A network slicing method based on data clustering

Abstract

The application discloses a network slice method based on data clustering, and comprises the following steps: firstly, a plurality of user data are acquired, the characteristics in the user data are normalized, and corresponding threshold values are set; it is judged whether the characteristic values are greater than the corresponding threshold values; if yes, 1 is assigned; if no, 0 is assigned; secondly, the characteristic frequent occurrence times of each characteristic are defined as the total times of the characteristic values being 1 in single user data, and the characteristic frequent occurrence times of each characteristic are obtained by scanning the user data; then, the minimum characteristic frequent occurrence time is set, the characteristics with the characteristic frequent occurrence times less than the minimum characteristic frequent occurrence time are deleted, and the remaining characteristics are arranged according to the characteristic frequent occurrence times; a tree diagram is constructed according to the arranged characteristics, frequent item sets are found in the tree diagram, and a normalized hypergraph is made; finally, the strong correlation between the characteristics is judged by using the obtained hypergraph, and a network slice resource allocation scheme is made. The application can reasonably, efficiently and fairly allocate the network slice resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of 5G network, and particularly relates to a network slicing method based on data clustering. BACKGROUND

[0002] With the development of wireless communication technology, in recent years, mobile communication system has gradually transitioned from the fourth generation mobile communication technology system to the fifth generation mobile communication technology era. Compared with 4G, 5G has higher uplink and downlink rates, greater connection capacity and extremely low end-to-end latency. It not only enhances the technology on the basis of 4G, but also supports more rich communication service application scenarios, such as augmented reality, virtual reality, high-definition video services and customized services for vertical industries and commercial fields. The "5G Spectrum Recommendation White Paper" published by IMT-2020 classifies these scenarios into three application scenarios, which are enhanced mobile broadband, ultra-high reliability and low latency communication and massive machine type communication. In order to better serve the application in the classic scenario, the next generation mobile network alliance proposes to use virtualization technologies such as software-defined network and network function virtualization to abstract a layer of virtual network on the public physical network infrastructure, integrate the hardware infrastructure together to form logically isolated network resources, which is called network slicing. Network slicing has become one of the basic concepts to improve the efficiency and plasticity of 5G network, and its core idea is to build multiple end-to-end logical networks on shared infrastructure to provide customized resources for different vertical industries. Network slicing can provide scalable and elastic network resource management and configuration. By customizing the functions of different network slices, operators can provide customized network slices to customers. For different service level agreements and quality of service, operators can meet the needs of different vertical industries by flexibly customizing and combining slice resources. Therefore, it is of great significance to study the resource management algorithm in the network slicing environment.

[0003] The existing dynamic allocation scheme of network slice resources mostly lacks the trade-off consideration of the two contradictory demands of resource utilization and tenant isolation. How to improve the resource utilization efficiency of the operator while flexibly ensuring the performance isolation of the network slice is one of the difficulties of the dynamic allocation of network slice resources. The essential reason for the above demand contradiction is the random variation characteristics of slice load, which leads to the actual demand of slice for resource quantity also randomly changes. The stereotyped resource fixed reservation scheme cannot adapt to the actual demand quantity of the time-varying resource of the slice. How to dynamically allocate resources for the slice according to the load variation characteristics is the second difficulty to be solved. And most of the existing network slice mapping problems are based on the assumption of fixed resource reservation to consider and research, lacking the optimization of resource elasticity and utilization. In view of the demand for improving the resource utilization efficiency and elastic allocation of the mapping problem, how to establish a flexible resource sharing scheme is one of the difficulties of the slice mapping problem facing resource sharing. Therefore, in order to meet the comfortable experience demand of users and improve the network resource utilization, a fair and easy-to-implement resource management mechanism needs to be established, so that the network slice can realize different business functions on the same physical network. SUMMARY

[0004] The purpose of the present application is to provide a network slice method based on data clustering. The present application can reasonably and efficiently and fairly allocate the resources of the network slice, so that the overall utility, operating cost benefit and resource utilization rate of the network reach the ideal effect, and the differentiated needs of users can be met.

