. If the decision has high effectiveness and accuracy, it is summarized as the final decision and distributed to the participating APs; otherwise, an evaluation opinion is output, returned to the agent pool, and the multi-AP collaboration scheme is re-executed until the evaluation is passed or the number of retries reaches 3. Multiple failed retries will terminate the task, awaiting the next process.

[0023] Furthermore, the execution flow of this architecture is as follows:

[0024] S1: The user agent collects and analyzes AP data through MCP and transmits it to the agent controller of MAPCoordinator via HTTP service;

[0025] S2: After collecting the needs and status of all APs, the agent controller analyzes the most suitable multi-AP collaboration scheme and calls the sub-agents in the agent pool according to the calling order and input requirements.

[0026] S3: After receiving the desired information from shared memory or context, each agent in the agent pool outputs the specified parameter adjustment or allocation strategy according to the type of task it is responsible for, and saves it to shared memory;

[0027] S4: The arbitrator agent analyzes the decision results passed from the agent pool and evaluates their effectiveness. If the decision fails, a suggestion is generated, saved to shared memory, and the S3 step is repeated. Otherwise, a final decision is generated and distributed to each AP.

[0028] S5: Each AP extracts its own strategy for the next cycle from the final decision through corresponding processing and strictly executes it; after one cycle, step S1 is restarted.

[0029] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0030] (1) Based on actual multi-AP collaborative scenarios, this invention designs a scalable and general framework applicable to various technologies.

[0031] In this invention, based on the characteristics of multi-AP collaborative mechanisms and intelligent agent systems, we designed a semi-centralized, semi-distributed general framework. Each AP can define its own user intelligent agent, transmitting its needs and status to the central network based on specific standards. The central network defines an intelligent agent pool, which contains various SubAgents responsible for specific tasks. Each SubAgent corresponds to one or more collaborative technologies, and an Agent Controller is responsible for task identification and assignment, i.e., determining the type and order of SubAgent invocation. This architecture can extend to new collaborative technologies by defining new SubAgents, and can also improve performance by updating SubAgents. It has high versatility and maintainability.

[0032] (2) This invention enables the coexistence and cooperation of multiple collaborative technologies, and achieves technology integration and joint optimization.

[0033] This invention supports various collaborative technologies while also offering high flexibility and autonomy. The AgentController is responsible for task identification and allocation, intelligently deciding which technology or hybrid technology to use, while the ArbiterAgent resolves decision conflicts and provides error correction mechanisms. This mechanism allows us to use different collaborative technologies depending on the situation, achieving a cooperative effect. In the complex scenarios of modern networks, the coexistence of different technologies is becoming increasingly prominent, often requiring the design of specific mechanisms for adjustment; our invention, however, eliminates this step.

[0034] (3) The multi-agent system used in this invention can significantly improve network optimization performance.

[0035] Resource allocation and conflict avoidance among multiple APs is a complex engineering problem. This invention introduces an intelligent agent system that can improve the effectiveness of decision-making through group decision-making, which has higher advantages over single intelligent agents and traditional algorithms.

[0036] (4) This invention has user semantic awareness capabilities. It perceives user needs through distributed sensing, reports multi-user needs, and designs a general decision-making framework based on a multi-agent system that is applicable to various collaborative technologies. It has high practicality and effectiveness.

[0037] As supporting evidence of the inventiveness of the claims of this invention, the advantages and positive effects of the technical solution to be protected by this invention are also reflected in the following important aspects:

[0038] (1) The expected benefits and commercial value of the technical solution of this invention after transformation are as follows:

[0039] The architecture proposed in this invention not only significantly improves the performance of existing networks, but also has a first-mover advantage and commercial potential for the next-generation Wi-Fi 8 (IEEE 802.11bn) standard. It can be directly deployed in future high-density Wi-Fi campuses, FTTR home scenarios, and other scenarios, providing a foundation for realizing various Wi-Fi multi-AP collaborative solutions and building intelligent AP networks.

[0040] By enabling automated collaboration among intelligent agents, this system replaces the inefficient traditional network model that relies on manual surveying and static configuration (such as fixed channel / power planning). Embedded as a software feature in the main AP or cloud controller, this system allows for upgrades to existing equipment without hardware changes, providing network equipment manufacturers with high-value-added software services.

[0041] The modular design of the "intelligent body pool" makes the architecture not only compatible with existing Co-SR and Co-OFDMA, but also seamlessly connect to new technologies such as Co-BF (Co-Banding) that may be introduced in future Wi-Fi 8, extending the life cycle of network equipment and protecting the equipment investment of operators and enterprises.

