A method and system for determining the resilience of an unmanned system based on combinatorial testing

Abstract

The present invention discloses a method and system for determining the resilience of an unmanned system based on combinatorial testing, which relates to the technical field of unmanned system testing. The method includes: constructing a hypernetwork model of the unmanned system; the hypernetwork model includes a physical layer network, a collaboration layer network, and a task layer network; constructing a physical perturbation model, a collaboration perturbation model, and a task perturbation model; encoding each sub-model in the constructed physical perturbation model, task perturbation model, and collaboration perturbation model and generating a plurality of combinatorial test cases; determining a plurality of resilience metric performance parameters of the unmanned system and the standard values of each resilience metric performance parameter; selecting combinatorial test cases from the plurality of combinatorial test cases according to the resilience test requirements to form a test case set; in the unmanned system, sequentially executing each combinatorial test case in the test case set to obtain the test values of the resilience metric performance parameters, and determining the resilience value of the unmanned system according to the test values and the standard values. The present invention improves the reliability of the resilience test of the unmanned system.

Description

Technical Field

[0001] The present invention relates to the technical field of elastic evaluation of unmanned systems, and particularly to a method and system for determining the elasticity of unmanned systems based on combinatorial testing. Background Art

[0002] With the development of technology, large-scale and distributed unmanned systems have emerged and been applied on a large scale, such as unmanned aerial vehicle swarm systems, large-scale logistics robot systems, etc. However, most of the existing evaluation methods only test individual agents, ignoring the collaborative relationship between agents and the structural model of the entire unmanned system. Considering the deficiencies of the existing technology, due to the complex architecture of distributed unmanned systems, it is difficult to model during the testing process, there are key factors affecting collaborative cooperation, and they have multiple characteristics such as diversity, emergence, vulnerability, and complexity. When unmanned systems execute different tasks, they are subject to different types and intensities of perturbations. How to test and measure the elasticity of unmanned systems is a key problem to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for determining the elasticity of unmanned systems based on combinatorial testing, which improves the reliability of unmanned system elasticity testing.

[0004] To achieve the above purpose, the present invention provides the following solutions:

[0005] The present invention discloses a method for determining the elasticity of unmanned systems based on combinatorial testing, including:

[0006] Construct a hypernetwork model of the unmanned system; the hypernetwork model includes a physical layer network, a collaborative layer network, and a task layer network;

[0007] Construct a physical perturbation model according to the physical layer of the unmanned system, construct a collaborative perturbation model according to the collaborative layer of the unmanned system, and construct a task perturbation model according to the task layer of the unmanned system. The physical perturbation model includes at least 1 physical layer perturbation category sub-model, the collaborative perturbation model includes at least 1 collaborative layer perturbation category sub-model, and the task perturbation model includes at least 1 task layer perturbation category sub-model;

[0008] Encode each sub-model in the physical perturbation model, the task perturbation model, and the collaborative perturbation model and generate multiple combinatorial test cases;

[0009] Determine multiple elastic metric performance parameters of the unmanned system and the standard values of each elastic metric performance parameter; the elastic metric performance parameters include physical layer performance parameters, collaborative layer performance parameters, and task layer performance parameters;

[0010] Select combination test cases from multiple combination test cases according to the elastic test requirements to form a test case set;

[0011] In the unmanned system, sequentially execute each combination test case in the test case set, obtain the test value of the elastic metric performance parameter, and determine the elasticity value of the unmanned system according to the test value and the standard value.

[0012] Optionally, the nodes in the physical layer network are unmanned execution entities, and the links in the physical layer network are information interactions between the unmanned execution entities; the nodes in the collaboration layer network are physical entity unit groups, and the physical entity unit group includes multiple unmanned execution entities that perform set tasks. The links in the collaboration layer network are the allocation and scheduling instructions of the collaboration layer of the unmanned system to the physical entity unit group; the nodes in the task layer network are the subtasks obtained by decomposing the top-level task target by the task layer of the unmanned system, and the links in the task layer network are the information flows of the decomposed subtasks.

[0013] Optionally, the physical entity unit group is determined by the collaboration layer using a collaboration layer algorithm model to dynamically integrate each unmanned execution entity based on the collected sensor parameters; the collaboration layer algorithm model includes a MADDPG algorithm model, and the sensor parameters include the speeds of the unmanned execution entities.

