CN113507729A - RAN side network slice management system and method based on artificial intelligence - Google Patents
RAN side network slice management system and method based on artificial intelligence Download PDFInfo
- Publication number
- CN113507729A CN113507729A CN202111060374.9A CN202111060374A CN113507729A CN 113507729 A CN113507729 A CN 113507729A CN 202111060374 A CN202111060374 A CN 202111060374A CN 113507729 A CN113507729 A CN 113507729A
- Authority
- CN
- China
- Prior art keywords
- base station
- slice
- slicing
- algorithm
- state information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
Abstract
The invention discloses an artificial intelligence based RAN side network slice management system and method, the system comprises an AI slice algorithm platform and a plurality of base stations; the AI slicing algorithm platform provides an AI slicing algorithm and initiates and terminates the AI slicing function, the base station side realizes the interaction between the base station and the AI slicing algorithm platform and the collection and report of the state information of the base station through an AI Adapter module, the AI slicing algorithm realizes a network resource slicing allocation scheme, and the base station executes network resource scheduling based on AI slicing. The invention realizes RAN side network slicing based on AI without adding hardware equipment of the existing base station, supports various artificial intelligence algorithms and is compatible with the traditional old base station which does not support AI slicing, thereby being beneficial to the rapid deployment and maintenance of the network; meanwhile, the AI slice algorithm platform is connected with the base stations, so that the wireless resources of the base stations can be managed and optimized in a combined manner, and the overall utilization rate of network resources and the user experience are improved on the basis of fully utilizing the artificial intelligence algorithm.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to an artificial intelligence based RAN side network slice management system and method.
Background
With the rapid promotion of 5G commercial, various novel services such as high-definition video, intelligent security, automatic driving and other application scenes continuously emerge, and the demands on the network are more differentiated. The service scenarios of 5G are mainly classified into three major categories, namely mobile broadband enhancement (eMBB), large-scale machine communication (mtc) and ultra-high-reliability ultra-low-delay communication (urrllc). The eMBB scene mainly aims at the requirement of high speed, such as real-time video conference, high-speed downloading and the like; the mMTC meets the networking requirements of a large number of Internet of things devices in a large-scale connection device scene, such as an industrial scene; the uRLLC scenario addresses the needs of low latency and high reliability, such as remote driving. The three major types of scene services have different technical requirements on 5G networks. As one of the core technologies of 5G, the network slicing technology can divide a same set of physical networks into a plurality of logically isolated virtual networks, and provide targeted services for different services in different virtual networks. Operators can create different networks according to the service and application scenarios of users, including delay, rate, connection density, spectrum efficiency, traffic capacity, network efficiency and the like, by only deploying one set of hardware network equipment, so that the requirements of different services and application scenarios on network performance are guaranteed. Due to the mutual isolation of the virtual networks, the fault, congestion and configuration adjustment of any one virtual network cannot affect other virtual networks. Therefore, the network slicing technology can improve the user experience satisfaction degree and the network resource utilization rate.
Network slicing is an end-to-end technology, and relates to a radio access network, a core network, a bearer network and the like. In the latest release specification release R16 of 3GPP, the specification related to network slicing is more directed to slice management and orchestration on the core network side. In the R17 specification being set, network slices on the ran (radio Access network) radio Access network side are still currently under investigation.
Due to the rapid development of artificial intelligence, particularly deep learning and reinforcement learning, the capability of human beings and even beyond human beings is already provided in many complex application scenes. Currently, many RAN-side network slicing algorithms based on artificial intelligence exist. The publication No. CN112600695A, "RAN side network slice resource allocation method, device, and electronic apparatus", and the publication No. CN 112543508 a, "5G network slice oriented wireless resource allocation method and network architecture", all propose artificial intelligence based wireless access network slice methods. But the current network slicing method based on artificial intelligence is difficult to realize on the real wireless access network base station equipment products. The main reasons are: 1) the network slicing method based on artificial intelligence does not consider the actual architecture and software and hardware system of the wireless access network base station equipment. The current base station equipment usually adopts a special hardware system and does not have an AI acceleration hardware module which is highly dependent on an artificial intelligence algorithm. Meanwhile, the software and hardware system of the base station equipment is not used for supporting the artificial intelligence algorithm due to the limitation of cost and power consumption. The energy consumption of the base station at the present stage has brought huge economic and social responsibility pressure to operators, and it is difficult to increase the energy consumption by expanding hardware equipment in the future to gain extra computing power. 2) The conventional artificial intelligent network slicing algorithm only aims at single base station network resources and does not consider performing combined management and optimization on the network resources of multiple base stations.
Therefore, how to design and effectively apply various AI slicing algorithms with superior performance to a real base station product becomes a technical problem to be solved urgently.
Disclosure of Invention
The invention aims to provide an artificial intelligence based RAN side network slice management system and method aiming at the defects of the prior art, an AI slice algorithm platform provides an AI slice algorithm and initiates and terminates an AI slice function, a base station side realizes the interaction between a base station and the AI slice algorithm platform and the collection and report of the state information of the base station through an AI Adapter, the AI slice algorithm realizes a network resource slice allocation scheme, and the base station executes network resource scheduling based on AI slices. On the basis of not increasing hardware equipment of the existing base station, the invention realizes RAN side network slicing based on AI, is compatible with the traditional old base station which does not support AI slicing while supporting various artificial intelligence algorithms, and is beneficial to rapid deployment and maintenance of the network; meanwhile, the AI slice algorithm platform is connected with the base stations, so that the wireless resources of the base stations can be managed and optimized in a combined manner, and the overall utilization rate of network resources and the user experience are improved on the basis of fully utilizing the bonus of the artificial intelligence algorithm.
