Security alarm intelligent troubleshooting method, network equipment and storage medium

By obtaining and marking security event data on the security management platform, and optimizing K-values ​​using classification models and unsupervised algorithms, the huge problem of manpower and material resources required for marking and classification in the existing technology is solved, and efficient identification and accuracy of security events are achieved. Predictions reduce the workload and stress of administrators.

CN115622750BActive Publication Date: 2024-12-03NEW H3C SECURITY TECH CO LTD
View PDF 2 Cites -1 Cited by

Patent Information

Application Number
CN202211178870.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-27
Publication Date
2024-12-03
Estimated Expiration
2042-09-27

AI Technical Summary

Technical Problem

When the existing security management platform handles a large number of security incidents, the manpower and material resources are huge, which leads to huge pressure on security administrators in investigation and disposal, and it is difficult to effectively identify alarm events.

Method used

By obtaining security event data, extracting K number of security events and marking them, training using classification model, calculating the alarm probability of each event, and iteratively adjusting the K value to reduce the misreport rate, combining unsupervised algorithms and feature exceptions The detection algorithm is sorted and marked, and the K value is dynamically adjusted to optimize the classification model prediction.

Benefits of technology

It greatly reduces the number of staff markings on safety events, improves the prediction accuracy and rapid convergence capabilities of the classification model, and reduces the work pressure of safety administrators.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115622750B_ABST
    Figure CN115622750B_ABST
Patent Text Reader

Abstract

The present application provides a method for intelligent troubleshooting of security alarms, network equipment and storage medium, wherein the method includes obtaining security event data; extracting K number of security events and marking alarm events; using a classification model to perform classification training on the marked K security events; obtaining the probability that each security event is an alarm event; calculating a new K value based on the probability of an alarm occurring for each of the K acquired security events; iteratively processing the original K value through the calculated new K value; calculating the false positive and false negative rate of security events; when the false positive and false negative rate is less than a preset value, the classification model converges; when the false positive and false negative rate is greater than or equal to a preset value, the classification model continues to be trained by iterating the K value. In the embodiment of the present application, the staff only needs to mark the K number of security event data sampled, which greatly reduces the workload of the staff.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of communication equipment, and in particular to a method for intelligently troubleshooting security alarms, network equipment, and storage media. Background Art

[0002] With the rapid development of the Internet industry and the rapid progress of application innovation, the types, evolution and complexity of network traffic are increasing with the continuous emergence of new network applications and network protocols. At the same time, network operators and network regulatory departments have stronger demands for understanding the composition of network traffic, implementing differentiated network services, and purifying the network environment. Among them, the security alarm of the security management platform is the basis of the security management platform analysis. Most devices report alarm logs based on rules and threshold settings, resulting in thousands of security alarm logs being reported every day, which brings huge work pressure to security administrators in troubleshooting and handling.

[0003] The attack judgment and classification method based on machine learning requires a large number of samples with real labels to train the classifier, but obtaining a large number of real labels requires a lot of manpower and material resources. Therefore, in actual information security application scenarios, labeling a large number of security events reported on the security management platform is a large workload and cannot be completed. The labeling here is whether these security events are alarm events. For example, if it is an alarm event, it is marked 1, and if it is not an alarm event, it is marked 0. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present application provides a security alarm intelligent troubleshooting method, network equipment and storage medium.

[0005] According to a first aspect of an embodiment of the present application, a security alarm intelligent troubleshooting method includes:

[0006] Obtain security incident data;

[0007] Extract K security events and mark them as alarm events;

[0008] Use the classification model to perform classification training on the marked K security events;

[0009] Obtain the probability that each security event is an alarm event;

[0010] Calculate a new K value based on the probability of each of the K acquired security events generating an alarm;

[0011] Iterate the original K value through the calculated new K value;

[0012] Calculate the false positive and false negative rates of security incidents;

[0013] When the false positive and false negative rate is less than the preset value, the classification model converges;

[0014] When the false alarm rate is greater than or equal to the preset value, the classification model continues to be trained by iterating the K value.

