Executor scheduling method and device, electronic equipment and storage medium
By constructing relationship heterogeneous and isoformed graphs, using graph attention networks to predict the probability of threat traffic breaking the execution body, and scheduling in advance to replace the high-risk execution body, solving the problem of resource loss of existing mimicry defense systems and coping with changing network environments, improving the problem of improving the Defensive capabilities and resource utilization efficiency.
Patent Information
- Application Number
- CN202410984678.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-07-22
AI Technical Summary
The executor scheduling method of the existing mimic defense system is prone to resource loss and cannot effectively deal with the changing network environment, resulting in frequent scheduling, increasing the difficulty of attacks but also increasing resource consumption.
By constructing relationship heterogeneous graphs and relationship isomorphic graphs, use graph attention network to predict the probability of threat traffic breaking the execution body, discover high-risk execution bodies in advance and perform scheduling replacement, improve defense capabilities and reduce resource losses.
It improves the defense ability of the mimic defense system against attacks, reduces the resource loss caused by frequent scheduling, and enhances the defense ability of unknown vulnerabilities and attacks.
Smart Images

Figure CN119011208A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of network security technology, and in particular to an execution body scheduling method, device, electronic device and storage medium. Background Art
[0002] Existing executor scheduling methods for mimetic defense systems often start from the dynamic, redundancy, heterogeneity and other characteristics of mimetic defense systems, and combine negative feedback characteristics to achieve high robustness and security of mimetic defense systems, reduce the probability of common vulnerabilities and backdoors in the system, and thus increase the difficulty and cost of attackers' intrusion. However, as the number of heterogeneous executors increases, the mimetic defense system will become more complex. In addition, current executor scheduling methods mostly select and schedule based on the executor's own attributes such as heterogeneity and confidence. The entire scheduling process is relatively frequent and easily causes resource loss. Summary of the invention
[0003] The present invention provides an execution body scheduling method, device, electronic device and storage medium to solve the defect that the execution body scheduling method of the existing mimicry defense system is prone to resource loss, realizes the combined construction of a relational heterogeneous graph and a relational isomorphic graph, and uses a graph attention network to predict the probability of the threat traffic in the current network breaking through the current working execution body, and discovers the execution body with a high probability of being broken in advance, so as to schedule and replace the corresponding execution body in advance, thereby improving the defense capability of the mimicry defense against attacks and reducing the resource loss caused by frequent scheduling.
[0004] The present invention provides an execution body scheduling method, which includes the following steps.
[0005] A relational heterogeneous graph is constructed based on first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and a set of edge relationships corresponding to the plurality of threat flows and the plurality of execution bodies; and a graph attention network is used to determine first feature vectors corresponding to each of the plurality of first nodes in the relational heterogeneous graph, and second feature vectors corresponding to each of the plurality of second nodes.
[0006] A relational isomorphism graph is constructed according to the multiple first nodes and the node similarities between the multiple first nodes; and the graph attention network is used to determine the third eigenvectors corresponding to each of the multiple first nodes in the relational isomorphism graph.
[0007] For each first node among the multiple first nodes and each second node among the multiple second nodes, determine the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
[0008] The present invention also provides an execution body scheduling device, comprising the following modules.
[0009] A data processing module is used to construct a relational heterogeneous graph based on first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and a set of edge relationships corresponding to the plurality of threat flows and the plurality of execution bodies; and to use a graph attention network to determine first feature vectors corresponding to each of the plurality of first nodes in the relational heterogeneous graph, and second feature vectors corresponding to each of the plurality of second nodes; to construct a relational isomorphic graph based on the node similarities between the plurality of first nodes and the plurality of first nodes; and to use the graph attention network to determine third feature vectors corresponding to each of the plurality of first nodes in the relational isomorphic graph.
[0010] A scheme determination module is used to determine, for each first node among the multiple first nodes and each second node among the multiple second nodes, the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; and determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
[0011] The present invention also provides an electronic 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, any of the above-mentioned execution body scheduling methods is implemented.
[0012] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the execution body scheduling method described in any one of the above is implemented.
[0013] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the execution body scheduling method described above is implemented.
[0014] The present invention provides an execution body scheduling method, device, electronic device and storage medium. The method combines the constructed relationship heterogeneous graph and relationship isomorphic graph, uses a graph attention network to predict the probability of threat traffic in the current network breaking through the current working execution body, and discovers the execution body with a high probability of being broken in advance, so as to schedule and replace the corresponding execution body in advance, thereby improving the defense capability of mimicry defense against attacks and reducing the resource loss caused by frequent scheduling. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0016] Figure 1 This is one of the flow charts of the execution body scheduling method provided by the present invention.
[0017] Figure 2 It is a schematic diagram of the relational heterogeneous graph provided by the present invention.
[0018] Figure 3 It is a schematic diagram of the relational isomorphism graph provided by the present invention.
[0019] Figure 4 This is the second flowchart of the execution body scheduling method provided by the present invention.
[0020] Figure 5 It is a structural schematic diagram of the execution body scheduling device provided by the present invention.
[0021] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] In order to better understand the embodiments of the present invention, the background technology is first described in detail.
[0024] With the rapid development of industrial digitalization, intelligence, and informatization, industrial control systems (hereinafter referred to as industrial control systems) are closely integrated with information technology (IT) systems. In particular, with the development of cloud computing and edge computing, industrial control systems are transforming towards digitalization, and a cloud-edge collaborative technical architecture has gradually been formed. However, while cloud-edge collaborative industrial control systems realize data sharing, knowledge sharing, and resource sharing, their security boundaries are becoming increasingly blurred. Due to the lack of security protection measures, industrial control systems are very vulnerable to network attacks. On the one hand, the flat hybrid connection and ubiquitous access of cloud-edge devices have led to an increase in the exposure of attacks; the public-private hybrid connection and blurred security boundaries of the network have led to random flickering of attack paths; the hidden transmission across cyber-physical domains and the unclear attack characteristics have made it extremely difficult to accurately identify attacks and avoid them in advance. On the other hand, in the scenario of cloud-edge collaboration, the demand for process updates, continuous iterative configuration development of industrial control programs, and dynamic evolution of industrial control logic is increasing, and traditional passive defense measures such as firewalls and patches are difficult to apply. To address these issues, it is necessary to innovate and integrate new active defense technologies to achieve defense against unknown vulnerabilities and attacks in the network.
