A network security analysis system and method based on intelligent learning

Through the intelligent learning network security analysis system, combined with the behavioral data of the server and user side, normal, malicious, abnormal and unknown behavior are identified, which solves the problem of inaccurate identification of abnormal behavior in the existing technology and improves the accuracy of identification of malicious behavior. .

CN118041677BActive Publication Date: 2024-11-22HUNAN HAICHUAN DIGITAL OPERATION MANAGEMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410331606.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-22
Publication Date
2024-11-22
Estimated Expiration
2044-03-22

AI Technical Summary

Technical Problem

The existing network behavior audit technology cannot accurately identify the nature of abnormal behavior, and lacks in-depth analysis of user-side behavior, resulting in a low recognition rate of user behavior detection.

Method used

Using a network security analysis system based on intelligent learning, the average and standard deviation are calculated by collecting behavior data from the server and the user side, using a pre-constructed behavior identification model and user side behavior analysis model, combining access behavior data and mobile behavior data, the mean value and standard deviation are calculated. Identify normal, malicious, abnormal, and unknown behavior.

Benefits of technology

It effectively improves the recognition accuracy of malicious data types and the identification accuracy of user behavior labels, and improves the recognition accuracy of malicious behavior by the network security analysis system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118041677B_ABST
    Figure CN118041677B_ABST
Patent Text Reader

Abstract

The present invention discloses a network security analysis system and method based on intelligent learning, the method comprising: collecting access behavior data of a user ID at a server within a unit time; inputting the collected access behavior data into a pre-constructed behavior recognition model, and outputting a behavior label corresponding to the user ID, the behavior label comprising normal behavior, malicious behavior, abnormal behavior and unknown behavior; collecting user-side behavior data within a preset time of the user ID corresponding to the abnormal behavior and the unknown behavior, the user-side behavior data comprising b types of mobile behavior data; establishing an analysis set corresponding to the b types of mobile behavior data one by one, and calculating the average value and standard deviation of each analysis set one by one, and inputting the average value and standard deviation of each analysis set into a pre-constructed user-side behavior analysis model to obtain a user-side behavior label; the accuracy of the network security analysis system in identifying malicious behavior can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power grid monitoring, and in particular to a network security analysis system and method based on intelligent learning. Background Art

[0002] At present, network security is facing a severe situation. Various network attack methods emerge in an endless stream, bringing huge security risks and economic losses to enterprises and individuals. Although the existing network behavior audit technology can detect abnormal network activities, it cannot accurately identify the nature of abnormal behavior.

[0003] Since network attack behaviors often manifest themselves in multiple stages, it is difficult to comprehensively analyze user behavior using only server-side network traffic data analysis; existing technologies lack in-depth analysis of user-side behavior and cannot fully utilize user-side interactive behavior feature data to assist in user behavior detection, resulting in a low user behavior detection recognition rate.

[0004] Therefore, it is an urgent problem for technical personnel in this field to propose a network security analysis system and method based on intelligent learning to solve the difficulties existing in the prior art. Summary of the invention

[0005] The purpose of the present invention is to provide a network security analysis system and method based on intelligent learning to solve the shortcomings of the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a network security analysis method based on intelligent learning, comprising:

[0007] Collect the access behavior data of the user ID per unit time on the server side;

[0008] Input the collected access behavior data into the pre-built behavior recognition model, and output the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior.

[0009] Collecting user-side behavior data within a preset time period of a user ID corresponding to abnormal behavior and unknown behavior, wherein the user-side behavior data includes b types of mobile behavior data, where b is an integer greater than 0;

[0010] Establish analysis sets corresponding to the b types of mobile behavior data one by one, calculate the mean and standard deviation of each analysis set one by one, and input the mean and standard deviation of each analysis set into a pre-built user-side behavior analysis model to obtain user-side behavior labels; user-side behavior labels include normal behavior and malicious behavior.

[0011] Preferably, the access behavior data includes IP address, data transmission mode, port, transmission data type, transmission data volume, log level characteristics, log message characteristics and user ID.

