Safety automatic arrangement method based on AI

Through the AI-based security automation orchestration method, using Euclidean distance and local density to detect abnormalities, randomly segment data and perform time series analysis, optimize model parameters, and solve the problem of untimely response to unknown threats in the existing technology, and achieve the latest update High threat identification accuracy and response speed to form a dynamic security policy system.

CN119051930APending Publication Date: 2024-11-29GUIZHOU QIANYUN NETWORK INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411128539.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Existing automated response technologies are difficult to accurately identify and quickly respond to unknown or rapidly changing threats in highly dynamic and complex network environments, especially for zero-day attacks and persistent threats, resulting in extended security vulnerabilities and threat spreading speed. Accelerate and bring greater losses to the organization.

Method used

Using AI-based security automation orchestration method, abnormal detection is performed by calculating the Euclidean distance and local density between data points, randomly segmenting the data and building an isolated tree structure, generating anomaly scores, performing time series analysis and model parameter optimization, and simulated security Dangerous development trends, formulate response strategies, and improve threat identification accuracy and response speed.

Benefits of technology

Enhanced the resolution ability of network behavior data, improve the accuracy of potential threat identification, reduce false alarm rate, enhance the flexibility and speed of the response mechanism, form a security strategy system that is dynamically adaptable and continuously optimized, and improve preventive measures. Prospective and targeted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119051930A_ABST
    Figure CN119051930A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of automatic response, in particular to a safe automatic arrangement method based on AI, which comprises the following steps: on the basis of network behavior data, calculating Euclidean distance and local density between data points by using AI, and screening data through distance and density anomaly detection to obtain a network behavior analysis result. According to the method, data are randomly segmented and an isolated tree is constructed, so that the isolation of abnormal points is more effective, the false alarm rate is reduced, the flexibility and response speed of a response mechanism are enhanced, and potential risks are predicted and analyzed by comprehensively applying time sequence analysis and support vector machine model adjustment; according to the method, coping strategies can be foreseen and prepared earlier on the data level, the perspectiveness and pertinence of prevention measures are improved, qualitative leap is brought to safety protection, and by refining data analysis and processing details, safety response is not limited to post-processing, and a set of dynamically-adapted and continuously-optimized safety strategy system is formed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of automated response technology, and in particular to an AI-based security automated orchestration method. Background Art

[0002] The field of automated response technology aims to optimize the response mechanism of automated network and information security. By automatically executing security policies and processes, it can quickly identify, evaluate and respond to potential security threats or intrusions. Core functions include real-time threat detection, automated threat disposal, and automated management of incident response processes. By reducing the need for human intervention and increasing response speed, automated response technology helps organizations reduce security risks and improve overall security protection capabilities.

[0003] Among them, the AI-based security automation orchestration method refers to the method of using artificial intelligence technology to enhance and optimize the security response process, using AI to analyze batches of data, identify potential threat patterns, and respond quickly and effectively. Its uses include achieving efficient intrusion detection, automated threat analysis, and the execution of response strategies, improving the efficiency and effectiveness of security operations, and is widely used in security operations centers (SOCs), enterprise network security, and other fields to cope with increasingly complex network security challenges.

[0004] Although the existing field of automated response technology has achieved a certain degree of automation and information security management, it still lacks the needs for accurate identification and rapid response in a highly dynamic and complex network environment. Existing technologies rely on preset security rules and response processes, which leads to untimely or inaccurate responses when facing unknown or rapidly changing threats. For example, for zero-day attacks and highly complex persistent threats (APTs), existing technologies cannot effectively identify or rely too much on manual intervention, missing the best time to respond. When processing batch data, the existing technology has low data processing efficiency due to the limitations of algorithms and processing logic, and cannot update security policies in real time, affecting the real-time and effectiveness of the entire security protection. The consequence in actual operation is that the existence of security vulnerabilities is prolonged, the spread of threats is accelerated, and a larger range of security incidents are triggered, causing more serious losses to the organization. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a security automation orchestration method based on AI.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solution: a security automation orchestration method based on AI, comprising the following steps:

[0007] S1: Based on network behavior data, AI is used to calculate the Euclidean distance and local density between data points, and the data is filtered through distance and density anomaly detection to obtain network behavior analysis results;

[0008] S2: Based on the network behavior analysis results, data features are randomly selected and randomly divided, and abnormal points are isolated by constructing an isolated tree structure to isolate the data and generate anomaly scores;

[0009] S3: using the anomaly score, comparing the data scores by setting a standard threshold, marking the data points whose scores deviate from the set threshold as anomalies, classifying and processing the marked data, and obtaining potential danger marks;

[0010] S4: Based on the potential danger markers, AI is used to perform time series analysis, analyze the fit between differentiated data points, evaluate the connection between potential threats, perform security automation orchestration, and generate behavior association pattern data;

[0011] S5: Using the behavior association pattern data, adjusting the support vector machine model parameters item by item, optimizing the parameters for the identified abnormal pattern, updating the model to optimize the sensitivity of abnormality detection, and obtaining the model parameter update result;

[0012] S6: Based on the model parameter update results, simulate the development trend of safety hazards, perform predictive analysis on hazards, perform safety automation orchestration, and formulate response strategies to obtain hazard solutions.