The purpose of the invention is realized by the following technical scheme:
an artificial intelligence-based RAN side network slice management system comprises an AI slice algorithm platform and a plurality of base stations;
the base station comprises an AI Adapter module, the AI Adapter module is used for mutual communication between the base station and the AI slicing algorithm platform, and comprises the steps of collecting state information of the base station, outputting the result of the AI slicing algorithm to a base station protocol stack, and realizing the resource scheduling of a wireless air interface of the base station;
the AI slicing algorithm platform is used for receiving state information sent by the base station, respectively creating a docker virtual environment for each base station supporting the AI network slices, and isolating the docker virtual environment from AI slicing algorithm environments of other base stations, wherein an AI slicing algorithm is operated in each docker virtual environment; the AI slicing algorithm platform feeds back the state information to the AI slicing algorithm corresponding to each base station, the resource allocation is carried out on the network slices by the algorithm, and the allocation result is sent to the base stations by the AI slicing algorithm platform as network slice indication information; the AI slice algorithm platform can also directly send an AI slice termination request to the base station, or feed back an exception according to the AI slice algorithm, so that the base station is restored to a non-AI slice resource scheduling mode.
Further, the AI slicing algorithm platform is a physical device or a virtual device with computing capability and network communication capability.
A RAN side network slice management method based on artificial intelligence is realized based on the management system, and the method comprises the following steps:
s1: the AI slicing algorithm platform initiates an AI slicing request message to the base station;
s2: the base station receives and processes the AI slice request message and sends the supported AI slice capability to the AI slice algorithm platform;
s3: after receiving AI slicing capability information of a base station, the AI slicing algorithm platform creates a docker virtual environment for each base station supporting AI network slices, starts an AI slicing algorithm in the docker virtual environment, and sends an AI slice starting message to the base station;
s4: after receiving the AI slice starting message, the base station periodically collects the state information of the base station and sends the state information to the AI slice algorithm platform;
s5: the AI slicing algorithm platform receives the state information sent by the base station and sends the state information to the corresponding AI slicing algorithm, the AI slicing algorithm carries out resource allocation on the network slices, and the allocation result is used as the network slice indication information of the AI slicing algorithm platform and sent to the corresponding base station;
s6: after receiving the network slice indication information, the base station executes network slice resource scheduling according to the network slice indication information and continuously and periodically reports state information to the AI slice algorithm platform;
when the terminal is required to be terminated, the AI slice algorithm platform sends an AI slice termination request to the base station, and the base station restores to the original non-AI slice resource scheduling mode.
Further, the AI slice capability message includes: AI supporting/rejecting flag bit, maximum supported slice number SliceNum _ max, slice type flag, slice attribute parameter corresponding to each slice, maximum terminal number supported by each slice, maximum resource number used for slicing algorithm, supported state information statistical minimum period Ts _ min, supported state information reporting minimum period Tr _ min and resource information of neighboring base station;
when the base station does not support AI network slices, setting an AI support/rejection flag bit as a rejection flag, simultaneously setting the supported maximum slice number SliceNum _ max to be 0, and not carrying the rest parameters; when the base station supports the AI network slice, the AI supporting/rejecting zone bit is set as the supporting zone, and other parameters are set according to the self supporting capability and the resource information of the adjacent base station.
Further, the S3 specifically includes:
(1) an AI slice algorithm platform receives AI slice capability information of a base station in the validity period of a timer Treq, verifies the AI slice capability information, creates a docker virtual environment for each base station supporting AI network slices, and operates an AI slice algorithm corresponding to the base station in each docker virtual environment;
(2) the AI slicing algorithm platform verifies whether the adjacent cell base station is communicated with the AI slicing algorithm platform or not according to the resource information of the adjacent cell base station carried in the AI slicing capability message; if the adjacent cell base station is communicated with the AI slice algorithm platform, performing joint management on the resources of the adjacent cell base station during network slice allocation; if the connection is not available, the adjacent base station is considered to be unavailable, and the resources of the adjacent base station are not subjected to joint management;
(3) the AI slicing algorithm platform initializes the AI slicing algorithm in the docker virtual environment and sends an AI slicing start message to the base station after the initialization is completed.
Further, the AI slice start message includes a slice number SliceNum, a state information list, a state information counting period Ts, and a state information reporting period Tr.
Further, the S4 specifically includes:
(1) after receiving an AI slice starting message, an AI Adapter module of the base station firstly verifies and corrects the AI slice starting message, namely when SliceNum is less than or equal to SliceNum _ max, Tr is more than Ts, Ts is more than or equal to Ts _ min, and Tr is more than or equal to Tr _ min, all parameters keep original values; when SliceNum is larger than SliceNum _ max, the SliceNum takes SliceNum _ max as the standard during slicing; when Ts is less than Ts _ min, the Ts is reset to Ts _ min; when Tr < Tr _ min, Tr is reset to Tr _ min; when Tr is less than or equal to Ts, resetting Tr = Ts + time for collecting data and uploading the data;
(2) the AI Adapter module periodically collects the self state information and sends the state information to the AI slicing algorithm platform within the Tr time; if the AI slice termination flag in the state information is true, it indicates that an abnormal scenario occurs in the protocol stack, and in order to ensure that normal service scheduling is not affected, the base station adopts a non-AI network slice scheduling mode.