[0012] Preferably, the method for comprehensively analyzing and identifying the transmission data collected within a unit time and outputting the transmission data type includes:

[0013] Step 1, copy the transmission data collected per unit time by the same user ID into the same document to form a comprehensive document;

[0014] Step 2: Input the comprehensive document into a pre-built transmission data type recognition model to obtain a corresponding transmission data type, where the transmission data type includes a virus, a worm, spyware, adware, a botnet, or ransomware.

[0015] Preferably, the method for training a transmission data type recognition model comprises:

[0016] A transmission data set is collected in advance, where the transmission data set includes j groups of comprehensive documents and transmission data types corresponding to the j groups of comprehensive documents, where j is an integer greater than 1; the transmission data set is divided into a training set and a test set, a classifier is constructed, the comprehensive documents in the training set are used as input data, and the transmission data type in the training set is used as output data, the classifier is trained to obtain an initial classifier, the initial classifier is tested using the test set, and a classifier that meets a preset accuracy is output as a transmission data type recognition model, wherein the transmission data type recognition model is one of a support vector machine, a decision tree, and a random forest.

[0017] Preferably, the training method for the behavior recognition model includes:

[0018] Pre-collecting network behavior identification data, the network behavior identification data including k groups of access behavior data and behavior labels corresponding to the access behavior data, where k is an integer greater than 1, and converting the k groups of access behavior data into k groups of access behavior vector feature data;

[0019] Each set of access behavior vector feature data is used as the input of a behavior recognition model. The behavior recognition model uses the behavior label corresponding to the access behavior vector feature data as the output, uses the behavior label actually corresponding to the access behavior vector feature data as the prediction target, and uses minimizing the loss function value of the behavior recognition model as the training target. When the loss function value of the behavior recognition model is less than or equal to a preset target loss value, the training is stopped to obtain a behavior recognition model, which is one of an SVM regression model, a random forest regression model, or a neural network regression model.

[0020] Preferably, the b-type movement behavior data includes the user's click interval data, mouse movement speed data, mouse acceleration data, scrolling distance data and trajectory curvature change rate within a preset time;

[0021] The analysis set includes a click analysis set, a moving speed set, an acceleration data set, a scroll distance data set and a track curvature set.

[0022] Preferably, the click analysis set is established based on q click interval data within a preset time; the movement speed set is established based on w mouse movement speeds within a preset time; the acceleration data set is established based on e mouse acceleration data within a preset time; the scrolling distance data set is established based on t scrolling distance data within a preset time; the trajectory curvature set is established based on y trajectory curvature change rates within a preset time; q, w, e, t and y are all integers greater than 0.

[0023] Preferably, the method for calculating the standard deviation of the click analysis set includes:

[0024]

[0025] Where PD is the standard deviation of the click analysis set, Da is the ath click interval data of the click analysis set, is the average of the click analysis set;

[0026] The calculation method of the standard deviation of the moving speed set includes:

[0027]

[0028] Where PY is the standard deviation of the moving speed set; Ys is the sth mouse moving speed data in the moving speed set; is the average value of the mouse movement speed data in the movement speed set;

[0029] The method for calculating the standard deviation of the acceleration data set includes:

[0030]

[0031] Where PJ is the standard deviation of the acceleration data set; J f is the fth mouse acceleration data in the acceleration data set; is the average value of the mouse acceleration data in the acceleration data set;

[0032] The method for calculating the standard deviation of the rolling distance data set includes:

[0033]

[0034] Where PG is the standard deviation of the rolling distance data set; Gh is the hth scrolling distance data in the scrolling distance data set; is the average value of the rolling distance data in the rolling distance data set;

[0035] The method for calculating the standard deviation of the trajectory curvature set includes:

[0036]

[0037] Where PL is the standard deviation of the trajectory curvature set; L c is the curvature change rate of the cth trajectory in the trajectory curvature set; is the average value of the trajectory curvature change rate in the trajectory curvature set.

