Training method of alarm recognition model, method and device for alarm recognition

By constructing the label feature matrix and association relationship matrix of alarm data, using graph convolutional neural network model to train the alarm recognition model, and integrating the basic alarm features and intelligence matching features, the problem of inaccurate alarm information recognition in the existing technology is solved, and the problem of inaccurate alarm information recognition is realized Accurate positioning and reduction of workload of abnormal equipment.

CN113497729BActive Publication Date: 2024-11-26CHINA MOBILE INFORMATION TECHNOLOGY CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010260565.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-04-03
Publication Date
2024-11-26
Estimated Expiration
2040-04-03

Smart Images

  • Figure CN113497729B_ABST
    Figure CN113497729B_ABST
Patent Text Reader

Abstract

The present invention discloses a training method for an alarm recognition model, an alarm recognition method and a device. Specifically, the method comprises: obtaining a plurality of alarm data samples; preprocessing each alarm data sample respectively to obtain a training sample corresponding to each of the alarm data samples, wherein each training sample comprises a plurality of sample alarm information and an alarm category of each sample alarm information; constructing a sample alarm information label feature matrix and an adjacency matrix corresponding to the correlation relationship between the plurality of sample alarm information according to the plurality of sample alarm information; using the label feature matrix and the adjacency matrix of the sample alarm information as the input of the alarm recognition model, using the alarm category as the output of the alarm recognition model, iteratively training the alarm recognition model to be trained, and obtaining a target alarm recognition model. According to an embodiment of the present invention, the accuracy of alarm information recognition can be improved to reduce the difficulty and workload of operation and maintenance personnel.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a training method for an alarm recognition model, an alarm recognition method and device, equipment and computer storage medium. Background Art

[0002] Many network security monitoring devices used in daily applications, such as botnet monitoring and disposal systems, intrusion detection systems, and intrusion prevention systems, will issue alarms when they detect abnormalities in asset equipment. However, the amount of alarm data sent by these systems is large, and some alarms are false alarms. Further analysis and identification of the alarm information sent by the system is required to obtain more effective and accurate alarm information.

[0003] At present, when alarm recognition models are usually used to identify alarm information, the alarm information involved in the analysis is not comprehensive enough, resulting in low accuracy of alarm information recognition by existing alarm recognition models, and the inability to accurately locate abnormal assets and equipment, which in turn increases the difficulty and workload of operation and maintenance personnel in troubleshooting problematic equipment.

[0004] Therefore, how to improve the recognition accuracy of the alarm recognition model and accurately locate abnormal assets and equipment is an urgent problem to be solved. Summary of the invention

[0005] The embodiments of the present invention provide a training method for an alarm recognition model, an alarm recognition method, an apparatus, a device and a computer storage medium, which can improve the accuracy of alarm information recognition, thereby accurately locating abnormal asset equipment, reducing the difficulty and workload of operation and maintenance personnel in troubleshooting problem equipment.

[0006] In a first aspect, an embodiment of the present invention provides a method for training an alarm recognition model, the method comprising:

[0007] Obtain multiple alarm data samples;

[0008] Preprocessing each of the alarm data samples respectively to obtain a training sample corresponding to each of the alarm data samples, wherein each of the training samples includes a plurality of sample alarm information and an alarm category of each sample alarm information, wherein the sample alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0009] Constructing, according to the plurality of sample alarm information, a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information;

[0010] The label feature matrix of the sample alarm information and the adjacency matrix are used as inputs of the alarm recognition model, and the alarm category is used as output of the alarm recognition model. The alarm recognition model to be trained is iteratively trained to obtain a target alarm recognition model.

[0011] In a second aspect, an embodiment of the present invention provides a method for alarm identification, the method comprising:

[0012] Get multiple alarm data;

[0013] Preprocessing each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data, wherein the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0014] According to the plurality of alarm information, construct a label feature matrix corresponding to each alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information;

[0015] The label feature matrix and the adjacency matrix are input into a target alarm recognition model to obtain an alarm category corresponding to the alarm data; the target alarm recognition model is a model trained using the alarm recognition model training method described in claim 1.

[0016] Optionally, the preprocessing each of the alarm data respectively to obtain the alarm information corresponding to each of the alarm data includes:

[0017] Extracting features from each of the alarm data respectively to obtain first alarm feature information corresponding to each of the alarm data;

[0018] Each of the alarm data is matched with the information of the preset threat intelligence library to obtain the second alarm feature information corresponding to each of the alarm data.

[0019] Optionally, performing feature extraction on each of the alarm data to obtain first alarm feature information corresponding to each of the alarm data includes:

[0020] Extracting features from basic information of each of the alarm data respectively to obtain first feature information of the first alarm feature information;

[0021] The time characteristic information of each of the alarm data is calculated using a time series analysis algorithm to obtain second characteristic information of the first alarm characteristic information.

[0022] Optionally, the second alarm characteristic information includes any one or more of the following:

[0023] Internet Protocol (IP) reputation information, IP-related sample information, and IP-related domain name information.

[0024] Optionally, the time characteristic information includes first time characteristic information and second time characteristic information, the first time characteristic information is single-day time characteristic information, and the second time characteristic information is time characteristic information within a preset time period;

[0025] The calculating the time characteristic information of each of the alarm data by using a time series analysis algorithm to obtain the second characteristic information of the first alarm characteristic information includes:

[0026] Using a time series analysis algorithm to calculate the first time characteristic information of each of the alarm data, respectively, to obtain second characteristic information corresponding to the first time characteristic information;

[0027] Using a time series analysis algorithm to calculate the second time characteristic information of each of the alarm data, respectively, to obtain second characteristic information corresponding to the second time characteristic information;

[0028] The second characteristic information corresponding to the first time characteristic information and the second characteristic information corresponding to the second time characteristic information are combined to obtain the second characteristic information of the first alarm characteristic information.

[0029] Optionally, the calculating the first time characteristic information of each alarm data by using a time series analysis algorithm to obtain the second characteristic information corresponding to the first time characteristic information includes:

[0030] The first time characteristic information of each of the alarm data is calculated using the least square method to obtain second characteristic information corresponding to the first time characteristic information.

[0031] Optionally, the calculating the second time characteristic information of each of the alarm data by using a time series analysis algorithm to obtain the second characteristic information corresponding to the second time characteristic information includes:

[0032] Selecting the first single-day characteristic information corresponding to the minimum value of the number of single-day alarms in the alarm data within a preset time period;

[0033] Calculate the mean of the first single-day characteristic information to obtain a time mean;

[0034] Matching the second daily characteristic information within the preset time period with the time mean to obtain a time point that meets a preset matching condition, wherein the second daily characteristic information is the daily characteristic information corresponding to each day other than the day corresponding to the first daily characteristic information;

[0035] The time points satisfying the preset condition are calculated using the least square method to obtain second feature information corresponding to the second time feature information.