[0014] Optionally, the physical layer perturbation category sub-model includes that the unmanned execution entity cannot participate in the resource integration and scheduling of the collaboration layer.

[0015] Optionally, the collaboration layer perturbation category sub-model includes that the collaboration layer algorithm cannot give action instructions, the physical entity unit group cannot correctly execute the instructions given by the collaboration layer algorithm, and there is no communication between any two physical entity unit groups.

[0016] Optionally, the task layer perturbation category sub-model includes that the task layer cannot decompose and allocate tasks and the task assignment is lost due to communication failures between the nodes in the task layer and the collaboration layer.

[0017] Optionally, determining multiple elastic metric performance parameters of the unmanned system and the standard value of each elastic metric performance parameter specifically includes:

[0018] For each elastic metric performance parameter, obtain the mean curve of the elastic metric performance parameter in the normal operation state of the unmanned system, and use the mean curve as the standard value of the elastic metric performance parameter.

[0019] The present invention discloses a system for determining the elasticity of an unmanned system based on combinatorial testing, including:

[0020] A hypernetwork model construction module for constructing a hypernetwork model of an unmanned system; the hypernetwork model includes a physical layer network, a collaboration layer network, and a task layer network;

[0021] A perturbation model construction module for constructing a physical perturbation model based on the physical layer of the unmanned system, constructing a collaboration perturbation model based on the collaboration layer of the unmanned system, and constructing a task perturbation model based on the task layer of the unmanned system. The physical perturbation model includes at least one physical layer perturbation category sub-model, the collaboration perturbation model includes at least one collaboration layer perturbation category sub-model, and the task perturbation model includes at least one task layer perturbation category sub-model;

[0022] A combined test case generation module for encoding each sub-model in the physical perturbation model, the task perturbation model, and the collaboration perturbation model and generating a plurality of combined test cases;

[0023] A standard value determination module for elastic metric performance parameters for determining a plurality of elastic metric performance parameters of the unmanned system and the standard values of each of the elastic metric performance parameters;

[0024] A test case set construction module for selecting combined test cases from a plurality of combined test cases according to elastic test requirements to form a test case set;

[0025] An elastic value determination module for the unmanned system, for sequentially executing each combined test case in the test case set in the unmanned system, obtaining test values of elastic metric performance parameters, and determining the elastic value of the unmanned system according to the test values and the standard values.

[0026] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0027] The present invention constructs a physical layer network, a collaboration layer network, and a task layer network, considers the collaboration relationship between each agent of the unmanned system, encodes each sub-model in the physical perturbation model, the task perturbation model, and the collaboration perturbation model and generates a plurality of combined test cases, can realize the testing of different types and intensities of perturbations, and selects combined test cases from a plurality of combined test cases according to elastic test requirements to form a test case set, improving the reliability and flexibility of the elastic testing of the unmanned system. Description of the Drawings

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 Schematic flow chart of a method for determining the resilience of an unmanned system based on combinatorial testing provided by an embodiment of the present invention;

[0030] Figure 2 Schematic structural diagram of a system for determining the resilience of an unmanned system based on combinatorial testing provided by an embodiment of the present invention. Detailed implementation manners

[0031] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0032] The object of the present invention is to provide a method and system for determining the resilience of an unmanned system based on combinatorial testing, which improves the reliability of the resilience testing of the unmanned system.

[0033] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0034] Embodiment 1

[0035] As Figure 1 shown, this embodiment discloses a method for determining the resilience of an unmanned system based on combinatorial testing, including the following steps.

[0036] Step 100: Construct a hypernetwork model of the unmanned system; the hypernetwork model includes a physical layer network, a coordination layer network, and a task layer network.

[0037] In this embodiment, the hypernetwork model refers to a network model that describes the structure and behavior of a system. The hypernetwork model models subsystems and the interactions between multiple subsystems to achieve the analysis, design, and optimization of the system.

[0038] The nodes in the physical layer network are unmanned execution entities, and the edges in the physical layer network are information interactions between the unmanned execution entities; the nodes in the coordination layer network are physical entity unit groups, and the physical entity unit groups include multiple unmanned execution entities that execute set tasks. The edges in the coordination layer network are allocation and scheduling instructions of the coordination layer of the unmanned system to the physical entity unit groups; the nodes in the task layer network are subtasks obtained by decomposing the top-level task objectives by the task layer of the unmanned system, and the edges in the task layer network are information flows of the decomposed subtasks.