[0042] (2) The technical solution of this invention fills a technical gap in the industry both domestically and internationally:

[0043] This invention fills the gap in the field of multi-AP collaboration by establishing a unified heterogeneous scheduling framework. Existing technologies typically design specific mathematical algorithms for a single optimization objective (such as optimizing power or time slots only), lacking a general architecture capable of dynamically integrating multiple collaborative mechanisms (Co-SR, Co-OFDMA, Co-Beamforming, etc.). This invention pioneers a dynamic orchestration mechanism based on "Agent Cards" and "Task Recognition," filling the architectural gap where heterogeneous collaborative algorithms cannot adaptively switch and coexist within the same network.

[0044] (3) The technical solution of the present invention overcomes technical bias:

[0045] This invention overcomes the prejudice that "distributed Wi-Fi networks cannot achieve the efficient centralized scheduling of cellular networks": For a long time, Wi-Fi has been viewed as a contention-based (CSMA / CA) best-effort network, making it difficult to achieve precise centralized control like 5G. This invention, through a hybrid architecture of "distributed sensing (User Agent) + centralized decision-making (MAP Coordinator)," demonstrates that while retaining the flexibility of Wi-Fi, deterministic QoS can be fully achieved through centralized intelligent coordination, breaking the preconceived notion that Wi-Fi technology cannot achieve fine-grained centralized coordination. Attached Figure Description

[0046] Figure 1 This is a system architecture diagram provided in an embodiment of the present invention;

[0047] Figure 2 This is the scenario described in Example 1 of the present invention;

[0048] Figure 3 These are the experimental results of Case 1 provided in the embodiments of the present invention;

[0049] Figure 4 This is scenario 2 provided in the embodiments of the present invention;

[0050] Figure 5 This is the experimental result of Case 2 provided in the embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] like Figure 1 As shown, this embodiment of the invention provides a multi-agent, multi-AP network collaborative optimization architecture. The architecture is characterized by comprising two systems: a distributed system and a centralized system. The distributed system includes AP devices, user agents, and terminal units (STAs) connected to the APs. The centralized system, MAP Coordinator, is deployed on the main AP and includes an agent coordinator, an agent pool, and an arbitration agent. Agents in the two systems communicate with each other via network communication protocols (such as HTTP), while agents in the centralized system communicate with each other via shared memory.

[0053] The user intelligent agent is deployed on each AP participating in the collaborative work. It collects data from each layer of the AP at regular intervals through the MCP protocol or other similar means, analyzes user needs based on the current situation, and transmits them to the centralized system using the A2A protocol.

[0054] The collected data should include: application layer traffic changes, service priorities, latency requirements, etc.; MAC layer queue status, throughput and latency changes in the current cycle, rate changes in the current cycle, etc.; PHY layer power and RSSI changes, interference time changes, etc.

[0055] User requirements should follow the JSON format and should include an analysis of whether each layer meets all business needs in the current cycle, as well as the resource allocation required for the next cycle.

[0056] The intelligent agent controller is a control intelligent agent deployed on the main AP. It is the first link of the centralized system. After receiving information from all user intelligent agents, it begins to identify suitable multi-AP collaborative tasks for this round and assigns them to the identified sub-intelligent agents.

[0057] Since each sub-agent carries its own agent profile, which includes the agent's capabilities, input / output requirements, and the multi-AP collaborative schemes it supports, the agent controller should scan all sub-agents in the agent pool before making a decision. This allows it to understand the multi-AP collaborative schemes it can use, select the scheme most likely to meet the requirements of all APs, and then call the sub-agents in the order specified by the scheme. This process includes two stages: task identification and task allocation. The former is the process of selecting a scheme, and the latter is the process of calling the sub-agents according to the input requirements of each agent.

[0058] The agent pool is a collection of agents deployed on the main AP and exposing a call interface, belonging to the second ring of the centralized system; each sub-agent should follow the following standards when defined:

[0059] (1) Provide an Agent Card to demonstrate your capabilities and scope of application, serving as a standard pattern that the intelligent agent controller can scan;

[0060] (2) Provide input requirements and formats, as well as output formats, so that the intelligent agent can call them according to the requirements;

[0061] (3) Provide a list of all supported multi-AP collaborative schemes; We can predefine the calling order of a set of agents that can cooperate with each other, and assign them labels and applicable scenarios, which is called a multi-AP collaborative scheme; This mechanism can reduce the robustness of the agent controller;

[0062] (4) Part of the output can be saved into shared memory named after itself, making it convenient for other intelligent agents to read; it can also read the shared memory and read the data saved by other intelligent agents' protocols.

