Routing model-based network slice recommendation method and system, device, and medium

Abstract

Disclosed in the present invention are a routing model-based network slice recommendation method and system, a device, and a medium. The method comprises: collecting statistics about real-time traffic to analyze network load and demand, obtaining traffic distribution characteristics of the whole network, and pre-constructing basic network slices on the basis of the traffic distribution characteristics; deploying the constructed basic network slices on a switching node in the form of OpenFlow protocol flow tables; on the basis of bandwidth and time delay classification thresholds, classifying the network slices deployed on the switching node; virtualizing physical resources by means of NFV to obtain associations between various network slices and the physical resources; an SDN controller generating a routing policy on the basis of network load and traffic conditions; deploying the routing policy and obtaining an optimal link from the existing routing policy; selecting, from the various network slices, network slices to be allocated to a user, sorting the network slices to be allocated for recommendation, and sequentially recommending same to the user by means of the optimal link. In the present invention, requirements of users of the same type can be met by means of recommendation priority-based sorting.

Description

Network slice recommendation method, system, device and medium combined with routing model TECHNICAL FIELD

[0001] The present application relates to the technical field of mobile communication, in particular to a network slice recommendation method, system, device and medium combined with routing model. BACKGROUND

[0002] 3GPP divides the main application scenarios of 5G into three categories, namely enhanced mobile broadband (eMBB), massive machine type communication (mMTC) and ultra-reliable and low latency communication (uRLLC). mMTC and uRLLC are communications between people and machines and machines and machines, which are the main application scenarios of the Internet of Things; and eMBB is mainly used to improve the communication experience between people, which is the further improvement of user experience and other performances on the basis of the existing mobile broadband service scenario. The three main service scenarios have different requirements and different performance indicators, so the arrangement of the slice will directly affect the load, resource utilization and energy consumption of the network under the network architecture of NFV and SDN. Based on SDN technology, researchers have done a lot of research on optimizing network slice arrangement and improving resource utilization. The existing technology is to optimize network resources in a data center with a relatively simple network state, without considering the complex requirements of application services in terms of bandwidth, latency and reliability, and most of them only aim at network resource utilization or QoS and other single targets. The arrangement algorithm of NS is also optimized from the local information of the network for a single target, lacking overall consideration of the network. SUMMARY

[0003] The present application provides a network slice recommendation method, system, device and medium combined with routing model to solve the problems in the prior art.

[0004] To achieve the above purpose, the present application adopts the following technical scheme:

[0005] A network slice recommendation method combined with a routing model, comprising the following steps:

[0006] Statistical real-time traffic to analyze network load and demand, obtain the traffic distribution characteristics of the whole network, and pre-construct basic network slices according to the traffic distribution characteristics, and deploy the constructed basic network slices on the switching nodes in the form of OpenFlow protocol flow table;

[0007] The network slices deployed on the switching nodes are classified according to bandwidth and time delay classification thresholds, virtualization of physical resources is realized by using NFV to obtain the association relationship between various network slices and physical resources, the SDN controller generates a routing strategy according to network load and traffic conditions, and the routing strategy is deployed to obtain an optimal link from the existing routing strategy;

[0008] The network slice allocated to the user is selected from various network slices, and the allocated network slice is recommended and sorted in order to be recommended to the user through the optimal link.

[0009] To optimize the above technical solutions, the specific measures taken also include:

[0010] Further, the method for classifying the network slices deployed on the switching nodes according to bandwidth and time delay classification thresholds is matrix decomposition, and the specific process is as follows:

[0011] There are m network slices, n bandwidth and time delay classification thresholds, and a slice-classification threshold interaction matrix R of mxn dimensions;

[0012] The slice-classification threshold interaction matrix R is decomposed into two low-rank matrices: a slice matrix U and a classification threshold matrix V, and the product of the two is used to approximate R, which is formalized as follows: R≈U T V

[0013] The column vector U of the slice matrix U i represents a slice latent vector, and the column vector V of the classification threshold matrix V j represents a classification threshold latent vector,

[0014] To learn the latent vectors, the idea of linear regression is adopted to minimize the squared error between the original score and the predicted score, and the objective function L(U, V) is obtained as follows:

[0015] In the formula, R ij represents the original score, represents the predicted score, K is the number of (i, j) pairs whose scores are known, and λ is a regularization factor, represents the matrix norm;

[0016] The method for finding the local minimum of the objective function is the gradient descent method or the alternating least squares method.