Further, the S5 specifically includes:
(1) after receiving the state information sent by the base station, the AI slicing algorithm platform takes the state information as the input of a corresponding AI slicing algorithm, runs an algorithm model, and calculates the output of the slicing algorithm, namely a specific network resource slicing allocation scheme;
(2) the AI slicing algorithm platform issues the network resource slicing allocation scheme as a slicing indication message to the corresponding base station; the slice indication message includes: the effective time Tsclce, the slicing scheme information, the state information statistical period Tsn and the state information reporting period Trn of the current slicing scheme are the same as the corresponding values in the previous slicing indication message, and the effective time Tsclce, the state information statistical period Tsn and the state information reporting period Trn of the current slicing scheme are not carried in the current slicing indication message;
the slicing scheme information comprises specific slicing number and size of resources distributed in each slice;
the state information statistical period Tsn and the state information reporting period Trn are used for updating a state information statistical period Ts and a state information reporting period Tr in S4, and the updated Ts and Tr still need to satisfy Tr > Ts, Ts is more than or equal to Ts _ min, and Tr is more than or equal to Tr _ min.
Further, the S6 specifically includes:
(1) after receiving the slice indication message, an AI Adapter module of the base station sends the number of slices and the size of the resources distributed in each slice to a protocol stack of the base station according to the content of the slice indication message;
(2) the base station protocol stack uses the new scheduling information to perform resource scheduling on the terminal user, and the effective duration of the scheduling scheme is within the Tsell; after the Tsclice time, if the base station does not receive a new slice indication message, the base station adopts the original non-AI network slice scheduling mode;
and the AI Adapter module of the base station collects and counts the state information of the base station by using the new state information counting period Tss, and sends the collected state information to the AI slicing algorithm platform in the new state information reporting period Trn after the collection is finished, wherein the specific information collection and sending modes are the same as those in S4.
Further, the AI slice algorithm platform sends an AI slice termination request to the base station, and the resource scheduling mode that the base station recovers to the original non-AI slice is specifically realized by the following sub-steps:
(1) the AI slicing algorithm platform sends a slicing termination message to one or more base stations;
(2) after receiving the message of terminating the slicing, the AI Adapter module of the base station stops the current information collection of the base station, and simultaneously sends the information of terminating the slicing to a protocol stack of the base station, stops the current AI slicing scheduling mode, and the base station returns to the original non-AI network slicing scheduling mode;
(3) an AI Adapter module of the base station sends a slicing termination completion message to an AI slicing platform;
(4) and after receiving the slicing termination completion message, the AI slicing algorithm platform stops the corresponding AI slicing algorithm and releases the docker virtual environment created in S3, and the AI slicing algorithm platform terminates the connection with the base station.
The invention has the following beneficial effects:
(1) the invention can realize the AI network slicing algorithm only by adding an AI Adapter to the bottom layer software of the base station on the basis of not changing the hardware of the existing base station by introducing the AI slicing algorithm platform; meanwhile, the AI slicing algorithm platform is used as a public open platform and can be connected with a plurality of base stations, so that the resource joint management among the base stations can be realized, and the overall utilization rate of network resources is improved.
(2) The method of the invention designs a set of data collection flow, is not limited to a certain fixed artificial intelligence algorithm, is convenient for customers to carry out customized artificial intelligence algorithm research and deployment, and meets the requirements of practical application scenes.
(3) The method and the system can be conveniently and rapidly deployed, the AI slicing algorithm supports the random opening and termination without influencing the original functions and performances of the base station, the high-efficiency utilization of artificial intelligence to resources is fully obtained, meanwhile, the negative influence on the existing base station and the existing network is avoided, and the deployment requirement of the actual network can be met.
Drawings
The present invention will be described in further detail with reference to the accompanying drawings and embodiments.
FIG. 1 is a system architecture diagram of the present invention.
Fig. 2 is a flowchart illustrating an artificial intelligence based RAN-side network slice management method according to the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the objects and effects of the invention will become more apparent. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, the RAN-side network slice management system based on artificial intelligence of the present invention is oriented to a wireless base station, and the system architecture is composed of an AI slice algorithm platform and a plurality of base stations.
The AI slicing algorithm platform is a device with computing capability and network communication capability, communicates with the base station through a network, can be an independent entity device deployed at the base station side, can also be a virtual device on a regional cloud platform, and can even be a part of a core network. The AI slicing algorithm platform is used for receiving state information sent by the base station, respectively creating a docker virtual environment for each base station supporting the AI network slices, and isolating the docker virtual environment from AI slicing algorithm environments of other base stations, wherein an AI slicing algorithm is operated in each docker virtual environment; the AI slicing algorithm platform feeds back the state information to the AI slicing algorithm corresponding to each base station, the resource allocation is carried out on the network slices by the algorithm, and the allocation result is sent to the base stations by the AI slicing algorithm platform as network slice indication information; the AI slice algorithm platform can also directly send an AI slice termination request to the base station, or feed back an exception according to the AI slice algorithm, so that the base station is restored to a non-AI slice resource scheduling mode.
The base station is formed by adding an AI Adapter module on the basis of the traditional base station, the AI Adapter module realizes the mutual communication between the base station and the AI slicing algorithm platform, the mutual communication comprises the steps of collecting the state information of the base station and outputting the result of the AI slicing algorithm to a base station protocol stack, and the wireless air interface resource scheduling of the base station is realized.