[0063] (5) It should be responsible for a specific task and should not have obvious functional overlap with other intelligent agents.

[0064] The arbitration agent is deployed on the main AP, receives the results of the multi-agent decision and evaluates them. If the decision has high effectiveness and accuracy, it is summarized as the final decision and distributed to each AP participating in the collaboration. Otherwise, the evaluation opinion is output, returned to the agent pool, and the multi-AP collaboration scheme is re-executed until the evaluation is passed or the number of retries reaches 3.

[0065] The execution flow of this architecture is as follows:

[0066] S1: The user agent collects and analyzes AP data through MCP and transmits it to the agent controller of MAPCoordinator via HTTP service;

[0067] S2: After collecting the needs and status of all APs, the agent controller analyzes the most suitable multi-AP collaboration scheme and calls the sub-agents in the agent pool according to the calling order and input requirements.

[0068] S3: After receiving the desired information from shared memory or context, each agent in the agent pool outputs the specified parameter adjustment or allocation strategy according to the type of task it is responsible for, and saves it to shared memory;

[0069] S4: The arbitrator agent analyzes the decision results passed from the agent pool and evaluates their effectiveness. If the decision fails, a suggestion is generated, saved to shared memory, and the S3 step is repeated. Otherwise, a final decision is generated and distributed to each AP.

[0070] S5: Each AP extracts its own strategy for the next cycle from the final decision through corresponding processing and strictly executes it; after one cycle, step S1 is restarted.

[0071] The following are two implementation examples of the present invention.

[0072] Implementation Case 1

[0073] like Figure 2 The diagram illustrates a test scenario for Coordinated Resource Sharing based on traffic semantic awareness. In this scenario, two APs (AP1 and AP2) with overlapping coverage and mutual interference are set up, each experiencing bursts of high-priority traffic at different times. To simulate a real-world environment, the Nist bit error rate model and Log-distance fading channel model are used, with a channel bandwidth of 20MHz. The traffic model is set as follows: 60Mbps for high-priority services (e.g., video calls) and 60Mbps for best-effort (BE) services (e.g., file downloads).

[0074] The implementation process is described below:

[0075] Step S1: Set up the simulation scenario as required, deploy the MAP Coordinator on a cloud server or the main AP, and connect each AP through a high-speed backhaul network. Set up the simulation phases: Phase 1: AP1 carries bursty high-priority traffic, and AP2 carries BE traffic; Phase 2: swap the traffic loads to test dynamic response capabilities.

[0076] Step S2: Deploy agents at all levels and establish the communication protocol between the MAP Coordinator and the end-user agents (aAgent). Load the EDCA Agent responsible for time-domain resource management and the Arbiter Agent responsible for conflict detection into the agent pool.

[0077] Step S3: Run the simulation script to start the simulation.

[0078] Step S3 specifically includes: Each AP's end-side agent (aAgent) continuously monitors the network status. When a high-priority video stream is detected on AP1, the aAgent generates a semantic requirement of "prioritizing low latency for real-time services" through the large language model interface and reports it. After receiving the report, the Agent Controller infers the task objective as "allocating EDCA parameters to meet video traffic requirements" and calls the EDCA Agent. Based on the current queue depth, throughput, and service type, the EDCA Agent outputs optimized EDCA parameters (such as AIFSN, CWmin, etc.) to grant AP1 higher priority channel access rights. This decision, after conflict detection and approval by the Arbiter Agent, is then distributed to each AP for execution.

[0079] Step S4: Calculate the throughput and end-to-end latency for both services. The results are as follows: Figure 3 As shown in the figure, the curve changes indicate that after AgentMAPC intervenes, the throughput of APs carrying high-priority services is significantly improved and the latency is greatly reduced. Furthermore, when the traffic in phase 2 is flipped, the system can quickly identify and readjust the parameters, ensuring the QoS requirements of newly initiated critical services.

[0080] Implementation Case 2

[0081] like Figure 4The diagram shows a test scenario for Co-Spatial Reuse (Co-SR). In this scenario, three access points (APs) are arranged in an equilateral triangle, operating in the same frequency band with overlapping coverage areas. To establish a high interference baseline, all APs initially operate at maximum transmit power (16.026 dBm), resulting in severe inter-BSS interference and reduced throughput.

[0082] The implementation process is described below:

[0083] Step S1: Set up a dense network simulation scenario as required and configure the data acquisition modules for each AP. Initially, due to high-power interference, the system reverts to the aggressive EDCA contention mode.

