A RAN Customized Slicing Management System and Method

Abstract

The first aspect of this invention discloses a RAN customized slice management system, characterized by comprising a model training module, a control inference agent module, and a slice real-time dynamic controller, respectively deployed in a non-real-time RIC control layer, a near-real-time RIC control layer, and a real-time RIC control layer. The second aspect of this invention provides a RAN customized slice management method. This invention addresses the challenges of new applications requiring SLA customized service assurance in open radio access network architectures. It introduces parametric / model predictive control theory and machine learning theory, proposing a CSM method and system with SLA assurance. In this invention, the CSM employs a three-layer collaborative intelligent controller (RIC) based on a secure learning model. The optimization strategy generated by the RIC is used to optimize the control parameters of the RIC, jointly achieving refined secure closed-loop control of intelligent slice management to obtain QoS / QoE performance optimization under SLA security assurance and improve SLA satisfaction.

Description

Technical Field

[0001] This invention relates to a method and system for RAN Customized Slicing Management (CSM) with Service Level Agreement (SLA) guarantees, belonging to the field of communication technology. Background Technology

[0002] To meet the evolving needs of future mobile internet, industrial IoT, intelligent vehicle-to-everything (V2X) communication, and machine-to-machine (M2M) communication applications, the B5G / 6G next-generation mobile communication technology standard defines typical communication scenarios such as high capacity, low latency, high reliability, and massive connectivity. In particular, the diverse and differentiated quality of service (QoS) requirements of various services pose new challenges to the radio resource management level of B5G / 6G networks, leading to widespread attention to new network architectures and technologies such as Software-Defined Networking (SDN), Functional Virtualization (VNF), and network slicing. Network slicing, in particular, can provide dedicated, customized, and guaranteed services for various services within public communication networks. Slicing management on the radio access side is a crucial measure for RAN controllers to implement multi-service radio resource management. Recently, the Service Level Agreement (SLA) defined by the 3GPP international standard has, for the first time, provided a mutually agreed-upon framework for service level and quality standards between service providers and users, or between service providers themselves. Guaranteed communication services that meet SLA requirements will bring new opportunities for the comprehensive digital upgrade of various vertical industries. This means that future communication networks will not only need to guarantee the QoS and QoE requirements of various services, but also the isolation requirements between different services. Currently, global research on SLAs is still in its early stages, and RAN slice management technology based on SLA requirements is even more cutting-edge and challenging.

[0003] This is because simultaneously supporting multiple services and meeting diverse service SLA requirements in cellular networks places higher demands on the shared use and management of wireless resources, which existing wireless resource management technologies struggle to address. On one hand, service traffic on the wireless access network side not only exhibits high dynamic uncertainty but also possesses diverse, flexible, and time-sensitive service SLA requirements, easily exacerbating the probability of packet collisions in wireless links. On the other hand, dynamic changes in wireless channel quality, diverse and dynamic uncertainties in service load types, and varying degrees of real-time and priority requirements for latency-sensitive services pose significant challenges to dynamic security management and real-time control in public networks. For example, in applications such as secure message transmission, real-time telemedicine, and intelligent machine collaboration in vehicle-to-everything (V2X) networks, the service quality requirements of URLLC (Ultra-Reliable Low-Latency Communications) services differ significantly from those of eMBB (Extremely High-Capacity Broadband) services, often prioritizing resource scheduling. However, for multiple concurrent low-latency-sensitive services or high-real-time services, priority scheduling optimization strategies cannot avoid inter-slice interference. Furthermore, existing capacity optimization theories cannot effectively address the slice isolation problem under dynamic service disturbances. Existing research on RAN slice resource dynamic allocation optimization models based on reinforcement learning mostly only meets the differentiated QoS / QoE requirements of URLLC and eMMB services, and still does not address the isolation guarantee between different service slices. According to publicly available literature, RAN slice resource dynamic allocation control models with SLA guarantees are rarely reported. Summary of the Invention

[0004] The technical problem this invention aims to solve is that existing RAN slice resource allocation models based on capacity optimization theory often cannot effectively handle dynamic slice management that includes multiple latency-sensitive or real-time services independently.