[0013] As a further solution of the present invention, the network behavior analysis results include distance metrics of abnormal behaviors, density distribution of data points and abnormal degree grading of data points, the abnormality score includes the number of segmentation times required for isolation, the abnormal probability of the data point and the position in the entire data set, the potential danger mark includes the score of the data point, the classified abnormal state and a list of data points marked as potential threats, the behavior association pattern data includes the relationship between differentiated time series data points, time dependency analysis results and connection evaluation of potential threats, the model parameter update results include adjusted parameter values, performance comparison before and after parameter optimization and optimization of anomaly detection sensitivity, and the danger solution includes predictive analysis results of security risks, formulated response strategies and security issues expected to be resolved.

[0014] As a further solution of the present invention, based on network behavior data, using AI, calculating the Euclidean distance and local density between data points, screening data through distance and density anomaly detection, and obtaining the network behavior analysis results are specifically as follows:

[0015] S101: based on the network behavior data, extract the coordinates of each data point, calculate the straight-line distance between each pair of data points, and record the distance value of each group of data points to obtain a data point distance matrix;

[0016] S102: using the data point distance matrix and AI to perform local density calculation on each data point, by calculating the average value of the distance between each data point and its neighboring points, identifying the density value of each data point, and generating a local density analysis result;

[0017] S103: Using the local density analysis results, the density values ​​are screened to identify data points that deviate from the average density value, and abnormal data points are screened using set density and distance thresholds to obtain network behavior analysis results.

[0018] As a further solution of the present invention, the data features are randomly selected and randomly segmented based on the network behavior analysis results, and the abnormal points are isolated by constructing an isolated tree structure to isolate the data. The steps of generating anomaly scores are specifically as follows:

[0019] S201: Based on the network behavior analysis result, randomly select features for data segmentation, select a target feature set of data points, including a timestamp, and perform data processing according to the features, adjust the original order, and generate a random feature selection result;

[0020] S202: using the random feature selection result, segmenting the data, grouping the data based on the value of the selected feature, each group including a subset of data points, and generating data segmentation buckets by discretizing the feature values;

[0021] S203: By dividing the data into buckets, an isolation tree is established for each data bucket according to the fit and difference between the data points, wherein the nodes in the isolation tree represent the data bucket and the degree of isolation, and an abnormality score is generated.

[0022] As a further solution of the present invention, the abnormality scoring is adopted, and the data scores are compared by setting a standard threshold, and the data points whose scores deviate from the set threshold are marked as abnormal, and the marked data are classified and processed to obtain the potential danger mark. Specifically, the steps are as follows:

[0023] S301: Based on the abnormal score, identify the statistical distribution of the score, calculate the average value and standard deviation of the data point score, set the threshold of abnormal detection, the threshold is the average value plus twice the standard deviation, and generate an abnormal threshold setting result;

[0024] S302: using the abnormal threshold setting result, comparing the score of each data point with the set threshold, marking the data points whose scores deviate from the threshold as abnormal, screening the data points whose scores exceed or do not reach the threshold, and generating a marked list of abnormal data points;

[0025] S303: Classify the abnormal data points through the abnormal data point mark list, assign differentiated danger levels to each type of abnormality through the score value and the degree of deviation from the threshold, and generate potential danger marks.

[0026] As a further solution of the present invention, the steps of performing time series analysis through AI, analyzing the fit between differentiated data points, evaluating the connection between potential threats, performing security automation orchestration, and generating behavior association pattern data are as follows:

[0027] S401: Based on the potential risk marker, identify data points marked as abnormal, sort the data in time series through AI, sort the data points according to the timestamps of the data, check the continuity of the time series, and generate a time series sorting result;

[0028] S402: using the time series sorting result, performing a fit analysis on the sorted data points, calculating the time and behavior differences between adjacent data points, evaluating the fit of the behavior patterns of the data points, and generating a data point fit analysis result;

[0029] S403: Based on the data point fit analysis results, a k-means clustering algorithm is used to evaluate the connections and patterns of potential threats, identify data groups with similar fitting patterns, analyze the behavioral correlations between groups, perform security automation orchestration, and cluster data points with similar behavioral patterns to generate behavioral correlation pattern data.

[0030] As a further solution of the present invention, the formula of the k-means clustering algorithm is as follows:

[0031]

[0032] Among them, S a It is cluster C a From the data point to the center point μ a The sum of the squares of the distances, x a is the data point in the cluster, μ a is the center point of the cluster, t a is the nonlinear adjustment factor of the distance, n a is the dimension of the data point,

[0033] W a (x a ,μ a ,t a ,n a ) is a weight function that is dynamically adjusted according to the data point position and the center point position, ∥x a -μ a ∥ 2t The calculation is point xa To center μ a The distance 2t a Power.