[0017] Further, the specific process of realizing virtualization of physical resources by using NFV is as follows:

[0018] A virtual link and the nodes at both ends of the link are defined as a component, and the component includes a free component, a restricted component and a fixed component;

[0019] determining whether there is a free component in the virtual network, if there is, mapping the free component, after a free component completes mapping, mapping all limited components connected with the free component, if there is not, mapping the limited components in the virtual network;

[0020] The mapping of the fixed component only needs to complete the mapping of the fixed link, and the fixed link is mapped to the two mapped virtual nodes M and M on the determined shortest path.

[0021] Further, the specific process of mapping the free component is:

[0022] In all free components of the virtual network, the free component with the largest link bandwidth requirement is selected as the free component to be mapped by using the greedy algorithm n v i and n v j represent the nodes at both ends of the link, specifically: in the free component set S, all node pairs are traversed in order to determine the shortest path until a path satisfying the node and link constraints of the free component is found; the node and link constraints of the free component include the hop limit of the physical path, the computing power constraints of the two physical nodes at both ends of the physical path, and the bandwidth constraint of the physical path.

[0023] Further, the method for obtaining the optimal link from the existing routing strategy is to perform network pruning by using NetSliming, specifically:

[0024] The Gamma parameter in the BN layer in the CNN network is used to learn the pruning parameter end-to-end, including the following steps:

[0025] First, add a regularization penalty to the Gamma parameter of the BN to train the model, and the loss function L bng of the channel pruning method of the BN layer Gamma parameter becomes:

[0026] In the formula, (x, y) represents the input and target of training, W represents the weight used for training, is the loss value of the normal training of the CNN network, g(γ) is the sparsity penalty term of the scaling factor, γ is the scaling factor, and λ is the balance coefficient;

[0027] Then, all the Gamma parameters in the CNN network are sorted, and according to the pruning ratio given by the human, the channels whose Gamma parameters are less than the set value are removed.

[0028] Further, the method for recommending and ranking the allocated network slices is to evaluate the network slice ranking recommendation index by using MAP, and the method is specifically as follows:

[0029] For N network slice recommendation lists, each list has n recommended network slices and different network slice rankings, the MAP of each network slice recommendation list is calculated, and the formula is as follows:

[0030] In the formula, MAP represents the average accuracy, P i (k) represents the accuracy of the i-th recommendation list at position k, δ ij is an indicative function, indicating whether the j-th network slice of the i-th recommendation list is a correct recommendation;

[0031] The network slice recommendation list with the highest MAP value is selected as the list recommended to the user.

[0032] Further, the method for recommending and ranking the allocated network slices is an NDCG ranking model, and the method is specifically as follows:

[0033] The NDCG value is calculated as an evaluation index for quantifying the recommendation ranking, and the calculation formula is as follows:

[0034] In the formula, IDCG i is the maximum DCG value that can be taken by the i-th network slice recommendation list, N is the total number of network slice recommendation lists, n is the total number of network slices in each network slice recommendation list, is an indicative function, indicating the score of the j-th candidate network slice in the i-th network slice recommendation list.

[0035] The application also provides a network slice recommendation system combined with a routing model, which comprises:

[0036] A network slice construction module is configured to count real-time traffic to analyze network load and demand, obtain traffic distribution characteristics of the whole network, and pre-construct basic network slices according to the traffic distribution characteristics;

[0037] A network slice deployment module is configured to deploy the constructed basic network slices on a switching node in the form of an OpenFlow protocol flow table;

[0038] A classification module is configured to classify the network slices deployed on the switching node according to bandwidth and time delay classification thresholds;

[0039] A virtualization module is configured to obtain the association between various network slices and physical resources by using NFV to realize virtualization of physical resources;

[0040] The SDN controller is used to generate routing policies based on network load and traffic conditions.

[0041] The optimal link selection module is used to deploy routing policies and obtain the optimal link from existing routing policies;

[0042] The recommendation and sorting module is used to select network slices from various types of network slices to allocate to users, sort the allocated network slices, and recommend them to users in order through the optimal link.

[0043] The present invention also proposes a computer-readable storage medium storing a computer program that causes a computer to execute the network slice recommendation method combined with the routing model as described above.