[0025] Cyberspace Mimic Defense (CMD) technology is a new type of active defense concept. This concept does not rely too much on prior information such as attack codes and attack behavior characteristics. It constructs an uncertain operating environment based on dynamic, redundancy, heterogeneity and other technical means, changes the static, deterministic and similarity of industrial control systems, disrupts the triggering mechanism of vulnerability backdoors, and makes it impossible for attackers to judge whether the attack is successful and whether the attack effect is achieved, which increases the difficulty and cost of the attack. Compared with most traditional network defense technologies, CMD technology does not require known vulnerabilities or risks, nor does it require the attacker's goals, characteristics, mechanisms and methods to respond to security threats. As the core technology of cyberspace mimic defense, the use of appropriate defense methods to realize the scheduling of mimic defense executors can better improve the defense capabilities of the mimic defense system against unknown vulnerabilities and attacks. Therefore, the executor scheduling method of the mimic defense system has always been the focus of research.
[0026] Existing executor scheduling methods for mimetic defense systems often start from the dynamic, redundancy, heterogeneity and other characteristics of mimetic defense systems, and combine negative feedback characteristics to achieve high robustness and security of mimetic defense systems, reduce the probability of common vulnerabilities and backdoors in the system, and thus increase the difficulty and cost of attackers' intrusion. However, as the number of heterogeneous executors increases, the mimetic defense system will become more complex. In addition, current executor scheduling methods mostly select and schedule based on the executor's own attributes such as heterogeneity and confidence. The entire scheduling process lacks the ability to perceive the current network environment, cannot cope with the changing network environment, and cannot dynamically schedule executors according to changes in the network environment, resulting in resource loss.
[0027] In order to solve the above technical problems, an embodiment of the present invention provides an execution body scheduling method. The method combines the constructed relationship heterogeneous graph and relationship isomorphic graph, and adopts a graph attention network to predict the probability of threat traffic in the current network breaking through the current working execution body, and discovers the execution body with a high probability of being broken in advance, so as to schedule and replace the corresponding execution body in advance, thereby improving the defense capability of mimicry defense against attacks and reducing the resource loss caused by frequent scheduling.
[0028] The following is a detailed description of the execution body scheduling method provided by the embodiment of the present invention.
[0029] Figure 1 is a flow chart of the execution body scheduling method provided by the present invention, such as Figure 1 As shown, the method includes the following: step 101-step 103.
[0030] Step 101: construct a relational heterogeneous graph based on the first nodes corresponding to each of the plurality of threat flows, the second nodes corresponding to each of the plurality of execution bodies, and the edge relationship sets corresponding to the plurality of threat flows and the plurality of execution bodies; and use a graph attention network to determine the first feature vectors corresponding to each of the plurality of first nodes and the second feature vectors corresponding to each of the plurality of second nodes in the relational heterogeneous graph.
[0031] Among them, threat traffic usually refers to malicious traffic attacks against a network or system, which is intended to exhaust bandwidth, system resources or cause service unavailability. The number of threat traffic is p, where p is an integer greater than 1. At this time, the first nodes corresponding to each of the p threat traffics can constitute a first node set, and the first node set can be T={t 1 ,t 2 ,…,t α ,…,t p} means, where t α It represents the αth threat traffic and the corresponding first statistical feature. Optionally, the first statistical feature may include: traffic size, traffic rate, traffic source and traffic distribution, etc.
[0032] The execution body refers to various computing systems that perform specific tasks. In the mimic defense architecture, the execution body is responsible for performing actual operations and computing tasks. Optionally, the execution body may include: computers, servers or network devices, etc. The number of the execution bodies is q, where q is an integer greater than 1. At this time, the second nodes corresponding to the q execution bodies can constitute a second node set, and the second node set can be represented by E={e 1 ,e 2 ,…,e β ,…,e q} means, where e β It represents the βth executable body and the corresponding second statistical feature. Optionally, the second statistical feature may include: activity, communication feature, resource occupancy feature, etc.
[0033] The edge relationship set may include multiple edge relationships, each edge relationship is used to characterize the attack relationship between the corresponding threat flow and the corresponding execution body. It should be noted that each edge relationship is established when the corresponding threat flow can break through the corresponding execution body. Among them, the threat flow can break through the execution body is determined in the record of the given threat flow breaking through the execution body.
[0034] A relational heterogeneous graph (heterogeneous graph for short) refers to a graph structure with more than one node type or edge type in the graph. It is mainly used to model multiple types of objects (i.e., threat traffic and execution bodies) and the complex associations between these multiple objects (i.e., the attack relationship between threat traffic and execution bodies). Specifically, this relational heterogeneous graph can be used in G te =(V te ,E te ) represents, where V te represents a plurality of first nodes and a plurality of second nodes; E te Indicates the attack relationship between threat traffic and execution entities.
[0035] The graph attention network includes multiple network layers and is a deep learning method based on graph neural networks. It introduces an attention mechanism to process node data with complex relationships (such as the multiple first nodes and the multiple second nodes mentioned above), and can adaptively learn node features in heterogeneous graphs to achieve more accurate probability predictions. The node features mentioned above contain rich threat traffic feature information (i.e., the first feature vector) and execution body feature information (i.e., the second feature vector).
[0036] After determining multiple threat flows and multiple execution bodies, the electronic device can generate a first node corresponding to each of the multiple threat flows and a second node corresponding to each of the multiple execution bodies; then determine the attack relationship between all first nodes and all second nodes to construct an edge relationship set; then, the electronic device constructs a relationship heterogeneous graph based on all first nodes, all second nodes and the edge relationship set; since the advantage of the graph attention network in the prediction task is that it can adaptively pay attention to the relationship between different nodes in the relationship heterogeneous graph and effectively capture the complex dependency relationship between nodes, the electronic device can use the graph attention network to extract features from all first nodes to obtain the first feature vectors corresponding to each of them, and extract features from all second nodes to obtain the second feature vectors corresponding to each of them, providing data support for subsequent probability prediction. In addition, the accuracy of all first feature vectors and all second feature vectors is also relatively high.
[0037] For example, Figure 2 is a schematic diagram of a relational heterogeneous graph provided by the present invention, such as Figure 2 As shown, there are four threat flows, corresponding to the first node p 1 , the first node p 2 , the first node p 3 and the first node p 4 ; There are five execution bodies, corresponding to the second node q 1 , the second node q 2 , the second node q 3 , the second node q 4 and the second node q 5 ; Since the first node p 1 For the second node q 1 and the second node q 2 All have an attack relationship, so the first node p 1 and the second node q 1 Corresponding to an edge relationship, the first node p 1 and the second node q 2 Corresponding to an edge relationship, and so on, the first node p 2 With the second node q 2 , the second node q 3 and the second node q 4 Each corresponds to an edge relationship, the first node p 3 With the second node q 4 and the second node q 5 Each corresponds to an edge relationship, the first node p 4 With the second node q 4 This corresponds to one edge relationship, based on which there are a total of eight edge relationships.