As shown in fig. 2, the RAN-side network slice management method based on artificial intelligence of the present invention includes the following steps:
s1: an AI slicing algorithm platform initiates an AI slicing request message to a base station;
in one embodiment, in step S1, the AI slicing algorithm platform is connected to one or more base stations through a network, and the AI slicing algorithm platform sends an AI slicing function request message to one or more base stations at a time according to the needs of the network operation manager. And after the AI slice algorithm platform sends the message, starting a timer Treq. As one embodiment, the AI slicing algorithm platform is connected to the core network, and the AI slicing function request message includes a recommended number of slices, a slice type flag, and a slice attribute parameter for each slice, where the slice type flag and the slice attribute parameter for each slice are provided by a slice template layout system on the core network side.
S2: the base station receives and processes the AI slice request message and sends the supported AI slice capability to the AI slice algorithm platform;
as one embodiment, the S2 is implemented by the following sub-steps:
s2.1: and after receiving the AI slice request message, an AI Adapter module of the base station verifies whether a protocol stack of the base station supports the AI network slice according to the current running state and the self capacity of the equipment. Since the AI slicing algorithm platform may send an AI slicing request message to a base station that does not conventionally support AI network slicing, in order not to cause an abnormality of the conventional base station, the AI network slicing capability needs to be verified before the AI network slicing is formally executed. When the base station is a traditional base station which does not support an AI network slice resource scheduling mode per se and cannot normally work due to faults, failure of connection to a core network and core network abnormity, the base station is considered not to support the AI network slice; but the self ability of the base station supports the AI network slice resource scheduling mode and the base station normally works on line, and then the base station is considered to support the AI network slice.
S2.2: and the base station sets self AI slicing capability information according to the relevant slicing information in the AI slicing request message and sends the AI slicing capability message to the AI slicing algorithm platform. As one embodiment, the AI slice capability message includes: AI support/rejection flag bit, maximum supported slice number SliceNum _ max, slice type flag, slice attribute parameter corresponding to each slice, maximum terminal number supported by each slice, maximum resource number used for slicing algorithm, supported state information statistical minimum period Ts _ min, supported state information reporting minimum period Tr _ min, and resource information of neighboring base station. When the base station does not support AI network slices, setting an AI support/rejection flag bit as a rejection flag, simultaneously setting the supported maximum slice number SliceNum _ max to be 0, and not carrying the rest parameters; when the base station supports the AI network slice, the AI supporting/rejecting zone bit is set as the supporting zone, and other parameters are set according to the self supporting capability and the resource information of the adjacent base station. Since the base station may partially support or not support the parameters carried in the AI slice request message, including the recommended number of slices, the slice attribute parameters of each slice, and the like, the base station needs to modify the parameters according to its own actual conditions. In consideration of the practical situation, different base stations have different capabilities, and the method does not limit the capabilities of the base stations.
S3: after receiving AI slicing capability information of a base station, the AI slicing algorithm platform creates a docker virtual environment for each base station supporting AI network slices, starts an AI slicing algorithm in the docker virtual environment, and sends an AI slice starting message to the base station;
in step S3, the AI slice algorithm platform receives the AI slice capability message of the base station within the validity period of the timer Treq, and performs corresponding operations according to the information in the AI slice capability message. As one embodiment, the method is specifically realized by the following sub-steps:
s3.1: the AI slicing algorithm platform verifies the AI slicing capability message, if the AI supporting/rejecting flag bit in the AI slicing capability message is a rejecting flag, the base station is considered not to support the AI slicing algorithm, the process is finished without the following S4, S5, S6 and S7, and the base station still adopts the traditional non-AI network slicing scheduling mode at the moment; if the AI supporting/rejecting flag bit in the AI slicing capability message is a supporting flag and the maximum supported slice number SliceNum _ max is not 0, the AI slicing algorithm platform respectively creates a docker virtual environment for each base station to be isolated from AI slicing algorithm environments of other base stations, so as to ensure that the slicing algorithms of a plurality of base stations do not conflict; operating an AI slicing algorithm corresponding to the base station in each docker virtual environment;
s3.2: and the AI slicing algorithm platform verifies whether the adjacent cell base station is communicated with the AI slicing algorithm platform or not according to the resource information of the adjacent cell base station carried in the AI slicing capability message. If the adjacent cell base station is communicated with the AI slice algorithm platform, performing joint management on the resources of the adjacent cell base station during network slice allocation; if the connection is not available, the adjacent base station is considered to be unavailable, and the resources of the adjacent base station are not subjected to joint management;
s3.3: the AI slicing algorithm platform initializes the AI slicing algorithm in the docker virtual environment and sends an AI slicing start message to the base station after the initialization is completed.
And selecting the AI slicing algorithm corresponding to each base station by a network operation manager according to the capability of the AI slicing algorithm platform and the actual scene. The AI slicing algorithm platform serves as a public platform and supports various AI algorithms. The specific AI slicing algorithm supported by the AI slicing algorithm platform and which algorithm to select under different scenarios are determined by the network operation manager, and the system and method are not limited. -
As one embodiment, the AI slice start message includes: the method comprises the following steps of slice number SliceNum, a state information list, a state information statistical period Ts and a state information reporting period Tr, wherein the slice number SliceNum represents the number of slices for logically isolating base station resources by a slicing algorithm.