[0044] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the network slicing recommendation method combined with the routing model as described above.

[0045] The beneficial effects of this invention are:

[0046] This invention prioritizes high-bandwidth links during virtual network mapping, controlling the hop count of the physical path mapped to these links, thus conserving link resources. It increases the success rate of mapping blocked and fixed components, avoiding iterations and fully exploring physical resources, ultimately improving the probability of successful virtual network request mapping. Furthermore, this invention's priority recommendation sorting can satisfy the needs of similar users. Attached Figure Description

[0047] Figure 1 is a flowchart of the network slice recommendation method combining a routing model proposed in this invention;

[0048] Figure 2 is a schematic diagram of virtual network mapping;

[0049] Figure 3 is a flowchart of the NetSliming pruning process. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0051] In one embodiment, the present invention proposes a network slice recommendation method combining a routing model. The flowchart of the method is shown in Figure 1, and includes the following steps:

[0052] Statistical real-time traffic to analyze network load and demand, get the whole network traffic distribution characteristics, according to the traffic distribution characteristics to construct a basic network slice, the constructed basic network slice is deployed in the form of OpenFlow protocol flow table on the switching node;

[0053] According to the bandwidth and delay classification threshold, the network slice deployed on the switching node is classified, and the method is matrix decomposition, which can reduce the matrix error and improve the model accuracy. Specifically as follows:

[0054] There are m network slices, n bandwidth and delay classification thresholds, and a m*n dimensional slice-classification threshold interaction matrix R:

[0055] The slice-classification threshold interaction matrix R is decomposed into two low-rank matrices: slice matrix U and classification threshold matrix V, and the product of the two is used to approximate R, which is formalized as follows: R≈U T V

[0056] The column vector U of the slice matrix U i represents the slice hidden vector, and the column vector V of the classification threshold matrix V j represents the classification threshold hidden vector,

[0057] To learn the hidden vector, the idea of linear regression is adopted to minimize the squared error between the original score and the predicted score, and the objective function L(UV) is obtained as follows:

[0058] In the formula, R ij represents the original score, represents the predicted score, K is the number of (i,j) pairs whose scores are known, and a regularization term is added to prevent overfitting, λ is the regularization factor, represents the matrix norm;

[0059] The method for finding the local minimum of the objective function is gradient descent or alternating least squares (ALS). The ALS method is conducive to distributed expansion and improves the computational efficiency.

[0060] The association between various network slices and physical resources is obtained by virtualizing physical resources using NFV (Network Function Virtualization). Network virtualization technology refers to: through abstraction, allocation and isolation mechanisms, a public physical network (Substrate Network, SN) supports multiple virtual networks, each virtual network can use independent protocol systems, and can reasonably allocate node resources and link resources in the entire network according to dynamic changes in user demand. Each virtual network is a resource piece of the underlying network, which is composed of virtual nodes (such as virtual routers) and virtual links.

[0061] Virtual network mapping can be decomposed into node mapping and link mapping. As shown in Figure 2. Since multiple virtual nodes belonging to the same virtual network cannot be mapped to the same underlay node, but multiple virtual links can be mapped to the same underlay path. The virtual network mapping problem can be described as: M: G v → (N', L', R N , R L )

[0062] Where R N and R L are the node and link resources requested by the underlay network for the virtual network G v Virtual network mapping can be decomposed into node mapping and link mapping.

[0063] The specific process of virtualizing physical resources by NFV is:

[0064] Define a virtual link and the nodes at both ends of the link as a component, which includes a free component, a restricted component and a fixed component;

[0065] Determine whether there is a free component in the virtual network, if there is a free component, map the free component, and the specific process of mapping the free component is:

[0066] In all free components of the virtual network, select the free component with the largest link bandwidth requirement as the free component to be mapped by using a greedy algorithm n v i and n v j Indicate the nodes at both ends of the link, specifically: traverse all node pairs in the free component set S in order to determine the shortest path, until a path that satisfies the node and link constraints of the free component is found; The node and link constraints of the free component include the hop limit of the physical path, the computing power constraints of the two physical nodes at both ends of the physical path and the bandwidth constraints of the physical path. Prioritize completing links with large bandwidth requirements, which can control the number of hops of the physical path to which the virtual link with large bandwidth requirements is mapped, thereby saving the consumption of link resources.