[0038] The electronic device can construct a heterogeneous relationship graph according to the first nodes corresponding to the four threat flows, the second nodes corresponding to the five execution bodies, and the edge relationship sets corresponding to the four threat flows and the five execution bodies.
[0039] Step 102: construct a relational isomorphism graph based on the multiple first nodes and the node similarities between the multiple first nodes; and use a graph attention network to determine the third eigenvectors corresponding to each of the multiple first nodes in the relational isomorphism graph.
[0040] Among them, the similarity between nodes refers to the similarity between the corresponding features of the nodes in structure, attributes or functions.
[0041] A relational isomorphic graph (abbreviated as isomorphic graph) refers to a graph structure with only one node type and one relationship type in the graph. It is mainly used to model one type of object (i.e. threat traffic) and the complex association relationship between such objects (i.e. similarity between nodes). Specifically, the relational isomorphic graph can be represented by G t =(V t ,E t ) represents, where V t Represents multiple first nodes; E t Indicates the similarity relationship between threat traffic.
[0042] It should be noted that the similarity between threat flows is used as the edge weight to measure the adjacency relationship between nodes, that is, as an edge relationship. The edge relationship in the relationship isomorphism graph is established when the similarity between the corresponding threat flows is greater than a preset similarity threshold. The preset similarity threshold can be set before the electronic device leaves the factory or can be customized, and is not specifically limited here.
[0043] After determining the first nodes corresponding to each of the multiple threat flows, the electronic device can determine the node similarity between all the first nodes to combine all the first nodes to construct a relational isomorphism graph; since the advantage of the graph attention network in the prediction task is that it can adaptively focus on the relationship between nodes in the relational isomorphism graph and effectively capture the complex dependencies between nodes, the electronic device can use the graph attention network to extract features from all the first nodes to obtain the corresponding third feature vectors, providing data support for subsequent probability predictions. In addition, the accuracy of all third feature vectors is also relatively high.
[0044] For example, Figure 3 is a schematic diagram of a relational isomorphism graph provided by the present invention, such as Figure 3 As shown, there are six threat flows, corresponding to the first node p 1 , the first node p 2 , the first node p 3 , the first node p 4, the first node p 5 and the first node p 6 ; Since the first node p 1 and the first node p 3 There is a node similarity between them, so the first node p 1 and the first node p 3 Corresponding to an edge relationship, and so on, the first node p 3 and the first node p 5 Corresponding to an edge relationship, the first node p 4 With the first node p 2 , the first node p 5 and the first node p 6 Each of them corresponds to an edge relationship, based on this, there are five edge relationships in total.
[0045] The electronic device can construct a relational isomorphism graph according to the first nodes corresponding to the six threat flows and the similarities between the nodes of the six threat flows.
[0046] Step 103: For each first node among the multiple first nodes and each second node among the multiple second nodes, determine the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
[0047] Among them, the attack probability refers to the probability that the attack behavior threatening the traffic successfully breaks through the execution body.
[0048] For any first node among multiple first nodes and any second node among multiple second nodes, the electronic device can determine the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first characteristic vector and the third characteristic vector of the first node and the second characteristic vector of the second node, and then determine the scheduling measures of the execution body. Since the accuracy of the first characteristic vector, the second characteristic vector and the third characteristic vector is relatively high, the attack probability and scheduling measures determined thereby are also relatively accurate.
[0049] Based on this, for a certain first node, the electronic device can obtain the same number of attack probabilities corresponding to the first node as the number of second nodes.
[0050] In an embodiment of the present invention, by combining the constructed relationship heterogeneous graph and relationship isomorphic graph, a graph attention network is used to predict the probability of threat traffic in the current network breaking through the current working execution body, and the execution body with a high probability of being broken through is discovered in advance, so as to schedule and replace the corresponding execution body in advance, thereby improving the defense capability of mimicry defense against attacks and reducing the resource loss caused by frequent scheduling.
[0051] In order to better understand the embodiments of the present invention, the execution body scheduling method is described in detail below.
[0052] Optionally, the electronic device constructs a relational heterogeneous graph based on first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and a set of edge relationships corresponding to the plurality of threat flows and the plurality of execution bodies, which may include: the electronic device determines the plurality of threat flows and the plurality of execution bodies based on record information of a given threat flow breaking through the execution body; the electronic device constructs a relational heterogeneous graph based on first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and a set of edge relationships corresponding to the plurality of threat flows and the plurality of execution bodies.
[0053] The record information of a given threat traffic breaking through an execution body mainly refers to the detailed data recorded when the threat traffic successfully attacks one or more execution bodies in a network environment.
[0054] After obtaining the record information of a given threat flow breaking through an execution body, the electronic device can parse the record information to obtain the first statistical characteristics corresponding to each of the multiple threat flows, the second statistical characteristics corresponding to each of the multiple execution bodies, and the attack relationship between the multiple threat flows and the multiple execution bodies; then, based on all the first statistical characteristics, determine the first node corresponding to each of the multiple threat flows, and based on all the second statistical characteristics, determine the second node corresponding to each of the multiple execution bodies; finally, the electronic device constructs a heterogeneous graph of the relationship between the threat flow and the execution body based on all the first nodes, all the second nodes, and the set of edge relationships corresponding to all the attack relationships, which provides strong decision support for the subsequent determination of scheduling measures for the execution body, so as to effectively improve network security.
[0055] It should be noted that when the network environment changes, that is, when multiple threat flows and / or multiple executors change, the above-mentioned relationship heterogeneous graph can be updated according to the nodes corresponding to the changed data; or, when multiple threat flows change, the above-mentioned relationship isomorphic graph can be updated according to the changed first node.
[0056] In some embodiments, an electronic device uses a graph attention network to determine a first eigenvector corresponding to each of multiple first nodes in a relational heterogeneous graph, which may include: performing the following operations for each first node: the electronic device determines a first neighborhood node set corresponding to the first node in the relational heterogeneous graph; the electronic device determines multiple first attention scores of the first node in the current network layer based on the first node, the first neighborhood node set and multiple second nodes; the electronic device determines a first eigenvector of the first node in the current network layer based on the multiple first attention scores and the first node.
[0057] The first neighbor node set may include at least one first neighbor node, the at least one first neighbor node includes the first node itself, and each first neighbor node is a first node having a direct edge relationship with the first node. The first neighbor node set may be represented by Nt.
[0058] The current network layer refers to the layer of the network where the node features are being calculated. l Network layer representation.
[0059] The first attention score is used to characterize the contribution degree of the first neighborhood node set to the first node, that is, the degree of attention of the first node to the first neighborhood node set.
