Task decision-making method of service network and related device

Abstract

The embodiment of the invention provides a task decision-making method of a service network and a related device, and relates to the technical field of data processing. The method comprises: obtaining a target topological graph corresponding to a service network, the target topological graph comprising a connection relationship among a plurality of nodes and a node feature of each node, the plurality of nodes comprising a first node corresponding to an edge server, the node feature of the first node being used for describing a condition of a corresponding slave device; performing node classification based on the target topological graph by using a node classifier to obtain sub-task categories corresponding to the first nodes; according to the sub-task category corresponding to each first node, multiple sub-tasks split from the target total task are allocated to the corresponding edge servers, and the category of the sub-tasks allocated to one edge server is the sub-task category corresponding to the edge server. Therefore, the analysis and control of the complex network can be realized under the conditions of occupying less computing resources and consuming less time.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a task decision method and related devices for a service network. Background Art

[0002] A general service network refers to a network structure composed of multiple service providers, users and other related parties, which aims to provide specific services through cooperation and collaboration. Its main purpose is to improve the efficiency and quality of services and meet the needs of users. Since the service network is a complex network with massive high-dimensional data and complex structure, the task intelligent decision-making of the service network has always attracted much attention, and related technical methods have been continuously proposed. Some people have proposed to use game theory to control the operation of complex networks and make task decisions, and some people have proposed to use reinforcement learning algorithms to make task decisions for service networks.

[0003] Due to the complexity of service networks in real-world applications, existing methods are still unable to respond to these situations in a timely manner, such as changes in overall task requirements, updates to service provider status, and inability to provide services due to problems with service providers. These problems require algorithms that can perform real-time analysis and decision-making, which is lacking in existing algorithms. Game theory and other methods target service networks that are mostly simple in structure and small in size, making it difficult to analyze and control more complex networks in actual situations. Although methods such as reinforcement learning have better effects, they often require a long time to analyze and make decisions for large networks, and the computing resources required for algorithm operation are also large, making them unsuitable for application in actual situations. Therefore, how to provide a method that occupies less computing resources, takes less time, and can achieve analysis and control of complex networks is a technical problem that technicians in this field urgently need to solve. Summary of the invention

[0004] The embodiments of the present application provide a task decision method and related devices for a service network, which can analyze and control a complex network while occupying less computing resources and taking less time.

[0005] The embodiments of the present application can be implemented as follows: In a first aspect, an embodiment of the present application provides a task decision method for a service network, wherein the service network includes a plurality of edge servers and a slave device that is communicatively connected to the edge servers, and the method includes: Obtaining a target topology map corresponding to the service network, wherein the target topology map includes connection relationships between multiple nodes and node characteristics of each node, the multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding slave device; Using a node classifier to classify nodes based on the target topology graph, obtaining a subtask category corresponding to each first node, wherein the subtask category corresponding to each first node is a type of multiple subtasks corresponding to the target total task to be assigned; According to the subtask categories corresponding to the first nodes, the multiple subtasks split from the target overall task are allocated to the corresponding edge servers, wherein the type of the subtask allocated to an edge server is the subtask category corresponding to the edge server.

[0006] In a second aspect, an embodiment of the present application provides a task decision device for a service network, wherein the service network includes a plurality of edge servers and a slave device that is communicatively connected to the edge servers, and the device includes: A topology map obtaining module, used to obtain a target topology map corresponding to the service network, wherein the target topology map includes a connection relationship between multiple nodes and node characteristics of each node, the multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding slave device; A classification module, used for performing node classification based on the target topology graph using a node classifier to obtain a subtask category corresponding to each first node, wherein the subtask category corresponding to each first node is a type of multiple subtasks corresponding to the target total task to be assigned; The allocation module is used to allocate multiple subtasks split from the target overall task to corresponding edge servers according to the subtask categories corresponding to each first node, wherein the type of subtask allocated to an edge server is the subtask category corresponding to the edge server.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, including a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor can execute the machine executable instructions to implement the task decision method of the service network described in the aforementioned embodiment.

[0008] In a fourth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the task decision method of the service network as described in the aforementioned embodiment is implemented.