[0038] Preferably, the method for constructing a user-side behavior analysis model includes:

[0039] Pre-collect v groups of user-side behavior data, establish analysis sets for b types of mobile behavior data in a group of user-side behavior data, calculate the average value and standard deviation of each analysis set, use the average value and standard deviation of each analysis set calculated one by one as user-side feature data, and set user-side behavior labels for the user-side feature data;

[0040] The user-side feature data is used as the input of the user-side behavior analysis model, the user-side behavior analysis model uses the user-side behavior label corresponding to the user-side feature data as the output, uses the user-side behavior label actually corresponding to the user-side feature data as the prediction target, and uses minimizing the loss function value of the user-side behavior analysis model as the training target; when the loss function value of the user-side behavior analysis model is less than or equal to a preset target loss value, the training is stopped to obtain the user-side behavior analysis model, and the user-side behavior analysis model is one of a deep belief network, a random forest regression or a naive Bayes model.

[0041] A network security analysis system based on intelligent learning, which executes a network security analysis method based on intelligent learning, including:

[0042] The first data collection module is used to collect the access behavior data of i user IDs per unit time on the server side;

[0043] The behavior recognition module inputs the collected access behavior data into the pre-built behavior recognition model and outputs the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior.

[0044] The second data collection module is used to collect user-side behavior data within a preset time period corresponding to the user ID of the abnormal behavior and the unknown behavior, and the user-side behavior data includes b types of mobile behavior data;

[0045] The user-side analysis module establishes analysis sets corresponding to the b types of mobile behavior data one by one, calculates the mean and standard deviation of each analysis set one by one, and inputs the mean and standard deviation of each analysis set into the pre-built user-side behavior analysis model to obtain user-side behavior labels; user-side behavior labels include normal behavior and malicious behavior.

[0046] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0047] By collecting the access behavior data of the user ID on the server within a unit time, the access behavior data includes IP address, data transmission mode, port, transmission data type, transmission data volume, log level characteristics, log message characteristics and user ID. First, the transmission data collected within a unit time is comprehensively analyzed and identified. This process can effectively improve the accuracy of malicious data type identification. After obtaining the transmission data type, it is combined with other access behavior data to further improve the accuracy of user ID behavior label identification on the server side. Behavior labels include normal behavior, malicious behavior, abnormal behavior and unknown behavior.

[0048] When abnormal and unknown behaviors occur, it may be the early stage of malicious behaviors, or it may be the abnormal operation of normal users, which triggers abnormal behavior warnings. At this time, combined with the user-side behavior data within the preset time of the user ID, the abnormal and unknown behaviors are judged and analyzed again, which can improve the accuracy of the network security analysis system in identifying malicious behaviors. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0050] Figure 1 A schematic diagram of a network security analysis system based on intelligent learning according to the present invention;

[0051] Figure 2 This is a flow chart of a network security analysis method based on intelligent learning of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0054] Example 1

[0055] like Figure 1 As shown, this embodiment is a network security analysis system based on intelligent learning, which is applied to the server side, including a first data acquisition module, a behavior recognition module, a second data acquisition module and a user-side analysis module. The first data acquisition module, the behavior recognition module, the second data acquisition module and the user-side analysis module are connected by wire and / or wirelessly.

[0056] The first data collection module is used to collect the access behavior data of the user ID on the server side within a unit time. The access behavior data includes IP address, data transmission mode, port, transmission data type, transmission data volume, log level characteristics, log message characteristics and user ID; the unit time is 1 minute. Within 1 minute, if the IP address of the same user ID changes frequently, there is a risk of leakage of the user ID and password, which is an abnormal behavior.

[0057] If the data transmission mode and port used by the same user ID are non-standard ports or ports other than commonly used ports for communication, the user ID is trying to bypass the network defense mechanism to make its behavior more covert. For example, common network attacks such as port scanning and malware propagation usually use unconventional ports for communication, which is a malicious behavior. There may be multiple data transmission modes per unit time, indicating that the user ID is trying to find attack vulnerabilities;

[0058] For example, when you open a browser and want to browse some web pages, normally the computer corresponding to the user ID will use standard ports and data transmission modes to communicate with the server. For example, HTTP data transmission mode uses port 80, and HTTPS data transmission mode uses port 443.

[0059] When malicious behavior occurs, the user ID corresponds to a client computer infected with malware that attempts to communicate through unconventional ports, such as port 8080 or port 9999, and data transmission modes such as ICMP or SSH, in order to avoid detection by firewalls or security software.

[0060] The inventors have discovered that when malicious behavior exists, both the malicious transmission data and the volume of the transmission data will change. For example, when attempting to transmit malicious code or instructions through the network, these data can be divided into small blocks for transmission, and only a small part is transmitted each time. The purpose is to avoid being discovered by the network monitoring system. Because the amount of data transmitted is very small, it is not easy to arouse suspicion. Therefore, by collecting the user ID's access behavior data per unit time, and comprehensively analyzing and identifying the transmission data collected per unit time, the accuracy of malicious data type identification can be effectively improved. Specifically, the method of comprehensively analyzing and identifying the transmission data collected per unit time and outputting the transmission data type is described below.

[0061] The log level characteristics and log message characteristics are shown below:

[0062] 2023-01-0110:00:00INFO User logged in successfully

[0063] 2023-01-0110:00:23ERROR Invalid input received

[0064] 2023-01-0110:00:45WARNING Disk space is running low

[0065] Among them, log level features include INFO, ERROR, and WARNING; log message features include User logged insuccessfully, ERROR Invalid input received, and Disk space is running low; before training the subsequent behavior model, the log level features and log message features must be converted into vector form, and the log level features are represented by a binary feature. For example, the "INFO" level is represented as [1, 0, 0], and the remaining levels are represented by other binary features; the log message features are vectorized and converted into numerical form using natural language processing (NLP) technology.

[0066] The method for comprehensively analyzing and identifying the transmission data collected within a unit time and outputting the transmission data type includes:

[0067] Step 1, copy the transmission data collected per unit time by the same user ID into the same document to form a comprehensive document, that is, copy the code and instructions into the same document;

[0068] Step 2: Input the comprehensive document into a pre-built transmission data type recognition model to obtain the corresponding transmission data type, where the transmission data type includes viruses, worms, spyware, adware, botnets, or ransomware.

[0069] Methods for training a transport data type recognition model include:

[0070] A transmission data set is collected in advance, and the transmission data set includes j groups of comprehensive documents and transmission data types corresponding to the j groups of comprehensive documents one by one. The transmission data types are annotated and classified by experts in advance for the comprehensive documents; the transmission data set is divided into a training set and a test set, and the ratio of the training set to the test set is 8:2. A classifier is constructed, and the comprehensive documents in the training set are used as input data, and the transmission data type in the training set is used as output data. The classifier is trained to obtain an initial classifier, and the initial classifier is tested using the test set, and a classifier that meets the preset accuracy is output as a transmission data type recognition model, and the transmission data type recognition model is one of support vector machine (SVM), decision tree, random forest, etc.

[0071] The behavior recognition module inputs the collected access behavior data into the pre-built behavior recognition model and outputs the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior.

[0072] Among them, normal behavior refers to normal network activities, including legitimate user access, normal data transmission and communication, etc.; malicious behavior refers to malicious network activities, including attack behavior, virus propagation and other malicious activities; abnormal behavior refers to behaviors that are inconsistent with normal behavior patterns or suspicious, but not necessarily malicious, including port scanning, abnormal data transmission patterns, failed login attempts, etc.; unknown behavior refers to certain uncertain or unclassifiable network activities, which can be marked as unknown behavior.

[0073] The training methods for the behavior recognition model include:

[0074] Pre-collecting network behavior identification data, the network behavior identification data including k groups of access behavior data and behavior labels corresponding to the access behavior data, and converting the k groups of access behavior data into k groups of access behavior vector feature data;

[0075] Each set of access behavior vector feature data is used as an input of a behavior recognition model, the behavior recognition model uses a behavior label corresponding to the access behavior vector feature data as an output, uses the behavior label actually corresponding to the access behavior vector feature data as a prediction target, and uses minimizing the loss function value of the behavior recognition model as a training target; when the loss function value of the behavior recognition model is less than or equal to a preset target loss value, the training is stopped to obtain a behavior recognition model, and the behavior recognition model is one of a SVM regression model, a random forest regression model, a neural network regression model, and the like;

[0076] The loss function value is mean square error;

[0077] Loss function formula The model is trained with minimization as the goal. In the loss function, MSE is the loss function value of the behavior recognition model, x is the access behavior data group number; k is the number of access behavior data groups; z x The behavior label predicted for the x-th group of access behavior data, The behavior label actually corresponding to the x-th group of access behavior data.