[0034] As a further solution of the present invention, the behavior association pattern data is used to adjust the support vector machine model parameters item by item, optimize the parameters for the identified abnormal pattern, update the model to optimize the sensitivity of abnormal detection, and obtain the model parameter update results in the following steps:

[0035] S501: Based on the behavior association pattern data, principal component analysis technology is used to identify key feature vectors from the data, analyze abnormal behavior features of differentiated data points, and evaluate the weight and influence of abnormal behavior features in the support vector machine model to generate feature impact analysis results;

[0036] S502: using the feature impact analysis results, adjusting the parameters of the support vector machine model item by item, including adjusting the parameters of the kernel function, the penalty coefficient and the boundary threshold, optimizing the sensitivity and discrimination of the model to abnormal behaviors, adjusting the target requirements of each abnormal mode, and generating a parameter adjustment record;

[0037] S503: Using the parameter adjustment record, the performance of the adjusted support vector machine model is evaluated, and the response speed of the model is evaluated by cross-validating the model to match the sensitivity requirements of anomaly detection, thereby obtaining a model parameter update result.

[0038] As a further solution of the present invention, the formula of the principal component analysis technology is as follows:

[0039] X proj =(X p -μ)·V p ·D p ·W p

[0040] Among them, X proj is the feature weight matrix, X p is the centered data matrix, μ is the average value of the feature, V p is the principal component vector, D p is the standard deviation matrix of the data, W p is the feature weight coefficient.

[0041] As a further solution of the present invention, based on the model parameter update result, the development trend of safety hazards is simulated, prediction analysis is performed on the hazards, safety automation is orchestrated, and response strategies are formulated to obtain the hazard solution steps as follows:

[0042] S601: Based on the model parameter update result, use the updated support vector machine model to simulate the trend of security events, predict the development trend by inputting abnormal behavior data, and generate a security trend prediction result;

[0043] S602: Using the security trend prediction results, performing risk assessment analysis, analyzing risk areas and potential security vulnerabilities in the prediction results, assessing the threat level to security, and generating risk assessment results;

[0044] S603: Based on the risk assessment results, formulate response strategies, design defense measures and response plans, and formulate associated response measures for differentiated levels of risks, including adjusting firewall settings, updating virus protection and performing network access control, to obtain risk solutions.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, AI is used to calculate the Euclidean distance and local density between data points, to screen out abnormal data, enhance the ability to analyze network behavior data, and improve the accuracy of potential threat identification. By randomly segmenting the data and constructing an isolated tree, the isolation of abnormal points is more effective, the false alarm rate is reduced, and the flexibility and response speed of the response mechanism are enhanced. The comprehensive use of time series analysis and support vector machine model adjustment to predict and analyze potential dangers can foresee and prepare response strategies earlier at the data level, improve the foresight and pertinence of preventive measures, and bring a qualitative leap in security protection. By refining data analysis and processing details, security response is not limited to post-processing, but forms a set of dynamically adaptive and continuously optimized security policy systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the main steps of the present invention;

[0048] Figure 2 It is a schematic diagram of the refinement of S1 of the present invention;

[0049] Figure 3 It is a schematic diagram of the refinement of S2 of the present invention;

[0050] Figure 4 It is a schematic diagram of the refinement of S3 of the present invention;

[0051] Figure 5 It is a schematic diagram of the refinement of S4 of the present invention;

[0052] Figure 6 It is a detailed schematic diagram of S5 of the present invention;

[0053] Figure 7 It is a detailed schematic diagram of S6 of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0056] See also Figure 1 The present invention provides a technical solution, a security automation orchestration method based on AI, comprising the following steps:

[0057] S1: Based on network behavior data, AI is used to calculate the Euclidean distance and local density between data points, and the calculation results are summarized and integrated. The data is filtered through distance and density anomaly detection to obtain network behavior analysis results.

[0058] S2: Based on the results of network behavior analysis, data features are randomly selected and randomly split, and anomalies are isolated by building an isolated tree structure to isolate data and generate anomaly scores;

[0059] S3: Use anomaly scoring to compare data scores by setting a standard threshold, mark data points whose scores deviate from the set threshold as anomalies, classify the marked data, and obtain potential danger marks;

[0060] S4: Based on potential danger markers, AI is used to perform time series analysis, evaluate time dependencies, analyze the fit between differentiated data points, evaluate the connection between potential threats, perform security automation orchestration, and generate behavioral association pattern data;

[0061] S5: Using the behavior association pattern data, adjust the support vector machine model parameters item by item, optimize the parameters for the identified abnormal patterns, update the model to optimize the sensitivity of abnormality detection, and obtain the model parameter update results;

[0062] S6: Based on the model parameter update results, simulate the development trend of safety hazards, conduct predictive analysis on hazards, perform safety automation orchestration, and formulate response strategies to obtain hazard solutions.

[0063] The results of network behavior analysis include distance measurement of abnormal behavior, density distribution of data points and abnormal degree classification of data points. Anomaly scores include the number of segmentations required for isolation, abnormal probability of data points and position in the entire data set. Potential danger markers include scores of data points, classified abnormal states and lists of data points marked as potential threats. Behavioral association pattern data include relationships between differentiated time series data points, time dependency analysis results and connection assessments of potential threats. Model parameter update results include adjusted parameter values, performance comparison before and after parameter optimization and optimization of anomaly detection sensitivity. Danger solutions include predictive analysis results of security risks, formulated response strategies and security issues expected to be resolved.