[0060] For a first node among multiple first nodes, the electronic device may first determine the first neighboring node set corresponding to the first node in the relational heterogeneous graph; then, the electronic device determines a first attention score of the first node in the current network layer based on the first node, the first neighboring node set and a second node, based on which the electronic device will obtain as many first attention scores as there are second nodes, and then calculate in combination with the first node to obtain the first feature vector of the first node in the current network layer. In other words, the above first feature vector is related to all first nodes adjacent to the first node.
[0061] For example, in combination with the above Figure 2 The heterogeneous graph shown in the figure shows that for the first node p 1 , the corresponding first neighboring nodes include the first node p 2 ; For the first node p 2 , the corresponding first neighboring nodes include the first node p 1 , the first node p 3 and the first node p 4 ; For the first node p 3 , the corresponding first neighboring nodes include the first node p 2 ; For the first node p 4 , the corresponding first neighboring nodes include the first node p 2 .
[0062] In some embodiments, the first neighborhood node set includes at least one first neighborhood node, and the electronic device determines multiple first attention scores of the first node in the current network layer based on the first node, the first neighborhood node set and multiple second nodes, which may include: for each second node, the electronic device uses a graph attention network to determine the first attention score of the first node in the current network layer based on the first attribute feature vector of the first node in the previous network layer, the second attribute feature vector corresponding to each of the at least one first neighborhood node in the previous network layer, and the third attribute feature vector of the second node in the previous network layer.
[0063] The previous network layer refers to the network layer before the current network layer. l -1 network layer representation. That is, the node features output by the previous network layer are the node features input by the current network layer.
[0064] The first attribute feature vector refers to the feature information of the first node in the relational isomorphism graph after being processed in the previous network layer.
[0065] The second attribute feature vector refers to the feature information of the first neighboring node corresponding to the first node in the relational isomorphism graph after being processed in the previous network layer.
[0066] The third attribute feature vector refers to the feature information of the second node in the relational isomorphism graph after being processed in the previous network layer.
[0067] For the above-mentioned second node, when determining the first attention score of the above-mentioned first node in the current network layer, the electronic device may first use a graph attention network to parse the relational heterogeneous graph to obtain the first attribute feature vector of the first node in the previous network layer, the second attribute feature vector corresponding to at least one first neighboring node in the previous network layer, and the third attribute feature vector of a second node in the previous network layer; then, the first attribute feature vector, all second attribute feature vectors and the third attribute feature vector are calculated to obtain a first attention score with higher accuracy.
[0068] That is to say, combining the above Figure 2 As can be seen from the relationship heterogeneous graph shown, the electronic device can determine 4 first attention scores.
[0069] Optionally, for the i-th first node among the multiple first nodes, the i-th first node is l -1 The first attribute feature vector in the network layer, using the first vector to determine the formula Calculated.
[0070] in, Indicates that the first node of the i-th l-1 The first attribute feature vector in the network layer also represents the l -1 The embedding representation of the first node of the i-th network layer; represents the i-th first node; represents the first encoding matrix to be learned, , Indicates the number of rows , the number of columns is A real matrix of .
[0071] In addition, the first attribute feature vector of the i-th first node in the first network layer is available Indicates that the first attribute feature vector is the initial attribute feature vector corresponding to the i-th first node .
[0072] Optionally, for the i-th first node and the j-th second node among the plurality of second nodes, the i-th first node is l The first attention score in the network layer, calculated using the first score formula Calculated.
[0073] in, Indicates that the first node of the i-th l The first attention score in the network layer; represents the activation function; express l The first weight of the network layer, , Indicates the number of rows , a real matrix with 1 columns; Indicated in l The first feature transformation weights shared in the network layer, , Indicates the number of rows , the number of columns is A real matrix of ; Indicates that the jth second node is l -1 The third attribute feature vector in the network layer; Indicates the first neighbor node set Nt. x The first neighbor node is l -1 The second attribute feature vector in the network layer; Represents a matrix connection.
[0074] In some embodiments, the electronic device determines the first feature vector of the first node in the current network layer based on multiple first attention scores and the first node, which may include: the electronic device determines the first feature vector of the first node in the current network layer based on multiple first attention scores and the first attribute feature vector of the first node in the previous network layer.
[0075] In the process of determining the first feature vector of the first node in the current network layer, the electronic device may calculate multiple first attention scores and the above-mentioned first attribute feature vector to obtain a first feature vector with higher accuracy.
[0076] Optionally, for the i-th first node, the i-th first node is l The first eigenvector in the network layer is generated using the first vector formula Calculated.
[0077] in, Indicates that the first node of the i-th l The first eigenvector in the network layer, , Indicates the number of rows , a real matrix with 1 columns; Represents a linear transformation function.
[0078] In some embodiments, the electronic device uses a graph attention network to determine the second eigenvector corresponding to each of multiple second nodes in a relational heterogeneous graph, which may include: performing the following operations for each second node: the electronic device determines a second neighborhood node set corresponding to the second node in the relational heterogeneous graph; the electronic device determines multiple second attention scores of the second node in the current network layer based on the second node, the second neighborhood node set and multiple first nodes; the electronic device determines the second eigenvector of the second node in the current network layer based on the multiple second attention scores and the second node.
[0079] The second neighbor node set may include at least one second neighbor node, the at least two first neighbor nodes include the second node itself, and each second neighbor node is a second node having a direct edge relationship with the second node. The second neighbor node set may be represented by Ne.
[0080] The second attention score is used to characterize the contribution degree of the second neighborhood node set to the second node, that is, the degree of attention of the second node to the second neighborhood node set.
[0081] For a second node among multiple second nodes, the electronic device may first determine the second neighboring node set corresponding to the second node in the relational heterogeneous graph; then, the electronic device determines a second attention score of the second node in the current network layer based on the second node, the second neighboring node set and a first node, based on which the electronic device will obtain as many second attention scores as there are first nodes, and then calculate in combination with the second node to obtain the second feature vector of the second node in the current network layer. In other words, the above second feature vector is related to all second nodes adjacent to the second node.
[0082] For example, in combination with the above Figure 2 The heterogeneous relationship graph shown in the figure, for the second node q 1 , the corresponding second neighborhood nodes include the second node q 2 ; For the second node q 2 , the corresponding second neighborhood nodes include the second node q 1 , the second node q 3 and the second node q 4 ; For the second node q 3 , the corresponding second neighborhood nodes include the second node q 2 and the second node q 4 ; For the second node q 4 , the corresponding second neighborhood nodes include the second node q 2 , the second node q 3 and the second node q 5 ; For the second node q 5 , the corresponding second neighborhood nodes include the second node q 4 .