[0005] To address the aforementioned technical problems, the first aspect of this invention discloses a RAN customized slice management system, comprising a model training module, a control inference agent module, and a slice real-time dynamic controller deployed on a non-real-time RIC control layer, a near-real-time RIC control layer, and a real-time RIC control layer, respectively. The control inference agent module performs inference based on the training results obtained from offline training at multiple time scales obtained by the model training module, generating a set of control parameter optimization instructions. The slice real-time dynamic controller utilizes the principle of parametric / model predictive control to solve a finite-time domain optimization problem with customized requirements as the optimization objective and slice buffer behavior dynamic equations and service level protocol guarantees as joint constraints. This yields an optimized control sequence for real-time inter-slice RB resource allocation, achieving real-time slice allocation control corresponding to the current customized orchestration service.

[0006] Preferably, the control inference agent module and the slice real-time dynamic controller constitute a slice security learning wireless intelligent controller, which performs fine-grained security closed-loop control on network slice resources according to the customized management requirements of intelligent slices, so as to optimize the service quality and / or user experience quality performance of different application service orchestrations under the service level protocol security guarantee.

[0007] Preferably, the control inference agent module employs a multi-agent reinforcement learning method: based on the current network slice access status, the control inference agent module configures a unique agent for each network slice in the near real-time RIC control layer and executes an inference output control parameter optimization strategy; in the non-real-time RIC control layer, the model training module uses a policy update method of a constrained multi-agent reinforcement learning model to optimize and update the parameters of the agents configured in the current network slice using a multi-agent constraint strategy.

[0008] Preferably, the control parameter optimization instruction set includes: parameter H, representing the time range for the parameterized / model predictive controller to predict the system state; parameter H c , represents the control time step within the prediction time range; parameter D, represents the system dynamic equation parameter of the slice buffer, which is the estimated RB resources required for new services arriving within the estimated prediction time range; parameter Req, represents the slice resource request instruction; parameter Rel, represents the slice release instruction.

[0009] Preferably, the finite-time optimization problem is expressed as: In the formula: t is the current time, T is the slice lifetime, c1, c2, and c3 represent constraints; x(t) is the state variable of the time t slice buffer, which represents the RB resources required for data packets arriving at different times as observed in the time t slice buffer, and is also expressed as Where i is the remaining delay and s is the slice set; u(t) is the action variable of time t on the slice buffer, which is the RB resource allocated in the slice buffer at time t for data packets arriving at different times, and also represents the resource allocation result between slices. l(x(t),u(t)) is the evaluation function for taking action u(t) in the current slice buffer state; maxJ(x,u) is the unified representation of the optimization objective; constraint c1 is the dynamic equation of the slice buffer, further expressed as: In the formula, The RB resources required by newly arriving data packets in the time slice set s of time t, d s It represents the business latency requirements in the slice set s; constraint c2 is the upper and lower bound constraints of u(t). , These are the upper and lower bounds of u(t), respectively; constraint c3 is the upper and lower bound constraint of x(t). , Let x(t) be the upper and lower bounds, respectively.

[0010] Preferably, the parameters of the finite-time optimization problem are intelligently configured according to the control parameter optimization instruction set, wherein parameter H is used to configure the slice lifecycle T, and parameter H c Parameter D is used to configure the range of control action time selection. The parameters Req and Rel are used to configure the slice set s.

[0011] The second aspect of the technical solution of this invention is to provide a RAN customized slice management method, which, based on the above-mentioned RAN customized slice management system, implements a slice resource allocation strategy through the following steps:

[0012] Step 1: Initialize time t=0 and slice lifecycle T, and update the service level protocol configuration based on customized requirements;

[0013] Step 2: The non-real-time RIC control layer builds a data warehouse based on the collected data and uses the model training module to train the model offline.

[0014] Step 3: The current control inference agent module takes the current network state and customized requirements as input to infer and generate a set of control parameter optimization instructions;

[0015] Step 4: The current slice real-time dynamic controller performs real-time collaborative dynamic control of the slice buffer based on the control parameter optimization instruction set, generating an optimized control sequence for RB resource allocation between slices, and controlling the current RB resource allocation quantity of each slice.