[0064] See also Figure 2 Based on network behavior data, AI is used to calculate the Euclidean distance and local density between data points, and the data is filtered through distance and density anomaly detection. The specific steps to obtain the network behavior analysis results are as follows:

[0065] S101: Based on the network behavior data, the coordinates of each data point are extracted, the straight-line distance between each pair of data points is calculated, and the distance value of each group of data points is recorded to obtain the execution flow of the data point distance matrix as follows;

[0066] Based on the network behavior data, the coordinates of each data point are extracted. The coordinates represent the location of an event or a specific point of an interaction in the network. The Euclidean distance between each pair of data points is calculated by measuring the straight-line distance between the two points. The distance between each pair of points in the data set is recorded for further density calculation and outlier identification. The data point distance matrix is ​​constructed using the formula:

[0067]

[0068] Among them, D ij represents the distance between data points i and j, x i ,y i 、x j and j are the coordinates of the corresponding data points.

[0069] S102: Using the data point distance matrix and AI, local density calculation is performed on each data point. By calculating the average value of the distance between each data point and its neighboring points, the density value of each data point is identified, and the execution flow of generating the local density analysis result is as follows;

[0070] The data point distance matrix is ​​used to calculate the local density, and AI technology is used to enhance the accuracy and efficiency of density estimation. The average distance between each data point and its neighboring points is calculated to estimate the local density of the data point. This involves summing and averaging the distances around each data point, taking into account a fixed number of nearest neighbors or all points within a specific distance range to form a local density analysis result.

[0071] S103: Through the local density analysis results, the density values ​​are screened, the data points that deviate from the average density value are identified, and the abnormal data points are screened through the set density and distance thresholds to obtain the network behavior analysis results. The execution process is as follows;

[0072] Through the results of local density analysis, abnormal data points are identified and data points whose density values ​​deviate significantly from the overall average density are screened, which involves setting a density threshold and a distance threshold to determine whether a point is considered abnormal. By comparing the density of each point with the mean or median of the overall density and considering the distance to the nearest neighbor, abnormal behavior or potential outliers can be effectively identified, which affects the accuracy and reliability of network behavior analysis and obtains network behavior analysis results.

[0073] See also Figure 3 ,According to the results of network behavior analysis, the data features are randomly selected and randomly divided, and the abnormal points are isolated by building an isolated tree structure to isolate the data. The specific steps for generating anomaly scores are as follows:

[0074] S201: Based on the network behavior analysis results, randomly select features for data segmentation, select the target feature set of data points, including timestamps, and process the data according to the features, adjust the original order, and generate the random feature selection results. The execution process is as follows;

[0075] Based on the results of network behavior analysis, the random feature selection method is used to process the network behavior data, and key features such as timestamps and related network behavior indicators are randomly selected as the basis for data segmentation. It can reduce bias and increase the generalization ability of the model because it does not rely on a specific set of data features. Each data point is reorganized and classified according to the features, thereby adjusting the original data order. It not only helps to identify hidden patterns in the data, but also provides a basis for subsequent data segmentation and anomaly detection. The random feature selection results are generated using the formula:

[0076]

[0077] Among them, F r A function representing a randomly selected set of features, represents the eigenvalue of data point i, S r is the number of features selected.

[0078] S202: Using the random feature selection result, the data is segmented, and the data is grouped based on the value of the selected feature, each group includes a subset of data points, and the execution process of generating data segmentation buckets by discretizing the feature values ​​is as follows;

[0079] Using the random feature selection results, the data set is further segmented, the data is logically grouped based on the values ​​of the selected features, and created based on the discretization of the feature values. Each bucket includes a subset of data points with similar feature values. This allows for a more detailed analysis of the behavior patterns and statistical characteristics within each group, improves the efficiency and accuracy of data processing, and forms data segmentation buckets using the formula:

[0080] G k ={i|i f ∈Bucket k}

[0081] Among them, G k represents the kth data segmentation bucket based on feature f, i f is the eigenvalue of data point i, Bucket k is the range of eigenvalues ​​of the kth bucket.

[0082] S203: By partitioning the data into buckets, an isolation tree is established for each data bucket according to the fit and difference between the data points. The nodes in the isolation tree represent the data bucket and the degree of isolation. The execution process of generating anomaly scores is as follows;

[0083] By partitioning the data into buckets and establishing an isolation tree, data anomalies can be effectively identified. The isolation tree recursively selects features and partitions data points according to the feature values. In the process, each node of the tree represents a data bucket and the degree of isolation of the data. The deeper the node, the higher the anomaly of the related data point. The anomaly score for each data point is generated using the formula:

[0084]

[0085] Among them, S iso represents the anomaly score of data point i, E(h(i)) is the average path length from the data point to the isolated leaf node, and c(D) is the normalization factor of the number of data points D.