The base station reports the maximum slice number SliceNum _ max supported by the AI slicing algorithm platform, so that the SliceNum is required to be less than or equal to SliceNum _ max; the state information counting period Ts is used for counting the state information of the base station which changes in real time by taking the period Ts as a unit, and the state information reporting period Tr is used for reporting the collected state information by taking the period Tr as a unit. Considering the actual situation, additional time is needed for collecting the state information and reporting the information, so the reporting period Tr should be greater than the statistical period Ts, and meanwhile, since the base station reports the supported state information statistical minimum period Ts _ min and the supported state information reporting minimum period Tr _ min in the AI slice capability message in S2, Ts must be equal to or greater than Ts _ min, and Tr must also be equal to or greater than Tr _ min; for the AI slicing algorithm, the state information will be used as the input of the algorithm model, the state information required by different AI algorithms will be different, the specific required state information is determined by the network operation manager according to the selected AI algorithm, and the method and the device are not limited. As one embodiment, the status information list includes: the method comprises the steps that the size of a data packet to be sent of each terminal user in a protocol stack PDCP layer in a slice, the size of the sent data packet, a time delay statistic value of each terminal user, the resource utilization rate of each slice (the resource utilization rate refers to the ratio of actually scheduled and used resources to resources allocated to the slice), and an AI slice termination mark; the state information statistical period Ts is 100 ms; the state information reporting period Tr is 150 ms.
S4, after receiving the AI slice starting message, the base station periodically collects the state information of the base station and sends the state information to an AI slice algorithm platform;
as one embodiment, the S4 is implemented by the following sub-steps:
s4.1: after receiving the AI slice start message, the AI Adapter of the base station verifies and corrects the indication message, which specifically includes:
1) the slice number SliceNum in the AI slice start message cannot be larger than the supported maximum slice number SliceNum _ max provided by the AI slice request response message in step S2. Due to limited resources of the base station, when the number of the maximum slices supported by the base station is exceeded, an error of insufficient resources occurs during actual resource allocation. For the AI slice starting message, the number of slices is greater than SliceNum _ max, and the number of slices is finally subject to SliceNum _ max;
2) a state information counting period Ts and a state information reporting period Tr in the AI slice starting message cannot be smaller than the corresponding Ts _ min and Tr _ min in the step S2, if the state information counting period Ts and the state information reporting period Tr cannot be met, the Ts is reset to be Ts _ min and Tr is set to be Tr _ min; meanwhile, Tr must be larger than Ts, if the Tr is not satisfied, Tr = Ts + time for collecting data and uploading the data is reset. In one embodiment, the time for collecting data and uploading data is 50 ms.
S4.2: an AI Adapter module of the base station collects the state information of the base station, and the statistical time of the state information is Ts. The size of the data packet to be sent and the size of the sent data packet are accumulated summation values within the time Ts; the delay statistic value is the average value of the delay values of all the data packets in the PDCP layer within the statistic period Ts. Further, the delay value of the data packet at the PDCP layer is defined as half of the difference between the time when the data packet arrives at the PDCP layer and the time when the PDCP layer receives the feedback that the data packet was normally received. The resource utilization rate of each slice is the average value of the resource utilization rate of each slice in the Ts time; because the slicing algorithm based on artificial intelligence is in actual execution, scenes of network resource slicing allocation schemes which are unreasonable in slicing allocation and cannot be executed by the base station exist, especially when algorithm training stages and network environments are changed greatly, a base station protocol stack is abnormal when the network resource slicing allocation schemes provided by the artificial intelligence algorithm are used, and for abnormal scenes of the protocol stack, an AI slicing termination flag needs to be set to be true.
S4.3: after an AI Adapter module of the base station collects the state information of the base station, the state information is sent to an AI slicing algorithm platform within the time Tr; if the AI slice termination mark in the state information is true, the abnormal scene of the protocol stack is shown, and in order to ensure that normal service scheduling is not influenced, the base station adopts a traditional non-AI network slice scheduling mode;
s5, the AI slice algorithm platform receives the state information sent by the base station and sends the state information to the corresponding AI slice algorithm, the AI slice algorithm carries out resource allocation on the network slice, and the allocation result is used as the network slice indication information of the AI slice algorithm platform and sent to the corresponding base station;
as one embodiment, the S5 is implemented by the following sub-steps:
s5.1: after the AI slice algorithm platform receives the state information sent by the base station, the state information (together with the state information of the neighboring base stations, if any) is used as the input of the trained AI slice algorithm, the algorithm model is run, and the output of the slice algorithm, namely the specific network resource slice allocation scheme, is calculated.
There are various implementation manners of the specific AI slicing algorithm. Taking the deep neural network method as an example, the deep neural network can be adopted, the state information is used as the input of the deep neural network, and the output is the network resource slice allocation scheme; meanwhile, the reinforcement learning is supported, wherein the reinforcement learning comprises deep reinforcement learning combined with the reinforcement learning, the state information is used as the reinforcement learning state (status), the delay statistic value of each terminal user and the resource utilization rate of each slice in the state information are used as the reward (reward) of the reinforcement learning, and the network resource slice allocation scheme is used as the action (action) of the reinforcement learning. In terms of specific algorithm implementation, there are many published papers, and the following examples are given in this embodiment:
（1）P. Du, "Deep Learning-based Application Specific RAN Slicing for Mobile Networks," 2018 IEEE 7th International Conference on Cloud Networking (CloudNet), 2018, pp. 1-3；
（2）C. Qi, et al.“Deep reinforcement learning with discrete normalized advantage functions for resource management in network slicing,” IEEE Communications Letters, vol. 23, no. 6, pp. 1337–1341, Aug. 2019；
(3) y, Hua, et al, "GAN-powered off distribution recovery for resource management in network sizing," IEEE J. Sel. area. Comm., vol. 38, No. 2, pp. 334-.
Other AI slicing algorithms may be used by those skilled in the art depending on the actual requirements.