[0067] After a free component completes mapping, map all restricted components connected to the free component.

[0068] If there is no free component, map the restricted components in the virtual network;

[0069] Map the fixed components in the virtual network, since the virtual nodes at both ends of the fixed component have been mapped, mapping the fixed component only needs to complete the mapping of the fixed link, and the fixed link is mapped to the two mapped virtual nodes M With M The shortest path is determined. If the bandwidth resource of this shortest path cannot satisfy the bandwidth requirement of the fixed link , iteration is needed.

[0070] Improved free link mapping:

[0071] When mapping the free component, only three constraints are considered for the candidate physical path, which are the hop constraint of the physical path, the bandwidth constraint of the physical path and the resource constraint of the nodes at both ends of the physical path. The improvement is to add a new constraint to the candidate physical path.

[0072] After the free component mapping is completed, the restricted component connected to the free component needs to be mapped, and if the restricted component mapping fails, iteration is needed. Similarly, after the mapping of the free component and the restricted component is completed, the mapping of the fixed component also needs to be completed, and if the fixed component mapping fails, iteration is needed. However, iteration will exclude the region node from the set S, but the resources near the region may not have been sequentially traversed. Therefore, if the success rate of the blocked component and the fixed component mapping can be increased to avoid iteration and fully discover physical resources, the probability of successful mapping of the virtual network request can ultimately be improved.

[0073] The SDN controller generates a routing strategy according to the network load and traffic conditions, deploys the routing strategy, and obtains the optimal link from the existing routing strategy. The method for obtaining the optimal link is to use NetSliming for network pruning. The main idea of network pruning is to remove redundant parameters in the neural network that contribute little to the output result, thereby improving network efficiency.

[0074] Network pruning is originally used to solve the problem of overfitting, and is now commonly used to reduce network complexity. Pruning generally consists of three steps: first, train an initial neural network model normally, then remove redundant connections and parameters in the initial model, i.e. the "pruning" operation, and then retrain the model to fine-tune the connections of the remaining neurons, thereby ensuring the performance of the model.

[0075] The network pruning using NetSliming is as follows:

[0076] The Gamma parameter in the BN layer of the CNN network is used to learn the pruning parameters end-to-end, including the following steps:

[0077] First, add a regularization penalty to the Gamma parameter of the BN to train the model. The loss function L of the channel pruning method of the Gamma parameter of the BN layer bng becomes:

[0078] In the formula, (x, y) represents the input and target of training, W represents the weight for training, The loss value for normal training of the CNN network, g(γ) is a sparse penalty term of a scaling factor, γ is a scaling factor, and λ is a balance coefficient.

[0079] Then, all the Gamma parameters in the CNN network are sorted, and the channels with Gamma parameters less than a set value are removed according to a pruning ratio given by a person.

[0080] Finally, model fine-tuning is performed. This process can be repeated multiple times to achieve better results, as shown in FIG. 3.

[0081] The network slice allocated to the user is selected from various network slices, and since the allocated slice can not be unique, the priority recommendation ranking becomes important. The allocated network slice is recommended and ranked in order, and is recommended to the user through an optimal link. Two methods are disclosed in the embodiment for recommendation ranking, one of which uses a MAP evaluation network slice ranking recommendation index, and the other is an NDCG ranking model. Those skilled in the art can realize that other recommendation ranking methods can also be used.

[0082] In fact, for the same recommended slice set, we want the correct item to be as early as possible. Therefore, a more reasonable evaluation method is to also consider the position information of the recommended item. MAP (Mean Average Precision) is designed to achieve this goal, which evaluates whether the recommended result that the user likes is as early as possible. The MAP evaluation network slice ranking recommendation index is as follows:

[0083] For N network slice recommendation lists, each list has n recommended network slices and different network slice rankings, and the MAP of each network slice recommendation list is calculated, as follows:

[0084] In the formula, MAP represents the average accuracy, P i (k) represents the accuracy of the i-th recommended list at position k, δ ij is an indicator function, indicating whether the j-th network slice of the i-th recommended list is a correct recommendation;

[0085] The network slice recommendation list with the highest MAP value is selected as the list recommended to the user.