[0086] See also Figure 4 , using anomaly scoring, by setting a standard threshold to compare data scores, marking data points whose scores deviate from the set threshold as abnormal, classifying and processing the marked data, and obtaining the potential danger mark steps are as follows:

[0087] S301: Based on the anomaly score, identify the statistical distribution of the score, and set the threshold of anomaly detection by calculating the average and standard deviation of the data point score. The threshold is the average plus twice the standard deviation. The execution process of generating the anomaly threshold setting result is as follows;

[0088] Based on the anomaly score, the statistical distribution of the score is identified. The threshold for anomaly detection is set by calculating the mean and standard deviation of the data point scores. The sum of the data points is calculated and divided by the number of data points to obtain the mean. The square of the difference between each data point and the mean is calculated and then divided by the number of data points to obtain the variance. The square root is the standard deviation. The anomaly detection threshold is set to the mean plus twice the standard deviation to generate the anomaly threshold setting result. The formula used is:

[0089] T v =μ v +k v ·σ

[0090] Among them, T v is the abnormal monitoring threshold, μ v represents the mean of the ratings, σ represents the standard deviation of the ratings, and k v is the regulating factor.

[0091] S302: Using the abnormal threshold setting result, comparing the score of each data point with the set threshold, marking the data point whose score deviates from the threshold as abnormal, screening the data points whose scores exceed or do not reach the threshold, and generating the abnormal data point mark list. The execution process is as follows;

[0092] The results of abnormal threshold setting are used to compare the scores of each data point to determine whether it exceeds or does not reach the threshold. If the data point deviates from the threshold, the data point is marked as abnormal. The specific operation is to traverse the data points, record all abnormal data points, ensure that all data points that deviate from the standard are accurately marked, provide necessary input for subsequent processing, and generate a list of abnormal data point marks.

[0093] S303: Classify the abnormal data points through the abnormal data point mark list, assign differentiated danger levels to each type of abnormality through the score value and the degree of deviation from the threshold, and generate the execution process of the potential danger mark as follows;

[0094] Through the abnormal data point mark list, the abnormal data points are classified and processed. According to the score of each data point and the degree of deviation from the threshold, a differentiated danger level is assigned to each type of abnormality. The danger level is assigned based on the deviation between the score and the threshold. A high degree of deviation means a higher risk. By accurately measuring the degree of deviation, a clear danger level is defined for each type of abnormality, and a list of potential danger marks is generated.

[0095] See also Figure 5 , using AI to perform time series analysis, analyze the fit between differentiated data points, evaluate the connection between potential threats, perform security automation orchestration, and generate behavioral association pattern data. The specific steps are:

[0096] S401: Based on the potential danger markers, identify the data points marked as abnormal, sort the data in time series through AI, sort the data points according to the timestamps of the data, check the continuity of the time series, and generate the execution process of the time series sorting results as follows;

[0097] Based on potential danger markers, identify data points marked as abnormal, sort the data in time series through AI, collect the timestamps of each data point and arrange them in chronological order. Through sorting, ensure the time continuity and correctness of the data. To ensure the continuity of the data, calculate the difference between the timestamps of adjacent data points, check for abnormal time jumps or data missing, and generate the time series sorting results. The formula used is:

[0098] δt i =t i+1 -t i

[0099] Among them, δt i Represents the time difference between two consecutive data points, t i and t i+1 Represents the timestamps of two consecutive data points.

[0100] S402: Using the time series sorting result, performing fit analysis on the sorted data points, calculating the time and behavior differences between adjacent data points, evaluating the fit of the behavior pattern of the data points, and generating the data point fit analysis result. The execution process is as follows;

[0101] Using the time series sorting results, we perform fit analysis on the sorted data points, including calculating the time difference and behavior difference between adjacent data points. The behavior difference is evaluated by the score difference. Based on the time and behavior difference, we evaluate the fit of the behavior pattern of each data point to determine whether the data point conforms to the expected behavior pattern. This helps identify the consistency or anomaly of the behavior pattern and generates the fit analysis results of the data point. The formula used is:

[0102]

[0103] Among them, H(x i ,t i ) represents the continuous data points x i to i The goodness of fit value, δx i and δt i Representing differences in behavior and timing, respectively.

[0104] S403: Based on the data point fit analysis results, the k-means clustering algorithm is used to evaluate the connection and pattern of potential threats, identify data groups with similar fitting patterns, analyze the behavioral correlation between groups, perform security automation orchestration, and cluster data points with similar behavioral patterns to generate behavioral correlation pattern data. The execution process is as follows;

[0105] Based on the results of the data point fit analysis, the k-means clustering method is used to evaluate the connections and patterns of potential threats. A new clustering standard is defined based on the similarity measurement of behavior patterns to identify data groups with similar behavior patterns. Further behavior correlation analysis is performed on the groups to identify common threats or abnormal patterns. Through cluster analysis, security automation orchestration is performed to cluster data points with similar behavior patterns together for centralized processing or monitoring, generating behavior correlation pattern data.

[0106] The formula for the k-means clustering algorithm is as follows:

[0107]

[0108] Among them, S a It is cluster C a From the data point to the center point μ a The sum of the squares of the distances, x a is the data point in the cluster, μ a is the center point of the cluster, t a is the nonlinear adjustment factor of the distance, n a is the dimension of the data point,

[0109] W a (x a ,μ a ,t a ,n a ) is a weight function that is dynamically adjusted according to the data point position and the center point position, ∥x a -μ a ∥ 2t The calculation is point x a To center μ a The distance 2t a Power.