[0083] In some embodiments, the second neighborhood node set includes at least one second neighborhood node, and the electronic device determines multiple second attention scores of the second node in the current network layer based on the second node, the second neighborhood node set and multiple first nodes, which may include: for each first node, the electronic device uses a graph attention network to determine the second attention score of the second node in the current network layer based on the third attribute feature vector of the second node in the previous network layer, the fourth attribute feature vector corresponding to each of the at least one second neighborhood node in the previous network layer, and the first attribute feature vector of the first node in the previous network layer.
[0084] The third attribute feature vector refers to the feature information of the second node in the relational isomorphism graph after being processed in the previous network layer.
[0085] The fourth attribute feature vector refers to feature information of the second neighboring node corresponding to the second node in the relational isomorphism graph after being processed in the previous network layer.
[0086] The first attribute feature vector refers to the feature information of the first node in the relational isomorphism graph after being processed in the previous network layer.
[0087] For the above-mentioned first node, when determining the second attention score of the above-mentioned second node in the current network layer, the electronic device may first use a graph attention network to parse the relational heterogeneous graph to obtain the third attribute feature vector of the second node in the previous network layer, the fourth attribute feature vector corresponding to at least one second neighboring node in the previous network layer, and the first attribute feature vector of a first node in the previous network layer; then, the third attribute feature vector, all fourth attribute feature vectors and the first attribute feature vector are calculated to obtain a second attention score with higher accuracy.
[0088] That is to say, combining the above Figure 2 As can be seen from the relationship heterogeneous graph shown, the electronic device can determine 5 second attention scores.
[0089] Optionally, for the mth second node among the multiple second nodes, the mth second node is l -1 The third attribute feature vector in the network layer, using the second vector to determine the formula Calculated.
[0090] in, Indicates that the mth second node is l -1 The third attribute feature vector in the network layer also represents the l -1 The mth second node embedding representation of the network layer; represents the mth second node; represents the second encoding matrix to be learned, , Indicates the number of rows , the number of columns is A real matrix of .
[0091] In addition, the third attribute feature vector of the mth second node in the first network layer is available Indicates that the third attribute feature vector is the initial attribute feature vector corresponding to the mth second node .
[0092] Optionally, for the mth second node and the nth first node among the plurality of first nodes, the mth second node is l The second attention score in the network layer is calculated using the second score formula Calculated.
[0093] in, The mth second node is l Second attention score in the network layer; express l The second weight of the network layer, , Indicates the number of rows , a real matrix with columns of 1, ; exist l The second feature transformation weights shared in the network layer, , Indicates the number of rows , the number of columns is A real matrix of ; Indicates that the nth first node is l -1 The first attribute feature vector in the network layer; Represents the first node in the second neighborhood node set Ne. y The second neighbor node is l -1 The corresponding fourth attribute eigenvector in the network layer.
[0094] In some embodiments, the electronic device determines the second feature vector of the second node in the current network layer based on multiple second attention scores and the second nodes, which may include: the electronic device determines the second feature vector of the second node in the current network layer based on multiple second attention scores and the third attribute feature vector of the second node in the previous network layer.
[0095] In the process of determining the second eigenvector of the second node in the current network layer, the electronic device may calculate multiple second attention scores and the third attribute eigenvector to obtain a second eigenvector with higher accuracy.
[0096] Optionally, for the mth second node, the mth second node is l The second eigenvector in the network layer is generated using the second vector generation formula Calculated.
[0097] in, Indicates that the mth second node is l The second eigenvector in the network layer, , Indicates the number of rows , the number of columns is A real matrix of .
[0098] In some embodiments, the electronic device uses a graph attention network to determine the third eigenvector corresponding to each of multiple first nodes in a relational isomorphism graph, which may include: performing the following operations for each first node: the electronic device determines at least one third neighboring node corresponding to the first node in the relational isomorphism graph; the electronic device determines at least one third attention score of the first node in the current network layer based on the first node and at least one third neighboring node; the electronic device determines the third eigenvector of the first node in the current network layer based on the at least one third attention score and the first node.
[0099] The at least one third neighboring node includes the first node itself, and each third neighboring node is a third node having a direct edge relationship with the third node. The third neighboring node set can be represented by Nz.
[0100] The third attention score is used to characterize the contribution degree of the third neighborhood node set to the third node, that is, the degree of attention of the third node to the third neighborhood node set.
[0101] For a first node among multiple first nodes, the electronic device may first determine at least one third neighboring node corresponding to the first node in the relational isomorphism graph; then, the electronic device determines a third attention score of the first node in the current network layer based on the first node, a third neighboring node and a second node, based on which the electronic device will obtain as many third attention scores as there are third neighboring nodes, and then calculate in combination with the first node to obtain the third eigenvector of the first node in the current network layer. In other words, the third eigenvector is related to all third nodes adjacent to the third node.
[0102] For example, in combination with the above Figure 3 The relationship isomorphism graph shown, for the first node p 1 , the corresponding third neighboring nodes include the first node p 3 ; For the first node p 2 , the corresponding third neighboring nodes include the first node p 4 ; For the first node p 3 , the corresponding third neighboring nodes include the first node p 1 , the first node p 4 and the first node p 5 ; For the first node p 4 , the corresponding third neighboring nodes include the first node p 2 , the first node p 5 and the first node p 6 ; For the first node p 5 , the corresponding third neighboring nodes include the first node p 3 and the first node p 4 ; For the first node p 6 , the corresponding third neighboring nodes include the first node p 4 .
[0103] In some embodiments, the electronic device determines at least one third attention score of the first node in the current network layer based on the first node and at least one third neighboring node, which may include: for each third neighboring node, the electronic device uses a graph attention network to determine the third attention score of the first node in the current network layer based on the fifth attribute feature vector of the first node in the previous network layer, the sixth attribute feature vector of the third neighboring node in the previous network layer, and the seventh attribute feature vector corresponding to each of at least one third neighboring node in the previous network layer.
[0104] The fifth attribute feature vector refers to the feature information of the first node in the relational isomorphism graph after being processed in the previous network layer.
[0105] The sixth attribute feature vector refers to feature information of the third neighboring node corresponding to the first node in the relational isomorphism graph after being processed in the previous network layer.
[0106] The seventh attribute feature vector refers to the feature information of the second node in the relational isomorphism graph after being processed in the previous network layer.
[0107] For a certain third-neighborhood node, when the electronic device determines the third attention score of a certain first node in the current network layer, it can first use the graph attention network to parse the relational isomorphism graph to obtain the fifth attribute feature vector of the first node in the previous network layer, the sixth attribute feature vector of the third neighboring node in the previous network layer, and the seventh attribute feature vector corresponding to at least one third neighboring node in the previous network layer; then, the fifth attribute feature vector, the sixth attribute feature vector and all seventh attribute feature vectors are calculated to obtain a third attention score with higher accuracy.