[0009] The task decision method and related device of the service network provided by the embodiment of the present application first obtain a target topology map corresponding to the service network including multiple edge servers and subordinate devices of the edge servers, the target topology map includes the connection relationship between multiple nodes and the node characteristics of each node, the multiple nodes include the first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding subordinate device; then, a node classifier is used to classify the nodes based on the above target topology map to obtain the subtask category corresponding to each first node, and the subtask category corresponding to each first node is the type of multiple subtasks corresponding to the target total task to be assigned; finally, according to the subtask category corresponding to each first node, the multiple subtasks split from the target total task are assigned to the corresponding edge server, and the type of the subtask assigned to an edge server is the subtask category corresponding to the edge server. In this way, the analysis and control of complex networks can be achieved while occupying less computing resources and taking less time. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 A block diagram of an electronic device provided in an embodiment of the present application; Figure 2 A flowchart of a task decision method for a service network provided in an embodiment of the present application; Figure 3 A schematic diagram of the corresponding relationship between the service network and the graph neural network provided in the embodiment of the present application; Figure 4 for Figure 2 A schematic flow chart of the sub-steps included in step S300; Figure 5 for Figure 2 A schematic flow chart of the sub-steps included in step S100; Figure 6 for Figure 5 A schematic flow chart of the sub-steps included in sub-step S120; Figure 7 for Figure 6 A schematic flow chart of the sub-steps included in sub-step S124; Figure 8 for Figure 7 A schematic flow chart of the sub-steps included in sub-step S1241; Fig. 9A block diagram of a task decision device for a service network provided in an embodiment of the present application.

[0012] Icons: 100 - electronic device; 110 - memory; 120 - processor; 130 - communication unit; 200 - task decision device of service network; 210 - topology map acquisition module; 220 - classification module; 230 - allocation module. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0014] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0015] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0016] The service network has a complex composition and structure, a large amount of sparse data with low information density, and is susceptible to subtle interference. These characteristics make data extraction and analysis in the service network difficult and the system difficult to manage. This application proposes a task allocation problem in a service network based on a graph neural network, aiming to propose an effective algorithm to solve the task allocation problem of a complex system on the basis of overcoming the above difficulties.

[0017] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0018] Please refer to Figure 1 , Figure 1 A block diagram of an electronic device 100 provided in an embodiment of the present application. The electronic device 100 may be a core server in a server network, and the electronic device 100 may split the target total task that needs to be processed by the service network into subtasks and divide them into edge servers in the service network. The electronic device 100 may include a memory 110, a processor 120, and a communication unit 130. The memory 110, the processor 120, and the communication unit 130 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these elements may be electrically connected to each other via one or more communication buses or signal lines.

[0019] The memory 110 is used to store programs or data. The memory 110 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0020] The processor 120 is used to read / write data or programs stored in the memory 110 and execute corresponding functions. For example, the memory 110 stores a task decision device 200 of a service network, and the task decision device 200 of the service network includes at least one software function module that can be stored in the memory 110 in the form of software or firmware. The processor 120 executes various functional applications and data processing by running software programs and modules stored in the memory 110, such as the task decision device 200 of the service network in the embodiment of the present application, that is, realizing the task decision method of the service network in the embodiment of the present application.

[0021] The communication unit 130 is used to establish a communication connection between the 110 and other communication terminals through a network, and to send and receive data through the network.

[0022] It should be understood that Figure 1 The structure shown is only a schematic diagram of the structure of the electronic device 100. The electronic device 100 may also include Figure 1 More or fewer components as shown, or with Figure 1Different configurations shown. Figure 1 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0023] Please refer to Figure 2 , Figure 2 A flow chart of a task decision method for a service network provided in an embodiment of the present application. The method can be applied to the above-mentioned electronic device. The specific flow of the task decision method for the service network is described in detail below. In this embodiment, the method may include steps S100 to S300.

[0024] Step S100, obtaining a target topology map corresponding to the service network.

[0025] Step S200: Using a node classifier to classify nodes based on the target topology graph, to obtain a subtask category corresponding to each first node.

[0026] Step S300: Allocate multiple subtasks split from the target overall task to corresponding edge servers according to the subtask categories corresponding to the first nodes.

[0027] Since topological graphs have the ability to store large amounts of sparse data and simplify complex relationships, abstracting service networks into topological graphs can facilitate their management. Graph neural networks can handle complex topological structures, and are good at analyzing data and extracting information from massive amounts of sparse data. Therefore, graph neural networks are widely studied and applied in the context of service networks, and have obvious advantages, such as intelligent management of transportation systems, resource scheduling, and so on.

[0028] In this embodiment, the service network is constructed as a multi-layer structure to facilitate network supervision. The service network is similar to the multi-layer structure of the industrial Internet. The top layer of the service network is the core server, the middle layer is the edge server, and the bottom layer is the functional module in the service network (i.e., the underlying hardware device). The entire service network is supervised by the core server (cloud server), which is responsible for receiving and processing the data sent by the edge server, and sending subsequent task instructions to the edge server after analyzing the data.