[0015] Preferably, K number of security events are extracted, including:

[0016] When it is judged as the initial iteration, an unsupervised algorithm is used to sort the security events;

[0017] Get the top K security events and mark whether they are warning events.

[0018] When it is determined that it is not the initial stage of the iteration, K security events after the iteration are obtained;

[0019] Use the K number of security events after iteration to perform classification model prediction and obtain the predicted probability of each security event;

[0020] The K number of security events after iteration are used to calculate and sort based on an unsupervised algorithm;

[0021] Based on the predicted probability of security events, the predicted probability is sorted and the entropy value of the predicted probability is calculated and sorted;

[0022] Extract the top K / 3 security events sorted by the iterative unsupervised algorithm;

[0023] Extract the top K / 3 security events with the highest probability after the prediction probability is sorted;

[0024] Extract K / 3 security events ranked top by entropy value;

[0025] The first K / 3 security events sorted by the iterative unsupervised algorithm, the first K / 3 security events with high probabilities sorted by predicted probabilities, and the first K / 3 security events with high entropy values ​​are combined into K security events.

[0026] Furthermore, unsupervised algorithm ranking can be based on feature anomaly detection algorithms or use graph ranking algorithms.

[0027] Further, the formula of the graph sorting algorithm is:

[0028]

[0029]

[0030] Among them, a represents the weight value, h represents the node, and the sorting is based on the h result.

[0031] Furthermore, the feature anomaly detection algorithm ranks events based on abnormal flag scores.

[0032] Preferably, the new K value is obtained by calculation, including:

[0033] The corresponding entropy value is calculated based on the probability of the event marked as an alarm event and the event not marked as an alarm event in the predicted probability;

[0034] Sorting the entropy values ​​of the K security events;

[0035] Assuming that there are m security events with a probability of 0.5, then K / 3+K6≈m;

[0036] Calculation shows that K≈2m.

[0037] Furthermore, the corresponding entropy value formula is calculated based on the probability of events marked as alarm events and not alarm events in the predicted probability:

[0038]

[0039] Among them, p(x i ) represents the probability of an alarm event occurring, n=1,2.

[0040] Preferably, calculating the false positive or false negative rate of security events includes:

[0041] Obtain a new batch of security incident data;

[0042] Input a new batch of security events into the classification model in a K-value iteration manner to predict alarm events;

[0043] Mark a new batch of security events as alarm events;

[0044] The predicted alarm events are compared with the marked alarm events, and the missed alarm and false alarm rates of the predicted alarm events are calculated.

[0045] A second aspect of an embodiment of the present application provides a network device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-mentioned intelligent security alarm troubleshooting method when executing the program.

[0046] A third aspect of an embodiment of the present application provides a theoretical storage medium on which computer program instructions are stored. When the program instructions are executed by a processor, they are used to implement the above-mentioned intelligent troubleshooting method for security alarms.

[0047] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0048] In the embodiment of the present application, the staff only needs to mark K number of security event data that are randomly inspected, which greatly reduces the workload of the staff.

[0049] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments consistent with the application and, together with the application, serve to explain the principles of the application.

[0051] Figure 1 It is a schematic diagram of the process flow of an embodiment of the present application;

[0052] Figure 2 Schematic diagram of a method for obtaining K number of security events in this application;

[0053] Figure 3 It is a schematic diagram of the graph sorting algorithm of this application;

[0054] Figure 4 It is a schematic diagram of the method of calculating and obtaining a new K value in this application;

[0055] Figure 5 This is a schematic diagram of the false positive and false negative rates of security incidents calculated in this application. DETAILED DESCRIPTION

[0056] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0057] To solve the problems existing in the background technology, the present application embodiment provides a security alarm intelligent troubleshooting method, such as Figure 1 As shown, including:

[0058] Obtain security incident data;

[0059] Extract K security events and mark them as alarm events;

[0060] Use the classification model to perform classification training on the marked K security events;

[0061] Obtain the probability that each security event is an alarm event;

[0062] Calculate a new K value based on the probability of each of the K acquired security events generating an alarm;

[0063] Iterate the original K value through the calculated new K value;

[0064] Calculate the false positive and false negative rates of security incidents;

[0065] When the false positive and false negative rate is less than the preset value, the classification model converges;

[0066] When the false alarm rate is greater than or equal to the preset value, it is necessary to continue training the classification model by iterating the K value.