[0016] Step 5: The slice scheduler in the centralized unit or distributed unit further schedules RB resources in the corresponding slice for the user based on the slice type to which the user service belongs;

[0017] Step 6 、 If the current time t has not reached the slice lifecycle T, then update t to t+1 and return to step 2. Otherwise, determine whether to end slice management. If yes, end the entire method. Otherwise, return to step 1.

[0018] Preferably, in step 2, the non-real-time RIC control layer collects measurement data from the near-real-time RIC control layer, the real-time RIC control layer, and the E2 node, as well as data transmitted by the network slice subnet management function in the service management and orchestration framework, and uses the data analysis function module to unify the data types of different interfaces to build a data warehouse.

[0019] Preferably, in step 3, if the current time t does not meet the control update time granularity corresponding to the near real-time RIC control layer, the control inference agent module updates the control parameter optimization instruction set after updating the offline training results output by the non-real-time RIC control layer.

[0020] Preferably, in step 4, if the current time t meets the control update time granularity corresponding to the real-time RIC control layer, then after updating the parameter configuration of the slice real-time dynamic controller according to the control parameter optimization instruction set, real-time collaborative dynamic control is performed.

[0021] This invention addresses the challenges of new applications requiring customized Service Level Agreements (SLAs) assurance in Open RAN architectures. It introduces Parametric / Model Predictive Control (PMPC) and machine learning theories to propose a CSM (Content Management System) method and system with SLA guarantees. In this invention, the CSM employs a three-layer collaborative intelligent controller (RIC) based on a secure learning model. The optimization strategy generated by the RIC is used to optimize the control parameters of the intelligent slice controller, jointly achieving refined secure closed-loop control of intelligent slice management. This results in QoS / QoE performance optimization under SLA security assurance, improving SLA satisfaction (SSR).

[0022] The advantages of the RAN-customized slice management method and system with SLA guarantees proposed in this invention compared to existing technologies are as follows:

[0023] (1) Based on the standard-defined open interface, it is well compatible with existing standard open RAN architectures, such as CIS-RAN and O-RAN;

[0024] (2) It provides a specific reference implementation model for slice intelligent coordination management based on security learning control, and realizes real-time fine control of slice management in less than 10 ms. It can provide QoS / QoE required for customized services with SLA guarantee for user applications. It is more suitable for application scenarios such as advanced automation and unmanned operation in vertical industries.

[0025] (3) By decoupling the time domain and coordinating the functions of non-real-time, near-real-time and real-time three-level wireless RICs, the complexity brought about by real-time training and learning is avoided, and the dynamic stability problem of RAN slice real-time control caused by dynamic complexity such as multiple application targets, heterogeneous multiple services and changes in access network load is effectively solved, and interference between slices is suppressed.

[0026] (4) Supports efficient access to application-customized orchestration services by SMO (Service Management and Orchestration). These advantages are conducive to the formation of a new model of customizable, programmable and quality-assured service orchestration services between mobile network operators (MNO) and service providers (SP), improving the efficiency of network resource utilization and the level of open sharing, with broad application prospects. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the RAN customized slice management method and system with SLA guarantee of the present invention, wherein, Figure 1 The RU, PHY-Low, RF, O1, E2, and UE shown are strictly defined English abbreviations in the 3GPP standard, and will not be elaborated here. In addition, nonRT-RIC means non-real-time RIC control layer, nearRT-RIC means near real-time RIC control layer, RT-RIC means real-time RIC control layer, SL-RIC means slice security learning wireless intelligent controller, Database means data warehouse, DAF means data analysis function module, and NSSMF means network slice subnet management function.

[0028] Figure 2 This is a system workflow diagram of the present invention;

[0029] Figure 3 The diagram illustrates the comparison results of the three slice comprehensive performance radar charts obtained by the present invention and existing methods in the customized orchestration business application scenario exemplified in Table 1. Detailed Implementation

[0030] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0031] This invention employs a three-layer collaborative intelligent controller (RIC) architecture based on a security learning model to implement a novel method and system for customized RAN slice management with SLA guarantees. Specifically, as follows: Figure 1As shown, the first aspect of this invention discloses a RAN customized slice management system, including a model training module, a control inference agent module, and a slice real-time dynamic controller deployed in a non-real-time RIC control layer, a near-real-time RIC control layer, and a real-time RIC control layer, respectively. This enables intelligent collaboration of offline training, online inference, and real-time control across multiple time scales. The control inference agent module of the near-real-time RIC control layer and the slice real-time dynamic controller of the real-time RIC control layer constitute a slice security learning wireless intelligent controller. Based on the customized intelligent slice management requirements, it performs fine-grained security closed-loop control on slice resources to achieve QoS / QoE performance optimization of different application service orchestrations under SLA security assurance.