[0036] Optionally, constructing, based on the plurality of alarm information, a label feature matrix corresponding to each alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information, comprises:

[0037] Constructing a label feature matrix corresponding to each of the alarm information according to the first alarm feature information and the second alarm feature information of the plurality of alarm information;

[0038] Constructing corresponding association relationship information between at least two pieces of alarm information according to the IP control relationship information of the plurality of alarm information;

[0039] According to the association relationship information, the corresponding adjacency matrix is ​​determined.

[0040] Optionally, the alarm recognition model includes a graph convolutional neural network model.

[0041] In a third aspect, an embodiment of the present invention provides a training device for an alarm recognition model, the device comprising:

[0042] An acquisition module, used to acquire multiple alarm data samples;

[0043] A preprocessing module, used to preprocess the plurality of sample alarm data to obtain training samples, wherein the training samples include a plurality of sample alarm information and an alarm category of each sample alarm information, wherein the sample alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0044] A construction module, configured to construct, based on the plurality of sample alarm information, a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information;

[0045] The training module is used to use the label feature matrix of the sample alarm information and the adjacency matrix as inputs of the alarm recognition model, and the alarm category as output of the alarm recognition model, and iteratively train the alarm recognition model to be trained to obtain a target alarm recognition model.

[0046] In a fourth aspect, an embodiment of the present invention provides a device for alarm identification, the device comprising:

[0047] An acquisition module is used to acquire multiple alarm data;

[0048] A preprocessing module, used to preprocess each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data, wherein the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0049] A construction module, used to construct, based on the plurality of alarm information, a label feature matrix corresponding to each of the alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information;

[0050] The recognition module is used to input the label feature matrix and the adjacency matrix into the target alarm recognition model to obtain the alarm category corresponding to the alarm data; the target alarm recognition model is a model trained using the alarm recognition model training method described in the first aspect.

[0051] In a fifth aspect, an embodiment of the present invention provides a training device for an alarm recognition model, the device comprising: a processor and a memory storing computer program instructions;

[0052] When the processor executes the computer program instructions, the training method of the alarm recognition model as described in the first aspect is implemented.

[0053] In a sixth aspect, an embodiment of the present invention provides a device for alarm identification, the device comprising: a processor and a memory storing computer program instructions;

[0054] When the processor executes the computer program instructions, the method for alarm identification as described in the second aspect and the optional aspect of the second aspect is implemented.

[0055] In the seventh aspect, an embodiment of the present invention provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the training method of the alarm recognition model as described in the first aspect, and / or the alarm recognition method as described in the second aspect and the optional aspect of the second aspect are implemented.

[0056] The training method of the alarm recognition model, the method, device, equipment and computer storage medium of the alarm recognition in the embodiment of the present invention, by combining the sample alarm information and the correlation between the sample alarm information, trains the target alarm recognition model. In addition, the sample alarm information not only involves the alarm statistical feature information, but also considers the intelligence matching feature information, and fuses the characteristics of the alarm information itself with the intelligence matching feature information. The training samples of the alarm recognition model include alarm information of multiple different dimensions and the correlation between each alarm information, which enables the model to conduct a more comprehensive analysis of the alarms in the network environment and identify the alarm information that actually has infection problems. Based on this, the technical solution of the embodiment of the present invention can improve the accuracy of alarm information recognition, and then accurately locate the abnormal asset equipment, reducing the difficulty and workload of the operation and maintenance personnel in troubleshooting the problem equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 It is a flowchart of a method for training an alarm recognition model provided by an embodiment of the present invention;

[0059] Figure 2 It is a flowchart of a method for alarm identification provided by an embodiment of the present invention;

[0060] Figure 3 It is a schematic diagram of the alarm information identification and analysis of the zombie worm system provided by an embodiment of the present invention;

[0061] Figure 4 It is a schematic diagram of a scenario graph of an alarm log recognition model provided by an embodiment of the present invention;

[0062] Figure 5 is a schematic diagram of a GCN neural network structure provided by an embodiment of the present invention;

[0063] Figure 6 It is a structural schematic diagram of a training device for an alarm recognition model provided by an embodiment of the present invention;

[0064] Figure 7 It is a structural diagram of an alarm identification device provided by an embodiment of the present invention;

[0065] Figure 8 It is a schematic diagram of the hardware structure of the training device of the alarm recognition model provided by an embodiment of the present invention;

[0066] Fig. 9 It is a schematic diagram of the hardware structure of the alarm identification device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present invention and are not configured to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0068] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0069] At present, network security monitoring systems such as botnet monitoring and disposal systems, intrusion detection systems (IDS), intrusion prevention systems (IPS) and website application-level intrusion prevention systems (WAF) are mainly used to monitor network devices or network transmissions in real time and issue alarms when suspicious transmission behaviors are detected.

[0070] Generally, although the above network security monitoring system can send an alarm to the infected host, the alarm false alarm rate is very high, and the alarm information sent by the system needs to be further analyzed and identified to obtain more effective and accurate alarm information. However, the existing alarm recognition model has a low accuracy rate for alarm information recognition and cannot accurately locate abnormal asset equipment.

[0071] In order to solve the problems of the prior art, the training method of the alarm recognition model, the method, device, equipment and computer storage medium of alarm recognition provided in the embodiments of the present invention can train the alarm recognition model by combining the sample alarm information and the correlation between the sample alarm information. The training samples of the alarm recognition model include alarm information of multiple different dimensions and the correlation between each alarm information, wherein the alarm information includes basic statistical feature information of the alarm and intelligence matching feature information, and the basic features of the alarm information and the intelligence matching feature information are integrated. This enables the model to conduct a more comprehensive analysis of the alarms in the network environment and more accurately identify the alarm information with infection problems. Based on this, the accuracy of the alarm information recognition can be improved, and then the abnormal asset equipment can be accurately located. Based on more accurate alarm information, the scope of abnormal equipment that needs to be troubleshooted can be narrowed, reducing the difficulty and workload of the operation and maintenance personnel in troubleshooting the problem equipment.

[0072] The following describes the training method of the alarm recognition model, the method, device, equipment and computer storage medium for alarm recognition provided by the embodiments of the present invention in conjunction with the accompanying drawings. It should be noted that these embodiments are not intended to limit the scope of the present invention.