[0029] In this embodiment, the last two layers of the service network are constructed as a topology map, each edge server is a node of the topology map, and the communication channels between the edge servers are set as edges. In addition to receiving instructions from the core server, the edge server can also exchange information with servers of the same level (i.e., other edge servers) as needed to achieve the effect of task cooperation. In addition, each edge server manages one or more functional modules, which are its subordinate modules (i.e., subordinate devices). Each edge server can send task instructions to its subordinate devices and receive their device information. Among them, the device information can be used to indicate the working status of the subordinate device (for example, the CPU occupancy of the subordinate device, etc.) and corresponding data (for example, what functional devices the subordinate device has, etc.). The edge server can determine which subordinate devices it is connected to, the working status of each subordinate device and the corresponding data, etc. based on the received device information. The device information of the subordinate devices of each edge server constitutes the node feature vector of the corresponding node of the edge server.

[0030] Since this embodiment solves the task allocation problem in the service network as a node classification problem, the classification method in the core server as the supervisor of the system can be regarded as a node classifier, while the edge server layer is the topology graph that needs to be learned, and the slave device layer appears as a graph feature. Figure 3 The schematic diagram of the correspondence between the service network and the graph neural network shown is that the core server layer corresponds to the node classifier, the edge server layer corresponds to the topology graph, the edge servers are the nodes in the topology graph, the interaction relationship between the edge servers corresponds to the edges in the graph, and the slave device layer corresponds to the node features.

[0031] Since the task allocation problem is handled by the node classification problem, we first briefly introduce the topological graph features and the graph neural network node classification model. Indicates that is the set of nodes in the graph, is the set of edges in the graph, N is the number of nodes in the graph, and M is the number of edges in the graph. Represents an undirected topological graph, where A is The adjacency matrix of , which is the connection relationship between node i and node j, is expressed as: , for The node feature matrix is ​​composed of the feature vectors of N nodes, and F is the feature dimension of each node.

[0032] In semi-supervised learning, the adjacency matrix A of the graph, the node features X and the label set are usually used. (p is the number of nodes with labels) Train the graph neural network model to get the model , is the set of learnable parameters of the model. Input the graph adjacency matrix A and node features X into the model You can get the prediction result set of the node ,in, (K is the number of categories of nodes in the graph, which can be expressed as , represents the probability that the node’s label is j ( ).here, It is a node classifier. In this embodiment, As a node classifier used when classifying nodes into subtask categories, it is understandable that the node classifier can also be obtained through a full supervision method, and the method for obtaining the node classifier is not specifically limited here.

[0033] In this embodiment, when it is necessary to allocate the target total task to be allocated, the target topology map corresponding to the service network can be obtained in any way. The target topology map includes multiple nodes, the connection relationship between the multiple nodes, and the node characteristics of each node. The multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the subordinate device of the edge server corresponding to the first node. The node characteristics of the first node are the node feature vectors of the nodes corresponding to the edge server corresponding to the first node. It can be understood that if there is an edge between two first nodes in the target topology, it means that there is an interactive relationship between the two edge servers corresponding to the two first nodes. The target topology map describes the edge servers in the service network, the connection status between the edge servers, the subordinate device status of each edge server, etc.

[0034] When the target topology map is obtained, the target topology map can be analyzed using a pre-obtained node classifier, thereby classifying each node in the target topology map to obtain the subtask category of each first node in the target topology map. The subtask category corresponding to each first node is the type of multiple subtasks corresponding to the target overall task to be assigned. That is, the multiple subtask categories corresponding to the target overall task can be determined in advance, and the node classifier can be used to determine which subtask category each first node corresponds to.

[0035] The input target topology can be recorded as , the node classifier is denoted as , where the communication relationship between nodes is A, the node state (and node characteristics) is X, and the subtask category set obtained by dividing the target total task is the label set That is, the label set used by the node classifier is the category set of subtasks obtained by dividing the target total task. In this way, the subtask category corresponding to each first node can be obtained.

[0036] In the case of determining the subtask category corresponding to each first node, multiple subtasks can be split out from the target overall task, and the multiple split subtasks can be distributed among the edge servers in the service network in the manner of allocating the subtasks to edge servers corresponding to the subtask category of the subtasks. That is, the type of subtasks assigned to an edge server is the subtask category corresponding to the edge server. The specific allocation method can be determined in combination with actual needs. For example, an edge server is assigned one subtask or multiple subtasks, and the specific allocation method is not limited here.