[0067] In the embodiment of the present application, since only K number of security events are extracted each time for prediction and security event marking through the model, the number of security events that need to be marked manually by the staff is greatly reduced. And since the K value is a dynamically adjusted data, the classification model prediction is made according to the adjusted K value, and the prediction data is more accurate and can achieve rapid convergence.

[0068] In this embodiment, it is necessary to explain how to extract K number of security events, which specifically includes the following two cases: Figure 2 As shown:

[0069] Case 1: When it is determined to be the initial iteration, because there is no new K value of the iteration that has been calculated at this time, an unsupervised algorithm is used to sort the security events; the top K security events are obtained and marked as alarm events.

[0070] Case 2: When it is determined that it is not the initial stage of the iteration, obtain K security events after the iteration; (1) Use the K security events after the iteration to perform classification model prediction and obtain the predicted probability of each security event; (2) Use the K security events after the iteration to perform calculations based on the unsupervised algorithm and sort them; (3) Perform predicted probability sorting and entropy value calculation and entropy value sorting based on the predicted probability of the security event. Extract the top K / 3 security events sorted by the unsupervised algorithm after the iteration; extract the top K / 3 security events with high probability after the predicted probability sorting; extract the top K / 3 security events with high entropy value sorting; combine the top K / 3 security events sorted by the unsupervised algorithm after the iteration, the top K / 3 security events with high probability after the predicted probability sorting and the top K / 3 security events with high entropy value sorting into K security events. It should be noted here that the top ranking refers to events with a high probability of alarm events.

[0071] The unsupervised algorithm here can be obtained based on a feature anomaly detection algorithm or a graph sorting algorithm.

[0072] The graph sorting algorithm is specifically implemented through the following formula: Among them, a represents the weight value, h represents the node, and the sorting is based on the h result. As shown in Figure 3, A(1)=H(2)+H(3)+H(4), H(1)=A(5)+A(6)+A(7). Each node has two attribute values, a value and h value. From a physical sense: when a host launches multiple attack events, it means that the host is likely to have been compromised (poisoned), which is represented in the figure as a node pointing to multiple nodes. Here, the sorting based on the h value of each node is used as the final sorting result.

[0073] The feature anomaly detection algorithm ranks events based on their abnormal flag scores.

[0074] In the embodiment of the present application, a new K value is calculated and obtained, such as Figure 4 As shown, including:

[0075] The corresponding entropy value is calculated based on the probability of the event marked as an alarm event and the event not marked as an alarm event in the predicted probability;

[0076] Sort the entropy values ​​of K security events;

[0077] The number of security events with a probability of 0.5 is preset. This is because through experiments, it is found that the entropy value of uncertain security events is around the predicted probability of 0.5, which is also consistent with its actual physical characteristics. Then K / 3+K6≈m; therefore, K≈2m is calculated. The prediction by entropy ranking mainly utilizes the stability characteristics of entropy.

[0078] The corresponding entropy value formula is calculated based on the probability of events marked as alarm events and non-alarm events in the predicted probability:

[0079]

[0080] Among them, p(x i ) represents the probability of an alarm event, n = 1, 2. The corresponding entropy value is calculated based on the probability of marking category 0 and category 1 in the predicted probability. In the embodiment of the present application, mark 0 can be a non-alarm event or a normal event, and mark 1 can be an alarm event. Of course, it can also be set in reverse.

[0081] The false positive and false negative rate of security events calculated in the embodiment of the present application is as follows: Figure 5 As shown, including:

[0082] Obtain a new batch of security incident data;

[0083] Input a new batch of security events into the classification model in a K-value iteration manner to predict alarm events;

[0084] Mark a new batch of security events as alarm events;

[0085] The predicted alarm events are compared with the marked alarm events, and the missed alarm and false alarm rates of the predicted alarm events are calculated.