[0108] Optionally, for the first of the multiple first nodes k The first node, k The first node is l -1 The fifth attribute feature vector in the network layer, using the third vector to determine the formula Calculated.
[0109] in, No. k The first node is l -1 The fifth attribute feature vector in the network layer; Indicates k The first node; represents the third encoding matrix to be learned, , Indicates the number of rows , the number of columns is A real matrix of .
[0110] In addition, k The fifth attribute feature vector of the first node in the first network layer is available Indicates that the first attribute feature vector is k The initial attribute feature vector corresponding to the first node .
[0111] Optionally, for at least one third neighboring node v The third neighbor node, v The third neighbor node is l -1 The sixth attribute feature vector in the network layer, using the fourth vector to determine the formula Calculated.
[0112] in, No. v The third neighbor node is l -1 The sixth attribute feature vector in the network layer; Indicates v third neighbor nodes; represents the fourth encoding matrix to be learned, , Indicates the number of rows , the number of columns is A real matrix of .
[0113] Optional, for k The first node and the v The third neighbor node, k The first node is l The third attention score in the network layer is calculated using the third score formula Calculated.
[0114] in, Indicates k The first node is l The third attention score in the network layer; express l The third weight of the network layer, , Indicates the number of rows , a real matrix with columns of 1, ; exist l The third feature transformation weights shared in the network layer, , Indicates the number of rows , the number of columns is A real matrix of ; Indicates v The third neighbor node isl -1 The sixth attribute feature vector in the network layer; Indicates z The third neighbor node is l -1 The corresponding seventh attribute feature vector in the network layer.
[0115] In some embodiments, the electronic device determines the third feature vector of the first node in the current network layer based on at least one third attention score and the first node, which may include: the electronic device determines the third feature vector of the first node in the current network layer based on at least one third attention score and the fifth attribute feature vector of the first node in the previous network layer.
[0116] In the process of determining the third eigenvector of the first node in the current network layer, the electronic device may calculate at least one third attention score and the fifth attribute eigenvector to obtain a third eigenvector with higher accuracy.
[0117] Optional, for k The first node, k The first node is l The third eigenvector in the network layer is generated using the third vector formula Calculated.
[0118] also, Indicates k The first node is l The third eigenvector in the network layer, , Indicates the number of rows , a real matrix with columns equal to 1.
[0119] In some embodiments, the electronic device determines the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector, the third feature vector and the second feature vector corresponding to the first node and the second feature vector corresponding to the second node, which may include: the electronic device determines the first target feature based on the first feature vector, the third feature vector and the first training parameter weight corresponding to the first node; the electronic device determines the second target feature based on the second feature vector and the second training parameter weight corresponding to the second node; the electronic device determines the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first target feature and the second target feature.
[0120] Among them, the first target feature can also be called the first latent feature, which is a vector; the second target feature can also be called the second latent feature, which is also a vector.
[0121] The electronic device can calculate the first feature vector, the third feature vector and the first training parameter weight corresponding to the first node to obtain the first target feature, and calculate the second feature vector and the second training parameter weight corresponding to the second node to obtain the second target feature, and then calculate in combination with the first target feature to obtain the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node, and the accuracy of the attack probability is relatively high.
[0122] It should be noted that there is no limit to the timing of the electronic device determining the first target feature and determining the second target feature.
[0123] Optionally, for the sth first node among the multiple first nodes, the first target feature adopts the first target formula in the fully connected neural network Calculated.
[0124] in, represents the first target feature; Represents a function; Indicates that the sth first node is l The first eigenvector in the network layer; Indicates that the sth first node is l The third eigenvector in the network layer; represents the first training parameter weight, , Indicates the number of rows , the number of columns is A real matrix of .
[0125] Optionally, for the oth second node among the plurality of second nodes, the second target feature adopts the second target formula in the fully connected neural network: Calculated.
[0126] in, represents the second target feature; Indicates that the oth second node is l The second eigenvector in the network layer; represents the second training parameter weight, , Indicates the number of rows , the number of columns is A real matrix of , .
[0127] Optionally, for the sth first node and the oth second node, the attack probability is calculated using the inner product, and the specific probability calculation formula is: Calculated.
[0128] in, It represents the attack probability of the threat traffic corresponding to the s-th first node on the executable body corresponding to the o-th second node.
[0129] Based on this, the electronic device can calculate the attack probability of any threat traffic on any executor.
[0130] In some embodiments, the electronic device determines the scheduling measures of the execution body corresponding to the second node based on the attack probability, which may include: when the attack probability is greater than a preset probability threshold, the electronic device determines the scheduling replacement of the execution body corresponding to the second node as the scheduling measure.
[0131] After determining the power probability, the electronic device can compare the attack probability with a preset probability threshold: if the attack probability is less than or equal to the preset probability threshold, it means that the corresponding threat traffic has not attacked the corresponding execution body, or the attack degree of the corresponding threat traffic on the corresponding execution body is small and cannot affect the normal operation of the execution body. At this time, no operation is performed; if the attack probability is greater than the preset probability threshold, it means that the corresponding threat traffic is attacking the corresponding execution body, and the attack degree is small, which has affected the normal operation of the execution body to a large extent. At this time, it is necessary to execute a scheduling plan to schedule and replace the execution body in advance, or schedule and clean it, thereby improving the safe operation level of the execution body.
[0132] It should be noted that the process in which the electronic device compares the attack probability with the preset probability threshold can be regarded as a binary classification problem of predicting whether the executor will be breached by the threat traffic. The whole process can be used as a prediction model. At this time, the binary cross entropy loss function commonly used in binary classification problems can be used to optimize the prediction model, update the model parameters, and obtain the trained prediction model. The trained prediction model has a higher accuracy.
[0133] Among them, the binary cross entropy loss function is: .
[0134] represents the loss value; N Represents the number of input samples. Each input sample is a pair of threat traffic and execution body (i.e., attack relationship) in the heterogeneous graph. Therefore, N can also represent the number of edges in the heterogeneous graph. Indicates the number of input samples N The true label corresponding to the i-th input sample in , i.e., attack or not attack; Represents the attack probability corresponding to the i-th true label.
[0135] For example, Figure 4 is a flow chart of the execution body scheduling method provided by the present invention, such as Figure 4As shown, the method includes: constructing a relational heterogeneous graph based on first nodes corresponding to each of the plurality of threat flows and second nodes corresponding to each of the plurality of execution bodies; determining initial attribute feature vectors corresponding to each of the plurality of threat flows and initial attribute feature vectors corresponding to each of the plurality of execution bodies; using a graph attention network, determining first feature vectors corresponding to each of the plurality of first nodes in the current network layer based on the initial attribute feature vectors corresponding to each of the plurality of threat flows, and determining second feature vectors corresponding to each of the plurality of second nodes in the current network layer based on the initial attribute feature vectors corresponding to each of the plurality of execution bodies.