[0037] In this embodiment, the service network is abstracted into a three-layer system, and the structure and data in the network are synthesized into a topological graph. A graph neural network node classifier is used to make decisions on task allocation, which reduces the complexity of the service network and facilitates the management of the service network.

[0038] As a possible implementation, the target topology map is a topology map generated directly from the service network, that is, the nodes in the target topology map are all nodes corresponding to the edge servers. In this way, the information of each edge server and subordinate device in the service network can be collected to generate the target topology map.

[0039] After obtaining the target topology map, the node classifier can be used to classify each node in the target topology map based on the target graph, so as to obtain the subtask category corresponding to each edge server. Figure 4 Please refer to Figure 4 , Figure 4 for Figure 2 Schematic diagram of the flow of sub-steps included in step S300. In this embodiment, step S300 may include sub-steps S310 to S320.

[0040] Sub-step S310, for each sub-task category, split the task content corresponding to the sub-task category in the target overall task according to the number of first nodes corresponding to the sub-task category to obtain sub-tasks.

[0041] Sub-step S320: for each edge server, assign a subtask under the subtask category of the edge server to the edge server.

[0042] In this embodiment, the target total task includes more task contents, and the more task contents correspond to different subtask categories. After the nodes are classified according to the actual situation of the service network, the task contents corresponding to each subtask category can be split into subtasks for each subtask category, and the number of subtasks corresponding to one subtask category is equal to the number of first nodes corresponding to the subtask category; then, one subtask is assigned to an edge server corresponding to the subtask category of the subtask, so that one edge server is only assigned one subtask, thereby avoiding poor task completion due to uneven task distribution. That is, one edge server is only assigned one subtask, and the type of the assigned subtask is the subtask category corresponding to the edge server.

[0043] As a possible implementation method, due to some special requirements, some edge servers may not be assigned subtasks under their corresponding subtask categories, so as to protect or make other arrangements for these edge servers. Figure 5 For example, the priority of the new target task to be processed next is higher than the priority of the current target task. Assuming that the subtask category corresponding to an edge server is a1, and the resources for processing subtasks of subtask category a1 are relatively few, if the subtask of subtask category a1 is assigned to the edge server this time, it may lead to poor execution effect of the new target task (for example, slow processing speed). To avoid this situation, Figure 6 The target topology map is obtained in the manner shown, which ensures that the edge server is not assigned subtasks of subtask category a1 this time, so that the subtasks of subtask category a1 split from the new target total task can be subsequently assigned to the edge server, so that the edge server can have sufficient resources to process the subtasks of subtask category a1 split from the new target total task.

[0044] Please refer to Figure 5 , Figure 5 for Figure 2 Schematic diagram of the flow of sub-steps included in step S100. In this embodiment, step S100 may include sub-steps S110 to S120.

[0045] Sub-step S110: determining at least one target attack node from the multiple first nodes corresponding to the multiple edge servers.

[0046] Sub-step S120, taking the at least one target attack node as a target of a counter-attack, adjusting the initial topology map by a counter-attack method, and obtaining the target topology map.

[0047] In this embodiment, at least one target attack node can be determined from a plurality of first nodes corresponding to a plurality of edge servers included in the service network according to actual needs. That is, the target attack node is a node for which the classification result of the subtask category is expected to be changed. The initial topology graph corresponding to the service network can be attacked to cause the node classifier to misclassify the at least one target attack node, thereby affecting the subsequent subtask allocation, so that the target attack node is not allocated a subtask under the subtask category corresponding to itself.

[0048] Optionally, the at least one target attack node is used as the target of the counterattack, and the initial topology map is adjusted by the counterattack method to obtain the target topology map, and then the nodes are classified according to the target topology map, and then the subtasks are assigned according to the subtask categories corresponding to each first node. The initial topology map is used to describe the situation of the edge server and the slave device in the server, that is, the initial topology map only describes the real situation of the service network. Optionally, after the nodes are classified based on the target topology map obtained by the counterattack, the target topology map can be used Figure 4 Subtasks are assigned using the subtask assignment method shown.

[0049] When adjusting the initial topology map, the adjustment can be made based on the following goal: for each target attack node, the subtask category of the target attack node obtained by the node classifier based on the initial topology map is different from the subtask category of the target attack node obtained based on the target topology map. The specific method of adjusting the initial topology map to obtain the target topology map by adversarial attack can be determined in combination with actual conditions and is not specifically limited here.

[0050] Optionally, the target topology map can be obtained by injecting a second node into the initial topology map and synthesizing the node features and adjacency relationship of the second node. The second node is connected to the target attack node. That is, using the injection attack algorithm, the node features and adjacency relationship of the injected node are synthesized to cause the node classifier to misclassify the specific target attack node.