[0032] The control inference agent module employs machine learning (ML) or reinforcement learning (RL) methods to perform inference based on the model training results of the non-real-time RIC control layer. It generates a set of control parameter optimization instructions for parameterized / model predictive control (PMPC) and outputs them to the sliced ​​real-time dynamic controller through the new interface rtE2 between the near-real-time RIC control layer and the real-time RIC control layer. This is used to update the real-time control of the real-time RIC control layer.

[0033] In a preferred embodiment of this invention, the control inference agent module employs a multi-agent RL method:

[0034] Based on the current network slice access status, a unique agent (abbreviated as "Actor") is configured for each slice in the near real-time RIC control layer to execute inference output control parameter optimization strategies;

[0035] In the non-real-time RIC control layer, the model training module adopts the policy update method of constrained multi-agent reinforcement learning model, namely Multi-Agent Constrained Policy Optimization (MACPO), to update the parameters of slice actors.

[0036] In another preferred embodiment of the present invention, the transmission method of the new interface rtE2 is similar to that of the existing E2 interface. The control parameter optimization instruction set is transmitted from the near real-time RIC control layer to the real-time RIC control layer through the rtE2 interface after the RIC-Control field in the extended E2SM message body is encapsulated by the E2AP protocol layer.

[0037] In another preferred embodiment of the present invention, the control parameter optimization instruction set includes:

[0038] The parameter H represents the time range within which the PMPC controller predicts the system state.

[0039] Parameter H c This represents controlling the time step within the predicted time range;

[0040] Parameter D represents the system dynamic equation parameter of the slice buffer, which is the estimated RB resources required for the arrival of new services within the estimated prediction time range;

[0041] The parameter Req represents the slice resource request command;

[0042] The parameter Rel represents the slice release command.

[0043] The slice real-time dynamic controller utilizes a PMPC principle. Its working mechanism is based on optimizing the instruction set of control parameters. By solving a finite-time domain optimization problem with customized requirements such as QoS / QoE performance as optimization objectives and slice buffer behavior dynamic equations and SLA guarantees as joint constraints, it obtains a real-time optimal control sequence for inter-slice resource allocation, namely the optimized control sequence for inter-slice RB resource allocation, thus realizing real-time slice allocation control corresponding to the current customized orchestration service. The slice resource intra-slice scheduler of CU (centralized unit) / DU (distributed unit) further schedules the corresponding RB resources within the current slice for the user.

[0044] In a preferred embodiment of the present invention, the finite-time optimization problem is expressed as follows:

[0045]

[0046] In the formula: t is the current time, T is the slice lifecycle, and c1, c2, and c3 represent constraints;

[0047] x(t) is the state variable of the time-t slice buffer. This state variable represents the RB resources required by data packets arriving at different times as observed in the time-t slice buffer, and is also expressed as... ;

[0048] u(t) is the action variable of time t on the slice buffer. This action variable is the RB resources allocated in the slice buffer at time t for data packets arriving at different times, and it also represents the resource allocation result between slices. ;

[0049] l(x(t), u(t)) is the evaluation function of action u(t) in the current slice buffer state, which evaluates the QoS / QoE performance of slice users such as latency, packet loss rate, and throughput, as well as QoE performance, SLA satisfaction, and isolation between slices.

[0050] J(x,u) is a unified representation of the optimization objective;

[0051] Constraint c1 is the dynamic equation of the slice buffer, which can be further expressed as:

[0052]

[0053] In the formula, It refers to the RB resources required for data packets with a remaining delay of i in the time slice set s of time t. It refers to the RB resources allocated to data packets with a remaining delay of i in the time slice set s of time t. The RB resources required by newly arriving data packets in the time slice set s of time t, d s It refers to the business latency requirements in the slice set s;

[0054] Constraint c2 is the upper and lower bound constraint of u(t). , Let be the upper and lower bounds of u(t), respectively;

[0055] Constraint c3 is the upper and lower bound constraint of x(t). , Let x(t) be the upper and lower bounds, respectively.