[0110] The execution flow is as follows:

[0111] Choose a suitable nonlinear factor t a , to enhance the penalty or reward for points far from the center, determine the dimension n of the data point a , used to adjust the complexity of weight calculation and calculate the 2t of the Euclidean distance from each point to the center point a Power, according to x a, μ a ,t a and n a The value of dynamically calculates the weight W of each point a , weight coefficient W a It is obtained through an optimized learning algorithm, such as using the gradient descent method to adjust W a To minimize the overall intra-cluster distance, all values ​​are merged and the cohesion S of each cluster is calculated by the formula a , in order to achieve a more refined clustering effect.

[0112] See also Figure 6 , using the behavior association pattern data, adjust the support vector machine model parameters item by item, optimize the parameters for the identified abnormal patterns, update the model to optimize the sensitivity of abnormal detection, and obtain the model parameter update results in the following steps:

[0113] S501: Based on the behavior association pattern data, the principal component analysis technology is used to identify key feature vectors from the data, analyze the abnormal behavior characteristics of the differentiated data points, and evaluate the weight and influence of the abnormal behavior characteristics in the support vector machine model. The execution process of generating the feature impact analysis results is as follows;

[0114] Based on the behavior association pattern data, the principal component analysis (PCA) technology is used to identify key eigenvectors from the data, calculate the covariance matrix of the data, record the eigenvalues ​​and corresponding eigenvectors, and select the eigenvector with the largest contribution as the principal component. The principal component helps to identify the key variation direction in the data points, analyze the abnormal behavior characteristics of the differentiated data points, evaluate the weight and influence of the abnormal behavior characteristics in the support vector machine (SVM) model, calculate the contribution of each principal component in the SVM model, and generate the feature impact analysis results.

[0115] The formula for the principal component analysis technique is as follows:

[0116] X proj =(X p -μ)·V p ·D p ·W p

[0117] Among them, X proj is the feature weight matrix, X p is the centered data matrix, μ is the average value of the feature, V p is the principal component vector, D p is the standard deviation matrix of the data, W p is the feature weight coefficient.

[0118] The execution flow is as follows:

[0119] Calculate the mean μ of each feature and subtract the mean from each data point to achieve data centering, calculate the covariance matrix of the centered data and extract the principal component vector V p , calculate the standard deviation of the original data and construct the diagonal matrix D p , the weight coefficient W is determined according to the variance explanation ratio of each principal component p The weight coefficient is obtained by normalizing the variance explanation ratio and multiplying it by an adjustment factor (such as 0.1) to enhance the model's sensitivity to small variance features and complete the calculation process of the feature weight matrix.

[0120] S502: Using the feature impact analysis results, adjust the parameters of the support vector machine model item by item, including adjusting the parameters of the kernel function, the penalty coefficient and the boundary threshold, optimize the sensitivity and discrimination of the model to abnormal behaviors, and make adjustments according to the target requirements of each abnormal mode. The execution process of generating parameter adjustment records is as follows;

[0121] Using the results of feature impact analysis, adjust the parameters of the support vector machine model item by item, including adjusting the parameters of the kernel function, penalty coefficients, and boundary thresholds. For each identified abnormal behavior pattern, adjust the parameters according to specific needs to optimize the model's sensitivity and discrimination to abnormal behavior. Ensure that the model can more accurately identify and respond to abnormal data points, and generate model parameter adjustment records. The formula used is:

[0122]

[0123] Among them, Φ(C,γ,θ) represents the optimization objective function of the model, C represents the penalty coefficient, γ represents the kernel function parameter, θ represents the boundary threshold, ξ r represents the slack variable, and w represents the normal vector of the splitting hyperplane.

[0124] S503: Using the parameter adjustment record, the performance of the adjusted support vector machine model is evaluated. By cross-validating the model, the response speed of the model is evaluated to match the sensitivity requirements of anomaly detection. The execution process of obtaining the model parameter update result is as follows;

[0125] The parameter adjustment records are used to evaluate the performance of the adjusted support vector machine model. The performance of the model on different data subsets is calculated by cross-validating the model, and the response speed of the model and the sensitivity requirements for matching anomaly detection are evaluated. The process includes multiple divisions of the training and validation data sets. After each adjustment, the validation set is used to test the accuracy and responsiveness of the model, and the model parameter update results are obtained. The formula used is:

[0126]

[0127] Among them, R(α,β) represents the average response speed and sensitivity of the model, α represents the model parameter set, β l represents different data subsets, k represents the number of cross-validation folds l, g(α,β i ) is a function used to calculate the accuracy of the model given the parameters and data subset.

[0128] See also Figure 7 Based on the model parameter update results, simulate the development trend of safety hazards, conduct predictive analysis on hazards, perform safety automation orchestration, and formulate response strategies. The specific steps to obtain hazard solutions are as follows:

[0129] S601: Based on the model parameter update result, the updated support vector machine model is used to simulate the trend of security events, and the development trend is predicted by inputting abnormal behavior data to generate the security trend prediction result. The execution process is as follows;

[0130] Based on the model parameter update results, the updated support vector machine (SVM) model is used to simulate the trend of security events and collect abnormal behavior data as model input. The data includes network traffic anomalies, unauthorized access attempts, and suspicious user behavior. The updated model parameters are used to predict the future development trend of abnormal behavior through the SVM model. Through the prediction function of the model, the evolution path of security threats is evaluated and the security trend prediction results are generated. The formula used is:

[0131] P(i,Θ)=sign(w·x+b)

[0132] Among them, P(i, Θ) is the function of outputting the predicted label, i and Θ are the input data and model parameter sets respectively, w and b are the weight and bias terms respectively, and sign is the function used to determine the category to which the data point i belongs.