[0051] Optionally, you can Figure 6 The node injection and information synthesis are performed in the manner shown, thereby obtaining the target topology diagram. Figure 6 , Figure 6 for Figure 5 Schematic diagram of the flow of sub-steps included in sub-step S120. In this embodiment, sub-step S120 may include sub-steps S121 to S124.

[0052] Sub-step S121, for each target attack node, with the goal of maximizing the impact on the classification result of the target attack node, synthesize the victim feature of a virtual node connected to the target attack node according to the node feature of the target attack node.

[0053] Sub-step S122, classifying the at least one target attack node according to the victim characteristics corresponding to each target attack node to obtain at least one attack node set.

[0054] Sub-step S123, for each attack node set, each target attack node in the attack node set is connected to the same second node, and the node characteristics of the second node to which each target attack node in the attack node set is connected are determined according to the victim characteristics corresponding to the attack node set to obtain an intermediate topology graph.

[0055] Sub-step S124, obtaining the target topology map according to the intermediate topology map.

[0056] In this embodiment, for each target attack node, a virtual node can be set to connect with the target attack node, with the goal of maximizing the impact on the subtask category to which the target attack node is classified, and synthesizing the victim characteristics of the virtual node based on the node characteristics of the target attack node. That is, it is expected that the synthesized victim characteristics can achieve the following goals: the subtask category obtained by the node classifier based on the node characteristics of the target attack node and the victim characteristics of the virtual node is different from the subtask category of the target attack node obtained based on the original topology map. The victim characteristics of a virtual node are the node characteristics of the virtual node.

[0057] After obtaining the victim features corresponding to each attack node, the target attack nodes determined can be classified according to the victim features corresponding to each target attack node to obtain at least one attack node set. The number of second nodes is preset, and the number of the attack node set is equal to the number of the second nodes; optionally, the number of the second nodes is less than the number of target attack nodes. Optionally, the Kmean algorithm or other unsupervised classification algorithms can be used to classify the attack nodes based on the victim features, and the specific method of classifying the attack nodes based on the victim features is not limited here.

[0058] Afterwards, for each attack node set, each target attack node in the attack node set can be connected to the same second node, and the victim feature corresponding to the attack node set can be used to determine the node feature of the second node connected to each target attack node in the attack node set. Optionally, a victim feature can be randomly selected from the victim features corresponding to the attack node set as the node feature of the second node connected to the target attack node in the attack node set. In this way, an intermediate topology map can be obtained.

[0059] Optionally, the intermediate topology map may be directly used as the target topology map. Figure 7 The intermediate topology is further adjusted in the manner shown to obtain the target topology. Figure 7 , Figure 7 for Figure 6 Schematic diagram of the flow of sub-steps included in sub-step S124. In this embodiment, sub-step S124 may include sub-steps S1241 and S1242.

[0060] Sub-step S1241, obtaining a proxy model corresponding to the node classifier.

[0061] In this embodiment, if you have the authority to obtain the node classifier, you can directly use the node classifier as the proxy model to be used in the counterattack. Figure 8 The proxy model is obtained in the manner shown. Figure 8 , Figure 8 for Figure 7 Schematic diagram of the flow of sub-steps included in sub-step S1241. In this embodiment, sub-step S1241 may include sub-steps S12411 and S12412.

[0062] Sub-step S12411, obtaining node information of each of the plurality of first nodes in the initial topology graph, and obtaining a sub-task category obtained by the node classifier based on the node information of each of the plurality of first nodes.

[0063] Sub-step S12412, fitting the proxy model according to the node information of each of the plurality of first nodes and the corresponding sub-task categories.

[0064] In this embodiment, the node information of each of the multiple first nodes in the initial topology diagram can be obtained, and the multiple first node information is sent to the node classifier for node classification to obtain the subtask categories corresponding to each of the multiple first nodes. Among them, the node information of each first node includes the node characteristics of the first node and the node characteristics of the first node that is connected to the first node. Optionally, the direct connection between the two can be determined as the existence of a connection relationship, and the direct connection and indirect connection between the two can also be determined as the existence of a connection relationship. For example, if the first node A and the first node C are both connected to the first node B, it can be considered that the first nodes B, C and the first node A are connected.

[0065] Afterwards, the node classifier may be fitted according to the node information of each of the plurality of first nodes and the corresponding subtask categories, thereby obtaining the proxy model.