[0136] According to the first nodes corresponding to each of the multiple threat flows, a relational isomorphism graph is constructed; the initial attribute feature vectors corresponding to each of the multiple threat flows are determined; and a graph attention network is used to determine the third feature vectors corresponding to each of the multiple first nodes in the current network layer according to the initial attribute feature vectors corresponding to each of the multiple threat flows.
[0137] For the s-th first node and the o-th second node, determine the first target feature corresponding to the s-th first node based on the first feature vector and the third feature vector corresponding to the s-th first node, and determine the second target feature corresponding to the o-th second node based on the second feature vector corresponding to the o-th second node; perform an inner product operation on the first target feature and the second target feature to determine the attack probability of the threat traffic corresponding to the s-th first node on the executable body corresponding to the o-th second node.
[0138] When the attack probability is greater than a preset probability threshold, the execution body corresponding to the o-th second node is scheduled to be replaced.
[0139] The entire process can perceive the current network environment in real time. When the network environment changes, by combining the constructed relationship heterogeneous graph and relationship isomorphic graph, the graph attention network is used to predict the probability of threat traffic in the current network breaking through the current working executor, and the executors with a high probability of being broken through are discovered in advance, so as to schedule and replace the corresponding executors in advance, thereby improving the defense capability of mimicry defense against attacks and reducing the resource loss caused by frequent scheduling.
[0140] The following is a description of the execution body scheduling device provided by the present invention. The execution body scheduling device described below and the execution body scheduling method described above can be referenced to each other.
[0141] Figure 5 is a schematic diagram of the structure of the execution body scheduling device provided by the present invention, such as Figure 5 As shown, the device includes the following: a data processing module 501 and a solution determination module 502.
[0142] The data processing module 501 is used to construct a relational heterogeneous graph based on the first nodes corresponding to each of the plurality of threat flows, the second nodes corresponding to each of the plurality of execution bodies, and the set of edge relationships corresponding to the plurality of threat flows and the plurality of execution bodies; and use a graph attention network to determine the first feature vectors corresponding to each of the plurality of first nodes in the relational heterogeneous graph, and the second feature vectors corresponding to each of the plurality of second nodes; construct a relational isomorphic graph based on the node similarities between the plurality of first nodes and the plurality of first nodes; and use the graph attention network to determine the third feature vectors corresponding to each of the plurality of first nodes in the relational isomorphic graph.
[0143] The plan determination module 502 is used to determine, for each first node among the multiple first nodes and each second node among the multiple second nodes, the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; and determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
[0144] Optionally, the data processing module 501 is specifically used to perform the following operations for each first node: in the relational heterogeneous graph, determine a first neighboring node set corresponding to the first node; determine a plurality of first attention scores of the first node in the current network layer based on the first node, the first neighboring node set and the plurality of second nodes; and determine a first feature vector of the first node in the current network layer based on the plurality of first attention scores and the first node.
[0145] Optionally, the first neighborhood node set includes at least one first neighborhood node, and the data processing module 501 is specifically used to determine, for each second node, a first attention score of the first node in the current network layer based on the first attribute feature vector of the first node in the previous network layer, the second attribute feature vector corresponding to each of the at least one first neighborhood node in the previous network layer, and the third attribute feature vector of the second node in the previous network layer, using the graph attention network.
[0146] Optionally, the data processing module 501 is specifically used to determine the first feature vector of the first node in the current network layer based on the multiple first attention scores and the first attribute feature vector of the first node in the previous network layer.
[0147] Optionally, the data processing module 501 is specifically used to perform the following operations for each second node: determining a second neighborhood node set corresponding to the second node in the relational heterogeneous graph; determining a plurality of second attention scores of the second node in the current network layer based on the second node, the second neighborhood node set and the plurality of first nodes; and determining a second feature vector of the second node in the current network layer based on the plurality of second attention scores and the second node.
[0148] Optionally, the second neighborhood node set includes at least one second neighborhood node, and the data processing module 501 is specifically used to determine, for each first node, a second attention score of the second node in the current network layer based on the third attribute feature vector of the second node in the previous network layer, the fourth attribute feature vector corresponding to each of the at least one second neighborhood node in the previous network layer, and the first attribute feature vector of the first node in the previous network layer, using the graph attention network.
[0149] Optionally, the data processing module 501 is specifically used to determine the second feature vector of the second node in the current network layer based on the multiple second attention scores and the third attribute feature vector of the second node in the previous network layer.
[0150] Optionally, the data processing module 501 is specifically used to perform the following operations for each first node: in the relationship isomorphism graph, determine at least one third neighboring node corresponding to the first node; based on the first node and the at least one third neighboring node, determine at least one third attention score of the first node in the current network layer; based on the at least one third attention score and the first node, determine a third eigenvector of the first node in the current network layer.
[0151] Optionally, the data processing module 501 is specifically used to determine, for each third neighboring node, a third attention score of the first node in the current network layer using the graph attention network according to the fifth attribute feature vector of the first node in the previous network layer, the sixth attribute feature vector of the third neighboring node in the previous network layer, and the seventh attribute feature vector corresponding to each of the at least one third neighboring node in the previous network layer.
[0152] Optionally, the data processing module 501 is specifically used to determine the third feature vector of the first node in the current network layer based on the at least one third attention score and the fifth attribute feature vector of the first node in the previous network layer.
[0153] Optionally, the scheme processing module 502 is specifically used to determine the first target feature based on the first feature vector, the third feature vector and the first training parameter weight corresponding to the first node; determine the second target feature based on the second feature vector and the second training parameter weight corresponding to the second node; and determine the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first target feature and the second target feature.
[0154] Optionally, the solution processing module 502 is specifically configured to determine, when the attack probability is greater than a preset probability threshold, a scheduling replacement of the execution body corresponding to the second node as the scheduling measure.
[0155] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the body scheduling method.
[0156] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0157] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the execution body scheduling method provided by the above methods.
[0158] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the execution body scheduling method provided by the above methods.