[0133] S602: Using the security trend prediction results, conduct risk assessment analysis, and evaluate the threat level to security by analyzing the risk areas and potential security vulnerabilities in the prediction results. The execution process of generating the risk assessment results is as follows;

[0134] The security trend prediction results are used to conduct risk assessment analysis, which includes identifying risk areas and potential security vulnerabilities in the prediction results. Statistical and model analysis methods are used to assess the degree of threat posed by risk areas to security. Based on the various risk levels obtained from the analysis, a basis is provided for subsequent security strategy formulation, and risk assessment results are generated using the formula:

[0135]

[0136] Among them, R c represents the overall risk score, r crepresents the severity score of a single risk factor c, p c represents the probability of the risk factor occurring, and U is the total number of risk factors.

[0137] S603: Based on the risk assessment results, formulate response strategies, design defense measures and response plans, and formulate relevant response measures for differentiated levels of risks, including adjusting firewall settings, updating virus protection and performing network access control. The execution process of the risk solution is as follows;

[0138] Based on the risk assessment results, formulate response strategies, design defense measures and response plans, and formulate corresponding response measures for different levels of risks, including adjusting firewall settings, updating virus protection software and strengthening network access control. According to the needs of each risk level, implement protection and response strategies item by item to ensure that potential security threats can be responded to quickly and effectively, integrate all measures, and generate risk solutions. The formula used is:

[0139]

[0140] Among them, S k represents the security defense score, K represents the type of defense measures, J represents the set of risk factors, and c jk is the coverage efficiency of defense measure k for risk factor c, a c is the activity and frequency of risk factor c.

[0141] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A security automation orchestration method based on AI, characterized in that: The following steps are involved: Based on network behavior data, AI is used to calculate the Euclidean distance and local density between data points, and the data is filtered through distance and density anomaly detection to obtain network behavior analysis results; Based on the network behavior analysis results, data features are randomly selected and randomly segmented, and anomaly points are isolated by constructing an isolated tree structure to isolate data and generate anomaly scores; Using the anomaly scoring, by setting a standard threshold to compare data scores, marking data points whose scores deviate from the set threshold as abnormal, classifying and processing the marked data, and obtaining potential danger marks; Based on the potential risk markers, AI is used to perform time series analysis, analyze the fit between differentiated data points, evaluate the connection between potential threats, perform security automation orchestration, and generate behavioral association pattern data; Using the behavior association pattern data, the support vector machine model parameters are adjusted item by item, the parameters are optimized for the identified abnormal pattern, the model is updated to optimize the sensitivity of abnormality detection, and the model parameter update result is obtained; Based on the update results of the model parameters, the development trend of safety hazards is simulated, prediction analysis is performed on the hazards, safety automation orchestration is performed, and response strategies are formulated to obtain hazard solutions.

2. The AI-based security automation orchestration method according to claim 1, characterized in that: The network behavior analysis results include distance metrics of abnormal behaviors, density distribution of data points, and abnormal degree grading of data points; the abnormality score includes the number of segmentations required for isolation, the abnormal probability of the data point, and the position in the entire data set; the potential danger mark includes the score of the data point, the classified abnormal state, and a list of data points marked as potential threats; the behavior association pattern data includes the relationship between differentiated time series data points, time dependency analysis results, and connection assessment of potential threats; the model parameter update results include adjusted parameter values, performance comparison before and after parameter optimization, and optimization of anomaly detection sensitivity; the danger solution includes predictive analysis results of security risks, formulated response strategies, and security issues expected to be resolved.

3. The AI-based security automation orchestration method according to claim 1, characterized in that: Based on network behavior data, AI is used to calculate the Euclidean distance and local density between data points, and the data is filtered through distance and density anomaly detection. The specific steps to obtain network behavior analysis results are as follows: Based on the network behavior data, the coordinates of each data point are extracted, the straight-line distance between each pair of data points is calculated, and the distance value of each group of data points is recorded to obtain the data point distance matrix; The data point distance matrix is ​​adopted, and AI is used to calculate the local density of each data point, and the density value of each data point is identified by calculating the average value of the distance between each data point and its neighboring points, thereby generating a local density analysis result; The density values ​​are screened through the local density analysis results to identify data points that deviate from the average density value, and abnormal data points are screened through set density and distance thresholds to obtain network behavior analysis results.

4. The AI-based security automation orchestration method according to claim 1, characterized in that: According to the network behavior analysis results, data features are randomly selected and randomly divided, and anomalies are isolated by building an isolated tree structure to isolate data. The steps of generating anomaly scores are as follows: Based on the network behavior analysis results, randomly select features for data segmentation, select a target feature set of data points, including timestamps, and process data according to the features to adjust the original order and generate random feature selection results; Using the random feature selection result, segmenting the data, grouping the data based on the value of the selected feature, each group includes a subset of data points, and generating data segmentation buckets by discretizing the feature values; Through the data segmentation buckets, an isolation tree is established for each data bucket according to the fit and difference between data points, and the nodes in the isolation tree represent the data buckets and the degree of isolation, thereby generating an anomaly score.