[0066] At this time, the node classifier in the core server is a black box. Each query to this black box will cause the risk of being analyzed by the core server. Therefore, it is necessary to fit this black box with as few queries as possible to reduce the risk of the attack being discovered. In this case, Bayesian regression can be used as a proxy model, so that a better proxy model can be fitted with a smaller number of queries.

[0067] Sub-step S1242, with the goal of increasing the loss corresponding to the proxy model, adjust the node characteristics and adjacency relationship of the second node in the intermediate topological graph to obtain the target topological graph.

[0068] Among them, the loss corresponding to the proxy model is the sum of the target differences corresponding to each target attack node, the target difference corresponding to a target attack node is the difference between the maximum probability corresponding to the target attack node and the probability of being classified as a label category, the maximum probability corresponding to the target attack node is the maximum value of the probabilities that the target attack node is classified by the proxy model to a category other than the label category, the probability that the target attack node is classified into the label category is the probability that the target attack node is classified by the proxy model to the label category, and the label category is the category of the target attack node obtained by the node classifier based on the initial topology graph.

[0069] With the goal of increasing the loss corresponding to the proxy model, the node features and adjacency relationship of the second node in the intermediate topology graph are iteratively adjusted for a limited number of times, and the topology graph when the iteration is stopped is used as the target topology graph. The target topology graph is a toxic graph, and then the node classifier is used to obtain the subtask category of each edge server based on the toxic graph, and then the subtask allocation is performed.

[0070] The following is an example of how to obtain a toxic target topology map.

[0071] In the node classifier If you do not have permission to obtain the results and parameters, you can use the proxy model That is, by fitting The adjacency relationship and characteristics of some nodes in the initial topology graph are input as sample nodes. The input of each sample node is recorded as a query, and the corresponding output is obtained. Estimate and generate proxy models The generated proxy model It can be used to estimate the node classifier well. The classification results of each node in the graph provide a reference for subsequent attacks.

[0072] The effect of the attack is to make the node classifier classify the target attack node as wrong as possible. In this embodiment, the method used is injection attack, which injects new nodes into the graph and synthesizes their node features and adjacency relationships. The classification result of the target attack node is wrong. Let the set of injection nodes (i.e., the second node) be , where B is the number of injected nodes, then the adjacency matrix and node feature matrix after injecting the nodes are: , ,in, , is the adjacency relationship of the injection node that needs to be generated, is the node feature of the injection node to be generated. The attack method used in this embodiment is to find an effective generation and The method is to make the injection node make the subtask category classification result of the target attack node produce errors as much as possible. Therefore, in this embodiment, the attack is defined as: , where the target attack node set is , T is the number of target attack nodes, is the probability that the node is classified as a label category (i.e., the classification result of the node classifier when it is not attacked), is the maximum probability that the node is classified as a category other than the label category. That is, the purpose of the adversarial attack is to transform the node classifier to the target attack node set. The classification of the nodes in is wrong, which has the effect of disrupting the task allocation decision. The loss function of the adversarial attack is: ,in, The proxy model The estimated probability of being classified into subtask category j is , the larger the value of the loss function, the better the attack effect.

[0073] When it is necessary to obtain a poisonous target topology, the proxy model can be fitted first. Then, the target attack node is determined, and a small number of nodes are injected into the initial topology to initialize the adjacency matrix and node features. Among them, the node features of the injected nodes and the initialization of the adjacency relationship are just to establish a vector or add rows and columns after the original matrix, and the contents are all 0.

[0074] Afterwards, for each target attack node, it is assumed that the target attack node is connected to a virtual node, with the goal of maximizing the impact on the classification result of the target attack node under the proxy model (that is, influencing the classification result output by it to not be the corresponding label category), and synthesizing the victim features of the virtual node connected to the target attack node.

[0075] Next, the selected target attack nodes are classified according to the victim characteristics of each target attack node, and the number of categories is the number of injection nodes. The target attack nodes of the same category are connected to the same injection node, and a victim characteristic is randomly selected from the victim characteristics of the target attack nodes of the same category and assigned to the injection node trained by the target attack node of the category, and other connection relationships are injected into the injection node (the other injected adjacency relationships can all represent no connection, or be set for the purpose of maximizing loss). Finally, the node characteristics and adjacency relationships of the injection node are further changed to maximize the loss function as much as possible, thereby obtaining the target topology map. Among them, when setting the adjacency relationship of the injection node, the number of nodes corresponding to the edge server that the injection node can directly connect to can be set to reduce the number of iterations. The target topology map can be obtained by adjustment when one target attack node is connected to only one injection node.

[0076] Wherein, when the node characteristics and adjacency relationship of each second node are determined, the nodes can be classified based on the target topology map when assigning tasks in a corresponding manner according to the authority possessed.