[0039] Among them, step 100 specifically includes:

[0040] Step 101: Model the physical layer. The physical layer refers to the physical layer network of the hypernetwork model. The specific steps of physical layer modeling are to abstract the nodes and edges of this layer of network. Its nodes are the smallest executable entities (unmanned executable entities) of this cluster, and the edges are the information interactions between the smallest executable entities. Taking the drone delivery system as an example, the smallest executable entity in this system refers to a single drone, and the edge is the information interaction between various sensors of the drones.

[0041] Step 102: Model the coordination layer. The coordination layer refers to the coordination layer network of the hypernetwork model. The coordination layer algorithm gives instructions based on the collected sensor parameters such as speed, carrying capacity, etc. and environmental observation values, and dynamically integrates and allocates the smallest executable entities of the hypernetwork physical layer to better meet the requirements of the task layer. The specific steps of coordination layer modeling are to abstract the nodes and edges of this layer of network. Its nodes are the physical entity unit groups integrated by the coordination layer algorithm model (specifically the MADDPG algorithm model). The physical entity unit group refers to the smallest executable entities teaming up to form a physical entity unit group in order to complete a certain task. The edge is the allocation and scheduling instruction of the coordination layer algorithm. Among them, the "coordination layer algorithm" refers to: a method used in the unmanned system to generate control and scheduling instructions for the physical layer nodes. For example, in the drone cluster system based on multi-agent reinforcement learning, the reinforcement learning model MADDPG is the coordination layer algorithm of this unmanned system. Among them, the "allocation and scheduling instruction" refers to the action instruction sent by the coordination layer algorithm to the physical entity unit group according to the collected information, such as the teaming instruction in the drone system.

[0042] The sensor parameters include the speeds of each unmanned executable entity.

[0043] Step 103: Model the task layer. The task layer refers to the task layer network of the hypernetwork model. The specific steps of task layer modeling are to abstract the nodes and edges of this layer of network. Its nodes are the objects that disassemble and allocate the top-level task objectives according to the information collected by themselves, such as the communication range of the drone, the number of drones within the communication range, the distance of the drone from the destination, etc. The edge is the information flow of the disassembled task objective allocation. Taking the drone delivery system as an example, among them, "disassembling and allocating the top-level task objective" means that the central scheduling system allocates the transportation tasks of each drone according to the cargo situation.

[0044] Step 200: Construct a physical perturbation model based on the physical layer of the unmanned system, construct a collaborative perturbation model based on the collaborative layer of the unmanned system, and construct a task perturbation model based on the task layer of the unmanned system. The physical perturbation model includes at least one physical layer perturbation category sub-model, the collaborative perturbation model includes at least one collaborative layer perturbation category sub-model, and the task perturbation model includes at least one task layer perturbation category sub-model.

[0045] The physical layer perturbation category sub-model includes that the unmanned execution entity cannot participate in the resource integration and scheduling of the collaborative layer (losing contact with the whole).

[0046] The collaborative layer perturbation category sub-model includes that the collaborative layer algorithm cannot give action instructions, the physical entity unit group cannot correctly execute the instructions given by the collaborative layer algorithm, and there is no communication between any two physical entity unit groups.

[0047] The task layer perturbation category sub-model includes that the task layer cannot disassemble and allocate tasks and the task assignment is lost due to communication failures between nodes and the collaborative layer in the task layer.

[0048] The unmanned system perturbation model extracts key elements for perturbation modeling according to the hypernetwork model in step 100; "key elements" here refer to: if the change of this element will affect the behavior of the object at this layer, then this element can be called a key element; taking the unmanned aerial vehicle system as an example, in the physical layer, analyze the sensor parameters that will directly affect the behavior of the unmanned aerial vehicle, and select some sensor parameters as key elements according to requirements. "Perturbation modeling" here refers to: clarifying the perturbations that affect the overall operation of the unmanned system under each layer, and then defining the corresponding perturbation injection method and the domain of the perturbation (the value range of the perturbed value after quantization).

[0049] Among them, step 200 specifically includes:

[0050] Step 201: Physical layer perturbation modeling, that is, constructing a physical perturbation model.