[0086] The NDCG ranking model is as follows:

[0087] The NDCG (Normalized Discounted Cumulative Gain) value is calculated as an evaluation index for quantifying the recommendation ranking, and the calculation formula is as follows:

[0088] In the formula, IDCG i is the maximum DCG value that can be taken by the i-th network slice recommendation list, which serves to ensure that the ideal value of NDCG is 1, thereby facilitating comparison. N is the total number of network slice recommendation lists, and n is the total number of network slices in each network slice recommendation list, is an indicator function, indicating the score of the j-th candidate network slice in the i-th network slice recommendation list.

[0089] In another embodiment, the present application proposes a network slice recommendation system combined with a routing model corresponding to the method of embodiment one, comprising:

[0090] A network slice construction module is configured to count real-time traffic to analyze network load and demand, obtain traffic distribution characteristics of the entire network, and pre-construct basic network slices according to the traffic distribution characteristics;

[0091] A network slice deployment module is configured to deploy the constructed basic network slices on a switching node in the form of an OpenFlow protocol flow table;

[0092] A classification module is configured to classify the network slices deployed on the switching node according to bandwidth and latency classification thresholds;

[0093] A virtualization module is configured to obtain an association relationship between various network slices and physical resources by using NFV to realize virtualization of physical resources;

[0094] An SDN controller is configured to generate a routing strategy according to network load and traffic conditions;

[0095] An optimal link selection module is configured to deploy the routing strategy and obtain an optimal link from the existing routing strategy;

[0096] A recommendation ranking module is configured to select and assign network slices to users from various network slices, recommend and rank the assigned network slices, and recommend the network slices to the users in order through the optimal link.

[0097] The implementation manners of the modules and the functions of the modules in the system are completely consistent with the steps of the method of embodiment one, and thus will not be described here.

[0098] In another embodiment, the present application proposes a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the network slice recommendation method combined with a routing model as described in embodiment one.

[0099] In another embodiment, the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the network slice recommendation method based on a combined routing model when executing the computer program.

[0100] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium which can contain or store programs for use by or in connection with an instruction execution system, apparatus or device. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the above. More specific examples of computer storage medium can include one or more wires, portable computer disks, hard drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0101] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0102] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled in the art, some improvements and refinements without departing from the principles of the present application shall be considered within the protection scope of the present application.