[0077] For example, when an attacker has communication rights to an edge server, the injected node represents a virtual edge server. By establishing false features (node ​​features) for the virtual server (injected node) and its communication with the edge server (adjacent relationship), the attacker can attack the task allocation decision (node ​​classifier) ​​of the core server, causing a specific edge server (victim node) to be assigned the wrong task. When an attacker has the ability to tamper with the core server information, he can directly tamper with the state data set collected by the core server, add the features and adjacency of the virtual server, and thus make the core server produce errors in the decision results of task allocation to a specific edge server.

[0078] In this embodiment, the service network is abstracted as a three-layer system, and the structure and data in the network are synthesized into a topology graph. The graph neural network node classifier is used to make decisions on task allocation, which reduces the complexity of the service network and facilitates the management of the service network. It is also possible to attack the initial topology graph corresponding to the service network according to actual needs, so that the target attack node is assigned to a subtask of a different category from the subtask category corresponding to it before the attack.

[0079] In order to execute the corresponding steps in the above embodiments and various possible methods, a method for implementing a task decision device 200 of a service network is given below. Optionally, the task decision device 200 of the service network can adopt the above Figure 1 The device structure of the electronic device 100 is shown. Fig. 9 , Fig. 9 A block diagram of a task decision device 200 for a service network provided in an embodiment of the present application. It should be noted that the basic principle and technical effects of the task decision device 200 for a service network provided in this embodiment are the same as those in the above-mentioned embodiments. For the sake of brief description, for parts not mentioned in this embodiment, reference may be made to the corresponding contents in the above-mentioned embodiments. The service network includes a plurality of edge servers and subordinate devices that are communicatively connected to the edge servers. In this embodiment, the task decision device 200 for the service network may include: a topology map acquisition module 210, a classification module 220, and an allocation module 230.

[0080] The topology map acquisition module 210 is used to obtain a target topology map corresponding to the service network, wherein the target topology map includes connection relationships between multiple nodes and node characteristics of each node, wherein the multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding subordinate device.

[0081] The classification module 220 is used to classify nodes based on the target topology graph using a node classifier to obtain subtask categories corresponding to each first node, wherein the subtask categories corresponding to each first node are types of multiple subtasks corresponding to the target overall task to be assigned.

[0082] The allocation module 230 is used to allocate multiple subtasks split from the target overall task to corresponding edge servers according to the subtask categories corresponding to each first node, wherein the type of subtask allocated to an edge server is the subtask category corresponding to the edge server.

[0083] Optionally, the above modules can be stored in the form of software or firmware. Figure 1The memory 110 shown in the figure may be fixed in the operating system (OS) of the electronic device 100 and may be Figure 1 Meanwhile, the data and program codes required for executing the above modules may be stored in the memory 110.

[0084] The embodiment of the present application also provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the task decision method of the service network is implemented.

[0085] In summary, the embodiment of the present application provides a task decision method and related device for a service network. First, a target topology diagram corresponding to a service network including multiple edge servers and subordinate devices of the edge servers is obtained. The target topology diagram includes the connection relationship between multiple nodes and the node characteristics of each node. The multiple nodes include a first node corresponding to the edge server. The node characteristics of the first node are used to describe the situation of the corresponding subordinate device; then, a node classifier is used to classify nodes based on the above target topology diagram to obtain the subtask category corresponding to each first node. The subtask category corresponding to each first node is the type of multiple subtasks corresponding to the target total task to be assigned; finally, according to the subtask category corresponding to each first node, the multiple subtasks split from the target total task are assigned to the corresponding edge server. The type of the subtask assigned to an edge server is the subtask category corresponding to the edge server. In this way, the analysis and control of complex networks can be achieved while occupying less computing resources and taking less time.

[0086] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0087] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0088] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. 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, and other media that can store program codes.

[0089] The above description is only an optional embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A task decision method for a service network, characterized in that: The service network includes a plurality of edge servers and slave devices in communication connection with the edge servers, and the method includes: Obtaining a target topology map corresponding to the service network, wherein the target topology map includes connection relationships between multiple nodes and node characteristics of each node, the multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding slave device; Using a node classifier to classify nodes based on the target topology graph, obtaining a subtask category corresponding to each first node, wherein the subtask category corresponding to each first node is a type of multiple subtasks corresponding to the target total task to be assigned; According to the subtask categories corresponding to the first nodes, the multiple subtasks split from the target overall task are allocated to the corresponding edge servers, wherein the type of the subtask allocated to an edge server is the subtask category corresponding to the edge server.