[0084] Step S2: Deploy agents at all levels, loading the SR Agent (spatial reuse agent) responsible for power control and the Arbiter Agent responsible for performance evaluation into the agent pool of the MAP Coordinator.

[0085] Step S3: Run the simulation script to start the simulation.

[0086] Step S3 specifically includes: Each AP's aAgent detects a low throughput state caused by high interference and reports the semantic goal of "improving overall throughput". The Agent Controller decomposes the task into "using the SR mechanism to improve network throughput" and assigns it to the SR Agent. Based on a global interference view, the SR Agent no longer makes adjustments in a single dimension, but generates a set of joint power control strategies through iterative optimization (e.g., reducing the power of interference sources in exchange for parallel transmission opportunities). The ArbiterAgent evaluates the contribution of this strategy to the total network capacity, and after multiple rounds of feedback and fine-tuning, finally stabilizes the transmit power of each AP in a coordinated manner at the optimal value (e.g., around 4 dBm).

[0087] Step S4: Analyze the convergence of transmit power for each AP and the total network throughput. Results are as follows: Figure 5 As shown, the data indicates that the SR Agent successfully guided the network to a low-interference, high-reuse operating point, increasing the average throughput of a single AP from 24.90 Mbps to 37.96 Mbps, and the total network throughput from 74.69 Mbps to 113.96 Mbps, achieving a significant increase in network capacity.

[0088] Example 1: Implementation of Multi-Access Point Collaborative Optimization for High-Density Office Scenarios

[0089] In high-density office environments, multiple access points simultaneously provide services for video conferencing, file transfer, and instant messaging. User agents deployed at each access point periodically collect data on application-layer load changes, queue congestion status, and physical-layer signal quality changes. They then perform local analysis to determine if the current latency and bandwidth requirements are met, generating a unified data description which is sent to the main access point. The agent controller at the main access point, based on the received multi-access-point requirements, scans the agent pool for available coordination schemes, selects a multi-agent scheme that simultaneously supports power adjustment and load migration, and sequentially calls relevant sub-agents to generate candidate policies. An arbitration agent evaluates the overall throughput and fairness of the policy, and after confirming its validity, distributes it uniformly, enabling multiple access points to complete coordinated adjustments within the same period.

[0090] Example 2: Cooperative Scheduling Implementation for Low-Latency Services

[0091] In scenarios involving voice calls and real-time control services, the user agent prioritizes identifying queue backlogs in latency-sensitive services during analysis and explicitly indicates the need for increased scheduling priority in the next cycle in the requirement description. Based on this, the agent controller in the centralized system prioritizes combinations of sub-agents with queue scheduling and latency guarantee capabilities, transmitting intermediate results via shared memory. The arbitration agent performs end-to-end latency prediction on the scheduling results output by multiple agents; if the predicted value meets the constraints, execution is directly initiated, thus achieving low latency guarantees without relying on a single algorithm model.

[0092] Example 3: Power Coordination Implementation for Complex Interference Environments

[0093] In environments with frequently changing wireless interference, the user agent focuses on collecting the proportion of time affected by interference and the trend of signal strength changes, and analyzes whether the current interference is causing service degradation. The centralized system, based on the interference-related requirements of multiple access points, selects a collaborative scheme that includes sub-agents for interference perception and power coordination. Each sub-agent provides power adjustment suggestions and writes them into shared memory. The arbitration agent comprehensively analyzes the mutual influence relationships between different access points to avoid local optima leading to global performance degradation, thereby achieving cross-access point interference suppression collaboration.

[0094] Example 4: Implementation of Adaptive Switching Between Multiple Schemes

[0095] In scenarios where network load and service types change frequently, the agent controller rescans the agent pool before each round of collaboration, dynamically selecting the most suitable collaboration scheme based on current needs, rather than using a fixed single scheme. When the initial scheme evaluation fails, the arbitrator agent generates improvement suggestions and writes them to shared memory, triggering a re-execution of the collaboration process. This allows the second collaboration to automatically adjust the order of sub-agent calls and parameter emphasis, thus demonstrating the adaptive capability at the scheme level.

[0096] Example 5: Implementation of finite retries to ensure stability

[0097] Under extreme network conditions, multiple collaborative results may fail to achieve the expected performance. The arbitrary agent records the reason for each failed evaluation and limits the number of collaborative retries. When the preset maximum number of retries is reached, the solution with the best overall performance is directly selected from the existing candidate results and issued as the final decision. This approach avoids the system getting stuck in infinite iterations while ensuring the predictability and engineering usability of collaborative decisions.