Claims

1. A network slicing recommendation method combining a routing model, characterized in that, The method comprises the following steps: statistically analyzing real-time traffic to analyze network load and demand, obtaining traffic distribution characteristics of the whole network, and pre-constructing basic network slices according to the traffic distribution characteristics; the constructed basic network slices are deployed on the switching nodes in the form of OpenFlow protocol flow tables; classifying the network slices deployed on the switching nodes according to bandwidth and delay classification thresholds, using NFV to realize virtualization of physical resources to obtain the association relationship between the network slices and the physical resources, generating a routing strategy by an SDN controller according to the network load and the traffic, and deploying the routing strategy to obtain an optimal link from the existing routing strategy; selecting network slices from the network slices to be distributed to users, and recommending and sorting the distributed network slices in order to recommend the network slices to the users through the optimal link. 2.The method of recommending network slices in combination with a routing model according to claim 1, wherein, The method for classifying the network slices deployed on the switching nodes according to the bandwidth and delay classification thresholds is matrix decomposition, and the method is as follows: there are m network slices, n bandwidth and delay classification thresholds, and an m x n-dimensional slice-classification threshold interaction matrix R; the slice-classification threshold interaction matrix R is decomposed into two low-rank matrices, a slice matrix U and a classification threshold matrix V, and the product of the two matrices is used to approximate R, and the formula is as follows: R≈U T V Column vectors U of the slice matrix U i Representative slice hidden vector, column vectors V of the classification threshold matrix V j Representative classification threshold hidden vector; To learn the latent vectors, the idea of linear regression is adopted to minimize the squared error between the original ratings and the predicted ratings, and the objective function L(U, V) is obtained as follows: wherein R ij denotes the original score, denotes the predicted score, K is the number of (i, j) pairs for which the score is known, and λ is a regularization factor, the matrix norm represents a matrix norm; the method for finding a local minimum of the objective function is a gradient descent method or an alternating least squares method. 3.The method of recommending network slices in combination with a routing model according to claim 1, wherein, The specific process of using NFV to realize virtualization of physical resources is as follows: a virtual link and nodes at both ends of the link are defined as a component, and the component includes a free component, a limited component and a fixed component; it is determined whether the free component exists in the virtual network, if the free component exists, the free component is mapped, after a mapping of the free component is completed, all limited components connected to the free component are mapped, and if the free component does not exist, the limited components in the virtual network are mapped; Mapping of a fixed component is completed by mapping of a fixed link, mapping the fixed link to two mapped virtual nodes on the virtual network With the determined shortest path. 4.The method of recommending a network slice in combination with a routing model according to claim 3, wherein, The specific process of mapping the free component is as follows: In all the free components of the virtual network, the free component with the largest link bandwidth requirement is selected as the free component to be mapped using a greedy algorithm and the nodes at both ends of the link are denoted as follows: in the set S of free components, all node pairs are sequentially traversed to determine the shortest path, until a path satisfying the node and link constraints of the free component is found; the node and link constraints of the free component include a hop limit of a physical path, a computing capacity constraint of two physical nodes at both ends of the physical path and a bandwidth constraint of the physical path. 5.The method of recommending network slices in combination with a routing model according to claim 1, wherein, The method for obtaining the optimal link from the existing routing strategy is to perform network pruning by using NetSliming, and the method is as follows: the BN layer in the CNN network is used to learn the pruning parameters end to end by using the Gamma parameter in the BN layer, and the method comprises the following steps: First, train the model with a regularized penalty on the BN's Gamma parameters, the loss function L of the channel pruning method for BN layer Gamma parameters becomes bng becomes where (x, y) represents the input and target of training, and W represents the weight for training, the loss value of the normal training of the CNN network is g(γ), the sparsity penalty term of the scaling factor is g(γ), the scaling factor is γ, and the balance coefficient is λ; then, all the Gamma parameters in the CNN network are sorted, and according to a pruning ratio given by a person, channels with Gamma parameters less than a set value are removed. 6.The method of recommending network slices in combination with a routing model according to claim 1, wherein, The method for recommending and sorting the distributed network slices is to evaluate network slice sorting recommendation indexes by using MAP, and the method is as follows: For N network slice recommendation lists, each list has n recommended network slices and the network slices are sorted differently, the MAP of each network slice recommendation list is calculated, and the formula is as follows: where MAP represents the mean accuracy, P i (k) represents the accuracy of the ith recommendation list at position k, δ ij is an indicator function that indicates whether the jth network slice of the ith recommendation list is a correct recommendation; the network slice recommendation list with the highest MAP value is selected as the list recommended to the user. 7.The method of claim 1, wherein, The method for recommending and ranking the allocated network slices is an NDCG ranking model, and specifically as follows: The NDCG value is calculated as an evaluation index for quantifying the recommendation ranking, and the calculation formula is as follows: where IDCG i is the maximum DCG value that can be taken by the i-th network slice recommendation list, N is the total number of network slice recommendation lists, n is the total number of network slices in each network slice recommendation list, is an indicator function, representing the score of the jth candidate network slice in the ith network slice recommendation list. 8.A network slice recommendation system combined with a routing model, characterized in that, It comprises: a network slice construction module, configured to count real-time traffic to analyze network load and demand, obtain traffic distribution characteristics of the whole network, and pre-construct basic network slices according to the traffic distribution characteristics; a network slice deployment module, configured to deploy the constructed basic network slices on switching nodes in the form of OpenFlow protocol flow tables; a classification module, configured to classify the network slices deployed on the switching nodes according to bandwidth and latency classification thresholds; a virtualization module, configured to obtain an association relationship between various network slices and physical resources by using NFV to realize virtualization of the physical resources; an SDN controller, configured to generate a routing strategy according to network load and traffic conditions; an optimal link selection module, configured to deploy the routing strategy and obtain an optimal link from the existing routing strategy; a recommendation and ranking module, configured to select network slices from various network slices to be allocated to users, recommend and rank the allocated network slices, and recommend the network slices to the users in sequence through the optimal link.

9. A computer readable storage medium storing a computer program, characterized in that, The computer program enables a computer to execute the network slice recommendation method combined with a routing model according to any one of claims 1-7.

10. An electronic device, comprising: It comprises: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the network slice recommendation method combined with a routing model according to any one of claims 1-7.

Images