[0005] The technical scheme of the present application: a network slice method based on data clustering, comprising the following steps:

[0006] Step 1, obtaining user data, normalizing each characteristic in the user data, and setting the threshold value corresponding to the normalized characteristic, judging whether the normalized characteristic value is greater than the corresponding threshold value, if greater, assigning a value of 1 to the characteristic, if less, assigning a value of 0, completing the initialization of user data;

[0007] Step 2, defining the characteristic frequent occurrence times as the total number of times that the characteristic value is 1 in a single user data, scanning all the initialized user data to obtain the characteristic frequent occurrence times of each characteristic;

[0008] Step 3, setting the minimum characteristic frequent occurrence times, deleting the characteristics with characteristic frequent occurrence times less than the minimum characteristic frequent occurrence times, and arranging the remaining characteristics according to the characteristic frequent occurrence times;

[0009] Step 4, constructing a tree diagram according to the arranged characteristics, finding the frequent item set in the tree diagram, and making a normalized hypergraph;

[0010] Step 5, using the obtained hypergraph to judge the strong correlation between characteristics and formulate a network slice resource allocation scheme.

[0011] The network slicing method based on data clustering, in step 1, the normalization is to use a n to represent the value of the feature a of user n before normalization, and a hou to represent the value of the feature a of user n after normalization, a max to represent the maximum value of feature a in all users, a min to represent the minimum value of feature a in all users, and the normalization formula is as follows:

[0012]

[0013] The network slicing method based on data clustering, in step 1, the feature with a value of 1 is a feature with prominent demand for the user, and the feature with a value of 0 is a feature with small demand for the user.

[0014] The network slicing method based on data clustering, in step 4, the construction of the tree diagram is to set the root node as null, connect the sorted features to the root node according to the first feature of the first user data, then connect the remaining features of the first user in the order from front to back, and mark the number 1 after each node, then connect the feature points of the second user, connect the same feature points and mark the number 1 after the node, connect a new branch node after the previous node to represent the feature with a different feature point and assign the initial value as 1, and so on to traverse all user data to complete the construction of the tree diagram.

[0015] The network slicing method based on data clustering, the finding of the frequent item set is to set a threshold G, select a feature node P, find all paths containing the feature node in the tree diagram and find all feature nodes on the same path as the feature node; then start from the end of P node and traverse to the root node, calculate the common path number containing node P and the traversed node, divide the common path number by the total number of users to get the weight value, if the obtained weight value is greater than the set threshold G, stop traversing forward, and traverse all features in the tree according to this method to get the frequent item set.

[0016] The network slicing method based on data clustering, the production of the normalized hypergraph is to put the features in a single frequent item set into the same hyperedge as different vertices, and all different hyperedges composed of frequent item sets form a hypergraph; wherein the weight value H x of the frequent item set is taken as the initial weight value of the hyperedge, Y x is the normalized weight value of the hyperedge, and Y x and H x have the following relationship:

[0017]

[0018] Finally, normalize all hyper-edges by the above formula to obtain the normalized hypergraph.

[0019] The aforementioned network slicing method based on data clustering, when the normalized hypergraph has the same node in different hyper-edges, if the weight of the hyper-edge is large, the characteristic nodes in different hyper-edges are put into the same network slice for resource allocation when the network slice is made;

[0020] If the hyper-edge weight is small and cannot be merged with other hyper-edges containing the same characteristics but with larger weights, the characteristics in all hyper-edges with small weights and the characteristics deleted in the early stage are merged into a network slice, and resource scheduling is uniformly performed.

[0021] The aforementioned network slicing method based on data clustering, if the clustering result of the network slice has a highest limit T in quantity, and the number of clustering results t exceeds the highest limit, the t-T hyper-edges with small weights in the current multiple hyper-edges are selected, one of them is taken, and it is judged which characteristic in the hyper-edge has a high frequency of occurrence by comparing the numerical results of the previous traversal of the user characteristics, and then all the characteristic nodes in this hyper-edge are put into the hyper-edges containing the characteristic nodes in the T hyper-edges with high weights, if the T hyper-edges with high weights do not contain this characteristic node, the nodes in this hyper-edge are randomly put into one of the T hyper-edges with high weights, and the remaining other hyper-edges with small weights are also merged according to the above method.

