A SDN link connectivity prediction method based on SMA-SVM

Abstract

The application discloses a kind of based on SMA-SVM's SDN link connectivity prediction method, comprising: constructing the link connectivity vector reflecting target network data plane;Target SDN network data plane is modeled;Construct target SDN network link connectivity similarity vector;Using difference integrated moving average autoregressive model ARIMA, network link connectivity similarity vector sequence is analyzed in time series, and the change rule of link connectivity similarity vector with time is extracted;Network link connectivity vector classifier is constructed and optimized using slime mould optimization algorithm SMA;According to the prediction value output by network link connectivity vector classifier, locate fragile link and node in target network.The method of the application can improve the prediction accuracy of network link connectivity based on SDN to some extent, compared with other link prediction methods, the method has the advantages of better accuracy and stability.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of network reliability, and particularly relates to a slime mold algorithm parameter optimization and SDN link connectivity prediction method based on SMA-SVM. BACKGROUND

[0002] With the development of space technology, the competition in the space field is increasingly fierce. Space-based networks are the research direction of the next generation of network communication due to their large communication capacity, high transmission rate and wide coverage. SDN technology is applied to space-based networks due to its more advanced network architecture concept. Due to the complex radiation environment in space and the mobility of the on-board system, the reliability of the SDN network link faces great challenges. Unreliability has a greater and greater impact on the overall network performance of space-based networks. As the most basic reliability indicator of the network, the connectivity of the link node is particularly important, so the prediction method and research of the connectivity of the link node are particularly important.

[0003] The research on the link node connectivity prediction of SDN network can provide support for the upper application of SDN network traffic control, routing selection, link reliability reinforcement, and improvement of network service quality and flexibility, and provide technical solutions for SDN network in dealing with sudden traffic request and link load pressure. In recent years, there have been numerous studies on network link prediction, such as link prediction for urban road network, link prediction based on social network, and link prediction for opportunistic network. Although there have been many studies on link prediction in different networks, there are still many deficiencies in the research on network link prediction under the architecture of SDN. For example, the Psor link prediction algorithm based on complex network, which defines the Psor index and Psor similarity index of the node to predict the link by comprehensively considering the degree of the node and its neighbor nodes. Another link prediction method based on Kshell decomposition and neighbor node degree denoising, which sorts the importance of all nodes in the network through kshell decomposition, then comprehensively judges the importance of the node from the local perspective combined with the degree of the neighbor node, and finally uses the optimized network data for link prediction. Both methods study the structural characteristics of nodes in the link to predict the link, but they do not dig out the importance of the interactive behavior characteristics between different nodes for link prediction. Link prediction is a method to estimate the probability of future link formation between network nodes. There are two main types of link prediction methods. The first type is to use the characteristics of the graph to use local and global network features to network entities. Most of this type of method outputs similarity evaluation scores. The second type of method uses information related to the available link to create a network representation, and this type is also considered as a learning-based problem. With the development of deep learning technology, using deep learning to study link prediction is a hot research direction. In the learning-based method, feature engineering is used to select better feature data, thereby improving the training effect of the model. Representation learning can be divided into supervised, semi-supervised and unsupervised learning. In supervised representation learning, feature engineering is used to extract features. In semi-supervised representation learning, part of the data is labeled and learned using domain knowledge, and the other part is automatically learned by algorithm. Unsupervised representation learning extracts effective features or representations from unlabeled data, mainly used for data preprocessing or data dimensionality reduction in supervised learning.

[0004] The purpose of network link prediction is to improve network robustness and reliability. Current research on network reliability mainly focuses on functional reliability and component reliability of the network. Because the network is a complex system, it cannot be measured by a single indicator. The current mainstream research framework for network reliability is hierarchical research. The research methods include neural network-based reliability research, fuzzy theory-based reliability research, and heuristic algorithms based on biological populations. When the network is running, its state may change over time according to the changes in the state of its components. From the perspective of reliability, the reliability of component state is based on a specific random mechanism, and specific network applications also affect the overall reliability of the network from the upper layer. For example, under the influence of cloud computing, big data, and mobile traffic, the traditional network architecture generates a large amount of diverse traffic. These data traffic is difficult to extract features using general methods, thereby impacting and challenging the prediction of network reliability.

[0005] One of the difficulties in performing link prediction tasks on complex networks is the dynamic change characteristics exhibited by the network. To address the dynamic nature and time-varying characteristics exhibited by the network, time series prediction models can more effectively extract the dynamic nature and time-varying characteristics, thereby improving the prediction accuracy of network link reliability. Time series prediction models rely on the order of event occurrence, so even the same data values will produce different results when input into the model after changing the order. Time series are divided into stationary and non-stationary sequences. Stationary sequences are sequences in which observations fluctuate around a fixed level, while non-stationary sequences contain long-term trends or cyclical fluctuations. Non-stationary sequences can be converted to stationary sequences through differencing. Time series prediction models include traditional time series modeling and machine learning-based time series prediction. Traditional time series prediction models include stationary sequence fitting models (ARMA), machine learning-based time series prediction models such as random forests, SVM, and xgboost. In current network link prediction methods, scholars have applied time series prediction models to link reliability prediction. In combination with related optimization algorithms, link reliability prediction has achieved relatively high accuracy, but it does not fully reflect the characteristics of SDN, and the accuracy is still not high enough. SUMMARY

[0006] The present application aims at the problems existing in the prior art, and provides an SDN link connectivity prediction method based on SMA-SVM, which introduces SMA to optimize parameters of a network link connectivity classifier, constructs an SDN link connectivity vector, and extracts features of the link connectivity vector changing over time by using an ARIMA time series prediction model, so as to improve the accuracy of link prediction in the SDN. In the present application, the constructed network link connectivity vector mainly reflects the features of the complex network structure and the node betweenness centrality of the SDN. The link connectivity feature vector is obtained by fusing the link connectivity vector and four similarity feature parameters, the time series features of the link connectivity feature vector are extracted, the feature vector sequence and the connection state are used as samples for training the network link connectivity classifier, the parameters of the network link connectivity classifier are optimized by SMA to improve the classification accuracy of the classifier, and finally the network link connectivity classifier performs a link prediction task according to the current link state information.