[0073] The following first introduces the training method of the alarm recognition model provided by the embodiment of the present invention.

[0074] In the embodiment of the present invention, Figure 1 As shown, Figure 1 1 is a flow chart of a method for training an alarm recognition model provided by an embodiment of the present invention. The method for training an alarm recognition model is specifically implemented as follows:

[0075] S101: Acquire multiple alarm data samples.

[0076] Specifically, the alarm data sample may include alarm log data obtained from a corresponding network security monitoring system.

[0077] Here, the acquired alarm log data may be cleaned first. Redundant fields in the alarm log data may be deleted to improve the efficiency of alarm log analysis and identification. Specifically, the redundant fields may include fields that are irrelevant to identifying alarm information, or repeated fields, etc.

[0078] Specifically, for the alarm time in the alarm log data, the discrete time format can also be converted into a continuous value from 0 to 24, so as to perform feature extraction operations based on the alarm time, thereby improving the efficiency of alarm log analysis and recognition. The specific conversion method may include but is not limited to existing discrete data conversion rules, which will not be repeated here.

[0079] S102: Preprocess each of the alarm data samples respectively to obtain a training sample corresponding to each of the alarm data samples.

[0080] Specifically, the training sample may include a plurality of sample alarm information and an alarm category of each sample alarm information. The sample alarm information may include first alarm feature information and second alarm feature information.

[0081] Here, the first alarm characteristic information may be basic alarm characteristic information. Specifically, feature extraction may be performed on each alarm data to obtain the first alarm characteristic information corresponding to each alarm data. The basic alarm characteristic information may include one or more characteristic information such as the number of alarms, the number of corresponding controlled IPs, the number of corresponding controlling IPs, the number of alarms of the controlling IPs, the number of alarms of the controlled IPs, the single-day regression error of the IPs, the historical horizontal regression error of the IPs, and the variance of the IP alarm interval. It is understandable that the first alarm characteristic information may be the basic statistical characteristic information of the alarm.

[0082] Specifically, the second alarm characteristic information is intelligence matching characteristic information obtained after intelligence matching. Specifically, each alarm data can be matched with information in a preset threat intelligence library to obtain the second alarm characteristic information corresponding to each alarm data.

[0083] Specifically, threat intelligence is evidence-based knowledge about existing or potential threats facing IT or information assets, including scenarios, mechanisms, indicators, inferences, and feasible recommendations, which can provide a basis for decision-making in threat response. According to the data itself, threat intelligence can include hash values, IP addresses, domain names, network or host characteristics, etc. The preset threat intelligence library can be a specialized database built based on threat intelligence. Users can build relevant threat intelligence libraries based on their own work data, or directly use various existing open source threat intelligence libraries.

[0084] Specifically, the second alarm feature information may include one or more of IP reputation information, IP associated sample information, IP associated domain name information, and the like.

[0085] Specifically, the alarm category of the sample alarm information may be a network security problem category corresponding to the sample alarm information, such as host infection and compromise, where label 1 may be used to represent host infection and compromise, and label 0 may be used to represent non-host infection and compromise.

[0086] S103: Constructing, according to the plurality of sample alarm information, a sample alarm information label feature matrix and an adjacency matrix corresponding to association relationships between the plurality of sample alarm information.

[0087] Specifically, for multiple sample alarm information in the training sample set, on the one hand, feature learning is performed on multiple sample alarm information to construct a label feature matrix corresponding to the alarm information; on the other hand, feature learning is performed on the correlation between multiple sample alarm information to construct a corresponding adjacency matrix.

[0088] S104: using the label feature matrix and the adjacency matrix of the sample alarm information as inputs of the alarm recognition model, using the alarm category as output of the alarm recognition model, iteratively training the alarm recognition model to be trained, and obtaining a target alarm recognition model.

[0089] Here, based on the label feature matrix of the sample alarm information, the adjacency matrix and the alarm category of each sample alarm information, an alarm recognition model can be constructed so that the alarm recognition model has the ability to output the alarm category of any alarm information based on the label feature matrix and the corresponding adjacency matrix of any input alarm information.

[0090] Specifically, the target alarm recognition model is a model that can be used to perform deep learning on feature information of nodes and structural relationship information between nodes, and may specifically include but is not limited to a graph convolutional neural network (GCN) model.

[0091] In summary, in the embodiment of the present invention, the method for training the alarm recognition model can use the sample alarm information and the correlation between the sample alarm information as training samples to train the target alarm recognition model. The alarm information used as the training sample not only has the basic feature information of the alarm, but also has the intelligence matching feature information. It can be seen that, because the alarm feature information of multiple dimensions of the alarm and the structural relationship between the alarms are considered when training the model, the corresponding feature information of the alarm with host infection problems is learned more accurately, so that the model can perform a more comprehensive and accurate identification and analysis of the alarms in the network environment, and then identify the alarms that may actually have host infection problems. Based on this, the method of the embodiment of the present invention can improve the accuracy of the alarm information identification, and then accurately locate the abnormal asset equipment, reducing the difficulty and workload of the operation and maintenance personnel in troubleshooting the problem equipment.

[0092] The following is a detailed introduction to the alarm identification method provided by the embodiment of the present invention.

[0093] Figure 2 FIG. 1 is a flow chart of a method for alarm identification provided by an embodiment of the present invention. Figure 2 As shown, in an embodiment of the present invention, the method for alarm identification may include the following steps:

[0094] S201: Acquire multiple alarm data.

[0095] Specifically, the alarm data is the alarm data to be identified. The alarm data may include alarm log data obtained from a network security monitoring system.

[0096] Specifically, the acquired multiple alarm data may be cleaned, redundant fields in the alarm log data may be deleted, and discrete alarm times in the alarm data may be processed continuously to improve the efficiency of alarm log analysis and identification.

[0097] S202: Preprocess each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data.

[0098] Here, the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching. Preprocessing the alarm data may include feature extraction of the alarm data and threat intelligence matching of the alarm data.

[0099] Specifically, the first alarm feature information corresponding to each alarm data can be obtained by extracting features from each alarm data. The first alarm feature information can be basic statistical feature information of the alarm.

[0100] Here, the first alarm characteristic information may include first characteristic information and second characteristic information.

[0101] Specifically, the first feature information may be obtained by extracting features from basic information of each alarm data. Here, the basic information of the alarm data may include field information related to the alarm data used for statistical analysis.