[0022] Compared with the prior art, the application provides a method that does not need to traverse the original data multiple times by applying the hypergraph clustering idea and using the improved tree diagram method, saves computer resources, puts characteristics with strong correlation into a network slice, and thus the resources of the network slice can be better utilized, realizing reasonable, efficient and fair allocation of the resources of the network slice. From the perspective of network operators, the overall utility of the system, the operating cost and benefit of the network, and the resource utilization rate achieve ideal effects, from the perspective of users, on the one hand, the user's demand is met and the service quality is improved, on the other hand, the user's personalized service is realized, and different users are faced with, according to the differentiated demand characteristics of the users, realizing customized and fair service. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a flowchart of the application;

[0024] Figure 2 is a schematic diagram of the normalization and value assignment process of the simulation sample of step 2 of the application;

[0025] Figure 3 A simulation sample for step 3 of the present application;

[0026] Figure 4 An example of a dendrogram obtained from a simulation sample of the present application;

[0027] Figure 5 A simulation sample of the present application obtained frequent item sets;

[0028] Figure 6 A simulation sample of the present application to the normalized hypergraph. DETAILED DESCRIPTION

[0029] The present application will be further described below in conjunction with the accompanying drawings and examples, but not as the basis for limiting the present application.

[0030] Example: a network slicing method based on data clustering, as shown in Figure 1 , comprising the following steps:

[0031] Step 1, obtain user data, normalize each characteristic in the user data, and set the threshold value corresponding to the normalized characteristic, judge whether the normalized characteristic value is greater than the corresponding threshold value, if greater than 1, if less than 0, complete the user data initialization; In this embodiment, the normalization is represented by a n , a hou represents the value of the feature a of user n before normalization, a max represents the maximum value of the characteristic a in all users, a min represents the minimum value of the characteristic a in all users, then the normalization formula is as follows:

[0032]

[0033] Set a threshold value a yu of a characteristic a, scan the normalized value of characteristic a in all user data samples, assign 0 to the characteristic a value of each user whose characteristic a value is less than the set threshold value, otherwise set the characteristic a value of the user to 1, the default value of 1 is the characteristic of the user with more prominent demand, the value of 0 is the characteristic of the user with less demand, after initializing all different characteristics according to the above rules, the initialization is completed.

[0034] Figure 2 is a simulation sample for this step to normalize and value assignment process, Figure 2The present example provides five characteristics with different needs for five different users. The initial data is normalized, and a threshold of 0.5 is set for each characteristic. If the normalized value of each characteristic is higher than the threshold, 1 is filled in the initialized table, representing that the user has high quality needs for the characteristic. Otherwise, 0 is filled, representing that the user has no high needs for the characteristic.

[0035] Step 2, define the frequency of characteristics as the total number of times that the value of the characteristic is 1 in the single user data. Scan all the initialized user data to obtain the frequency of each characteristic.

[0036] Step 3, set the minimum frequency of characteristics. Remove the characteristics with a frequency less than the minimum frequency, and arrange the remaining characteristics according to the frequency. In this step, the minimum frequency of characteristics (indicated as minimum support in the figure) is set to 1. If the frequency of a characteristic is less than the minimum frequency, the characteristic is removed. Compare the frequency of all characteristics with the minimum frequency, remove the characteristics with a frequency less than the minimum frequency, and arrange the normalized user data of the remaining characteristics according to the frequency from large to small.

[0037] Figure 3 For the schematic diagram of this step, the minimum support in the example is set to 1. Since the frequency of all characteristics is greater than or equal to 1, no characteristics are filtered out. Then, the characteristics are reordered from front to back according to the frequency, preparing for the subsequent tree making.

[0038] Step 4, construct a tree diagram according to the arranged characteristics, find frequent item sets in the tree diagram, and make a normalized hypergraph. The construction of the tree diagram is to set the root node as null, connect the first characteristic of the first user data to the root node, then connect the remaining characteristics of the first user in the order from front to back, and mark the number 1 after each node. Then, connect the characteristic points of the second user, if the first characteristic of the second user is the same as the first characteristic of the first user, add 1 to the value of the characteristic point on the tree. The second characteristic point of the second user is the same as the second characteristic point of the first user, and the value of the characteristic point on the original tree is added by 1. Until the characteristic points of the second user and the characteristic points of the first user appear different, a branch node is connected after the previous node to represent the characteristic, and the initial value is assigned to 1. In this way, the tree diagram is constructed until all users are traversed.