[0056] SLA requirements are used as finite domain constraints c2 and c3 to obtain a real-time control strategy for sliced ​​resources with SLA guarantees, and to ensure that the real-time dynamic control of the PMPC controller is maintained under safe operating conditions.

[0057] In another preferred embodiment of the present invention, the parameters of the aforementioned finite-time optimization problem can be intelligently configured according to a control parameter optimization instruction set, wherein parameter H is used to configure the slice lifecycle T, and parameter H c Parameter D is used to configure the range of control action time selection. The parameters Req and Rel are used to configure the slice set s.

[0058] In another preferred embodiment of the present invention, the solution method for the above-mentioned finite-time optimization problem is the branch and bound method, the simplification method, the cutting plane method, the implicit enumeration method, the Hungarian method, the enumeration method, or a heuristic method.

[0059] like Figure 2 As shown, a second aspect of this invention is to provide a RAN customized slice management method based on the above-described RAN customized slice management system, which implements a slice resource allocation strategy using the following steps:

[0060] Step S1, initialization time t=0 and slice lifecycle T, update SLA configuration and other customized requirements.

[0061] Step S2: The non-real-time RIC control layer collects measurement data from the near-real-time RIC control layer, the real-time RIC control layer, and the E2 node, as well as data transmitted by the network slice subnet management function in the SMO framework. The data analysis function module unifies the data types of different interfaces to build a data warehouse, and the model training module performs offline training of the model. The data includes, but is not limited to, service traffic, service type, service SLA requirements, service QoS / QoE requirements, channel information, etc.

[0062] Step S3: Determine whether the current time t meets the control update time granularity corresponding to the near real-time RIC control layer. If it does, proceed to step S4; otherwise, proceed to step S5.

[0063] Step S4: After the control inference agent module updates the offline training results output by the non-real-time RIC control layer, proceed to step S5.

[0064] In one preferred embodiment of the present invention, the offline training results include the weight matrix optimized by the neural network, the agent optimization configuration parameters, etc.

[0065] Step S5: The current control inference agent module operates on the current network status collected by the E2 interface and the SLA requirements of SMO transmission, and infers to generate a control parameter optimization instruction set for PMPC.

[0066] Step S6: Determine whether the current time t meets the control update time granularity corresponding to the real-time RIC control layer. If it does, proceed to step S7; otherwise, proceed to step S8.

[0067] Step S7: Update the parameter configuration of the real-time dynamic controller of the slice according to the control parameter optimization instruction set, and proceed to step S8.

[0068] In a preferred embodiment of the present invention, the parameter H in the instruction set is optimized using control parameters. c To update and optimize the parameter configuration of the slice real-time dynamic controller.

[0069] Step S8: The current slice real-time dynamic controller optimizes the instruction set based on the control parameters learned by the current control inference agent module and performs real-time collaborative dynamic control of the slice buffer according to the PMPC principle. It generates an optimized control sequence for RB resource allocation between slices, controls the current RB resource allocation quantity of each slice, and can meet the customized performance requirements of each slice for QoS / QoE under SLA guarantee.

[0070] In one preferred embodiment of the present invention, the slice real-time dynamic controller performs real-time collaborative dynamic control based on parameters D, Req, and Rel in the control parameter optimization instruction set.

[0071] In step S9, at the Media Access Control (MAC) layer of the DU, the slice scheduler further schedules RB resources within the corresponding slice for the user based on the slice type to which the user service belongs.

[0072] In a preferred embodiment of the present invention, the intra-slice scheduler used is a proportional fair scheduler, a round-robin scheduler, or a max-min scheduler.

[0073] Step S10: If the current slice management lifecycle has not ended, update t to t+1 and return to step S2; otherwise, determine whether slice management has ended. If yes, end the entire method; otherwise, return to step S1.

[0074] Table 1 below illustrates the SLA requirements for customized business orchestration.

[0075] Table 1. Case Studies of SLA Requirements for Customized Businesses

[0076]

[0077] As shown in Table 1 above, the SMO of the open RAN network has customized the slice SLA guarantee requirements for a certain application orchestration service. For example, the three slices have customized QoS requirements such as different priorities, latency, and throughput. Among them, there are two latency-sensitive URLLC slices of 10ms and 5ms and one non-latency-sensitive high-traffic eMBB slice. For details, please refer to the configuration of the 3GPP standard.

[0078] Figure 3 The comparison results of the three slice comprehensive performance radar charts obtained by the present invention and existing methods in the customized orchestration business application scenario shown in Table 1 are as follows: The specific advantages are analyzed below:

[0079] According to the SLA guarantee requirements of mixed service orchestration in Table 1, the ULLRC latency sensitivity requirement is slice 1 < slice 2, while the priority requirement is slice 1 > slice 2. Slice 3 requires high traffic and is not latency sensitive. Radar chart comparison results: (1) The existing method (purple) cannot guarantee the SLA satisfaction (SSR) and throughput performance required for slice 2, but can only guarantee the high-priority slice 1. This shows that the existing priority allocation strategy may affect applications with relatively low priority but higher real-time requirements, and is not suitable for flexible demand customization; (the priority allocation strategy often assumes that high real-time performance means high priority, but the 3GPP standard defines the requirements of various applications, which is not necessarily the case). The present invention (blue) can guarantee all the requirements of slice 1 and slice 2 at the same time. (2) At the same time, although the throughput of slice 3 of the present invention is slightly lower than that of the existing method, its SLA isolation performance can still be guaranteed. The SLA isolation performance requires that the throughput and other QoS performance will not be violated by other services. Therefore, the above results show that after ensuring the throughput of slice 3 required for the customization in Table 1, the present invention does not allocate more resources to slice 3 as in the existing methods, thus avoiding excessive waste and achieving higher slice allocation efficiency.

[0080] In summary, while increasing throughput can be achieved through various mature physical layer technologies, such as high-frequency bands, massive MIMO, and beamforming, few can flexibly support multiple mixed latency slices, especially those containing real-time slices with latency <10ms. As can be seen from the comparison, this invention can flexibly customize the requirements for different slice isolation, QoS, and priorities.

Claims

1. A RAN customized slicing management system, characterized in that, A three-layer collaborative intelligent controller (RIC) based on a secure learning model is adopted, comprising a model training module, a control inference agent module, and a slice real-time dynamic controller deployed on the non-real-time RIC control layer, near-real-time RIC control layer, and real-time RIC control layer, respectively. Specifically: the control inference agent module of the near-real-time RIC control layer infers from the training results obtained by the model training module of the non-real-time RIC control layer across multiple time scales, generating a control parameter optimization instruction set. This instruction set is used to intelligently configure the parameters of the finite-time domain optimization problem solved by the slice real-time dynamic controller. The intelligent configuration includes constraints on the slice buffer behavior dynamic equation and SLA requirement constraints, ensuring that the real-time dynamic control of the slice real-time dynamic controller remains under safe operating conditions. The slice real-time dynamic controller of the real-time RIC control layer utilizes parametric / model predictive control principles to solve a finite-time domain optimization problem with customized requirements as the optimization objective and the slice buffer behavior dynamic equation and service level agreement guarantees as joint constraints. This yields an optimized control sequence for real-time inter-slice RB resource allocation, achieving real-time slice allocation control corresponding to the current customized orchestration service. The control inference agent module and the slice real-time dynamic controller constitute the slice safety learning intelligent controller SL-RIC. The control inference agent module of the near real-time RIC control layer adopts machine learning or reinforcement learning methods to infer based on the model training results of the non-real-time RIC control layer, generate a set of control parameter optimization instructions for parameterized / model predictive control, and output it to the slice real-time dynamic controller through the new interface rtE2 between the near real-time RIC control layer and the real-time RIC control layer to update the real-time control of the real-time RIC control layer. The control parameter optimization instruction set includes: parameter H, which represents the time range for the parameterized / model predictive controller to predict the system state; parameter H c , represents the time step size controlled within the predicted time range; parameter D, represents the system dynamic equation parameters of the slice buffer, which is the estimated RB resources required for new services arriving within the estimated predicted time range; parameter Req, represents the slice resource request instruction; parameter Rel, represents the slice release instruction; The parameters of the finite-time optimization problem are intelligently configured according to the control parameter optimization instruction set, wherein parameter H is used to configure the slice lifecycle T, and parameter H c Parameter D is used to configure the range of control action time selection. The parameters Req and Rel are used to configure the slice set s, where, It refers to the RB resources required by newly arriving data packets in the time slice set s of time t.

2. The RAN customized slice management system as described in claim 1, characterized in that, The control inference agent module employs a multi-agent reinforcement learning method: based on the current network slice access status, the control inference agent module configures a unique agent for each network slice in the near real-time RIC control layer and executes an inference output control parameter optimization instruction set; in the non-real-time RIC control layer, the model training module updates the parameters of the agent configured in the current network slice using a policy update method of a constrained multi-agent reinforcement learning model.

3. The RAN customized slice management system as described in claim 1, characterized in that, The finite-time optimization problem is expressed as: In the formula: t is the current time, T is the slice lifetime, c1, c2, and c3 represent constraints; x(t) is the state variable of the time t slice buffer, which represents the RB resources required for data packets arriving at different times as observed in the time t slice buffer, and is also expressed as Where i is the remaining delay and s is the slice set; u(t) is the action variable of time t on the slice buffer, which is the RB resource allocated in the slice buffer at time t for data packets arriving at different times, and also represents the resource allocation result between slices. l(x(t),u(t)) is the evaluation function for taking action u(t) in the current slice buffer state; maxJ(x,u) is the unified representation of the optimization objective; constraint c1 is the dynamic equation of the slice buffer, further expressed as: In the formula, The RB resources required by newly arriving data packets in the time slice set s of time t, d s It represents the business latency requirements in the slice set s; constraint c2 is the upper and lower bound constraints of u(t). , These are the upper and lower bounds of u(t), respectively; constraint c3 is the upper and lower bound constraint of x(t). , Let x(t) be the upper and lower bounds, respectively.

4. A RAN customized slice management method, characterized in that, Based on the RAN customized slice management system of claim 1, the slice resource allocation strategy is implemented using the following steps: Step 1: Initialize time t=0 and slice lifecycle T, and update the service level protocol configuration based on customized requirements; Step 2: The non-real-time RIC control layer constructs a data warehouse based on the collected data and performs offline training of the model using the model training module; Step 3: The current control inference agent module uses the current network state and customized requirements as input to infer and generate a set of control parameter optimization instructions; Step 4: The current slice real-time dynamic controller performs real-time collaborative dynamic control of the slice buffer based on the set of control parameter optimization instructions, generating an optimized control sequence for RB resource allocation between slices, and controlling the current RB resource allocation quantity of each slice; Step 5: The slice scheduler in the centralized unit or distributed unit further schedules RB resources in the corresponding slice for the user according to the slice type to which the user service belongs; Step 6: 、 If the current time t has not reached the slice lifecycle T, then update t to t+1 and return to step 2. Otherwise, determine whether to end slice management. If yes, end the entire method. Otherwise, return to step 1.

5. The RAN customized slice management method as described in claim 4, characterized in that, In step 2, the non-real-time RIC control layer collects measurement data from the near-real-time RIC control layer, the real-time RIC control layer, and the E2 node, as well as data transmitted by the network slice subnet management function in the service management and orchestration framework. The data analysis function module is used to unify the data types of different interfaces to build a data warehouse.

6. The RAN customized slice management method as described in claim 4, characterized in that, In step 3, if the current time t does not meet the control update time granularity corresponding to the near real-time RIC control layer, the control inference agent module updates the control parameters based on the offline training results output by the non-real-time RIC control layer and then generates the control parameter optimization instruction set.

7. The RAN customized slice management method as described in claim 4, characterized in that, In step 4, if the current time t meets the control update time granularity corresponding to the real-time RIC control layer, then after updating the parameter configuration of the slice real-time dynamic controller according to the control parameter optimization instruction set, real-time collaborative dynamic control is performed.

Images