[0007] The technical solution for achieving the object of the present application is as follows: an SDN link connectivity prediction method based on SMA-SVM, which comprises the following steps:

[0008] Step 1: constructing a link connectivity vector reflecting a target network data plane according to the complex network structure and the betweenness centrality features in the software defined network (SDN);

[0009] Step 2: modeling the target SDN network data plane: constructing a three-layer data forwarding model based on the functional attributes of the data plane of the OpenFlow general forwarding framework, which comprises an analysis layer, a flow table matching layer and a forwarding scheduling layer;

[0010] Step 3: the SDN control layer dynamically collects global network topology information of the target SDN network data plane by using a network snapshot, and constructs a target SDN network link connectivity similarity vector according to the flow rule features in the three-layer data forwarding model structure and the global network topology information;

[0011] Step 4: performing time series analysis on the network link connectivity similarity vector sequence by using the autoregressive integrated moving average (ARIMA) model, and extracting the change rule of the link connectivity similarity vector over time;

[0012] Step 5: constructing and optimizing the network link connectivity classifier by using the slime mold optimization (SMA) algorithm;

[0013] Step 6: positioning the fragile links and nodes in the target network according to the prediction value output by the network link connectivity classifier.

[0014] Further, the step 1 of constructing a link connectivity vector reflecting a target network data plane according to the complex network structure and the betweenness centrality features in the software defined network (SDN) comprises the following specific process:

[0015] Step 1.1, using network node intermediary influence coefficient As a reflection of the connectivity and robustness between data plane nodes, Calculated by the following formula (1):

[0016]

[0017] Wherein, The number of shortest paths between nodes x i , x j , The number of all shortest paths in x i , x j , V is the node set;

[0018] Step 1.2, using node pair common connection node influence coefficient As a reflection of the connectivity and robustness between data plane nodes, Calculated by the following formula (2):

[0019]

[0020] Wherein, F L (x i ) is the function of the node set connected to node x i , F L (x j ) is the function of the node set connected to node x j ;

[0021] Step 1.3, according to the network node intermediary influence coefficient and the node pair common connection node influence coefficient to construct the link connectivity vector The calculation method is shown in the following formula (3):

[0022]

[0023] Wherein, the parameter α ∈ [0, 1], used to control the weight of intermediary influence coefficient and node pair common connection node influence coefficient.

[0024] Further, step 2 said to the target SDN network data plane modeling, the specific process includes:

[0025] Step 2.1, based on OpenFlow general forwarding framework to build target SDN network three layer data forwarding model, respectively, the analysis layer, flow table matching layer and forwarding scheduling layer, wherein the analysis layer is responsible for parsing packet field, and the parsed field is passed to the flow table matching layer;

[0026] Step 2.2: extract the flow rule R in the flow entry and construct a triple representation, i.e. R = <M, P, I>, where M represents the matching field, P represents the priority, and I represents the instruction;

[0027] Step 2.3, according to the horizontal structural characteristics of the flow rule R shown between the flow tables, analyze the pre- and post-connection relationship between the nodes in the data plane, and set the data plane link connectivity state according to the pre- and post-connection relationship;

[0028] Step 2.4, the forwarding scheduling layer executes data forwarding task according to the flow table matching operation result, abstracts the data packet flow direction characteristics between different nodes in the data plane as different link connectivity similarity attributes, and constructs a network link connectivity similarity vector based on link connectivity similarity as a feature.

[0029] Further, step 2.3 specifically includes:

[0030] (1) According to the behavior of the flow rule, the dynamic nature of the flow rule is affected by the context of the current state, and a rule is affected by the pre- and post- rules in the table, which is embodied as the vertical structure relationship of the rules in the table;

[0031] (2) In the forwarding process, a flow rule R will be associated with a pre-remote table or a post-remote table, and this association is reflected in the connectivity between different nodes in the data plane, which is determined by the horizontal relationship between the flow tables.

[0032] Further, the SDN control layer of step 3 utilizes network snapshot to dynamically collect the global topology information of the target SDN network data plane, and constructs a network link connectivity similarity vector based on the flow rule characteristics in the three-layer model structure and the global topology information, the specific process includes:

[0033] Step 3.1, the SDN controller obtains the global network topology information through packet-out and packet-in, and performs a shortest path algorithm on the information, iterates through the members in the shortest path set p in the current topology, and records the information of each node acting as an intermediate node; Where p = {x i，j ...}, x i，j is the shortest path between node i and node j, and p is the shortest path set;

[0034] Step 3.2, fuse the link connectivity vector with the four similarity feature parameters reflecting the target network topology information, i.e. the Jaccard coefficient, the common neighbor weight coefficient AA, the resource allocation coefficient RA, and the preferred connection coefficient PA, to obtain the SDN link connectivity similarity vector SDN_LCV; Specifically includes:

[0035] Step 3.2.1, calculate the improved Jaccard feature vector

[0036]

[0037] where the Jaccard coefficient is calculated as:

[0038]

[0039] where Γ(x i ), Γ(x j ) represent the neighbor set of node x i , x j respectively, and |P| represents the cardinality of set P;

[0040] Step 3.2.2, calculate the improved preference connection feature vector

[0041]

[0042] where the preference connection coefficient is calculated as:

[0043]

[0044] where are the degrees of node x i , x j respectively;

[0045] Step 3.2.3, calculate the improved common neighbor weight feature vector

[0046]

[0047] where the common neighbor weight coefficient is calculated as:

[0048]

[0049] Step 3.2.4, calculate the improved resource allocation feature vector

[0050]

[0051] where the resource allocation coefficient is calculated as:

[0052]

[0053] Step 3.3, construct the target SDN network link connectivity similarity vector V LCV :

[0054]

[0055] Furthermore, step 4 involves using the Differential Integrated Moving Average Autoregressive Model (ARIMA) to perform time series analysis on the network link connectivity similarity vector sequence, extracting the changing patterns of the link connectivity similarity vector over time. The specific process includes:

[0056] Step 4.1: For node pair t in the target network, collect and extract link connectivity vector information within each time interval ε to form a link connectivity feature vector sequence V. LCVS :

[0057]

[0058] Step 4.2: Use the unit root test (ADF) method to perform stationarity analysis on the network link connectivity similarity vector sequence. If it is not stationary, perform difference operations to convert it into a stationary sequence; otherwise, proceed to the next step.

[0059] Step 4.3: Analyze the output of the ACF function. If it is zero, introduce the autocorrelation function to determine the order q of the autocorrelation model; where the autocorrelation function is defined as:

[0060]

[0061] in, It is a function of the covariance value with respect to k. Let be the link connectivity similarity vector corresponding to node pair t;

[0062] Step 4.4, calculate the order p of the moving average model according to equation (15):

[0063]

[0064] in, yes The partial autocorrelation coefficient, which represents the sequence right The intermediate variable series was removed when calculating the correlation. right The transmission effect of short-term correlation;

[0065] Step 4.5: Optimize the parameters of the ARIMA model according to the minimum information criterion;

[0066] Step 4.6: Use a sliding window to extract the link connectivity feature vector sequence obtained in Step 4.2. Use the extracted sequence as input to the ARIMA model. Predict the link connectivity feature vector at the next time step according to Equation (16). Use the features as labels to form a sample set F = [f1, f2, ..., f n ];

[0067]

[0068] Further, step 5 described the use of slime mold optimization algorithm SMA to construct and optimize network link connectivity classifier, the specific process includes:

[0069] Step 5.1, initialize the population size N, the maximum number of iterations and individual position, each individual in the population represents a set of hyperplane weight vectors in the network link connectivity classifier, denoted as w = (w1, w2,..., w4), where the hyperplane represents the classification plane of the network connectivity classifier;

[0070] Step 5.2, construct the network link connectivity classifier based on SVM, the link connectivity classifier hyperplane calculation is shown in formula (17);

[0071] f(w) = sign(δy label k(V LCV , w) + b) (17)

[0072] Where δ is the Lagrange multiplier, which is a coefficient variable, k(*, *) is used to map data to feature space, and linear learners are used in feature space for data classification processing, y label is the sample label, y label = 1 represents connectivity, y label = -1 represents unconnected, w is the weight vector of the hyperplane, b is the base of the hyperplane, and sign(·) is the sign function, which returns 1 when δy label V LCV ·w + b is greater than 0, otherwise 0;

[0073] Step 5.3, calculate the fitness value of all hyperplane weight vectors, the calculation method is shown in formula (18);

[0074]

[0075] Where N correct represents the number of correctly classified samples, N total is the total number of samples, and S(i) represents the fitness of the individual of the hyperplane weight vector;

[0076] Step 5.4, sort the fitness values of all hyperplane weight vector individuals, the sorting method is shown in formula (19),

[0077] SmellIndex = sort(S) (19)

[0078] Where SmellIndex represents the ascending order of individual fitness values;

[0079] Step 5.5, record the best fitness DF and the best position of the hyperplane weight vector individual in the current iteration

[0080] Step 5.6, update the hyperplane weight vector individual moving parameter Parameter Parameter a, b are calculated by formula (20) and (21) respectively

[0081]

[0082]

[0083] where t is the current iteration number, and max_t represents the maximum iteration number

[0084] Step 5.7, calculate the moving parameter of each hyperplane weight vector individual Parameter The calculation method of is shown in formula (22)

[0085]

[0086] where B(i) represents the median of SmellIndex, bF represents the best fitness value obtained in the current iteration process, wF represents the worst fitness value obtained in the current iteration process, and the component r is a moving random value, which is a random value in the interval [0, 1];

[0087] Step 5.8, update the individual position representing the hyperplane weight vector in the current population, and the individual position calculation method is shown in formula (23)

[0088]

[0089] where represents the weight vector of the hyperplane, and represent the weight vectors of two randomly selected hyperplanes, represents the hyperplane weight vector with the highest current fitness value, r is a random number in the interval [0, 1], LB and UB represent the lower limit and upper limit of the search, rand is a random number in the interval [0, 1], and the calculation method of p is shown in formula (24);

[0090] p = tanh |S(i) - DF|, i ∈ 1, 2...n (24)

[0091] Step 5.9, use the individual with the best fitness value obtained as the hyperplane weight vector of the network link connectivity classifier, and complete the construction and optimization process of the network link connectivity classifier.

[0092] Further, the step 6 is to locate the fragile link and node in the target network according to the predicted value output by the network link connectivity classifier, and the specific process includes:

[0093] Step 6.1, the control layer finds the fragile link according to the mapping table, defines the node at the end of the fragile link as a fragile node, and constructs an adjacency matrix A about the fragile node;

[0094] Step 6.2, according to the eigenvector centrality of the network fragile node, the vulnerability of the node is quantified, the node with the lowest eigenvector centrality in the control domain is marked as the fragile node in the domain, and the calculation method of the eigenvector centrality of the fragile node is shown in formula (25);

[0095] V c (v i )=λ -1 A V c (25)

[0096] Where V c is the vulnerability eigenvector of node v i , and λ is the corresponding eigenvalue;

[0097] Step 6.3, formula (25) is iterated to a convergent non-zero solution, the vulnerability eigenvalue of all nodes is obtained, the node with the lowest vulnerability eigenvalue in each control domain is marked, and the controller in the domain to which the marked node belongs is found through the mapping table;

[0098] Step 6.4, the flow table is issued to the data plane containing the intradomain fragile node in the controller mapping range, and the local routing between the intradomain fragile nodes in the control domain is re-planned.

[0099] Compared with the prior art, the present application has the following advantages: 1) compared with the method based on the basic similarity index, the present method combines the self-defined link connectivity vector with four similarity feature parameters according to the characteristics of SDN, and the SDN link connectivity similarity feature vector formed can more comprehensively reflect the characteristics of SDN. 2) the SMA algorithm with higher efficiency is used to optimize the parameters of the network link connectivity classifier, so that the prediction accuracy of the network link connectivity classifier is significantly improved.

[0100] The present application will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0101] Figure 1 It is a flow chart of the SDN link connectivity prediction method based on SMA-SVM of the present application.

[0102] Figure 2 It is a three-layer data plane model diagram.

[0103] Figure 3 Optimization process flowchart of network link connectivity classifier based on slime optimization algorithm.

[0104] Figure 4 Time series analysis and feature extraction process flowchart based on ARIMA model.

[0105] Figure 5 Optimal fitness value change comparison analysis chart for finding hyperplane weight vector based on slime optimization algorithm. DETAILED DESCRIPTION

[0106] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0107] It should be noted that if the present application examples involve "first", "second" and the like, the "first", "second" and the like are only for description purposes and should not be understood as indicating or implying the relative importance of the technical features indicated or the number of technical features indicated. Therefore, the features with "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of various embodiments can be combined with each other, but it must be based on the realization of ordinary skilled in the art, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, nor within the scope of protection claimed by the present application.

[0108] In one embodiment, in combination Figure 1 , a SMA-SVM-based SDN link connectivity prediction method is provided, comprising the following steps:

[0109] Step 1, according to the complex network structure and betweenness centrality characteristics in software defined network (SDN), a link connectivity vector reflecting the target network data plane is constructed; the specific process includes:

[0110] Step 1.1, the network node betweenness influence coefficient as an attribute reflecting the connectivity and robustness between data plane nodes, is calculated by the following formula (26):

[0111]

[0112] wherein, is the number of shortest paths between nodes x i , x j composed of node e, is x i, x j the number of all shortest paths, V is the node set;

[0113] Step 1.2, adopt the node pair common connection node influence coefficient as a reflection of the data plane node connectivity propagation properties, calculated by the following formula (27):

[0114]

[0115] where F L (x i ) is a function of the node set connected to node x i , F L (x j ) is a function of the node set connected to node x j ;

[0116] Step 1.3, according to the network node betweenness influence coefficient and the node pair common connection node influence coefficient to construct the link connectivity vector The calculation method is shown in the following formula (28):

[0117]

[0118] Where the parameter α∈[0, 1], used to control the weight of betweenness influence coefficient and node pair common connection node influence coefficient.

[0119] Step 2, modeling the target SDN network data plane: based on the functional properties of the OpenFlow general forwarding framework data plane, a three-layer data forwarding model is constructed, as shown in Figure 2 , respectively, the parsing layer, flow table matching layer and forwarding scheduling layer; the specific process includes:

[0120] Step 2.1, based on the OpenFlow general forwarding framework to build a three-layer data forwarding model of the target SDN network, respectively, the parsing layer, flow table matching layer and forwarding scheduling layer, wherein the parsing layer is responsible for parsing the data packet field, and passing the parsed field to the flow table matching layer;

[0121] Step 2.2: extract the flow rule R in the flow table item and construct a three tuple representation, that is, R = <M, P, I>, where M represents the matching domain, P represents the priority, and I represents the instruction;

[0122] Step 2.3, according to the horizontal structural characteristics shown by the flow rule R between the flow tables, analyze the preposition and postposition relationship of the connection between the nodes in the data plane, and set the data plane link connectivity state according to the preposition and postposition connection relationship; including:

[0123] (1) According to the behavior of flow rules, the dynamic of flow rules is affected by the context of the current state, and a rule is affected by the pre- and post- rules in the table, which is embodied in the vertical structure relationship of the rules in the table;

[0124] (2) In the forwarding process, a flow rule R will be associated with a pre- or post- remote table, and this association is reflected in the connectivity between different nodes in the data plane, which is determined by the horizontal relationship between flow tables.

[0125] Step 2.4, the forwarding scheduling layer executes data forwarding tasks according to the flow table matching operation results, abstracts the characteristics of data packet flow direction between different nodes in the data plane into link connectivity similarity properties, and constructs a network link connectivity similarity vector based on link connectivity similarity.

[0126] Step 3, the SDN control layer dynamically collects the global network topology information of the target SDN network data plane using network snapshots, and constructs the target SDN network link connectivity similarity vector based on the flow rule characteristics in the three-layer data forwarding model structure and the global network topology information; the specific process includes:

[0127] Step 3.1, the SDN controller obtains the global network topology information through packet-out and packet-in, and performs a shortest path algorithm on the information, iterates through the members in the shortest path set p in the current topology, and records the information of each node acting as an intermediate node; where p = {x i，j ...}, x i，j is the shortest path between node i and node j, and p is the shortest path set;

[0128] Step 3.2, fuse the link connectivity vector with the four similarity feature parameters reflecting the target network topology information, namely the Jaccard coefficient, the common neighbor weight coefficient AA, the resource allocation coefficient RA, and the preferred connection coefficient PA, to obtain the SDN link connectivity similarity vector SDN_LCV; specifically including:

[0129] Step 3.2.1, calculate the improved Jaccard feature vector

[0130]

[0131] wherein the Jaccard coefficient is calculated as follows:

[0132]

[0133] wherein Γ(x i ), Γ(x j ) respectively represent the node x i, x j , |P| represents the cardinality of set P;

[0134] Step 3.2.2, calculate the improved preference connection feature vector

[0135]

[0136] wherein, the preference connection coefficient The calculation formula is:

[0137]

[0138] wherein, respectively, the degree of node x i , x j

[0139] Step 3.2.3, calculate the improved common neighbor weight feature vector

[0140]

[0141] wherein, the common neighbor weight coefficient The calculation formula is:

[0142]

[0143] Step 3.2.4, calculate the improved resource allocation feature vector

[0144]

[0145] wherein, the resource allocation coefficient The calculation formula is:

[0146]

[0147] Step 3.3, construct the target SDN network link connectivity similarity vector V LCV :

[0148]

[0149] Step 4, combined with Figure 4 , use the difference integrated moving average autoregressive model ARIMA to analyze the time series of the network link connectivity similarity vector sequence, and extract the change rule of the link connectivity similarity vector with time; The specific process includes:

[0150] Step 4.1, for node pair t in the target network, collect the link connectivity vector information in each time interval ε, and form the link connectivity feature vector sequence V​LCVS :

[0151]

[0152] Step 4.2, stationary analysis is performed on the network link connectivity similarity vector sequence using the Augmented Dickey-Fuller test (ADF) method. If it is not stationary, difference operation is performed to convert it into a stationary sequence, otherwise the next step is performed;

[0153] Step 4.3, analyze the output of the ACF function. If it is zero, introduce the autocorrelation function to determine the order q of the autocorrelation model; wherein the autocorrelation function is defined as:

[0154]

[0155] wherein, is the covariance value about k, is the link connectivity similarity vector corresponding to node pair t;

[0156] Step 4.4, calculate the order p of the moving average model according to formula (40):

[0157]

[0158] wherein, is the partial autocorrelation coefficient of , which represents the sequence removes the intermediate variable sequence from the short-term correlation of ; Step 4.5, parameter optimization of ARIMA model according to the minimum information criterion;

[0159] Step 4.6, sliding window is used to intercept the link connectivity feature vector sequence obtained in step 4.2. The intercepted sequence is used as the input of the ARIMA model, and the link connectivity feature vector at the next time is predicted according to formula (41). The features are used as labels to form a sample set F = [f1, f2,..., f n ];

[0160]

[0161]

[0162] Step 5, combining Figure 3 , a mycoparasitic optimization algorithm SMA is used to construct and optimize the network link connectivity classifier; the specific process includes:

[0163] ​Step 5.1, initialize population size N, maximum iteration number and individual position, each individual in the population represents a set of hyperplane weight vector in network link connectivity classifier, denoted as w = (w1, w2,..., w4), where hyperplane represents the classification plane of network connectivity classifier;

[0164] Step 5.2, construct SVM-based network link connectivity classifier, according to the output of AIRMA model, it is judged that the sequence sample of link connectivity feature vector is nonlinearly separable, and the hyperplane of link connectivity classifier is calculated as formula (42);

[0165] f(w) = sign(δy label k(V LCV , w) + b) (42)

[0166] Where δ is the Lagrange multiplier, which is a coefficient variable, k(*, *) is used to map data to feature space, and linear learners are used in feature space for data classification processing, y label is the sample label, y label = 1 represents connectivity, y label = -1 represents non-connectivity, w is the weight vector of hyperplane, b is the base of hyperplane, and sign(·) is the sign function, which returns 1 when δy label V LCV ·w + b is greater than 0, otherwise returns 0;

[0167] Step 5.3, calculate the fitness value of all hyperplane weight vectors, the calculation method is shown in formula (43);

[0168]

[0169] Where N correct represents the number of correctly classified samples, N total is the total number of samples, and S(i) represents the fitness of the individual of hyperplane weight vector;

[0170] Step 5.4, sort the fitness values of all hyperplane weight vector individuals, the sorting method is shown in formula (44),

[0171] SmellIndex = sort(S) (44)

[0172] Where SmellIndex represents the ascending order of individual fitness value;

[0173] Step 5.5, record the best fitness DF and the best position

[0174] Step 5.6, update the moving parameters of hyperplane weight vector individuals Parameter Parameter a, b are calculated by formula (45) and (46) respectively.

[0175]

[0176]

[0177] where t is the current iteration number, and max_t represents the maximum iteration number.

[0178] Step 5.7, calculate the individual moving parameter of each hyperplane weight vector Parameter The calculation method is shown in formula (47).

[0179]

[0180] where B(i) represents the median of SmellIndex, bF represents the best fitness value obtained in the current iteration process, wF represents the worst fitness value obtained in the current iteration process, and the component r is a moving random value, which is a random value in the interval [0, 1];

[0181] Step 5.8, update the individual position representing the hyperplane weight vector in the current population, and the individual position calculation method is shown in formula (48);

[0182]

[0183] where represents the weight vector of the hyperplane, and represent the weight vectors of two randomly selected hyperplanes, represents the hyperplane weight vector with the highest current fitness value, r is a random number in the interval [0, 1], LB and UB represent the lower limit and upper limit of the search, rand is a random number in the interval [0, 1], and the calculation method of p is shown in formula (49);

[0184] p = tanh | S(i) - DF |, i ∈ 1, 2...n (49)

[0185] Step 5.9, use the individual with the best fitness value obtained as the hyperplane weight vector of the network link connectivity classifier to complete the construction and optimization process of the network link connectivity classifier.

[0186] Step 5.10, Figure 5The optimal fitness value convergence speed is to obtain the fitness of the best solution of each iteration. For the change of the optimal fitness value, the convergence curve of the improved algorithm based on myxomycete optimization is faster than that of the other two algorithms. In 100 iterations, thanks to the high efficiency of the local search of the optimal solution of the myxomycete optimization algorithm, the convergence efficiency is obviously higher than that of the other two algorithms in the same iteration step, and the particle swarm optimization algorithm and the random search algorithm are prone to fall into local extreme points, so the overall convergence performance is not as good as that of the myxomycete optimization algorithm, which proves that the myxomycete optimization algorithm has better ability to converge to the global optimum in the iteration process.

[0187] Step 6, locating the fragile link and node in the target network according to the prediction value output by the network link connectivity classifier. The specific process includes:

[0188] Step 6.1, the control layer finds the fragile link according to the mapping table, defines the node at the end of the fragile link as a fragile node, and constructs an adjacency matrix A about the fragile node;

[0189] Step 6.2, quantifying the vulnerability of the node according to the eigenvector centrality of the network fragile node, marking the node with the lowest eigenvector centrality in the control domain as the fragile node in the domain, and the calculation method of the eigenvector centrality of the fragile node is shown in formula (50);

[0190] V c (v i )=λ -1 A V c (50)

[0191] Where V c is the vulnerability eigenvector of node v i , and lambda is the corresponding eigenvalue;

[0192] Step 6.3, iterating formula (50) to a convergent non-zero solution to obtain the vulnerability eigenvalue of all nodes, marking the node with the lowest vulnerability eigenvalue in each control domain, and finding the controller in the domain to which the marked node belongs through the mapping table;

[0193] Step 6.4, issuing a flow table to the data plane containing the domain-internal fragile node in the mapping range of the controller, and re-planning the local routing between the fragile nodes in the control domain.

[0194] In one embodiment, an SMA-SVM-based SDN link connectivity prediction system is provided, which comprises:

[0195] The first module is configured to construct a link connectivity vector reflecting a data plane of a target network according to a complex network structure and betweenness centrality features in a software defined network (SDN);

[0196] The second module is configured to model the data plane of the target SDN network: based on functional attributes of the data plane of the OpenFlow general forwarding framework, construct a three-layer data forwarding model, which includes a parsing layer, a flow table matching layer, and a forwarding scheduling layer;

[0197] The third module is configured to realize that the SDN control layer dynamically collects global network topology information of the data plane of the target SDN network by using a network snapshot, and constructs a link connectivity similarity vector of the target SDN network according to flow rule features in the three-layer data forwarding model structure and the global network topology information;

[0198] The fourth module is configured to perform time series analysis on the link connectivity similarity vector sequence by using an autoregressive integrated moving average (ARIMA) model, and extract a change rule of the link connectivity similarity vector over time;

[0199] The fifth module is configured to construct and optimize a network link connectivity vector classifier by using a slime mold algorithm (SMA).

[0200] The sixth module is configured to locate fragile links and nodes in the target network according to a predicted value output by the network link connectivity vector classifier.

[0201] The specific limitations of the SMA-SVM-based SDN link connectivity prediction system can be referred to the limitations of the SMA-SVM-based SDN link connectivity prediction method described above, and will not be repeated here. Each module in the SMA-SVM-based SDN link connectivity prediction system described above can be realized by software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above-mentioned modules.

[0202] In one embodiment, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the following steps when executing the computer program:

[0203] Step 1: Construct a link connectivity vector reflecting a data plane of a target network according to a complex network structure and betweenness centrality features in a software defined network (SDN);

[0204] Step 2: Model the data plane of the target SDN network: based on functional attributes of the data plane of the OpenFlow general forwarding framework, construct a three-layer data forwarding model, which includes a parsing layer, a flow table matching layer, and a forwarding scheduling layer.

[0205] Step 3, the SDN control layer dynamically collects the global network topology information of the target SDN network data plane by using a network snapshot, and constructs a target SDN network link connectivity similarity vector according to the flow rule features in the three-layer data forwarding model structure and the global network topology information;

[0206] Step 4, a differential integrated moving average autoregressive model ARIMA is used to perform time series analysis on the network link connectivity similarity vector sequence, and the change rule of the link connectivity similarity vector over time is extracted;

[0207] Step 5, a slime mold optimization algorithm SMA is used to construct and optimize a network link connectivity classifier;

[0208] Step 6, according to the predicted value output by the network link connectivity classifier, the fragile links and nodes in the target network are located.

[0209] For specific definitions of each step, refer to the above definitions of the SMA-SVM-based SDN link connectivity prediction method, which will not be repeated here.

[0210] The SMA-SVM-based SDN link connectivity prediction method provided by the present application can improve the prediction accuracy of the network link connectivity based on SDN to a certain extent, and has the advantages of better accuracy and stability compared with other link prediction methods.

[0211] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above examples, and the above examples and descriptions in the specification are only to illustrate the principles of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting SDN link connectivity based on SMA-SVM, characterized in that, The method includes the following steps: Step 1: Based on the complex network structure and betweenness centrality characteristics in Software-Defined Networks (SDN), construct the link connectivity vectors that reflect the data plane of the target network. Step 2, Model the target SDN network data plane: Based on the functional attributes of the data plane of the OpenFlow general forwarding framework, construct a three-layer data forwarding model, namely the parsing layer, the flow table matching layer, and the forwarding scheduling layer; Step 3: The SDN control layer dynamically collects global network topology information of the target SDN network data plane using network snapshots, and constructs a target SDN network link connectivity similarity vector based on the flow rule characteristics in the three-layer data forwarding model structure and the global network topology information. Step 4: Use the Differential Integrated Moving Average Autoregressive Model (ARIMA) to perform time series analysis on the network link connectivity similarity vector sequence and extract the variation pattern of the link connectivity similarity vector over time. Step 5: Construct and optimize a network link connectivity classifier using the slime mold optimization algorithm (SMA). Step 6: Locate vulnerable links and nodes in the target network based on the predicted values ​​output by the network link connectivity classifier. Step 5 describes the construction and optimization of a network link connectivity classifier using the slime mold optimization algorithm (SMA). The specific process includes: Step 5.1: Initialize the population size N, the maximum number of iterations, and the individual positions. Each individual in the population represents a weight vector of a set of hyperplanes in the network link connectivity classifier, denoted as... , where the hyperplane represents the classification plane of the network connectivity classifier; Step 5.2, construct a network link connectivity classifier based on SVM. The hyperplane calculation of the link connectivity classifier is shown in Equation (1). （1） in, These are Lagrange multipliers, which are coefficient variables. This is used to map data to a feature space, where a linear learner is used for data classification. For sample labels, Represents connectivity. denoted as disconnected, w is the weight vector of the hyperplane. Let sign( be a basis for the hyperplane) ) is a sign function, when Returns 1 if the value is greater than 0, otherwise returns 0; Step 5.3, calculate the fitness value of all hyperplane weight vectors, the calculation method is shown in Equation (2); （2） in This represents the number of correctly classified samples. The total number of samples, Individual representing the hyperplane weight vector The fitness of; Step 5.4: Sort the fitness values ​​of all hyperplane weight vector individuals, and the sorting method is shown in Equation (3). （3） in Indicates the ascending order of individual fitness values; Step 5.5: Record the best fitness of the current iteration hyperplane weight vector individual. and the best location ; Step 5.6, update the individual movement parameters of the hyperplane weight vector. , ,parameter ,parameter The calculation methods for a and b are shown in equations (4) and (5) respectively. （4） （5） Where t is the current iteration number. Represents the maximum number of iterations; Step 5.7: Calculate the individual movement parameters of each hyperplane weight vector. ,parameter The calculation method is shown in equation (6); （6） in, represent The median of the two values, bF represents the best fitness value obtained in the current iteration, wF represents the worst fitness value obtained in the current iteration, and the component r is a moving random value, which is taken from a random value between [0,1]. Step 5.8, update the position of the individual representing the hyperplane weight vector in the current population. The method for calculating the individual position is shown in Equation (7). （7） in The weight vector representing the hyperplane, and The weight vectors represent the weights of two randomly selected hyperplanes. The hyperplane weight vector represents the highest fitness value, r is a random number in the interval [0,1], LB and UB represent the lower and upper limits of the search, rand is a random number in the interval [0,1], and the calculation method of p is shown in equation (8). ,i （8） Step 5.9: Use the individuals that obtain the best fitness value as the hyperplane weight vector of the network link connectivity classifier to complete the construction and optimization process of the network link connectivity classifier.

2. The SDN link connectivity prediction method based on SMA-SVM according to claim 1, characterized in that, Step 1 describes constructing link connectivity vectors that reflect the target network data plane based on the complex network structure and betweenness centrality characteristics of Software-Defined Networking (SDN). The specific process includes: Step 1.1: Use the network node intermediary degree influence coefficient. As a property reflecting the connectivity and robustness between nodes in the data plane Calculated using the following formula (9): （9） in, The nodes formed by node e The number of times the shortest path between them is found. for The number of all shortest paths in the array, where V is the set of nodes; Step 1.2, using the influence coefficient of nodes on commonly connected nodes. As a property reflecting the propagation of connectivity between nodes in the data plane Calculated using the following formula (10): （10） in, For nodes A function of the set of connected nodes. For nodes A function for a set of connected nodes; Step 1.3: Construct the link connectivity vector based on the network node intermediary influence coefficient and the node-to-co-connected node influence coefficient. The calculation method is shown in the following formula (11): （11） Among them, parameters This is used to control the weights of the intermediation influence coefficient and the influence coefficient of a node on its co-connected nodes.

3. The SDN link connectivity prediction method based on SMA-SVM according to claim 1 or 2, characterized in that, Step 2, which involves modeling the target SDN network data plane, specifically includes the following process: Step 2.1: Construct a three-layer data forwarding model for the target SDN network based on the OpenFlow general forwarding framework, namely the parsing layer, the flow table matching layer, and the forwarding scheduling layer. The parsing layer is responsible for parsing the data packet fields and passing the parsed fields to the flow table matching layer. Step 2.2: Extract the flow rule R from the flow table entry and construct its triple representation, i.e. , where M represents the matching field, P represents the priority, and I represents the instruction; Step 2.3: Based on the horizontal structured characteristics exhibited by flow rule R between flow tables, analyze the pre- and post-connection relationships between nodes in the data plane, and set the data plane link connectivity status according to the pre- and post-connection relationships; Step 2.4: The forwarding scheduling layer executes data forwarding tasks based on the flow table matching operation results, abstracts the characteristics of data packet flow between different nodes in the data plane into different link connectivity similarity attributes, and constructs a network link connectivity similarity vector based on link connectivity similarity.

4. The SDN link connectivity prediction method based on SMA-SVM according to claim 3, characterized in that, Step 2.3 specifically includes: (1) According to the behavior of flow rules, the dynamics of flow rules are affected by the context of the current state. A rule in the table is affected by the preceding and following rules, and the table reflects the vertical structural relationship of rules. (2) During the forwarding process, a flow rule R will be associated with the preceding remote table or the following remote table. This association is reflected in the data plane as the connectivity between different nodes. The connectivity between different nodes in the data plane is determined by the horizontal relationship between the flow tables.

5. The SDN link connectivity prediction method based on SMA-SVM according to claim 3, characterized in that, Step 3 describes the SDN control layer dynamically collecting global topology information of the target SDN network data plane using network snapshots. Based on the flow rule features in the three-layer model structure and the global topology information, a network link connectivity similarity vector is constructed. The specific process includes: Step 3.1: The SDN controller obtains global network topology information through packet-out and packet-in, and performs a shortest path algorithm on this information, sequentially traversing the shortest path set in the current topology. The members of the system are recorded, and information about each node acting as an intermediary node is also recorded. , Find the shortest path between node i and node j. It is the set of shortest paths; Step 3.2 involves fusing the link connectivity vector with four similarity feature parameters reflecting the target network topology: Jaccard coefficient, common neighbor weight coefficient (AA), resource allocation coefficient (RA), and preferred connectivity coefficient (PA), to obtain the SDN link connectivity similarity vector SDN_LCV; specifically including: Step 3.2.1, calculate the improved Jaccard eigenvectors. : （12） Among them, the Jaccard coefficient The calculation formula is: （13） in, , Representing nodes respectively , The neighborhood set of set P, where |P| represents the cardinality of set P; Step 3.2.2: Calculate the improved preference connectivity feature vector. : （14） Among them, the preference connection coefficient The calculation formula is: （15） in, , They are nodes , The degree; Step 3.2.3: Calculate the improved common neighbor weight feature vector. : （16） Among them, the common neighbor weight coefficient The calculation formula is: （17） Step 3.2.4: Calculate the improved resource allocation feature vector. : （18） Among them, the resource allocation coefficient The calculation formula is: （19） Step 3.3: Construct the target SDN network link connectivity similarity vector. : （20）。 6. The SDN link connectivity prediction method based on SMA-SVM according to claim 5, characterized in that, Step 4 describes using the Differential Integrated Moving Average Autoregressive Model (ARIMA) to perform time series analysis on the network link connectivity similarity vector sequence, extracting the changing patterns of the link connectivity similarity vector over time. The specific process includes: Step 4.1: For node pair t in the target network, collect and extract link connectivity vector information within each time interval ε to form a link connectivity feature vector sequence. : （21） Step 4.2: Use the unit root test (ADF) method to perform stationarity analysis on the network link connectivity similarity vector sequence. If it is not stationary, perform difference operations to convert it into a stationary sequence; otherwise, proceed to the next step. Step 4.3: Analyze the output of the ACF function. If it is zero, introduce the autocorrelation function to determine the order q of the autocorrelation model; where the autocorrelation function is defined as: （22） in, It is a function of the covariance value with respect to k. Let be the link connectivity similarity vector corresponding to node pair t; Step 4.4, calculate the order p of the moving average model according to equation (23): （23） in, yes The partial autocorrelation coefficient, which represents the sequence right The intermediate variable series was removed when calculating the correlation. right The transmission effect of short-term correlation; Step 4.5: Optimize the parameters of the ARIMA model according to the minimum information criterion; Step 4.6: Use a sliding window to extract the link connectivity feature vector sequence obtained in Step 4.

2. Use the extracted sequence as input to the ARIMA model. Predict the link connectivity feature vector at the next time step according to Equation (24). Use the features as labels to form a sample set. ; （24）。 7. The SDN link connectivity prediction method based on SMA-SVM according to claim 6, characterized in that, Step 6, which involves locating vulnerable links and nodes in the target network based on the predicted values ​​output by the network link connectivity classifier, specifically includes the following process: Step 6.1: The control layer searches for vulnerable links according to the mapping table, defines the nodes at the endpoints of the vulnerable links as vulnerable nodes, and constructs an adjacency matrix for the vulnerable nodes. ; Step 6.2: Quantify the vulnerability of nodes based on the eigenvector centrality of network vulnerable nodes, and mark the node with the lowest eigenvector centrality value in the control domain as a vulnerable node in the domain. The calculation method of the eigenvector centrality of vulnerable nodes is shown in Equation (25). （25） in For nodes Vulnerability feature vector, For the corresponding eigenvalues; Step 6.3 Iterate Equation (25) to a convergent non-zero solution to obtain the vulnerability feature values ​​of all nodes. Mark the node with the lowest vulnerability feature value in each control domain and find the controller in the domain to which the marked node belongs through the mapping table. Step 6.4: Issue flow tables to the data plane containing vulnerable nodes within the domain within the controller's mapping range, and re-plan the local routes between vulnerable nodes within its control domain.

8. The SDN link connectivity prediction system based on SMA-SVM according to any one of claims 1 to 7, characterized in that, The system includes: The first module is used to construct link connectivity vectors that reflect the data plane of the target network, based on the complex network structure and betweenness centrality characteristics of software-defined networks (SDN). The second module is used to model the target SDN network data plane: based on the functional attributes of the OpenFlow general forwarding framework data plane, a three-layer data forwarding model is constructed, namely the parsing layer, the flow table matching layer, and the forwarding scheduling layer. The third module is used to enable the SDN control layer to dynamically collect global network topology information of the target SDN network data plane using network snapshots, and to construct a target SDN network link connectivity similarity vector based on the flow rule characteristics in the three-layer data forwarding model structure and the global network topology information. The fourth module is used to perform time series analysis on the network link connectivity similarity vector sequence using the differential integrated moving average autoregressive model ARIMA, and to extract the variation pattern of the link connectivity similarity vector over time. The fifth module is used to construct and optimize a network link connectivity classifier using the slime mold optimization algorithm (SMA). The sixth module is used to locate vulnerable links and nodes in the target network based on the predicted values ​​output by the network link connectivity classifier.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images