[0051] Among them, step 201 includes:

[0052] Step 2011: Analyze the key elements of the physical layer and determine the perturbation categories according to the characteristics of the key elements.

[0053] The perturbation categories include: 1. The physical layer entity unit cannot participate in the resource integration and scheduling of the collaborative layer (losing contact with the whole).

[0054] Step 2012: Conduct perturbation modeling for different perturbation categories in the physical layer.

[0055] Among them, the specific steps of "conducting perturbation modeling for different perturbation categories in the physical layer" are as follows: Model the following perturbations for the perturbation categories in step 2011: Perturbation Dp1 : Paralyze the physical layer nodes with a probability of X1. The injection object is the nodes of the physical layer, and the domain R of X1 is 0 - 100%. p1 It is 0 - 100%.

[0056] Step 202: Cooperative layer perturbation modeling, that is, constructing a cooperative perturbation model.

[0057] Among them, Step 202 includes:

[0058] Step 2021: Analyze the key elements of the cooperative layer, and determine the perturbation categories according to the characteristics of the key elements;

[0059] Step 2022: Conduct perturbation modeling for different perturbation categories in the cooperative layer;

[0060] Among them, the specific content of the "cooperative layer perturbation category" in Step 2021 is: 1. The cooperative layer algorithm cannot give reasonable action instructions; 2. The physical entity unit group cannot correctly execute the instructions given by the cooperative layer algorithm; 3. Communication cannot be established between any two physical layer entity units.

[0061] Among them, the specific steps of the "conducting perturbation modeling for different perturbation categories in the cooperative layer" in Step 2022 are as follows: Model the following perturbation for the perturbation category 1 in Step 2021: Perturbation D c1 : For the key elements affecting the generation of instructions by the cooperative layer algorithm, make it deviate by X3 with a probability of X3. The injection object is the cooperative layer algorithm, and the domain R of X3 is 0 - 100%; Model the following perturbation for the perturbation category 2 in Step B21: Perturbation D c1 : For the action instructions sent by the cooperative layer algorithm to the physical entity unit group, make the physical entity unit group execute incorrectly with a probability of X4. The injection object is the physical entity unit group, and the domain R of X4 is 0 - 100%; Model the following perturbation for the perturbation category 3 in Step 2021: Perturbation D c2 : Make the communication between physical layer nodes interrupted with a probability of X2. The injection object is the connection edge of the physical layer, and the domain R of X2 is 0 - 100%. c2 It is 0 - 100%. c3 : Make the communication between physical layer nodes interrupted with a probability of X2. The injection object is the connection edge of the physical layer, and the domain R of X2 is 0 - 100%. p2 It is 0 - 100%.

[0062] Step 203: Task layer perturbation modeling, that is, constructing a task perturbation model.

[0063] Among them, Step 203 specifically includes:

[0064] Step 2031: Analyze the key elements of the task layer, and determine the perturbation categories according to the characteristics of the key elements;

[0065] Step 2032: Conduct perturbation modeling for different perturbation categories in the task layer;

[0066] Among them, the specific content of the "task layer disturbance category" described in step 2031 is as follows: 1. Node failure at the task layer cannot disassemble and allocate tasks; 2. Communication failure between the task layer node and the collaboration layer results in task allocation loss.

[0067] Among them, the specific steps of the "perturbation modeling for different disturbance categories at the task layer" described in step 2032 are as follows: Model the following perturbations for the disturbance category 1 in step 2031: Perturbation D a1 : Make the task layer node paralyzed with a probability of X5, and the injection object is the node at the task layer. The domain R a1 of X5 is 0 - 100%; Model the following perturbations for the disturbance category 2 in step 2031: Perturbation D a2 : For the information flow of task target allocation after the task layer is disassembled, make the information flow loss occur with a probability of X6, and the injection object is the edge at the task layer. The domain R a2 of X6 is 0 - 100%.

[0068] Step 300: Encode each sub - model in the physical perturbation model, the task perturbation model, and the collaboration perturbation model, and generate multiple combined test cases.