[0086] The second aspect of the embodiment of the present application further provides a network device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-mentioned intelligent security alarm troubleshooting method when executing the program.

[0087] A third aspect of an embodiment of the present application provides a storage medium on which computer program instructions are stored. When the program instructions are executed by a processor, they are used to implement the above-mentioned intelligent troubleshooting method for security alarms.

[0088] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

[0089] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A security alarm intelligent troubleshooting method, characterized in that: include: Obtain security incident data; Extract K security events and mark them as alarm events; Use the classification model to perform classification training on the marked K security events; Obtain the probability that each security event is an alarm event; Calculate the new K value based on the probability of each of the K acquired security events generating an alarm; Iterate the original K value through the calculated new K value; Calculate the false reporting and omission rates of security incidents; When the misreporting and omission rate is less than the preset value, the classification model converges; When the false positive and false negative rate is greater than or equal to the preset value, the classification model is trained by iterating the K value.

2. The intelligent security alarm troubleshooting method according to claim 1, characterized in that: Extract K security events, including: When it is judged to be the initial stage of iteration, an unsupervised algorithm is used to sort security events; Get the top K security events and mark them as alarm events. If it is determined that it is not the initial stage of the iteration, obtain K security events after the iteration; Use the K number of security events after iteration to perform classification model prediction and obtain the predicted probability of each security event; The K number of security events after iteration are calculated and sorted based on the unsupervised algorithm; the predicted probability of the security events is sorted and the entropy value of the predicted probability is calculated and sorted; Extract the top K / 3 security events sorted by the iterative unsupervised algorithm; Extract the top K / 3 security events with the highest probability after sorting the predicted probabilities; Extract the K / 3 security events ranked by the highest entropy value; The top K / 3 security events sorted by the iterative unsupervised algorithm, the top K / 3 security events with high probabilities sorted by predicted probabilities, and the top K / 3 security events with high entropy values ​​are combined into K security events.

3. The intelligent security alarm troubleshooting method according to claim 2, characterized in that: The unsupervised algorithm ranking is based on a feature anomaly detection algorithm or uses a graph ranking algorithm.

4. The intelligent security alarm troubleshooting method according to claim 3, characterized in that: The formula of the graph sorting algorithm is: Among them, a represents the weight value, h represents the node, and the sorting is based on the h result.

5. The intelligent security alarm troubleshooting method according to claim 3, characterized in that: The feature anomaly detection algorithm ranks events based on their flag scores.

6. The intelligent troubleshooting method for security alarms according to claim 1, characterized in that: Obtain a new K value through calculation, including: Calculate the corresponding entropy value based on the probability of the event marked as an alarm event and the event not being an alarm event in the predicted probability; Sorting the entropy values ​​of the K security events; Assuming that there are m security events with a probability of 0.5, then K / 3+K6≈m; Calculation shows that K≈2m.

7. The intelligent security alarm troubleshooting method according to claim 6, characterized in that: The corresponding entropy value formula is calculated based on the probability of events marked as alarm events and non-alarm events in the predicted probability: Among them, p(x i ) represents the probability of an alarm event occurring, n=1,2.

8. The intelligent security alarm troubleshooting method according to claim 1, characterized in that: The calculation of the false reporting and missing reporting rate of security incidents includes: Obtain a new batch of security event data; Input a new batch of security events into the classification model in a K-value iteration manner to predict alarm events; Mark a new batch of security events as alarm events; The predicted alarm events are compared with the marked alarm events, and the missed alarm and false alarm rates of the predicted alarm events are calculated.

9. A network device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it executes the security alarm intelligent troubleshooting method described in any one of claims 1-8.

10. A storage medium having computer program instructions stored thereon, characterized in that When the program instructions are executed by the processor, they are used to implement the security alarm intelligent troubleshooting method described in any one of claims 1 to 8.

Citation Information

Patent Citations

  • Enterprise security operation center (SOC) Bot

    CN106850293A

  • A security event identification reporting system and method

    CN109409266A