[0159] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0160] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for scheduling an execution body, characterized in that: include: Constructing a heterogeneous relationship graph according to first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and edge relationship sets corresponding to the plurality of threat flows and the plurality of execution bodies; And using a graph attention network, determine a first eigenvector corresponding to each of the plurality of first nodes in the relational heterogeneous graph, and a second eigenvector corresponding to each of the plurality of second nodes; Constructing a relational isomorphism graph according to the plurality of first nodes and the node similarities between the plurality of first nodes; and using the graph attention network to determine a third eigenvector corresponding to each of the plurality of first nodes in the relational isomorphism graph; For each first node among the multiple first nodes and each second node among the multiple second nodes, determine the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
2. The execution body scheduling method according to claim 1, characterized in that: The method of using a graph attention network to determine first feature vectors corresponding to each of the plurality of first nodes in the relational heterogeneous graph includes: The following operations are performed for each first node: In the relational heterogeneous graph, determining a first neighborhood node set corresponding to the first node; Determine, according to the first node, the first neighborhood node set, and the plurality of second nodes, a plurality of first attention scores of the first node in the current network layer; According to the multiple first attention scores and the first node, a first feature vector of the first node in the current network layer is determined.
3. The execution body scheduling method according to claim 2, characterized in that: The first neighborhood node set includes at least one first neighborhood node, and determining a plurality of first attention scores of the first node in the current network layer according to the first node, the first neighborhood node set, and the plurality of second nodes includes: For each of the second nodes, the graph attention network is used to determine the first attention score of the first node in the current network layer according to the first attribute feature vector of the first node in the previous network layer, the second attribute feature vector corresponding to each of the at least one first neighboring nodes in the previous network layer, and the third attribute feature vector of the second node in the previous network layer.
4. The execution body scheduling method according to claim 3, characterized in that: The determining, according to the plurality of first attention scores and the first node, a first feature vector of the first node in the current network layer comprises: According to the multiple first attention scores and the first attribute feature vector of the first node in the previous network layer, a first feature vector of the first node in the current network layer is determined.
5. The execution body scheduling method according to claim 1, characterized in that: Using a graph attention network, determining second feature vectors corresponding to each of the plurality of second nodes in the relational heterogeneous graph includes: The following operations are performed for each second node: In the relational heterogeneous graph, determining a second neighborhood node set corresponding to the second node; Determine, according to the second node, the second neighborhood node set, and the plurality of first nodes, a plurality of second attention scores of the second node in the current network layer; According to the multiple second attention scores and the second node, a second feature vector of the second node in the current network layer is determined.
6. The execution body scheduling method according to claim 5, characterized in that: The second neighborhood node set includes at least one second neighborhood node, and determining a plurality of second attention scores of the second node in the current network layer according to the second node, the second neighborhood node set, and the plurality of first nodes includes: For each of the first nodes, the graph attention network is used to determine the second attention score of the second node in the current network layer according to the third attribute feature vector of the second node in the previous network layer, the fourth attribute feature vector corresponding to each of the at least one second neighboring nodes in the previous network layer, and the first attribute feature vector of the first node in the previous network layer.
7. The execution body scheduling method according to claim 5, characterized in that: The determining, according to the plurality of second attention scores and the second node, a second feature vector of the second node in the current network layer comprises: A second feature vector of the second node in the current network layer is determined according to the multiple second attention scores and the third attribute feature vector of the second node in the previous network layer.
8. The execution body scheduling method according to claim 1, characterized in that: The step of using the graph attention network to determine the third feature vectors corresponding to each of the plurality of first nodes in the relational isomorphism graph comprises: The following operations are performed for each first node: In the relational isomorphism graph, determining at least one third neighboring node corresponding to the first node; Determining at least one third attention score of the first node in the current network layer according to the first node and the at least one third neighboring node; Determine a third feature vector of the first node in the current network layer based on the at least one third attention score and the first node.
9. The execution body scheduling method according to claim 8, characterized in that: The determining, according to the first node and the at least one third neighboring node, at least one third attention score of the first node in the current network layer includes: For each third neighboring node, the graph attention network is used to determine the third attention score of the first node in the current network layer according to the fifth attribute feature vector of the first node in the previous network layer, the sixth attribute feature vector of the third neighboring node in the previous network layer, and the seventh attribute feature vector corresponding to each of the at least one third neighboring node in the previous network layer.
10. The execution body scheduling method according to claim 8, characterized in that: The determining, according to the at least one third attention score and the first node, a third feature vector of the first node in the current network layer comprises: A third feature vector of the first node in the current network layer is determined based on the at least one third attention score and the fifth attribute feature vector of the first node in the previous network layer.
11. The execution body scheduling method according to any one of claims 1 to 10, characterized in that: The determining, according to the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node, the attack probability of the threat flow corresponding to the first node on the executable body corresponding to the second node includes: Determine a first target feature according to a first feature vector, a third feature vector and a first training parameter weight corresponding to the first node; Determine a second target feature according to a second feature vector corresponding to the second node and a second training parameter weight; An attack probability of the threat traffic corresponding to the first node on the executable body corresponding to the second node is determined according to the first target feature and the second target feature.
12. The execution body scheduling method according to any one of claims 1 to 10, characterized in that: The determining, according to the attack probability, a scheduling measure of an execution body corresponding to the second node includes: When the attack probability is greater than a preset probability threshold, scheduling and replacing the execution body corresponding to the second node is determined as the scheduling measure.
13. An execution body scheduling device, characterized in that: include: A data processing module, configured to construct a heterogeneous relationship graph according to first nodes corresponding to each of the plurality of threat flows, second nodes corresponding to each of the plurality of execution bodies, and edge relationship sets corresponding to the plurality of threat flows and the plurality of execution bodies; A graph attention network is used to determine first feature vectors corresponding to each of the plurality of first nodes in the relational heterogeneous graph, and second feature vectors corresponding to each of the plurality of second nodes; a relational isomorphic graph is constructed according to the node similarities between the plurality of first nodes and the plurality of first nodes; and the graph attention network is used to determine third feature vectors corresponding to each of the plurality of first nodes in the relational isomorphic graph; A scheme determination module is used to determine, for each first node among the multiple first nodes and each second node among the multiple second nodes, the attack probability of the threat traffic corresponding to the first node on the execution body corresponding to the second node based on the first feature vector and the third feature vector corresponding to the first node and the second feature vector corresponding to the second node; and determine the scheduling measures for the execution body corresponding to the second node based on the attack probability.
14. An electronic 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, the execution body scheduling method according to any one of claims 1 to 12 is implemented.
15. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the execution body scheduling method according to any one of claims 1 to 12 is implemented.
Citation Information
Patent Citations
Attack scene reconstruction system based on heterogeneous graph attention network
CN116074092A
Threat intelligence knowledge graph processing method and device, equipment and storage medium
CN116150392A
Encrypted malicious traffic detection method and device, equipment and storage medium
CN117176432A
Grid attack node selection method based on graph attention network
CN117834295A
APT attack tracing method and system based on heterogeneous graph attention network
CN117938424A