[0078] The second data collection module is used to collect user-side behavior data within a preset time period corresponding to the user ID of the abnormal behavior and the unknown behavior, and the user-side behavior data includes b types of mobile behavior data;

[0079] b. Mobile behavior data includes the user's click interval data, mouse movement speed data, mouse acceleration data, scrolling distance data and trajectory curvature change rate within a preset time; only collecting the user-side behavior data of the user ID corresponding to the abnormal behavior and unknown behavior can effectively reduce the amount of data analysis. Secondly, the reason is that when the abnormal behavior and the unknown behavior correspond to the user, they may also be caused by the user's normal operation. For example, the user tells the account ID to different friends for assistance, which is easy to trigger abnormal behavior; in addition, the access behavior analyzed by the server side is combined with the user-side behavior data to further enhance the user-side user behavior analysis and ensure the use of normal users.

[0080] User-side behavior data can use front-end technologies such as JavaScript to obtain various interactive behaviors of the user on the web page; the original data such as the timestamp when the behavior occurs, page coordinates, click interval data, mouse movement speed data, mouse acceleration data, scrolling distance data and trajectory curvature change rate can be recorded in the event callback function, and more details such as device motion sensor data can be collected with the help of front-end JS libraries (such as fingerprintjs2); user-side behavior data can be uploaded to the server through XHR or WebSocket.

[0081] The user-side analysis module establishes an analysis set for each of the b types of mobile behavior data, calculates the mean and standard deviation of each analysis set, and inputs the mean and standard deviation of each analysis set into a pre-built user-side behavior analysis model to obtain a user-side behavior label; the user-side behavior label includes normal behavior and malicious behavior. When malicious behavior is identified, the network security analysis system issues an early warning, restricts the corresponding user ID, blocks access to the session, or performs other restrictive operations such as verification codes.

[0082] The analysis set includes a click analysis set, a moving speed set, an acceleration data set, a scroll distance data set, and a trajectory curvature set.

[0083] The click analysis set is established based on q click interval data within a preset time; the movement speed set is established based on w mouse movement speeds within a preset time; the acceleration data set is established based on e mouse acceleration data within a preset time; the scrolling distance data set is established based on t scrolling distance data within a preset time; the trajectory curvature set is established based on y trajectory curvature change rates within a preset time; q, w, e, t and y are all integers greater than 0.

[0084] The standard deviation of the click analysis set is calculated by:

[0085]

[0086] Where PD is the standard deviation of the click analysis set, Da is the ath click interval data of the click analysis set, It is the average value of the click analysis set; it is worth noting that normal user clicks usually have a certain random time interval, while the click intervals of the user end controlled by the virus are often highly regular, the interval time is close to a constant value, and is too concentrated, so PD is small.

[0087] The calculation method of the standard deviation of the moving speed set includes:

[0088]

[0089] Where PY is the standard deviation of the moving speed set; Ys is the sth mouse moving speed data in the moving speed set; It is the average value of the mouse movement speed data in the movement speed set; the normal user-side mouse movement speed has a certain random fluctuation; while the movement trajectory of the user-side controlled by the virus is usually a straight line or a smooth curve, and the mouse movement speed distribution is too concentrated. Therefore, the user-side behavior can be reflected by the PY value.

[0090] The calculation method of the standard deviation of the acceleration data set includes:

[0091]

[0092] Where PJ is the standard deviation of the acceleration data set; J f is the fth mouse acceleration data in the acceleration data set; It is the average value of the mouse acceleration data in the acceleration data set. When a normal user moves the mouse, there will be some sudden changes in acceleration. The mouse movement trajectory of the user controlled by the virus is smooth, the acceleration changes slowly, and there is a lack of high acceleration components. Therefore, the PJ value can reflect the user behavior.

[0093] The calculation method of the standard deviation of the rolling distance data set includes:

[0094]

[0095] Where PG is the standard deviation of the rolling distance data set; G h is the hth scrolling distance data in the scrolling distance data set; It is the average value of the scrolling distance data in the scrolling distance data set. When a normal user scrolls a page, the distance and time are somewhat random and have a large degree of discreteness. The scrolling of a user controlled by a virus usually pauses at a specific position. Under abnormal behavior, the scrolling distance distribution shows an overly concentrated peak value and a small degree of discreteness. Therefore, the PG value can reflect the user behavior.

[0096] The calculation method of the standard deviation of the trajectory curvature set includes:

[0097]

[0098] Where PL is the standard deviation of the trajectory curvature set; L c is the curvature change rate of the cth trajectory in the trajectory curvature set; It is the average value of the trajectory curvature change rate in the trajectory curvature set; the curvature in the normal user-side mouse motion trajectory has a large change, and the user-side mouse motion trajectory controlled by the virus automation program is smooth and the curvature change is small. Therefore, under abnormal behavior, the trajectory curvature change rate is generally small, that is, the PL value is small; there is a lack of large curvature changes.

[0099] In the above formula, the three average values ​​can reflect the corresponding mean values ​​of most normal user-side behavior data, and combined with the corresponding standard deviation to reflect the distribution of user-side behavior data, the accuracy of abnormal behavior identification can be further improved.

[0100] Methods for building user-side behavior analysis models include:

[0101] Pre-collect v groups of user-side behavior data, establish analysis sets for b types of mobile behavior data in a group of user-side behavior data, calculate the average value and standard deviation of each analysis set, use the average value and standard deviation of each analysis set calculated one by one as user-side feature data, and set user-side behavior labels for the user-side feature data;

[0102] The user-side feature data is used as the input of the user-side behavior analysis model, the user-side behavior analysis model uses the user-side behavior label corresponding to the user-side feature data as the output, uses the user-side behavior label actually corresponding to the user-side feature data as the prediction target, and uses minimizing the loss function value of the user-side behavior analysis model as the training target; when the loss function value of the user-side behavior analysis model is less than or equal to a preset target loss value, the training is stopped to obtain the user-side behavior analysis model, and the user-side behavior analysis model is one of the models such as deep belief network, random forest regression or naive Bayes.

[0103] Example 2

[0104] like Figure 2 As shown, this embodiment provides a network security analysis method based on intelligent learning, including:

[0105] Collect the access behavior data of the user ID per unit time on the server side;

[0106] Input the collected access behavior data into the pre-built behavior recognition model, and output the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior.

[0107] Collecting user-side behavior data within a preset time period corresponding to the user ID of abnormal behavior and unknown behavior, wherein the user-side behavior data includes b types of mobile behavior data;

[0108] Establish analysis sets corresponding to the b types of mobile behavior data one by one, calculate the mean and standard deviation of each analysis set one by one, and input the mean and standard deviation of each analysis set into a pre-built user-side behavior analysis model to obtain user-side behavior labels; user-side behavior labels include normal behavior and malicious behavior.

[0109] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A network security analysis method based on intelligent learning, characterized in that: include: Collect the access behavior data of the user ID per unit time on the server side; Input the collected access behavior data into the pre-built behavior recognition model, and output the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior. Collecting user-side behavior data within a preset time period of a user ID corresponding to abnormal behavior and unknown behavior, wherein the user-side behavior data includes b types of mobile behavior data, where b is an integer greater than 0; Establish analysis sets for the b types of mobile behavior data one by one, calculate the mean and standard deviation of each analysis set one by one, and input the mean and standard deviation of each analysis set into the pre-built user-side behavior analysis model to obtain user-side behavior labels; user-side behavior labels include normal behavior and malicious behavior; The b-type movement behavior data includes the user's click interval data, mouse movement speed data, mouse acceleration data, scrolling distance data and trajectory curvature change rate within a preset time; The method for constructing a user-side behavior analysis model comprises: Pre-collect v groups of user-side behavior data, establish analysis sets for b types of mobile behavior data in a group of user-side behavior data, calculate the average value and standard deviation of each analysis set, use the average value and standard deviation of each analysis set calculated one by one as user-side feature data, and set user-side behavior labels for the user-side feature data; The user-side feature data is used as the input of the user-side behavior analysis model, the user-side behavior analysis model uses the user-side behavior label corresponding to the user-side feature data as the output, uses the user-side behavior label actually corresponding to the user-side feature data as the prediction target, and uses minimizing the loss function value of the user-side behavior analysis model as the training target; when the loss function value of the user-side behavior analysis model is less than or equal to a preset target loss value, the training is stopped to obtain the user-side behavior analysis model, and the user-side behavior analysis model is one of a deep belief network, a random forest regression or a naive Bayes model; The analysis set includes a click analysis set, a moving speed set, an acceleration data set, a scroll distance data set and a track curvature set; The click analysis set is established based on q click interval data within a preset time; the movement speed set is established based on w mouse movement speeds within a preset time; the acceleration data set is established based on e mouse acceleration data within a preset time; the scrolling distance data set is established based on t scrolling distance data within a preset time; the trajectory curvature set is established based on y trajectory curvature change rates within a preset time; q, w, e, t and y are all integers greater than 0.

2. According to the network security analysis method based on intelligent learning according to claim 1, it is characterized in that: The access behavior data includes IP address, data transmission mode, port, transmission data type, transmission data volume, log level characteristics, log message characteristics and user ID.

3. A network security analysis method based on intelligent learning according to claim 2, characterized in that: The method for comprehensively analyzing and identifying the transmission data collected within a unit time and outputting the transmission data type includes: Step 1, copy the transmission data collected per unit time by the same user ID into the same document to form a comprehensive document; Step 2: Input the comprehensive document into a pre-built transmission data type recognition model to obtain a corresponding transmission data type, where the transmission data type includes a virus, a worm, spyware, adware, a botnet, or ransomware.

4. A network security analysis method based on intelligent learning according to claim 3, characterized in that: Methods for training a transport data type recognition model include: A transmission data set is collected in advance, where the transmission data set includes j groups of comprehensive documents and transmission data types corresponding to the j groups of comprehensive documents, where j is an integer greater than 1; the transmission data set is divided into a training set and a test set, a classifier is constructed, the comprehensive documents in the training set are used as input data, and the transmission data type in the training set is used as output data, the classifier is trained to obtain an initial classifier, the initial classifier is tested using the test set, and a classifier that meets a preset accuracy is output as a transmission data type recognition model, wherein the transmission data type recognition model is one of a support vector machine, a decision tree, and a random forest.

5. The network security analysis method based on intelligent learning according to claim 2 is characterized in that: The training methods for the behavior recognition model include: Pre-collecting network behavior identification data, the network behavior identification data including k groups of access behavior data and behavior labels corresponding to the access behavior data, where k is an integer greater than 1, and converting the k groups of access behavior data into k groups of access behavior vector feature data; Each set of access behavior vector feature data is used as the input of a behavior recognition model. The behavior recognition model uses the behavior label corresponding to the access behavior vector feature data as the output, uses the behavior label actually corresponding to the access behavior vector feature data as the prediction target, and uses minimizing the loss function value of the behavior recognition model as the training target. When the loss function value of the behavior recognition model is less than or equal to a preset target loss value, the training is stopped to obtain a behavior recognition model, which is one of an SVM regression model, a random forest regression model, or a neural network regression model.

6. A network security analysis system based on intelligent learning, executing a network security analysis method based on intelligent learning according to any one of claims 1 to 5, characterized in that: include: The first data collection module is used to collect the access behavior data of i user IDs per unit time on the server side; The behavior recognition module inputs the collected access behavior data into the pre-built behavior recognition model and outputs the behavior label corresponding to the user ID. The behavior label includes normal behavior, malicious behavior, abnormal behavior and unknown behavior. The second data collection module is used to collect user-side behavior data within a preset time period corresponding to the user ID of the abnormal behavior and the unknown behavior, and the user-side behavior data includes b types of mobile behavior data; The user-side analysis module establishes analysis sets corresponding to the b types of mobile behavior data one by one, calculates the mean and standard deviation of each analysis set one by one, and inputs the mean and standard deviation of each analysis set into the pre-built user-side behavior analysis model to obtain user-side behavior labels; user-side behavior labels include normal behavior and malicious behavior.