S5.2: and after obtaining the network resource slice allocation scheme, the AI slice algorithm platform encapsulates the slice indication message and issues the encapsulated slice indication message to the base station. The slice indication message includes: when the effective time Tslice, the state information statistics period Tsn, and the state information reporting period Trn of the current slicing scheme are the same as the corresponding values in the previous slicing indication message, the effective time Tslice, the state information statistics period Tsn, and the state information reporting period Trn of the current slicing scheme may not be carried in the current slicing indication message. The slicing scheme information comprises specific slicing number and the size of resources distributed in each slice; the state information counting period Tsn and the state information reporting period Trn are used for updating the state information counting period Ts and the state information reporting period Tr in S4, and since the state information counting period and the state information reporting period affect the feature extraction of the AI algorithm and the response time of the network resource slice, the network operation manager flexibly modifies and configures the values according to the actual debugging condition of the AI slice algorithm; the state information counting period Tsn and the state information reporting period Trn are updated in the slice indication message according to actual situations, but still the requirements for the state information counting period and the state information reporting period in S3 and S4 need to be met.
S6: after receiving the network slice indication information, the base station executes network slice resource scheduling according to the network slice indication information and continuously and periodically reports state information to the AI slice algorithm platform;
as one embodiment, step S6 is specifically realized by the following sub-steps:
s6.1, after receiving the slice indication message, the AI Adapter module of the base station sends the number of slices and the size of the resource distributed in each slice to a protocol stack of the base station according to the content of the slice indication message;
s6.2: the base station protocol stack uses the new scheduling information to perform resource scheduling on the terminal user, and the effective duration of the scheduling scheme is within the Tsell; after the Tsclice time, if the base station does not receive a new slice indication message, i.e. S5, the base station will adopt a conventional non-AI network slice scheduling manner;
s6.3: the base station AI Adapter uses the new state information statistic period Tss to collect and count the state information of the base station, and sends the collected state information to the AI slice algorithm platform in the new state information reporting period Trn after the collection is completed, the specific collected information and sending mode are the same as those in S4, and the only difference is that the base station performs resource scheduling by using the network resource slice allocation scheme indicated by the AI slice platform at this time, instead of the traditional non-AI network slice scheduling mode in S4.
When the network slice is to be terminated, the AI slice algorithm platform sends an AI slice termination request to the base station, and the base station returns to the traditional non-AI network slice scheduling mode.
The AI slicing algorithm platform may actively terminate AI slicing resource management for a base station and/or multiple base stations, including but not limited to: 1) the network operation manager terminates slice management for the needs; 2) the state information reported by the base station in S4 shows that the resource utilization rate is not high; 3) the AI slice termination flag is true when K status messages are received consecutively. When the termination slice message is triggered, the termination process is realized by the following sub-steps:
(1) the AI slicing algorithm platform sends a slicing termination message to one or more base stations;
(2) after receiving the message of terminating the slicing, the AI Adapter module of the base station stops the current information collection of the base station, and simultaneously sends the information of terminating the slicing to a protocol stack of the base station, and stops the current AI slicing scheduling mode, and the base station returns to the traditional non-AI network slicing scheduling mode;
(3) an AI Adapter module of the base station sends a slicing termination completion message to an AI slicing platform;
(4) and after the AI slicing platform receives the slicing termination completion message, stopping the AI slicing algorithm and releasing the creation of the docker virtual environment in S3, and terminating the connection between the AI slicing algorithm platform and the base station.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and although the invention has been described in detail with reference to the foregoing examples, it will be apparent to those skilled in the art that various changes in the form and details of the embodiments may be made and equivalents may be substituted for elements thereof. All modifications, equivalents and the like which come within the spirit and principle of the invention are intended to be included within the scope of the invention.
Claims (10)
1. An artificial intelligence-based RAN side network slice management system is characterized by comprising an AI slice algorithm platform and a plurality of base stations;
the base station comprises an AI Adapter module, the AI Adapter module is used for mutual communication between the base station and the AI slicing algorithm platform, and comprises the steps of collecting state information of the base station, outputting the result of the AI slicing algorithm to a base station protocol stack, and realizing the resource scheduling of a wireless air interface of the base station;
the AI slicing algorithm platform is used for receiving state information sent by the base station, respectively creating a docker virtual environment for each base station supporting the AI network slices, and isolating the docker virtual environment from AI slicing algorithm environments of other base stations, wherein an AI slicing algorithm is operated in each docker virtual environment; the AI slicing algorithm platform feeds back the state information to the AI slicing algorithm corresponding to each base station, the resource allocation is carried out on the network slices by the algorithm, and the allocation result is sent to the base stations by the AI slicing algorithm platform as network slice indication information; the AI slice algorithm platform can also directly send an AI slice termination request to the base station, or feed back an exception according to the AI slice algorithm, so that the base station is restored to a non-AI slice resource scheduling mode.
2. The artificial intelligence based RAN-side network slice management system of claim 1, wherein the AI-slice algorithm platform is a physical device or a virtual device with computing capability and network communication capability.
3. A RAN-side network slice management method based on artificial intelligence, characterized in that the method is implemented based on the management system of claim 1, and the method comprises the following steps:
s1: the AI slicing algorithm platform initiates an AI slicing request message to the base station;
s2: the base station receives and processes the AI slice request message and sends the supported AI slice capability to the AI slice algorithm platform;
s3: after receiving AI slicing capability information of a base station, the AI slicing algorithm platform creates a docker virtual environment for each base station supporting AI network slices, starts an AI slicing algorithm in the docker virtual environment, and sends an AI slice starting message to the base station;
s4: after receiving the AI slice starting message, the base station periodically collects the state information of the base station and sends the state information to the AI slice algorithm platform;
s5: the AI slicing algorithm platform receives the state information sent by the base station and sends the state information to the corresponding AI slicing algorithm, the AI slicing algorithm carries out resource allocation on the network slices, and the allocation result is used as the network slice indication information of the AI slicing algorithm platform and sent to the corresponding base station;
s6: after receiving the network slice indication information, the base station executes network slice resource scheduling according to the network slice indication information and continuously and periodically reports state information to the AI slice algorithm platform;
when the terminal is required to be terminated, the AI slice algorithm platform sends an AI slice termination request to the base station, and the base station restores to the original non-AI slice resource scheduling mode.