[0098] Example 6: Implementation method based on access point granularity

[0099] During the final decision-making phase, each access point only parses its own relevant parameter configurations, such as power adjustment values, queue weights, or resource allocation ratios, without needing to understand the complete global coordination logic. Each access point strictly executes the corresponding parameters in the next scheduling cycle, and the user agent collects the results again at the end of the cycle and feeds them back to the centralized system, thus forming a closed-loop optimization process that combines distributed execution with centralized decision-making.

[0100] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0101] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-agent, multi-access point network cooperative optimization system, characterized in that, The system includes distributed systems and centralized systems. in: The distributed system is deployed across multiple access points and includes at least user agents that describe the service requirements of each access point. The centralized system is deployed at the main access point and includes at least an agent controller, an agent pool, and an arbitration agent; The user intelligent agent is used to periodically collect and analyze the multi-layer operation status of the access point, form a structured demand description, and send the demand description to the centralized system; The agent controller is used to select and sequentially call multiple sub-agents from the agent pool based on a predefined multi-agent collaboration scheme after receiving demand descriptions from multiple user agents, so as to form candidate collaborative decisions. The arbitration agent is used to evaluate the effectiveness of the candidate collaborative decisions, and generates the final collaborative decision and sends it to each access point for execution when the evaluation passes; and triggers re-collaborative calculation when the evaluation fails. In this system, the distributed system and the centralized system communicate through network services. In a centralized system, multiple agents interact with each other by sharing memory. This results in a multi-access point network collaborative optimization system composed of distributed sensing, centralized multi-agent collaborative decision-making, and arbitration feedback.

2. The system according to claim 1, characterized in that, The data collected by the user intelligent agent includes at least application layer service load changes, service priority information, and latency constraint information. The queue status, throughput changes, and transmission latency changes of the media access control layer. In addition, the physical layer's transmit power changes, receive signal strength changes, and interference duration changes.

3. The system according to claim 1, characterized in that, The user agent's output of the requirement description adopts a unified data structure format. The data structure at least represents whether each service is satisfied in the current cycle and the tendency to adjust network resources in the next cycle.

4. A centralized multi-agent scheduling system for collaborative optimization of multi-access point networks, characterized in that, The system is deployed at the main access point and includes: An intelligent agent controller is used to identify an appropriate multi-agent collaborative task after receiving a request description from multiple access points, and to determine a collaborative scheme for completing the collaborative task. The agent pool is used to store multiple sub-agents with complementary functions. Each sub-agent is associated with its capability description, input and output constraints, and the multi-agent cooperative schemes it can participate in. in, Before executing collaborative decisions, the agent controller scans the capability descriptions of each sub-agent in the agent pool. Based on the scan results, an executable multi-agent cooperative scheme is determined. And according to the sub-agent invocation order defined in the aforementioned collaborative scheme, the corresponding sub-agents are invoked sequentially to complete the collaborative decision-making. This avoids relying on a single optimization model to complete the collaborative optimization of multiple access points.

5. The system according to claim 4, characterized in that, The collaborative decision-making process of the intelligent agent controller includes a task identification stage and a task allocation stage. The task identification phase is used to determine the target solution that meets the current requirements from multiple collaborative solutions. The task allocation phase is used to invoke sub-agents according to the order defined in the target scheme and the input requirements.

6. The system according to claim 4, characterized in that, Each sub-agent provides agent description information that describes its own capabilities, input format, output format, and applicable scenarios. The agent description information serves as the basis for the agent controller to select and schedule schemes.

7. A multi-agent decision-making arbitration system for collaborative optimization of multi-access point networks, characterized in that, The system includes an arbitration agent. The arbitration agent is used to evaluate candidate decision results generated collaboratively by multiple sub-agents. Once the preset validity conditions are met, a final collaborative decision is generated and distributed to each access point for execution. If the evaluation fails to meet the preset validity conditions, improvement suggestions will be output and the multi-agent cooperative scheme will be re-executed. This improves the stability and accuracy of collaborative decision-making among multiple access points without relying on human intervention.

8. The system according to claim 7, characterized in that, When the arbitration agent fails the evaluation, it writes the evaluation opinion into shared memory. And serve as the input constraint for the next round of multi-agent collaborative computation.

9. The system according to claim 7, characterized in that, The arbitration agent limits the number of retries for the multi-agent collaborative scheme to no more than 3. Once the maximum number of retries is reached, the current best candidate decision is output as the final decision.

10. The system according to claim 7, characterized in that, After receiving the final collaborative decision, each access point extracts only the parameter configurations relevant to itself from the final collaborative decision. And network resource scheduling will be performed strictly in accordance with the parameter configuration in the next scheduling cycle.

Images