2. The method according to claim 1, characterized in that The step of allocating the multiple subtasks split from the target overall task to the corresponding edge servers according to the subtask categories corresponding to the first nodes includes: For each subtask category, the task content corresponding to the subtask category in the target overall task is split according to the number of first nodes corresponding to the subtask category to obtain subtasks, wherein the number of subtasks corresponding to each subtask category is equal to the number of corresponding first nodes; For each edge server, a subtask under the subtask category of the edge server is allocated to the edge server.

3. The method according to claim 1 or 2, characterized in that: The obtaining of the target topology map corresponding to the service network includes: Determine at least one target attack node from the plurality of first nodes corresponding to the plurality of edge servers; The at least one target attack node is taken as the target of the adversarial attack, and the initial topology map is adjusted by the adversarial attack method to obtain the target topology map; wherein the initial topology map is used to describe the situation of the edge server and the subordinate devices in the server; the initial topology map is adjusted based on the following goals: for each target attack node, the subtask category of the target attack node obtained by the node classifier based on the initial topology map is different from the subtask category of the target attack node obtained based on the target topology map.

4. The method according to claim 3, characterized in that The step of taking the at least one target attack node as a target of the counter-attack and adjusting the initial topology map in a counter-attack manner to obtain the target topology map includes: The target topology graph is obtained by injecting a second node into the initial topology graph and synthesizing node features and adjacency relationships of the second node, wherein the second node is connected to the target attack node.

5. The method according to claim 4, characterized in that The method of injecting the second node into the initial topology map and synthesizing the node features and adjacency relationship of the second node to obtain the target topology map includes: For each target attack node, with the goal of maximizing the impact on the classification result of the target attack node, based on the node features of the target attack node, synthesize the victim features of a virtual node connected to the target attack node; Classifying the at least one target attack node according to the victim characteristics corresponding to each target attack node to obtain at least one attack node set, wherein the number of the attack node sets is equal to the number of the second nodes; For each attack node set, each target attack node in the attack node set is connected to the same second node, and the node characteristics of the second node to which each target attack node in the attack node set is connected are determined according to the victim characteristics corresponding to the attack node set, so as to obtain an intermediate topology graph; The target topology map is obtained according to the intermediate topology map.

6. The method according to claim 5, characterized in that The step of obtaining the target topology map according to the intermediate topology map comprises: Obtaining a proxy model corresponding to the node classifier; With the goal of increasing the loss corresponding to the proxy model, the node features and adjacency relationships of the second node in the intermediate topology graph are adjusted to obtain the target topology graph, wherein the loss corresponding to the proxy model is the sum of the target difference values ​​corresponding to each target attack node, the target difference value corresponding to a target attack node is the difference between the maximum probability corresponding to the target attack node and the probability of being classified as a label category, the maximum probability corresponding to the target attack node is the maximum value of the probabilities of the target attack node being classified by the proxy model to a category other than the label category, the probability of the target attack node being classified to the label category is the probability of the target attack node being classified to the label category by the proxy model, and the label category is the category of the target attack node obtained by the node classifier based on the initial topology graph.

7. The method according to claim 6, characterized in that The obtaining of the proxy model corresponding to the node classifier comprises: Obtaining node information of each of the plurality of first nodes, and obtaining a subtask category obtained by the node classifier based on the node information of each of the plurality of first nodes, wherein the node information of each of the first nodes includes a node feature of the first node and a node feature of a first node that is connected to the first node; The proxy model is obtained by fitting according to the node information of each of the plurality of first nodes and the corresponding subtask categories.

8. A task decision device for a service network, characterized in that: The service network includes a plurality of edge servers and slave devices connected to the edge servers for communication, and the device includes: A topology map obtaining module, used to obtain a target topology map corresponding to the service network, wherein the target topology map includes a connection relationship between multiple nodes and node characteristics of each node, the multiple nodes include a first node corresponding to the edge server, and the node characteristics of the first node are used to describe the situation of the corresponding slave device; A classification module, used to classify nodes based on the target topology graph using a node classifier to obtain subtask categories corresponding to each first node, wherein the subtask categories corresponding to each first node are types of multiple subtasks corresponding to the target total task to be assigned; The allocation module is used to allocate multiple subtasks split from the target overall task to corresponding edge servers according to the subtask categories corresponding to each first node, wherein the type of subtask allocated to an edge server is the subtask category corresponding to the edge server.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor can execute the machine executable instructions to implement the task decision method of the service network as described in any one of claims 1-7.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the task decision method for a service network as described in any one of claims 1 to 7 is implemented.

Images