4. The artificial intelligence based RAN-side network slice management method of claim 3, wherein the AI slice capability message comprises: AI supporting/rejecting flag bit, maximum supported slice number SliceNum _ max, slice type flag, slice attribute parameter corresponding to each slice, maximum terminal number supported by each slice, maximum resource number used for slicing algorithm, supported state information statistical minimum period Ts _ min, supported state information reporting minimum period Tr _ min and resource information of neighboring base station;
when the base station does not support AI network slices, setting an AI support/rejection flag bit as a rejection flag, simultaneously setting the supported maximum slice number SliceNum _ max to be 0, and not carrying the rest parameters; when the base station supports the AI network slice, the AI supporting/rejecting zone bit is set as the supporting zone, and other parameters are set according to the self supporting capability and the resource information of the adjacent base station.
5. The RAN-side network slice management method according to claim 3 or 4, wherein the S3 specifically is:
(1) an AI slice algorithm platform receives AI slice capability information of a base station in the validity period of a timer Treq, verifies the AI slice capability information, creates a docker virtual environment for each base station supporting AI network slices, and operates an AI slice algorithm corresponding to the base station in each docker virtual environment;
(2) the AI slicing algorithm platform verifies whether the adjacent cell base station is communicated with the AI slicing algorithm platform or not according to the resource information of the adjacent cell base station carried in the AI slicing capability message; if the adjacent cell base station is communicated with the AI slice algorithm platform, performing joint management on the resources of the adjacent cell base station during network slice allocation; if the connection is not available, the adjacent base station is considered to be unavailable, and the resources of the adjacent base station are not subjected to joint management;
(3) the AI slicing algorithm platform initializes the AI slicing algorithm in the docker virtual environment and sends an AI slicing start message to the base station after the initialization is completed.
6. The RAN-side network slice management method according to claim 5, wherein the AI slice start message includes slice number SliceNum, status information list, status information statistics period Ts, and status information reporting period Tr.
7. The RAN-side network slice management method based on artificial intelligence of claim 6, wherein the S4 specifically is:
(1) after receiving an AI slice starting message, an AI Adapter module of the base station firstly verifies and corrects the AI slice starting message, namely when SliceNum is less than or equal to SliceNum _ max, Tr is more than Ts, Ts is more than or equal to Ts _ min, and Tr is more than or equal to Tr _ min, all parameters keep original values; when SliceNum is larger than SliceNum _ max, the SliceNum takes SliceNum _ max as the standard during slicing; when Ts is less than Ts _ min, the Ts is reset to Ts _ min; when Tr < Tr _ min, Tr is reset to Tr _ min; when Tr is less than or equal to Ts, resetting Tr = Ts + time for collecting data and uploading the data;
(2) the AI Adapter module periodically collects the self state information and sends the state information to the AI slicing algorithm platform within the Tr time; if the AI slice termination flag in the state information is true, it indicates that an abnormal scenario occurs in the protocol stack, and in order to ensure that normal service scheduling is not affected, the base station adopts a non-AI network slice scheduling mode.
8. The RAN-side network slice management method based on artificial intelligence of claim 3, wherein the S5 specifically is:
(1) after receiving the state information sent by the base station, the AI slicing algorithm platform takes the state information as the input of a corresponding AI slicing algorithm, runs an algorithm model, and calculates the output of the slicing algorithm, namely a specific network resource slicing allocation scheme;
(2) the AI slicing algorithm platform issues the network resource slicing allocation scheme as a slicing indication message to the corresponding base station; the slice indication message includes: the effective time Tsclce, the slicing scheme information, the state information statistical period Tsn and the state information reporting period Trn of the current slicing scheme are the same as the corresponding values in the previous slicing indication message, and the effective time Tsclce, the state information statistical period Tsn and the state information reporting period Trn of the current slicing scheme are not carried in the current slicing indication message;
the slicing scheme information comprises specific slicing number and size of resources distributed in each slice;
the state information statistical period Tsn and the state information reporting period Trn are used for updating a state information statistical period Ts and a state information reporting period Tr in S4, and the updated Ts and Tr still need to satisfy Tr > Ts, Ts is more than or equal to Ts _ min, and Tr is more than or equal to Tr _ min.
9. The RAN-side network slice management method based on artificial intelligence of claim 8, wherein the S6 specifically is:
(1) after receiving the slice indication message, an AI Adapter module of the base station sends the number of slices and the size of the resources distributed in each slice to a protocol stack of the base station according to the content of the slice indication message;
(2) the base station protocol stack uses the new scheduling information to perform resource scheduling on the terminal user, and the effective duration of the scheduling scheme is within the Tsell; after the Tsclice time, if the base station does not receive a new slice indication message, the base station adopts the original non-AI network slice scheduling mode;
and the AI Adapter module of the base station collects and counts the state information of the base station by using the new state information counting period Tss, and sends the collected state information to the AI slicing algorithm platform in the new state information reporting period Trn after the collection is finished, wherein the specific information collection and sending modes are the same as those in S4.