[0039] Figure 4As an example of the tree diagram obtained in this step, the tree diagram is sequentially constructed according to different characteristic values of the example sample and different high-demand characteristic combinations of different users, to prepare for the subsequent search for frequent item sets.

[0040] After obtaining the tree diagram, a threshold G is set, which is 0.4 in this embodiment, and a characteristic node P is selected, all paths containing the characteristic node P in the tree diagram are found, and all characteristic nodes on the same path as the characteristic node P are found; then starting from the end close to the P node, the traversal is performed in the direction of the root node, the number of common paths containing the node P and the traversed node is calculated, and the weight value is obtained by dividing the number of common paths by the total number of users, if the obtained weight value is greater than the set threshold G, the traversal is stopped, and all characteristics in the tree are traversed according to this method to obtain the frequent item set.

[0041] Figure 5 As an example of the frequent item set obtained in this step, Figure 5 shows the tree diagram according to Figure 4 After the traversal of each characteristic, the frequent item set and its weight are found, and the weight is obtained by dividing the number of paths containing the characteristics in the frequent item set in the tree diagram by the total number of users. (For example, the weight value of the third frequent item set is obtained as )

[0042] After obtaining the frequent item set, the production of the normalized hypergraph is to put the characteristics in a single frequent item set into the same hyperedge as different vertices, and the set of different hyperedges composed of all frequent item sets to form a hypergraph; wherein the weight H x of the frequent item set is taken as the initial weight of the hyperedge, Y x is the normalized weight of the hyperedge, and the relationship between Y x and H x is as follows:

[0043]

[0044] Finally, all hyperedges are normalized by the above formula to obtain the normalized hypergraph.

[0045] Figure 6 As an example of the normalized hypergraph obtained in this step, the left graph is the hypergraph before normalization, and the right graph is the hypergraph after normalization.

[0046] Step 5, using the obtained hypergraph to judge the strong correlation between characteristics and formulate a network slice resource allocation scheme, in this step, the manager can obtain intuitive judgment through the normalized hypergraph, and formulate a suitable network slice resource allocation method.

[0047] In the embodiment, there are some special cases, for example, when the same node appears in different hyperedges after normalization, if the weight of the hyperedge is large, the characteristic nodes in different hyperedges are put into the same network slice for resource allocation when the network slice is made;

[0048] If the weight of the hyperedge is small and cannot be combined with other hyperedges containing the same characteristics but with larger weights, the characteristics in all hyperedges with small weights and the characteristics deleted in the early stage are combined into a network slice for unified resource scheduling.

[0049] If the number of clustering results of the network slice has a highest limit T, and the number of clustering results t exceeds the highest limit, t-T hyperedges with small weights are selected from the current multiple hyperedges, one of the hyperedges is taken, and it is judged which characteristic in the hyperedge has a high frequency of occurrence by comparing the numerical results of the traversal of the user characteristics, then all the characteristic nodes in the hyperedge are put into the hyperedge containing the characteristic node in the T hyperedges with high weights, if the T hyperedges with high weights do not contain the characteristic node, the nodes in the hyperedge are randomly put into one of the T hyperedges with high weights, and the remaining other hyperedges with small weights are also combined according to the above method.

[0050] In summary, the application provides a method for not traversing the original data multiple times by applying the hypergraph clustering idea and using the improved tree diagram method, which puts the characteristics with strong correlation into a network slice under the premise of saving computer resources, so that the resources of the network slice can be better utilized, and the resources of the network slice are reasonably and efficiently and fairly allocated. From the perspective of network operators, the overall utility of the system, the operating cost and benefit of the network and the resource utilization rate achieve ideal effects, from the perspective of users, on the one hand, the user's demand is met and the service quality is improved, on the other hand, the user's individualized service is realized, and different users are faced with, according to the differentiated demand characteristics of the users, the customized and fair service is realized.