[0102] Specifically, the first characteristic information may include one or more of the following characteristic information: number of alarms, number of corresponding controlled IPs, number of corresponding controlling IPs, number of alarms of controlling IPs, number of alarms of controlled IPs, and variance of IP alarm intervals.

[0103] Specifically, the time characteristic information of each alarm data is calculated using a time series analysis algorithm to obtain the second characteristic information.

[0104] Here, the time characteristic information may be obtained by performing serial processing on the alarm time in the alarm data.

[0105] Specifically, the time characteristic information may include first time characteristic information and second time characteristic information. The first time characteristic information may be time characteristic information of a single day, and the second time characteristic information may be time characteristic information within a preset time period, that is, time characteristic information of historical data.

[0106] Specifically, the first time characteristic information of each alarm data is calculated by using the time series analysis algorithm to obtain the second characteristic information corresponding to the first time characteristic information. Optionally, the first time characteristic information is regressed by using the least square method to obtain the second characteristic information corresponding to the first time characteristic information, that is, the IP single-day regression error E of the alarm data. d .

[0107] Specifically, by using a time series analysis algorithm, the second time characteristic information of each alarm data is calculated respectively, and the second characteristic information corresponding to the second time characteristic information can be obtained.

[0108] Specifically, the second time characteristic information may be time characteristic information within a preset time period. First, the first single-day characteristic information corresponding to the minimum value of the number of single-day alarms in the alarm data within the preset time period may be selected. The first single-day characteristic information may be a time value after continuous processing, or may be multiple time values. The number of single-day alarms in the alarm data may be the number of single-day alarms corresponding to any host.

[0109] Secondly, the feature information of the first single day is averaged to obtain the time mean.

[0110] Thirdly, the second single-day characteristic information within the preset time period is matched with the time mean to obtain a time point that meets the preset matching condition.

[0111] Here, the second single-day characteristic information is the single-day characteristic information corresponding to each single day other than the single day corresponding to the first single-day characteristic information. The second single-day characteristic information can be understood as the single-day characteristic information corresponding to other dates except the date with the least number of alarms within the preset time period. These second single-day characteristic information are matched with the time mean to determine the time point that meets the preset matching condition. The preset matching condition can be that the time point in the second single-day characteristic information is consistent with or close to the time mean. Select the time point that meets the preset matching condition.

[0112] Then, the least square method can be used to calculate the time points that meet the preset conditions to obtain the second feature information corresponding to the second time feature information, that is, the IP historical horizontal regression error E of the alarm data. t .

[0113] By the above operation, the second characteristic information corresponding to the first time characteristic information and the second characteristic information corresponding to the second time characteristic information can be obtained respectively. The second characteristic information corresponding to the first time characteristic information and the second characteristic information corresponding to the second time characteristic information are combined to obtain the second characteristic information of the first alarm characteristic information, that is, the total second characteristic information.

[0114] Specifically, the second alarm feature information is the intelligence matching feature information obtained after intelligence matching. Specifically, each alarm data can be matched with the information of the preset threat intelligence library to obtain the second alarm feature information corresponding to each alarm data. The information of the preset threat intelligence library is all existing information threatening network security. Matching the alarm data with the threat intelligence in the threat intelligence library can increase the analysis dimension of the alarm data to identify whether the alarm is really an alarm of network security problems.

[0115] Specifically, the preset threat intelligence library can be a special database built based on threat intelligence. Threat intelligence can include HASH values, IP addresses, domain names, network or host features, etc. It is understandable that users can build relevant threat intelligence libraries based on threat data accumulated in actual work, or directly use various existing open source threat intelligence libraries, which will not be elaborated here.

[0116] Therefore, by performing the above operations, feature information of multiple dimensions of alarm data can be obtained, including basic feature information of alarm data and feature information after threat intelligence matching. Based on the identification and analysis of rich alarm data dimensions, the probability of identifying correct alarms can be increased, most false alarms can be screened out, and the false alarm rate can be reduced.

[0117] S203: constructing, based on the plurality of alarm information, a label feature matrix corresponding to each alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information.

[0118] Here, the corresponding association relationship information between at least two alarm information is constructed based on the IP control relationship information of the multiple alarm information. Specifically, since there may be a certain association relationship between different alarm information, such as the IP control relationship information of the alarm information, that is, the IP of a part of the alarm information is the control IP, and the IP of a part of the alarm information is the controlled IP, according to the IP control relationship information, an IP structure relationship diagram of the alarm information can be constructed to determine the corresponding association relationship information between at least two alarm information, and according to the association relationship information, feature learning is performed to determine the corresponding adjacency matrix.

[0119] In addition, a label feature matrix corresponding to each alarm information is constructed according to the first alarm feature information and the second alarm feature information of the plurality of alarm information. Specifically, feature learning can be performed on the plurality of alarm information to obtain a label feature matrix corresponding to each alarm information.

[0120] S204: Input the label feature matrix and the adjacency matrix into the target alarm recognition model to obtain the alarm category corresponding to the alarm data.

[0121] Specifically, the target alarm recognition model is a model trained using the alarm recognition model training method described in the above embodiment. Inputting multiple alarm information and the corresponding association relationship information between at least two alarm information into the alarm recognition model can obtain an alarm recognition result, that is, the alarm category corresponding to the alarm data.

[0122] It can be understood that the alarm recognition model can be a graph convolutional neural network model.

[0123] Specifically, graph convolutional neural network is an algorithm that can perform deep learning on graph data. In practical applications, many data do not have regular spatial structures, such as graphs abstracted from recommendation systems, electronic trading systems, computational geometry, brain signals, molecular structures, etc. The connections of each node in these graph structures are different. Some nodes have three connections, and some nodes have two connections, which are irregular data structures. Therefore, a method that can perform deep learning on graph data came into being, namely graph convolutional neural network. It can be understood that a graph is a data format that can be used to represent social networks, communication networks, protein molecular networks, etc. Generally speaking, the nodes in the graph are used to represent individuals in the network, and the edges in the graph are used to represent the connection relationship between individuals in the network. Therefore, we can think that graph data has two characteristics. One is that each node has its own feature information, and the other is that each node in the graph also has structural information. Among them, the GCN model can perform node classification or edge prediction based on the node feature information and node structure information in the input graph data. In the embodiment of the present invention, the GCN model is used to combine the association relationship between each alarm information (ie, node structure information) and the alarm feature information (node ​​feature information) of each alarm information to determine the alarm information category of each alarm information.