10. The RAN-side network slice management method based on artificial intelligence of claim 8, wherein the AI slice algorithm platform sends an AI slice termination request to the base station, and the resource scheduling manner for the base station to restore to the original non-AI slice is specifically implemented by the following sub-steps:
(1) the AI slicing algorithm platform sends a slicing termination message to one or more base stations;
(2) after receiving the message of terminating the slicing, the AI Adapter module of the base station stops the current information collection of the base station, and simultaneously sends the information of terminating the slicing to a protocol stack of the base station, stops the current AI slicing scheduling mode, and the base station returns to the original non-AI network slicing scheduling mode;
(3) an AI Adapter module of the base station sends a slicing termination completion message to an AI slicing platform;
(4) and after receiving the slicing termination completion message, the AI slicing algorithm platform stops the corresponding AI slicing algorithm and releases the docker virtual environment created in S3, and the AI slicing algorithm platform terminates the connection with the base station.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111060374.9A CN113507729B (en) | 2021-09-10 | 2021-09-10 | RAN side network slice management system and method based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111060374.9A CN113507729B (en) | 2021-09-10 | 2021-09-10 | RAN side network slice management system and method based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113507729A true CN113507729A (en) | 2021-10-15 |
| CN113507729B CN113507729B (en) | 2021-12-28 |
Family
ID=78017140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111060374.9A Active CN113507729B (en) | 2021-09-10 | 2021-09-10 | RAN side network slice management system and method based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113507729B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170318468A1 (en) * | 2016-04-27 | 2017-11-02 | Kabushiki Kaisha Toshiba | Radio resource slicing in a radio access network |
| CN109819452A (en) * | 2018-12-29 | 2019-05-28 | 上海无线通信研究中心 | A kind of Radio Access Network construction method calculating virtual container based on mist |
| CN111587601A (en) * | 2018-01-12 | 2020-08-25 | 华为技术有限公司 | Network slice provisioning and operation |
| CN112512057A (en) * | 2019-09-16 | 2021-03-16 | 中国移动通信集团河北有限公司 | Network slice abnormity identification method, device, equipment and computer storage medium |
| US20210243770A1 (en) * | 2018-05-08 | 2021-08-05 | Nokia Solutions And Networks Oy | Method, computer program and circuitry for managing resources within a radio access network |
-
2021
- 2021-09-10 CN CN202111060374.9A patent/CN113507729B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170318468A1 (en) * | 2016-04-27 | 2017-11-02 | Kabushiki Kaisha Toshiba | Radio resource slicing in a radio access network |
| CN111587601A (en) * | 2018-01-12 | 2020-08-25 | 华为技术有限公司 | Network slice provisioning and operation |
| US20210243770A1 (en) * | 2018-05-08 | 2021-08-05 | Nokia Solutions And Networks Oy | Method, computer program and circuitry for managing resources within a radio access network |
| CN109819452A (en) * | 2018-12-29 | 2019-05-28 | 上海无线通信研究中心 | A kind of Radio Access Network construction method calculating virtual container based on mist |
| CN112512057A (en) * | 2019-09-16 | 2021-03-16 | 中国移动通信集团河北有限公司 | Network slice abnormity identification method, device, equipment and computer storage medium |
Non-Patent Citations (2)
| Title |
|---|
| TÜRK TELEKOM: "Network Slicing Solution with RAN Slicing", 《TSG SA MEETING #SP-73 SP-160534》 * |
| 王东升等: "电力5G业务切片全生命周期研究", 《电力信息与通信技术》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113507729B (en) | 2021-12-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3439359B1 (en) | Method of generating radio access network slice, radio access network, and slice manager | |
| CN108512760B (en) | Routing method based on SDN guarantee service QoS | |
| CN110572481A (en) | intelligent electromechanical equipment data interaction method and system based on LoRa communication | |
| Azari et al. | On the latency-energy performance of NB-IoT systems in providing wide-area IoT connectivity | |
| CN108718283B (en) | TCP congestion control method for centralized end network coordination in data center network | |
| CN101502144A (en) | Element management system in wireless communication network | |
| JP4474217B2 (en) | Packet communication device | |
| CN109429244B (en) | Method and device for isolating management data of network slice subnet instances | |
| CN110225512B (en) | Wireless resource control method and system for cellular Internet of things | |
| CN109587782B (en) | Synchronization method and device | |
| KR102178880B1 (en) | Network system and data trasmission method based on device clustering in lorawan communication | |
| CN109842507A (en) | A kind of network slice management method and equipment | |
| CN104160735A (en) | Packet processing method, forwarder, packet processing device and packet processing system | |
| CN113507729B (en) | RAN side network slice management system and method based on artificial intelligence | |
| Halim et al. | Congestion control mechanism for Internet-of-Things (IOT) paradigm | |
| Ganjalizadeh et al. | Interplay between distributed AI workflow and URLLC | |
| CN105262682A (en) | Software defined network for power data communication and traffic grooming method thereof | |
| CN116506860A (en) | Base station network slice management system based on dynamic resource allocation | |
| CN102378233A (en) | Method and system for improving reliability of wireless communication network system | |
| CN110708710B (en) | 5G small base station and slice management module thereof | |
| CN114245317B (en) | Data processing method, device and computer readable storage medium | |
| CN110958162A (en) | BLE cascade network, link detection method, equipment and terminal thereof | |
| CN114143830B (en) | Flow optimization method and device, equipment and storage medium | |
| CN101651605A (en) | Method and system for transmitting and receiving data | |
| CN101179833B (en) | Congestion control method between base station and wireless network controller |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |