Network flow identification method based on fuzzy reasoning

Through fuzzy inference technology and multi-dimensional dispersed fuzzy inference unit, combining KL distance and autoregressive moving average model, the abnormal patterns in network traffic are identified and filtered, and the problem of the existing technology being unable to identify unknown attacks is solved, and the system's adaptability and Robustness enables dynamic response to new threats.

CN119071010APending Publication Date: 2024-12-03LINHUAN COAL MINE OF HUAIBEI MINING CO LTD +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202410913001.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The prior art relies on hard rules based on pattern matching libraries, and cannot identify unknown abnormal patterns or new cyber attacks, and the rule database needs to be updated frequently to deal with emerging threats, resulting in the system being passive and vulnerable in the face of new attacks.

Method used

The network traffic recognition method of fuzzy inference is adopted. By calculating the KL distance value and the autoregressive moving average model, combining multiple network feature measurements, the network traffic is comprehensively analyzed, and the uncertainty in the traffic data is used to process the uncertainty in the traffic data to reduce the The dependency of the static rule base, dynamically identifying and filtering exception traffic.

Benefits of technology

Real-time identification and filtering of complex and variable network environments is realized, the system is improved, and the system is robust and adaptable, and the potential abnormal situation can be identified more accurately, the false alarm rate and missed alarm rate are reduced, and the rule base maintenance is reduced. frequency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119071010A_ABST
    Figure CN119071010A_ABST
Patent Text Reader

Abstract

The invention discloses a network flow identification method based on fuzzy reasoning. The method comprises the following steps: 1, calculating KL distance values of network flow on each measure; 2, calculating a predicted value of the KL distance on each measure by adopting an autoregressive integral moving average model, and calculating a difference between the predicted value and an actual value of the KL distance; 3, continuously monitoring the average queue length of the network equipment, and calculating the discarding probability; 4, determining the membership degree of the difference value and the discarding probability to each fuzzy subset, formulating a fuzzy rule based on the logical relationship between the membership degree and the network abnormal state value, performing defuzzification operation, and reasoning out the network abnormal state value; and 5, carrying out weighted integration on the network abnormal state value of each measure, and identifying the network flow according to a monitoring threshold value. The invention aims to identify and filter the abnormal traffic by integrating multiple measurement indexes, so as to guarantee the real-time performance of the key service flow in the communication network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of network security, and in particular to a network traffic identification method using fuzzy reasoning. Background Art

[0002] With the rapid development of information technology, the Internet has profoundly affected all aspects of social economy and daily life. From home to business, from education to medical care, the popularity of the Internet has changed the way people communicate, work and entertain. However, the problems of network traffic management and security have become increasingly prominent. Abnormal traffic may cause data delays, information loss, network congestion, service interruptions and network attacks, causing serious economic losses and privacy leakage risks to enterprises and individuals. For example, distributed denial of service attacks occupy network resources through malicious traffic, causing network services to fail to operate normally; botnets infect a large number of computer devices and carry out large-scale network attacks, which seriously threaten network security.

[0003] After searching, the Chinese patent publication number is CN112965970B, which discloses a parallel detection method for abnormal traffic based on a hash algorithm, including: the device source initiates a request to the target server; the central management core intercepts and cleans the data packet, and distributes the data packet information to be detected to each detection core; each detection core divides the total rule base equally to obtain the sub-rule base responsible for matching; the detection core obtains the data packet number of each detection according to the hash algorithm, wherein the key value of the hash function is calculated by the core number, the total number of detection cores, the total number of data packets, the number of data packets detected by the core, and the number of conflicts.

[0004] The existing technology, similar to the above-mentioned traffic detection method, mainly relies on hard rules based on the pattern matching library, which means that it can only defend against known network attacks and cannot identify and detect unknown abnormal patterns or new attack types. This dependence makes the system passive and vulnerable when facing new attacks. Secondly, the rule library needs to be constantly updated and maintained in order to capture new abnormal patterns and attack types in a timely manner. If the rule library is not updated in a timely manner, the system will not be able to effectively detect new threats, which may lead to serious security vulnerabilities. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a fuzzy reasoning network traffic identification method. Its advantage is that it integrates multiple measurement indicators to identify and filter abnormal traffic and ensure the real-time performance of key business flows in the communication network.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A fuzzy reasoning network traffic identification method comprises the following steps:

[0008] Step 1: Calculate the KL distance value of this section of network traffic on multiple measures;

[0009] Step 1.1: Let the measurement set of network traffic be X = {x1, x2, ..., x u ,...,x U},1<u<U; among them, x u represents the u-th measure of network traffic, U represents the maximum measure of network traffic; based on the u-th measure x u Classify the packets or flows in normal network traffic and obtain the classified data packet set of normal network traffic. 1<i≤I; where, Indicates the normal network traffic at the uth measure x u The data packets of the i-th category, I represents the total number of categories;

[0010] Step 1.2: Classify the normal traffic packets Perform statistics to obtain the normal network traffic at the uth measurement x u The number of packets of the next i-th category num i (u), calculate the normal network traffic at the uth measure x u The proportion of packets of the i-th category p i (u), and then we can get the normal network traffic at the uth measurement x u The proportion set P(u) = {p1(u), p2(u), …, p i (u),…,p I (u)};

[0011] Step 1.3: Process the network traffic to be detected at time t according to the process of step 1.1 and step 1.2 to obtain the proportion set of the network traffic to be detected at time t at the u-th measure; wherein represents the proportion of the data packets of the i-th category of the network traffic to be detected at time t under the u-th measure;

[0012] Step 1.4: Calculate the u-th measure x at time t using the following formula: u The KL distance D of all categories of data packets KL (u,t):

[0013]

[0014] Step 2: Use the autoregressive moving average model to calculate the predicted value of the KL distance on each measure at time t;

[0015] Step 2.1: Assume the size of the sliding window is w, and use the sliding window to detect the network traffic in a period at the uth measurement xu The KL distance under the sliding window is processed to obtain the u-th measure x in the current sliding window u The KL distance subsequence under u ={D KL (u,tw),D KL (u,t-w+1),...,D KL (u,l),…,D KL (u,t)}, where D KL (u,l) represents the u-th measure x in the current sliding window u The KL distance value at the next l-th moment, l∈[tw,t-1];

[0016] Step 2.2: After completing ARIMA modeling, use the following formula to calculate the u-th dimension x at time t: u Upper relative entropy D KL The predicted value of (u,t)

[0017]

[0018] Among them, φ i represents the parameter estimate of the i-th lag of the autoregressive part (AR), θ i represents the parameter estimate of the ith lag of the moving average part (MA), ε(ti) represents the random error term at time ti;

[0019] Step 2.3: Calculate the u-th measure x at time t according to the following formula: u The actual value of the KL distance D KL (u,t) and predicted value The difference ΔD KL (u,t):

[0020]

[0021] Step 3: Continuously monitor the average queue length of the network device and calculate the drop probability p drop (t);

[0022] Step 3.1: Continuously monitor the queue length of the network device over a period of time and calculate the average queue length q(t);

[0023] in, represents the average queue length at time t, w q Indicates the queue weight. Usually a smaller value is selected to balance fast response and stability.

[0024] Step 3.2: Calculate the queue drop probability p at the current moment according to the following formula: drop (t):

[0025]

[0026] Among them, q min Indicates the minimum queue length threshold. If the average queue length is less than q min If q is set, no data packets will be dropped; max Indicates the maximum queue length threshold. If the average queue length at this time is greater than q max If the queue length is close to the maximum threshold, all packets will be discarded; p0 represents the basic discard probability. When the queue length is close to the maximum threshold, the discard probability is close to p0;

[0027] Step 4: Infer the network abnormality status value of the network traffic to be detected in each measurement;

[0028] Step 4.1: ΔD KL (u,t),p drop (t) and network abnormal state value S(u,t) are divided into R fuzzy subsets respectively;

[0029] Step 4.2, set ΔD KL (u, t) is the membership function of the rth fuzzy subset; set p drop (t) Membership function for the cth fuzzy subset Calculate S u,t For the membership function of the vth fuzzy subset

[0030] Step 4.3: Calculate the u-th measure x at time t using the following formula: u Below the trigger weight ω between the rth fuzzy subset and the cth fuzzy subset r,c (u,t):

[0031]

[0032] Step 4.4: Use formula (10) to get the kth measure x at time t u The network abnormal state value S(u,t) under:

[0033]

[0034] Step 5: Perform weighted synthesis of the network abnormal status values ​​of each measurement, and determine whether the network traffic to be detected is an abnormal flow based on the monitoring threshold, so as to filter the abnormal flow;

[0035] Step 5.1: Calculate the weighted network abnormal state value S(t) under all measurements at time t using the following formula:

[0036]

[0037] Step 5.2: Determine S t >S th Is it true? If true, it means that the network traffic to be detected at time t is an abnormal flow and is filtered. Otherwise, it means that the network traffic to be detected at time t is a normal flow. Among them, S th Indicates the monitoring threshold of network traffic.

[0038] The present invention is further configured that in step 1, the normal network traffic is at the uth measurement x u The proportion of packets of the i-th category p i The calculation formula for (u) is:

[0039]

[0040] The present invention is further configured such that the calculation formula for calculating the average queue length q(t) in step 3 is:

[0041] in, represents the average queue length at time t, w q Indicates the queue weight. Usually a smaller value is selected to balance fast response and stability.

[0042] The present invention is further configured such that ΔD in step 4 KL The membership function of (u, t) for the rth fuzzy subset is:

[0043] In the formula, std(D(u)) represents the standard deviation of the data in the sliding window, Denotes ΔD KL The membership degree of (u,t) to the rth fuzzy subset The variance of Denotes ΔD KL The membership degree of (u,t) to the rth fuzzy subset The center of , r = 1, 2,…, R.

[0044] The present invention is further configured that in step 4, p drop (t) Membership function for the cth fuzzy subset for:

[0045]

[0046] In the formula, express The variance of express The center of; c = 1, 2,…, R.

[0047] The present invention is further configured that in step 4, S u,t For the membership function of the vth fuzzy subset for:

[0048]

[0049] In the formula, S max Indicates the maximum value of the network abnormal status value. express The variance of express The center of , v=1,2,…,R.

[0050] The present invention is further configured that in step 4, p drop (t) Membership function for the cth fuzzy subset for:

[0051] in, express The left boundary parameter, express The central parameter of express The right boundary parameter of .

[0052] The present invention is further configured that in step 4, S u,t For the membership function of the vth fuzzy subset for:

[0053] Among them, a S(u,t) express The left boundary parameter, b S(u,t) express The central parameter, c S( u,t) The right boundary parameter of .

[0054] The beneficial effects of the present invention are:

[0055] 1. The present invention quantifies the difference between normal traffic and abnormal traffic by calculating the KL (Kullback-Leibler) distance; as a quantitative indicator, the KL distance can more accurately reflect the characteristics of abnormal traffic, thereby helping the system to accurately identify potential abnormal situations.

[0056] 2. The present invention can combine multiple network feature measurements, such as port number, data packet size, frame arrival time, etc., to conduct a comprehensive and detailed analysis of network traffic, and can identify more complex and hidden abnormal network traffic and filter it, thereby ensuring the security and real-time performance of the network.

[0057] 3. The present invention introduces fuzzy reasoning technology, which enables the system to handle the uncertainty and ambiguity in traffic data; fuzzy reasoning enhances the system's adaptability to complex and changeable network environments, and improves the flexibility and intelligence level of detection; compared with the hard rule method based on the pattern matching library, it is less dependent on static rule library updates and maintenance, and can more dynamically adapt to and respond to new network attacks and abnormal patterns, thereby improving the robustness and adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of the operation flow structure of a fuzzy reasoning network traffic identification method proposed by the present invention;

[0059] Figure 2 A schematic diagram of a multi-dimensional decentralized fuzzy reasoning unit of a fuzzy reasoning network traffic identification method proposed by the present invention. DETAILED DESCRIPTION

[0060] The technical solution of this patent is further described in detail below in conjunction with specific implementation methods.

[0061] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0062] Embodiment 1

[0063] Reference Figure 1 ,A fuzzy reasoning network traffic identification method includes the following steps:

[0064] Step 1: Calculate the KL distance value of network traffic on each measure;

[0065] Step 1.1, select multiple key indicators for network traffic identification, and classify the data packets according to each indicator. The network traffic measurement describes the attributes that can be used to identify network traffic in the traffic data, including network protocol, packet length information, port number, and frame arrival time. Let the network traffic measurement set be X = {x1, x2, ..., x u ,...,x U},1<u<U; among them, x u represents the u-th measure of network traffic, U represents the maximum measure of network traffic; based on the u-th measure x u Classify the packets or flows in normal network traffic and obtain the classified data packet set of normal network traffic. 1<i≤I; where, Indicates the normal network traffic at the uth measure x u The data packets of the i-th category are as follows: I represents the total number of categories. Common measurements are shown in Table 1:

[0066] Table 1 Common measurements in network traffic

[0067]

[0068] Step 1.2: Classify the normal traffic data packets Perform statistics to obtain the normal network traffic at the uth measurement x u The number of packets of the next i-th category num i (u), and thus use formula (1) to calculate the normal network traffic at the uth measure x u The proportion of packets of the i-th category p i (u), and then we can get the normal network traffic at the uth measurement x u The proportion set P(u) = {p1(u), p2(u), …, p i (u),…,p I (u)};

[0069]

[0070] Step 1.3: Process the network traffic to be detected at time t according to the process of steps 1.1 to 1.2 to obtain the network traffic to be detected at time t at the uth measure x u The proportion set Q(u,t) = {q1(u,t),q2(u,t),...,q i (u,t),...,q I (u,t)}; where q i (u,t) represents the network traffic to be detected at time t at the uth measurement x u The proportion of packets in the i-th category;

[0071] Step 1.4: Use formula (2) to calculate the u-th measure x at time t u The KL distance D of all categories of data packets KL (u, t): KL distance can be used to measure the distance between two probability distributions P(x u )={p1,p2,…,p i …,p m} and Q(x u )={q1,q2,…,q i …,q m}Difference:

[0072]

[0073] In formula (2), when q i = 0, the value of KL distance cannot be solved, so formula (2) is redefined and formula (3) is used to calculate the measure x u KL distance at time t:

[0074]

[0075] Step 2. Use the autoregressive integrated moving average model to calculate the predicted value of the KL distance on each measure, and calculate the difference between the predicted value and the actual value of the KL distance;

[0076] Step 2.1: Assume the size of the sliding window is w, and use the sliding window to measure the network traffic to be detected in a period at the uth measurement x u The KL distance under the sliding window is processed to obtain the u-th measure x in the current sliding window u The KL distance subsequence under u ={D KL (u,tw),D KL (u,t-w+1),...,D KL (u,l),...,D KL (u,t)}, where D KL (u,l) represents the u-th measure x in the current sliding window u The KL distance value at the next l-th moment, l∈[tw,t-1];

[0077] Step 2.2: Establish an ARIMA prediction model and calculate the KL distance prediction value

[0078] Step 2.2.1: Data preprocessing: First, use the ADF test to check the stationarity of the original data. If the ADF test passes, proceed to the next step. If the ADF test fails, perform a difference on the original data to convert it into stationary data, and then perform the ADF test. If the ADF test passes, perform zero mean processing on the sequence, that is, subtract the mean of the sequence to make it centered.

[0079] Step 2.2.2: Construct the ARIMA model, calculate the p and q parameters of the ARIMA model according to the Bayesian Information Criterion, and determine the order of the model. The Bayesian Information Criterion formula is shown in formula (4):

[0080]

[0081] in, represents the residual variance of the model, N represents the sample length, p and q represent the autoregressive order and moving average order of the model respectively. Find the p and q with the smallest BIC(p,q), which are the optimal order of the model.

[0082] Step 2.2.3, model testing: Calculate the mean square error of the autocorrelation function of the predicted sequence and the actual sequence to evaluate the fitting accuracy and prediction accuracy of the FARIMA model. The formula is shown in formula (5):

[0083]

[0084] Among them, ACF m (i) is the i-th order autocorrelation function of the prediction sequence, ACF s (i) is the i-th order autocorrelation function of the actual sequence,

[0085] Table 2. Model parameter configuration

[0086]

[0087] Step 2.2.4: After completing the ARIMA modeling, use formula (6) to calculate the u-th dimension x at time t u Upper relative entropy D KL The predicted value of (u,t)

[0088]

[0089] Among them, φ i represents the parameter estimate of the i-th lag of the autoregressive part (AR), θ i represents the parameter estimate of the ith lag of the moving average part (MA), ε(ti) represents the random error term at time ti;

[0090] Step 2.3: Calculate the uth measure x at time t according to formula (7): u The actual value of the KL distance D KL (u,t) and predicted value The difference ΔD KL (u,t):

[0091]

[0092] Step 3: Continuously monitor the average queue length of the network device and calculate the drop probability;

[0093] Step 3.1, continuously monitor the queue length of the network device over a period of time, and calculate the average queue length q(t) according to formula (8);

[0094]

[0095] in, represents the average queue length at time t, w qIndicates the queue weight. Usually a smaller value is selected to balance fast response and stability.

[0096] Step 3.2: Calculate the queue drop probability p at the current moment according to formula (9): drop (t)

[0097]

[0098] Among them, q min Indicates the minimum queue length threshold. If the average queue length is less than q min If q is set, no data packets will be dropped; max Indicates the maximum queue length threshold. If the average queue length at this time is greater than q max If the queue length is close to the maximum threshold, all packets will be discarded; p0 represents the basic discard probability. When the queue length is close to the maximum threshold, the discard probability is close to p0.

[0099] 5. The abnormal network traffic method based on multi-dimensional decentralized fuzzy reasoning according to claim 4 is characterized in that the step 4 comprises the following steps:

[0100] Step 4.1: ΔD KL (u,t),p drop (t) and the network abnormal state value S(u,t) are divided into R fuzzy subsets respectively. For example, ΔD KL (u,t),p drop (t) and S(u,t) are divided into five fuzzy sets {NB, NS, ZE, PS, PB}, where NB means very small, NS means small, ZE means medium, PS means large, and PB means very large;

[0101] Step 4.2, select a suitable membership function to describe the degree of belonging of KL distance difference and bandwidth occupancy to each fuzzy set. In traditional logic, an element either completely belongs to a set or does not belong to the set at all. In fuzzy logic, the membership of an element to a set is a continuous value between 0 and 1, indicating the similarity or belonging degree between the element and the set. In this study, a Gaussian membership function is used, and ΔD is set using formula (10). KL (u, t) membership function for the rth fuzzy subset:

[0102]

[0103] In formula (6), Represents the u-th measure x in the current sliding window u The mean KL distance under Denotes ΔD KLThe membership degree of (u,t) to the rth fuzzy subset The variance of Denotes ΔD KL The membership degree of (u,t) to the rth fuzzy subset The center of , r = 1, 2, ..., R;

[0104] Using formula (11) to set B oc,t For the membership function of the cth fuzzy subset

[0105] In formula (11), express The variance of express The center of; c = 1, 2, ..., R;

[0106] Calculate S using formula (12) u,t For the membership function of the vth fuzzy subset

[0107] In formula (12), S max Indicates the maximum value of the network abnormal status value. express The variance of express The center of , v = 1, 2, ..., R;

[0108] Step 4.3: After the precise parameters are fuzzified, according to the input variables Δh, B occupy and output variable Score i The fuzzy logic relationship between them builds a set of rules and stores them in the fuzzy rule base to build a fuzzy reasoning unit, such as Figure 2 As shown, the fuzzy rule takes the form of "If A Then B", which can be expressed as Relu(i,j): Then y(k)is B i,j ,i,j=1,2,···,l, where Relu(i,j) represents the (i×j)th rule and l represents the total number of rules. The fuzzy rule table is set as shown in Table 2:

[0109] Table 2. Fuzzy rules table

[0110]

[0111] Using formula (13) to calculate the u-th measure x at time t u Below the trigger weight ω between the rth fuzzy subset and the cth fuzzy subset r,c (u,t):

[0112]

[0113] Step 4.4: Use formula (14) to get the kth measure x at time t u The network abnormal state value S(u,t) under:

[0114]

[0115] Step 5: Perform a weighted synthesis of the anomaly scores of each dimension, and determine whether the network traffic to be detected is an abnormal flow based on the monitoring threshold, so as to filter the abnormal flow.

[0116] Step 5.1: The degree of network anomaly reflected by different measurement changes is different. The flow data of the substation communication network is statistically analyzed to obtain the degree of influence of network anomaly on different measurements, thereby amplifying the weight of the measurement with the least impact by the attack, so that each measurement anomaly can be better responded to. Calculate the uth measurement x according to formula (15): u The weighting coefficient α under u :

[0117]

[0118] In formula (15), ψ represents the maximum average influence, ψ u represents the uth measure x u The average influence under the above conditions is shown in Table 3.

[0119] Table 3 Statistics of the impact of each attack on each measure

[0120]

[0121] Step 5.2: Use formula (16) to calculate the weighted network abnormal state value S under all measurements at time t t :

[0122]

[0123] Step 5.3: Determine S t >S th Is it true? If true, it means that the network traffic to be detected at time t is an abnormal flow and is filtered. Otherwise, it means that the network traffic to be detected at time t is a normal flow. Among them, S thIndicates the monitoring threshold of network traffic. The threshold can be determined based on the network anomaly status value between the known abnormal data and the normal data statistics. Calculate the network anomaly status value between each measure of the known abnormal data and the normal measurement statistics, and then obtain the weighted network anomaly status value. Such data can be performed multiple times to obtain multiple groups of weighted network anomaly status values. Finally, the smallest data is selected as the detection threshold to obtain a better detection rate.

[0124] In order to verify the advantages of the flow identification method disclosed in the above embodiment over the prior art, a comparative test is conducted between the first embodiment and the prior art, and the test results are recorded as shown in the table:

[0125] Way Example 1 Prior art Accuracy 92.34% 80.43% Recall 85.97% 70.5% False Positive Rate 3.19% 3.17% False negative rate 7.61% 12.68%

[0126] It can be seen from the test results disclosed in the above table that the technical solution disclosed in Example 1 has significantly improved precision compared with the prior art; the recall rate is significantly improved; the false alarm rate is basically the same; and the missed alarm rate is greatly reduced.

[0127] Embodiment 2

[0128] Reference Figure 1 , a fuzzy reasoning network traffic identification method, which is different from the first embodiment in that: in step 4, p drop (t) The membership function for the cth fuzzy subset is: for:

[0129] in, express The left boundary parameter, b pdr op(t) means The central parameter of express The right boundary parameter of step 4 is S u,t For the membership function of the vth fuzzy subset for:

[0130] Among them, a S(u,t) express The left boundary parameter, b S(u,t) express The central parameter, c S(u,t) express The right boundary parameter of the system; by adopting the above calculation function, the calculation can be simplified and the system response can be made more timely.

[0131] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A fuzzy reasoning network traffic identification method, characterized in that: The following steps are involved: Step 1: Calculate the KL distance value of this section of network traffic on multiple measures; Step 1.1: Let the measurement set of network traffic be X = {x1, x2, ..., x u ,...,x U },1<u<U; among them, x u represents the u-th measure of network traffic, U represents the maximum measure of network traffic; based on the u-th measure x u Classify the packets or flows in normal network traffic and obtain the classified data packet set of normal network traffic. in, Indicates the normal network traffic at the uth measure x u The data packets of the i-th category, I represents the total number of categories; Step 1.2: Classify the normal traffic packets Perform statistics to obtain the normal network traffic at the uth measurement x u The number of packets of the next i-th category num i (u), calculate the normal network traffic at the uth measure x u The proportion of packets of the i-th category p i (u), and then we can get the normal network traffic at the uth measurement x u The proportion set P(u) = {p1(u), p2(u), …, p i (u),...,p I (u)}; Step 1.3: Process the network traffic to be detected at time t according to the process of step 1.1 and step 1.2 to obtain the proportion set of the network traffic to be detected at time t at the u-th measure; wherein represents the proportion of the data packets of the i-th category of the network traffic to be detected at time t under the u-th measure; Step 1.4: Calculate the u-th measure x at time t using the following formula: u The KL distance D of all categories of data packets KL (u,t): ; Step 2: Use the autoregressive moving average model to calculate the predicted value of the KL distance on each measure at time t; Step 2.1: Assume the size of the sliding window is w, and use the sliding window to detect the network traffic in a period at the uth measurement x u The KL distance under the sliding window is processed to obtain the u-th measure x in the current sliding window u The KL distance subsequence under u ={D KL (u,tw),D KL (u,t-w+1),...,D KL (u,l),...,D KL (u,t)}, where D KL (u,l) represents the u-th measure x in the current sliding window u The KL distance value at the next l-th moment, l∈[tw,t-1]; Step 2.2: After completing ARIMA modeling, use the following formula to calculate the u-th dimension x at time t: u Upper relative entropy D KL The predicted value of (u,t) Among them, φ i represents the parameter estimate of the i-th lag of the autoregressive part (AR), θ i represents the parameter estimate of the ith lag of the moving average part (MA), ε(ti) represents the random error term at time ti; Step 2.3: Calculate the u-th measure x at time t according to the following formula: u The actual value of the KL distance D KL (u,t) and predicted value The difference ΔD KL (u,t): Step 3: Continuously monitor the average queue length of the network device and calculate the drop probability p drop (t); Step 3.1: Continuously monitor the queue length of the network device over a period of time and calculate the average queue length q(t); in, represents the average queue length at time t, w q Indicates the queue weight. Usually a smaller value is selected to balance fast response and stability. Step 3.2: Calculate the queue drop probability p at the current moment according to the following formula: drop (t): Among them, q min Indicates the minimum queue length threshold. If the average queue length is less than q m i n If q is set, no data packets will be dropped; max Indicates the maximum queue length threshold. If the average queue length at this time is greater than q max If the queue length is close to the maximum threshold, all packets will be discarded; p0 represents the basic discard probability. When the queue length is close to the maximum threshold, the discard probability is close to p0; Step 4: Infer the network abnormality status value of the network traffic to be detected in each measurement; Step 4.1: ΔD KL (u,t),p drop (t) and network abnormal state value S(u,t) are divided into R fuzzy subsets respectively; Step 4.2, set ΔD KL (u, t) is the membership function of the rth fuzzy subset; set p drop (t) Membership function for the cth fuzzy subset Calculate S u,t For the membership function of the vth fuzzy subset Step 4.3: Calculate the u-th measure x at time t using the following formula: u Below the trigger weight ω between the rth fuzzy subset and the cth fuzzy subset r,c (u,t): Step 4.4: Use formula (10) to get the kth measure x at time t u The network abnormal state value S(u,t) under: Step 5: Perform weighted synthesis of the network abnormal status values ​​of each measurement, and determine whether the network traffic to be detected is an abnormal flow based on the monitoring threshold, so as to filter the abnormal flow; Step 5.1: Calculate the weighted network abnormal state value S(t) under all measurements at time t using the following formula: Step 5.2: Determine S t >S th Is it true? If true, it means that the network traffic to be detected at time t is an abnormal flow and is filtered. Otherwise, it means that the network traffic to be detected at time t is a normal flow. Among them, S th Indicates the monitoring threshold of network traffic.

2. According to the fuzzy reasoning network traffic identification method of claim 1, it is characterized in that: In step 1, the normal network traffic is measured at the uth measurement x u The proportion of packets of the i-th category p i The calculation formula of (u) is:

3. According to the fuzzy reasoning network traffic identification method of claim 1, it is characterized in that: The calculation formula for calculating the average queue length q(t) in step 3 is: in, represents the average queue length at time t, w q Indicates the queue weight. Usually a smaller value is selected to balance fast response and stability.

4. The fuzzy reasoning network traffic identification method according to claim 1 is characterized in that: ΔD in step 4 KL The membership function of (u, t) for the rth fuzzy subset is: In the formula, std(D(u)) represents the standard deviation of the data in the sliding window, Denotes ΔD KL The membership degree of (u,t) to the rth fuzzy subset The variance of Denotes ΔD KL The membership degree of (u,t) to the rth fuzzy subset The center of , r = 1, 2,…, R.

5. The fuzzy reasoning network traffic identification method according to claim 1 is characterized in that: In step 4, drop (t) Membership function for the cth fuzzy subset for: In the formula, express The variance of express The center of; c = 1, 2,…, R.

6. The fuzzy reasoning network traffic identification method according to claim 1 is characterized in that: In step 4, S u,t For the membership function of the vth fuzzy subset for: In the formula, S max Indicates the maximum value of the network abnormal status value. express The variance of express The center of , v=1,2,…,R.

7. A fuzzy reasoning network traffic identification method according to claim 6, characterized in that: In step 4, drop (t) Membership function for the cth fuzzy subset for: Among them, a pdrop(t) express The left boundary parameter, b pdrop(t) express The central parameter, c pdrop(t) express The right boundary parameter of .

8. A fuzzy reasoning network traffic identification method according to claim 7, characterized in that: In step 4, S u,t For the membership function of the vth fuzzy subset for: Among them, a S(u,t) express The left boundary parameter, b S(u,t) express The central parameter, c S(u,t) express The right boundary parameter of .

Citation Information

Patent Citations

  • A method and system for parallel detection of abnormal traffic based on hash algorithm

    CN112965970B