[0124] In summary, in the embodiments of the present invention, the method of alarm identification can accurately identify alarms of network security issues such as host infection that actually exist based on the alarm feature information of multiple dimensions of the alarm information and the structural relationship between the alarm information, and can achieve more comprehensive and accurate identification and analysis of alarms in the network environment, thereby improving the accuracy of alarm information identification.

[0125] Moreover, through the above-mentioned alarm identification method, most of the false alarms can be screened out, and the correct alarms can be identified more accurately, so as to accurately locate the infected and compromised hosts and other network asset devices, thereby reducing the number of abnormal asset devices that need to be troubleshooted, and reducing the difficulty and workload of operation and maintenance personnel in troubleshooting problem equipment.

[0126] In order to better understand the method of the present invention, the method for identifying alarm information is now described in detail in conjunction with an application example.

[0127] Optionally, in some embodiments of the present invention, taking the botnet monitoring and disposal system as an example, in the actual application scenario, the alarms of network security issues such as host infection can be more accurately analyzed in the manner shown below to achieve alarm identification.

[0128] Specifically, the zombie worm monitoring and disposal system (hereinafter referred to as the zombie worm system) is a security device that monitors abnormal behavior of the host. It collects all data in the host network through traffic mirroring, performs security detection and analysis on the data, discovers malicious files transmitted in the host network and abnormal events, and issues alarms in a timely manner. However, the zombie worm system has problems such as a large number of alarms, a high false alarm rate, and a small number of alarm content dimensions. These problems greatly increase the workload of operation and maintenance personnel, making it difficult to carry out asset inspections, and the threat of infected assets cannot be cleared in a timely manner.

[0129] In the implementation of the present invention, Figure 3 As shown, Figure 3 This is a schematic diagram of an embodiment of the present invention for identifying and analyzing alarm information of a zombie worm system. By using the method for identifying and analyzing alarm information of a zombie worm system, the alarms of infected compromised hosts in a large amount of alarm log data can be identified.

[0130] First, data processing is performed on the alarm log generated by the zombie worm system, which may include: deleting redundant fields in the alarm log and performing serial processing on the alarm time, etc.

[0131] Specifically, the daily alarm time is processed continuously, and the discrete time format is converted into a continuous value from 0 to 24. The format conversion can be performed using the following formula:

[0132]

[0133] Where T hour is the current hour, T min is the current minute, T s is the current second, T c The converted time value.

[0134] Next, the processed data can be subjected to feature extraction and intelligence matching fusion operations.

[0135] Specifically, on the one hand, feature mining is performed based on the alarm data. First, the basic features of the alarm data are extracted through feature mining.

[0136] On the other hand, the time information of the alarm log data is analyzed based on the time series linear regression method. Specifically, it can include longitudinal analysis of the single-day data of the alarm data and horizontal analysis of the historical data.

[0137] Specifically, the longitudinal analysis of single-day data can be directly implemented through least squares regression, and the output regression error E d .

[0138] Specifically, since the same asset, such as host resources, may have multiple alarms every day, the alarm data needs to be processed first before performing horizontal time series analysis based on historical data. The specific implementation steps are as follows:

[0139] S11, count the number of alarms of asset i per day;

[0140] S12, select the date d with the least number of alarms in the time period T of asset i min , take the average value t of the time after continuous processing avg ;

[0141] S13, select the alarm time and mean value t of other dates in the time period T of asset i avg Close to time point t i ;

[0142] S14, for the selected multiple time points t i , perform least squares regression analysis and output the error E T .

[0143] Here, for example, according to the above methods S11 to S14, when the selected time period is 7 days, 7 time points t can be selected. i For least squares regression analysis, output E T .

[0144] Specifically, the results of feature mining and extraction based on alarm data are shown in Table 1 below:

[0145]

[0146]

[0147] Table 1

[0148] Specifically, intelligence matching fusion combines the basic information in the alarm log with the threat intelligence information in the threat intelligence library to perform intelligence analysis and obtain intelligence matching feature information. Here, intelligence matching analysis can be performed based on the control IP and controlled IP in the alarm log data to obtain the intelligence matching feature information shown in Table 2:

[0149]

[0150] Table 2

[0151] Intelligence matching and fusion can not only enrich the scene map of the subsequent GCN model, but also add new alarm dimensions based on the original alarm log to facilitate staff to investigate.

[0152] Next, we can use the warning information obtained from the above analysis to construct a scene graph. Figure 4 As shown, Figure 4 It is a schematic diagram of a scenario graph of an alarm log recognition model provided by an embodiment of the present invention.

[0153] Specifically, the characteristic information of each alarm IP node in the alarm information can be obtained by the above-mentioned identification and analysis operation.

[0154] Specifically, by extracting features from the alarm data, basic feature statistics and time series analysis information are obtained. Basic feature statistics may include: number of alarms per day, number of controlled IPs, number of controlled IPs, number of alarms within time period T, etc. Time series analysis may include: single-day regression error E d , regression error E in time period T t wait.

[0155] Specifically, through the intelligence matching and fusion operation, IP-associated sample information, IP reputation information and IP-associated domain name information are obtained. IP-associated sample information may include: MD5 value, total number of associated samples, associated sample threat type, IP reputation information may include: threat type, family, threat level, attack area, etc.; IP-associated domain name information may include: domain name, number of associated domain names, associated domain name threat type, etc.

[0156] Specifically, Figure 4 As shown, the alarm log recognition model scenario graph also has a structural relationship between alarm IP nodes. The structural relationship between IP1, IP2, IP3, IP4 and IP5 is determined according to the control relationship of the alarm IP.

[0157] according to Figure 4 The scene graph can divide the input of the GCN neural network model into two types of data sets, namely the alarm feature data set and the structural relationship data set between alarms, as shown in Table 3:

[0158]

[0159]

[0160] Table 3

[0161] Next, the alarm recognition model can be trained based on the graph convolutional neural network. Specifically, the input data sets are the processed alarm feature information and the structural relationship between alarms. After feature learning, the label matrix corresponding to the alarm feature information and the adjacency matrix corresponding to the structural relationship between alarms are calculated respectively.

[0162] Specifically, label matrix: Each node, such as alarm IP, has its characteristic information X i , we can use the matrix X (N*D) Indicates. Where N represents the number of nodes, and D represents the number of feature information of each node.

[0163] Specifically, the adjacency matrix refers to the data set of graph structure information and the structural relationship graph between nodes, which can be represented by matrix A.

[0164] Specifically, the node propagation rule trained by the GCN model is as follows:

[0165]

[0166] Let the central node be i, where is the feature expression of node i at layer l, c ij Normalization factor (the inverse of the node degree), N i Neighbors of node i (including itself), R i The type of node i, Represents R i Node transformation weight parameter. This formula can be expressed as an operator of the GCN neural network. When the GCN model training achieves the expected effect, the corresponding parameters are obtained to determine the GCN alarm recognition model. The expected effect can be specifically the recall rate predicted by the model, that is, the recall rate reaches 90%. The GCN neural network structure can be as follows Figure 5 As shown, Figure 5 It is a schematic diagram of a GCN neural network structure provided by an embodiment of the present invention.

[0167] Finally, the trained GCN alarm recognition model is used to identify the alarms to be detected. By obtaining the alarm recognition results, the infected compromised hosts in the alarm log can be determined.

[0168] It is understandable that the alarm identification method and the alarm model training method are not only applicable to the analysis of the alarm log of the zombie worm system, but can also be applied to the alarm log analysis of other types of network security monitoring systems, such as IDS, IPS and WAF.

[0169] In summary, the alarm identification method of the embodiment of the present invention can be based on the alarm feature information of multiple dimensions of the alarm information and the structural relationship between the alarm information, and the alarm feature information of multiple dimensions involves analysis based on the alarm log itself, combined with threat intelligence, to accurately identify alarms of network security issues such as host infection, etc., and can achieve more comprehensive and accurate identification and analysis of alarms in the network environment, which can improve the accuracy of alarm information identification.

[0170] In addition, in the embodiment of the present invention, when analyzing alarm identification, it is not only based on the analysis of the alarm log itself, but also combined with threat intelligence to enrich the dimension of alarm information, so as to improve the accuracy of alarm log identification.

[0171] Moreover, through the above-mentioned alarm identification method, most of the false alarms can be screened out, and the correct alarms can be identified more accurately, so as to accurately locate the infected and compromised hosts and other network asset devices, thereby reducing the number of abnormal asset devices that need to be troubleshooted, and reducing the difficulty and workload of operation and maintenance personnel in troubleshooting problem equipment.

[0172] Based on the training method of the alarm recognition model provided in the above embodiment, accordingly, the present application also provides a specific implementation of the training device of the alarm recognition model. Please refer to the following embodiment.

[0173] In one embodiment of the present invention, Figure 6 As shown, Figure 6 : is a structural diagram of a training device for an alarm recognition model provided by another embodiment of the present invention, wherein the training device for an alarm recognition model specifically comprises:

[0174] An acquisition module 601 is used to acquire multiple alarm data samples;

[0175] A preprocessing module 602 is used to preprocess the plurality of sample alarm data to obtain training samples, wherein the training samples include a plurality of sample alarm information and an alarm category of each sample alarm information, wherein the sample alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0176] A construction module 603 is used to construct a sample alarm information label feature matrix and an adjacency matrix corresponding to the association relationship between the multiple sample alarm information according to the multiple sample alarm information;

[0177] The training module 604 is used to use the label feature matrix of the sample alarm information and the adjacency matrix as inputs of the alarm recognition model, and the alarm category as output of the alarm recognition model, and iteratively train the alarm recognition model to be trained to obtain a target alarm recognition model.

[0178] In summary, in an embodiment of the present invention, the training device of the alarm recognition model can be used to implement the method of alarm recognition model training, and can use sample alarm information and the correlation between sample alarm information as training samples to train a target alarm recognition model. The alarm information used as training samples not only has basic alarm feature information, but also intelligence matching feature information. It can be seen that, because the alarm feature information of multiple dimensions of the alarm and the structural relationship between the alarms are considered when training the model, the corresponding feature information of the alarm with host infection problems is learned more accurately, so that the model can perform a more comprehensive and accurate identification and analysis of the alarms in the network environment, and then identify the alarms that may actually have host infection problems. In this way, the accuracy of alarm information recognition can be improved, and then the asset equipment with abnormalities can be accurately located, reducing the difficulty and workload of operation and maintenance personnel in troubleshooting problem equipment.

[0179] Based on the alarm identification method provided in the above embodiment, the present application also provides a specific implementation of the alarm identification device. Please refer to the following embodiment.

[0180] In one embodiment of the present invention, Figure 7 As shown, Figure 7 : is a structural diagram of an alarm identification device provided by another embodiment of the present invention, and the alarm identification device specifically includes:

[0181] An acquisition module 701 is used to acquire multiple alarm data;

[0182] The preprocessing module 702 is used to preprocess each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data, wherein the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching;

[0183] A construction module 703 is used to construct a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information according to the multiple sample alarm information;

[0184] The identification module 704 is used to input the label feature matrix and the adjacency matrix into the target alarm recognition model to obtain the alarm category corresponding to the alarm data; the target alarm recognition model is a model trained using the alarm recognition model training method described in the above embodiment.

[0185] Optionally, in some embodiments, the preprocessing module 702 includes:

[0186] The extraction unit is used to extract features from each of the alarm data to obtain first alarm feature information corresponding to each of the alarm data.

[0187] The matching unit is used to match each of the alarm data with the information of the preset threat intelligence library respectively to obtain the second alarm feature information corresponding to each of the alarm data.

[0188] Optionally, the second alarm characteristic information includes any one or more of the following: Internet Protocol IP reputation information, IP associated sample information and IP associated domain name information.

[0189] Optionally, in some embodiments, the extraction unit includes:

[0190] An extraction subunit, used for performing feature extraction on basic information of each of the alarm data to obtain first feature information of the first alarm feature information;

[0191] The calculation subunit is used to calculate the time characteristic information of each of the alarm data using a time series analysis algorithm to obtain the second characteristic information of the first alarm characteristic information.

[0192] Optionally, in some embodiments, the calculation subunit is further used to calculate the first time feature information of each of the alarm data using a time series analysis algorithm to obtain second feature information corresponding to the first time feature information; calculate the second time feature information of each of the alarm data using a time series analysis algorithm to obtain second feature information corresponding to the second time feature information; combine the second feature information corresponding to the first time feature information and the second feature information corresponding to the second time feature information to obtain the second feature information; wherein the time feature information includes the first time feature information and the second time feature information, the first time feature information is the time feature information of a single day, and the second time feature information is the time feature information within a preset time period;

[0193] Optionally, in some embodiments, the calculation subunit is further used to calculate the first time characteristic information of each alarm data using a least square method to obtain second characteristic information corresponding to the first time characteristic information.

[0194] Optionally, in some embodiments, the calculation subunit is also used to select the first single-day characteristic information corresponding to the minimum value of the number of single-day alarms in the alarm data within a preset time period; perform mean calculation on the first single-day characteristic information to obtain a time mean; match the second single-day characteristic information within the preset time period with the time mean to obtain a time point that meets a preset matching condition, the second single-day characteristic information being the single-day characteristic information corresponding to each single day other than the single day corresponding to the first single-day characteristic information; and use the least squares method to calculate the time point that meets the preset condition to obtain the second characteristic information corresponding to the second time characteristic information.

[0195] Optionally, in some embodiments, the construction module 703 is also used to construct a label feature matrix corresponding to each of the alarm information based on the first alarm feature information and the second alarm feature information of the multiple alarm information; construct corresponding association relationship information between at least two alarm information based on the IP control relationship information of the multiple alarm information; perform feature learning based on the association relationship information to determine the corresponding adjacency matrix.

[0196] Optionally, in some embodiments, the alarm recognition model in the alarm recognition device may be a graph convolutional neural network model.

[0197] In summary, in the embodiment of the present invention, the alarm identification device can be used to implement the alarm identification method in the above embodiment, and can accurately identify alarms of network security problems such as host infection based on the alarm feature information of multiple dimensions of the alarm information and the structural relationship between the alarm information, and can realize a more comprehensive and accurate identification and analysis of alarms in the network environment, and can improve the accuracy of alarm information identification. In addition, through the above alarm identification method, most false alarms can be screened out, and correct alarms can be identified more accurately, so as to accurately locate infected and compromised hosts and other network asset devices, thereby reducing the number of asset devices that need to be troubleshooted, and reducing the difficulty and workload of operation and maintenance personnel in troubleshooting problem equipment.

[0198] Based on the alarm recognition model training method and alarm recognition method provided in the above embodiments, the present application also provides a specific hardware structure description of the alarm recognition model training device and the alarm recognition device. Please refer to the following embodiments.

[0199] Figure 8 A schematic diagram of the hardware structure of a training device for an alarm recognition model provided by an embodiment of the present invention is shown.

[0200] The training device of the alarm recognition model may include a processor 801 and a memory 802 storing computer program instructions.

[0201] Specifically, the processor 801 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present invention.

[0202] The memory 802 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 802 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 802 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 802 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 802 is a non-volatile solid-state memory. In a specific embodiment, the memory 802 includes a read-only memory (ROM). In appropriate cases, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM) or a flash memory or a combination of two or more of these.

[0203] The processor 801 implements any one of the alarm recognition model training methods in the above embodiments by reading and executing computer program instructions stored in the memory 802 .

[0204] In one example, the training device of the alarm recognition model may further include a communication interface 803 and a bus 810. Figure 8 As shown, the processor 801, the memory 802, and the communication interface 803 are connected via a bus 810 and communicate with each other.

[0205] The communication interface 803 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiment of the present invention.

[0206] Bus 810 includes hardware, software or both, and the parts of the training equipment of the alarm identification model are coupled to each other. For example, but not limitation, the bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 810 may include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0207] The training device of the alarm recognition model can execute the alarm recognition method in the embodiment of the present invention, thereby realizing the combination of Figure 1 Describes the training method of the alarm recognition model.

[0208] Fig. 9 A schematic diagram of the hardware structure of the alarm identification device provided by an embodiment of the present invention is shown.

[0209] The device for alarm identification may include a processor 901 and a memory 902 storing computer program instructions.

[0210] Specifically, the processor 901 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present invention.

[0211] The memory 902 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 902 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 902 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 902 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 902 is a non-volatile solid-state memory. In a specific embodiment, the memory 902 includes a read-only memory (ROM). In appropriate cases, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM) or a flash memory or a combination of two or more of these.

[0212] The processor 901 implements any one of the alarm identification methods in the above embodiments by reading and executing computer program instructions stored in the memory 902 .

[0213] In one example, the alarm identification device may further include a communication interface 903 and a bus 910. Fig. 9 As shown, the processor 901, the memory 902, and the communication interface 903 are connected via a bus 910 and communicate with each other.

[0214] The communication interface 903 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiment of the present invention.

[0215] Bus 910 includes hardware, software or both, and the components of the equipment of alarm identification are coupled to each other. For example, but not limitation, the bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 310 may include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0216] The alarm identification device can execute the alarm identification method in the embodiment of the present invention, thereby realizing the combination of Figure 2 Describes the method of alarm identification.

[0217] In addition, in combination with the training method of the alarm recognition model in the above embodiment, the embodiment of the present invention can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the training methods of the alarm recognition model in the above embodiment is implemented.

[0218] In addition, in combination with the alarm identification method in the above embodiment, the embodiment of the present invention can provide a computer storage medium to implement. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the alarm identification methods in the above embodiment is implemented.

[0219] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0220] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0221] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0222] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.

Claims

1. A training method for an alarm recognition model, It is characterized in that include: Acquire multiple alarm data samples, wherein the alarm data samples are obtained through a network security monitoring system; Preprocessing each of the alarm data samples respectively to obtain a training sample corresponding to each of the alarm data samples, wherein the training sample includes a plurality of sample alarm information and an alarm category of each sample alarm information, wherein the sample alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching; Constructing, according to the plurality of sample alarm information, a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information; The label feature matrix of the sample alarm information and the adjacency matrix are used as inputs of the alarm recognition model, the alarm category is used as output of the alarm recognition model, and the alarm recognition model is iteratively trained to obtain a target alarm recognition model; The step of constructing a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information according to the multiple sample alarm information includes: Performing feature learning based on the first alarm feature information and the second alarm feature information of the plurality of sample alarm information to construct a label feature matrix corresponding to each of the sample alarm information; According to the IP structure relationship diagram of the control IP and the controlled IP in the plurality of the sample alarm information, obtain the IP control relationship information of the plurality of the sample alarm information; Perform feature learning according to the IP control relationship information to construct corresponding association relationship information between at least two sample alarm information; Feature learning is performed based on the association relationship information to determine the corresponding adjacency matrix.

2. A method for alarm identification, It is characterized in that include: Acquire multiple alarm data, wherein the alarm data is obtained through a network security monitoring system; Preprocessing each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data, wherein the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching; According to the plurality of alarm information, construct a label feature matrix corresponding to each alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information; Inputting the label feature matrix and the adjacency matrix into a target alarm recognition model to obtain an alarm category corresponding to the alarm data; The target alarm recognition model is a model trained using the alarm recognition model training method described in claim 1; The step of constructing a label feature matrix corresponding to each of the warning information and an adjacency matrix corresponding to the association relationship information between at least two warning information according to the plurality of warning information includes: Performing feature learning based on the first alarm feature information and the second alarm feature information of the plurality of alarm information to construct a label feature matrix corresponding to each of the alarm information; According to the IP structure relationship diagram of the control IP and the controlled IP in the plurality of alarm information, obtain the IP control relationship information of the plurality of alarm information; Perform feature learning according to the IP control relationship information to construct corresponding association relationship information between at least two alarm information; Feature learning is performed based on the association relationship information to determine the corresponding adjacency matrix.

3. The method according to claim 2, It is characterized in that The preprocessing of each of the alarm data to obtain the alarm information corresponding to each of the alarm data comprises: Extracting features from each of the alarm data respectively to obtain first alarm feature information corresponding to each of the alarm data; Each of the alarm data is matched with the information of the preset threat intelligence library to obtain the second alarm feature information corresponding to each of the alarm data.

4. The method according to claim 3, It is characterized in that The extracting features of each of the alarm data to obtain first alarm feature information corresponding to each of the alarm data comprises: Extracting features from basic information of each of the alarm data respectively to obtain first feature information of the first alarm feature information; The time characteristic information of each of the alarm data is calculated using a time series analysis algorithm to obtain second characteristic information of the first alarm characteristic information.

5. The method according to claim 3, It is characterized in that The second alarm characteristic information includes any one or more of the following: Internet Protocol (IP) reputation information, IP-associated sample information, and IP-associated domain name information.

6. The method according to claim 4, It is characterized in that The time characteristic information includes first time characteristic information and second time characteristic information, the first time characteristic information is single-day time characteristic information, and the second time characteristic information is time characteristic information within a preset time period; The calculating the time characteristic information of each of the alarm data by using a time series analysis algorithm to obtain the second characteristic information of the first alarm characteristic information includes: Using a time series analysis algorithm to calculate the first time characteristic information of each of the alarm data, respectively, to obtain second characteristic information corresponding to the first time characteristic information; Using a time series analysis algorithm to calculate the second time characteristic information of each of the alarm data, respectively, to obtain second characteristic information corresponding to the second time characteristic information; The second characteristic information corresponding to the first time characteristic information and the second characteristic information corresponding to the second time characteristic information are combined to obtain the second characteristic information of the first alarm characteristic information.

7. The method according to claim 6, It is characterized in that The calculating the first time characteristic information of each of the alarm data by using a time series analysis algorithm to obtain the second characteristic information corresponding to the first time characteristic information includes: The first time characteristic information of each of the alarm data is calculated using the least square method to obtain second characteristic information corresponding to the first time characteristic information.

8. The method according to claim 6, It is characterized in that The calculating the second time characteristic information of each of the alarm data by using a time series analysis algorithm to obtain the second characteristic information corresponding to the second time characteristic information includes: Selecting the first single-day characteristic information corresponding to the minimum value of the number of single-day alarms in the alarm data within a preset time period; Calculate the mean of the first single-day characteristic information to obtain a time mean; Matching the second daily characteristic information within the preset time period with the time mean to obtain a time point that meets a preset matching condition, wherein the second daily characteristic information is the daily characteristic information corresponding to each day other than the day corresponding to the first daily characteristic information; The time points satisfying the preset condition are calculated using the least square method to obtain second feature information corresponding to the second time feature information.

9. The method according to any one of claims 2 to 8, It is characterized in that The alarm recognition model includes a graph convolutional neural network model.

10. A training device for an alarm recognition model, It is characterized in that The device comprises: An acquisition module, used for acquiring a plurality of alarm data samples, wherein the alarm data samples are obtained through a network security monitoring system; A preprocessing module, used to preprocess the plurality of alarm data to obtain training samples, wherein the training samples include a plurality of sample alarm information and an alarm category of each sample alarm information, wherein the sample alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching; A construction module, configured to construct, based on the plurality of sample alarm information, a label feature matrix of the sample alarm information and an adjacency matrix corresponding to the association relationship between at least two of the sample alarm information; A training module, used to use the label feature matrix of the sample alarm information and the adjacency matrix as inputs of an alarm recognition model, and the alarm category as output of the alarm recognition model, and iteratively train the alarm recognition model to obtain a target alarm recognition model; Among them, the construction module is also used to perform feature learning based on the first alarm feature information and the second alarm feature information of multiple sample alarm information, and construct a label feature matrix corresponding to each sample alarm information; obtain the IP control relationship information of multiple sample alarm information based on the IP structure relationship diagram of the control IP and the controlled IP in the multiple sample alarm information; perform feature learning based on the IP control relationship information, and construct the corresponding association relationship information between at least two sample alarm information; perform feature learning based on the association relationship information to determine the corresponding adjacency matrix.

11. A device for alarm identification, It is characterized in that The device comprises: An acquisition module, used for acquiring a plurality of alarm data, wherein the alarm data is obtained through a network security monitoring system; A preprocessing module, used to preprocess each of the alarm data respectively to obtain alarm information corresponding to each of the alarm data, wherein the alarm information includes first alarm feature information and second alarm feature information, wherein the first alarm feature information is basic alarm feature information, and the second alarm feature information is intelligence matching feature information obtained after intelligence matching; A construction module, used to construct, based on the plurality of alarm information, a label feature matrix corresponding to each of the alarm information and an adjacency matrix corresponding to the association relationship information between at least two alarm information; An identification module, used for inputting the label feature matrix and the adjacency matrix into a target alarm identification model to obtain an alarm category corresponding to the alarm data; the target alarm identification model is a model trained using the alarm identification model training method described in claim 1; Among them, the construction module is also used to perform feature learning based on the first alarm feature information and the second alarm feature information of multiple alarm information to construct a label feature matrix corresponding to each alarm information; obtain the IP control relationship information of multiple alarm information based on the IP structure relationship diagram of the control IP and the controlled IP in the multiple alarm information; perform feature learning based on the IP control relationship information to construct the corresponding association relationship information between at least two alarm information; perform feature learning based on the association relationship information to determine the corresponding adjacency matrix.

12. A training device for an alarm recognition model, It is characterized in that The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the training method of the alarm recognition model as claimed in claim 1 is implemented.

13. A device for alarm identification, It is characterized in that The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for alarm identification as described in any one of claims 2 to 9 is implemented.

14. A computer storage medium, It is characterized in that The computer storage medium stores computer program instructions, which, when executed by a processor, implement the training method for the alarm recognition model as claimed in claim 1 and / or the method for alarm recognition as claimed in any one of claims 2 to 9.