5. The AI-based security automation orchestration method according to claim 1, characterized in that: The steps of using the anomaly score to compare data scores by setting a standard threshold, marking data points whose scores deviate from the set threshold as abnormal, and classifying the marked data to obtain potential danger marks are as follows: Based on the anomaly scores, the statistical distribution of the scores is identified, and a threshold for anomaly detection is set by calculating the average and standard deviation of the scores of the data points, where the threshold is the average plus twice the standard deviation, to generate an anomaly threshold setting result; Using the abnormal threshold setting result, comparing the score of each data point with the set threshold, marking the data points whose scores deviate from the threshold as abnormal, screening the data points whose scores exceed or do not reach the threshold, and generating a marked list of abnormal data points; The abnormal data points are classified and processed through the abnormal data point mark list, and a differentiated danger level is assigned to each type of abnormality through the score value and the degree of deviation from the threshold to generate a potential danger mark.

6. The AI-based security automation orchestration method according to claim 1, characterized in that: According to the potential danger markers, AI is used to perform time series analysis, analyze the fit between differentiated data points, evaluate the connection between potential threats, perform security automation orchestration, and generate behavior association pattern data in the following specific steps: Based on the potential risk markers, identify data points marked as abnormal, sort the data in time series through AI, sort the data points according to the timestamps of the data, check the continuity of the time series, and generate a time series sorting result; Using the time series sorting results, performing fit analysis on the sorted data points, calculating the time and behavior differences between adjacent data points, evaluating the fit of the behavior patterns of the data points, and generating data point fit analysis results; Through the data point fitting analysis results, the k-means clustering algorithm is used to evaluate the connections and patterns of potential threats, identify data groups with similar fitting patterns, analyze the behavioral correlations between groups, perform security automation orchestration, and cluster data points with similar behavioral patterns to generate behavioral correlation pattern data.

7. The AI-based security automation orchestration method according to claim 6, characterized in that: The formula of the k-means clustering algorithm is as follows: Among them, S a It is cluster C a From the data point to the center point μ a The sum of the squares of the distances, x a is the data point in the cluster, μ a is the center point of the cluster, t a is the nonlinear adjustment factor of the distance, n a is the dimension of the data point, W a (x a ,μ a ,t a ,n a ) is a weight function that is dynamically adjusted according to the data point position and the center point position, ||x a -μ a || 2t The calculation is point x a To center μ a The distance 2t a Power.

8. The AI-based security automation orchestration method according to claim 1, characterized in that: Using the behavior association pattern data, the support vector machine model parameters are adjusted item by item, the parameters are optimized for the identified abnormal pattern, and the model is updated to optimize the sensitivity of abnormal detection. The steps of obtaining the model parameter update result are specifically as follows: Based on the behavior association pattern data, principal component analysis technology is used to identify key feature vectors from the data, analyze the abnormal behavior characteristics of differentiated data points, and evaluate the weight and influence of the abnormal behavior characteristics in the support vector machine model to generate feature impact analysis results; Using the feature impact analysis results, adjust the parameters of the support vector machine model item by item, including adjusting the parameters of the kernel function, the penalty coefficient and the boundary threshold, optimize the sensitivity and discrimination of the model to abnormal behavior, make adjustments according to the target requirements of each abnormal mode, and generate parameter adjustment records; The parameter adjustment records are used to evaluate the performance of the adjusted support vector machine model, and the response speed of the model is evaluated by cross-validating the model to match the sensitivity requirements of anomaly detection, thereby obtaining the model parameter update results.

9. The AI-based security automation orchestration method according to claim 8, characterized in that: The formula for the principal component analysis technique is as follows: X proj =(X p -μ)·V p ·D p ·W p Among them, X proj is the feature weight matrix, X p is the centered data matrix, μ is the average value of the feature, V p is the principal component vector, D p is the standard deviation matrix of the data, W p is the feature weight coefficient.

10. The AI-based security automation orchestration method according to claim 1, characterized in that: Based on the model parameter update results, the development trend of safety hazards is simulated, prediction analysis is performed on the hazards, safety automation orchestration is performed, and response strategies are formulated. The specific steps for obtaining a hazard solution are as follows: Based on the model parameter update result, the updated support vector machine model is used to simulate the trend of security events, and the development trend is predicted by inputting abnormal behavior data to generate security trend prediction results; Using the security trend prediction results, conduct risk assessment analysis, analyze risk areas and potential security vulnerabilities in the prediction results, evaluate the threat level to security, and generate risk assessment results; Based on the risk assessment results, a response strategy is formulated, defense measures and response plans are designed, and associated response measures are formulated for differentiated levels of risks, including adjusting firewall settings, updating virus protection, and performing network access control to obtain risk solutions.

Citation Information

Cited By

  • Method for detecting multicellular factors in blood sample

    CN119716035A

  • Distributed storage node fault detection system

    CN119718741A