Claims

1. A network slicing method based on data clustering, characterized in that: The method comprises the following steps: Step 1, obtaining user data, normalizing each characteristic in the user data, setting the threshold value corresponding to the normalized characteristic, judging whether the normalized characteristic value is greater than the corresponding threshold value, if greater, assigning a value of 1 to the characteristic, if less, assigning a value of 0, and completing user data initialization; Step 2, defining the characteristic frequent occurrence times as the total number of times that the characteristic value is 1 in single user data, scanning all initialized user data to obtain the characteristic frequent occurrence times of each characteristic; Step 3, setting the minimum characteristic frequent occurrence times, deleting the characteristics with characteristic frequent occurrence times less than the minimum characteristic frequent occurrence times, and arranging the remaining characteristics according to the characteristic frequent occurrence times from large to small; Step 4, constructing a tree diagram according to the arranged characteristics, finding frequent item sets in the tree diagram, and making a normalized hypergraph; Step 5, using the obtained hypergraph to judge the strong correlation between characteristics and develop a network slice resource allocation scheme; In step 4, the construction of the tree diagram is to set the root node as null, connect the first characteristic of the first user data to the root node according to the arranged characteristics, then connect the remaining characteristics of the first user in the order from front to back, and mark the number 1 after each node, then connect the characteristics of the second user, connect the same characteristic points, mark the number 1 after the node, connect a branch node representing the characteristic after the previous node for different characteristic points, and assign an initial value of 1, and so on to traverse all user data to complete the construction of the tree diagram; The search for frequent itemsets involves setting a threshold. Select a feature node In the tree diagram, find all paths containing the node with that characteristic, and find all characteristic nodes on the same path as the node with that characteristic; then, self-destruct... Traverse from the nearest node towards the root node, calculating the number of nodes that are simultaneously contained within the root node. The weight is calculated by dividing the number of common paths between the visited nodes and the total number of users. If the weight is greater than a set threshold, the user is considered to have a weight. Then stop traversing forward and traverse all the features in the tree in this way to obtain frequent itemsets; The normalization hypergraph is made by putting the characteristics of single frequent item set into the same hyperedge as different vertices, different hyperedge sets composed of all frequent item sets, and thus the hypergraph is formed; wherein the weight of the frequent item set As the initial weight of the hyperedge, let The normalized weight of the hyperedge is The relationship between is shown as follows: ； Finally, normalize all hyperedges by the formula to obtain a normalized hypergraph.

2. The method of claim 1, wherein: In step 1, the normalization is given by where x represents the value of the feature of the user before normalization, and where x represents the value of the feature of the user after normalization, and where x represents the value of the feature of the user before normalization, and where x represents the value of the feature of the user after normalization, and where x represents the value of the feature of the user before normalization, and where x represents the value of the feature of the user after normalization, and where x represents the value of the feature of the user before normalization, and where x represents the value of the feature of the user after normalization, and where x represents the value of the feature of the user before normalization, and where x represents the value of the feature of the user after normalization, and 。 3.The network slicing method based on data clustering according to claim 1, characterized in that: In step 1, the characteristic with a value of 1 is the characteristic with a more prominent demand for the user, and the characteristic with a value of 0 is the characteristic with a smaller demand for the user. 4.The network slicing method based on data clustering according to claim 1, characterized in that: When the same node appears in different hyperedges in the normalized hypergraph, if the weight of the hyperedge is large, the characteristic nodes in different hyperedges are put into the same network slice for resource allocation when making the network slice; If the hyperedge weight is small and cannot be combined with other hyperedges containing the same characteristic but with larger weights, the characteristics in all hyperedges with small weights and the characteristics deleted in the early stage are combined into a network slice for unified resource scheduling.

5. The method of claim 1, wherein: If there is a maximum limit to the number of clustering results for network slices. And the number of clustering results If the maximum limit is exceeded, then select the superedges with smaller weights from the current multiple superedges. Take one of the hyperedges and determine which characteristic of that hyperedge appears most frequently by comparing it with the numerical results of the previous traversal of user characteristics. Then, place all the characteristic vertices of that hyperedge into the hyperedge with the highest weight. Among the superedges containing vertices with this characteristic, if the superedge weight is higher... If a hyperedge does not contain a node with this characteristic, then all nodes in that hyperedge are randomly placed into the hyperedge with higher weights. One of the superedges selected is the one with the smaller weight. The remaining superedges with smaller weights are also merged using the same method.

Images