[0069] By encoding the perturbations D ai in step 200, D pi in step 200, D ci within their domains, and combining the encoded perturbations at different levels through orthogonal table design to generate combined test cases to evaluate the impact on the unmanned system. The specific steps are as follows:

[0070] Step 301: Encode the perturbations D ai in step 200, D pi in step 200, D ci within their domain R;

[0071] Step 302: Design the orthogonal table according to the number of perturbations;

[0072] Step 303: Generate a combined test case set according to the test requirements;

[0073] Among them, the "encoding of perturbations" described in step 301 is specifically as follows: Regard the different - level perturbations in step 200 as the test input Inputi for generating test cases, and determine the test input within its domain; at the same time, according to the test requirements, give the time range for applying perturbations as the test input Inputt to prepare for the design of the orthogonal table. Take the perturbation D p1 at the physical layer as an example: Select three values of 0, 50%, and 90% as the test input according to the requirements within its domain R p1 , and form the encoding shown in Table 1 after integration.

[0074] Table 1 First Coding Table

[0075]

[0076]

[0077] The Inputt generated considering the test requirements is integrated to form the coding shown in Table 2 (where t0, t1, and t2 are determined according to specific test requirements).

[0078] Table 2 Second Coding Table

[0079]

[0080] Among them, the "orthogonal array" mentioned in step 302 has the following specific content: Each parameter is arranged in a matrix according to a certain rule, where the "rule" means that in each column, the number of times different numbers appear is equal, and the arrangement of numbers in any two columns is complete and balanced. Each row in the orthogonal array represents a set of test cases, and each column represents a test parameter. Through the orthogonal array, the perturbations at different levels of the super network are combined, and the test can cover the coupling perturbations between system levels.

[0081] Among them, the "designing the orthogonal array according to the number of perturbations" mentioned in step 302 has the following specific steps:[[]]

[0082] Step 3021: Determine the number of test factors;

[0083] Step 3022: Determine the number of factor levels;

[0084] Step 3023: Design the orthogonal array according to the number of test factors and the number of levels;

[0085] Among them, the "test factors" in step 3021 have the following content: The test input Inputi generated in step 301, and determine the number m of test inputs according to the test requirements;

[0086] Among them, the "number of factor levels" in step 3022 has the following content: Determine the number k of optional test inputs for each Inputi in step 301;

[0087] Among them, the "designing the orthogonal array" in step 3023 has the following content:

[0088] Design the Ln(m**k) orthogonal array according to the number m of test factors in step 3021 and the number k of factor levels in step 3022, where n represents the number of test cases that can be generated in the ideal state of the orthogonal array, and its calculation method is n = k * (m - 1) + 1.

[0089] Step 400: Determine multiple resilience metric performance parameters of the unmanned system and the standard values of each resilience metric performance parameter.

[0090] The resilience metric performance parameters include physical layer performance parameters, collaboration layer performance parameters, and mission layer performance parameters.

[0091] Among them, step 400 specifically includes:

[0092] Clarify the resilience metric performance parameters for resilience assessment at each level of the unmanned system. The resilience metric performance parameters include the communication reach rate between each unmanned execution entity.

[0093] For each resilience metric performance parameter, obtain the mean curve of the resilience metric performance parameter under the normal operating state of the unmanned system, and use the mean curve as the standard value of the resilience metric performance parameter.

[0094] Among them, the "performance parameter" mentioned in step 401 refers to: a quantitative indicator used to describe the performance of a product, device, or system; based on the system modeling of the supernetwork, there are multiple dimensions to describe the system, and specific performance parameters should be selected for evaluation according to different dimensions; taking the unmanned aerial vehicle system as an example, the physical layer performance parameter can be the communication reach rate between unmanned aerial vehicles, etc., the collaboration layer performance parameter can be the scale of the unmanned aerial vehicle team formation cluster, etc., and the mission layer performance parameter can be the mission completion rate, etc.;

[0095] Among them, the "constructing the resilience assessment standard" mentioned in step 402 includes the following steps:

[0096] Step 4021: Collect the values of the performance parameters determined in step 401 of the unmanned system under the normal operating state multiple times;

[0097] Step 4022; Construct a normal performance curve based on the system parameter values in 4021;

[0098] Among them, the "values of the performance parameters determined by multiple calculations" mentioned in step 4021 are specifically as follows: Run the unmanned system multiple times and collect data without any perturbation to establish a basis for the system resilience assessment.

[0099] Among them, the "constructing a normal performance curve based on the system parameter values in 4021" mentioned in step 4022 is specifically as follows: Calculate the average value for the data of each indicator collected under the non-perturbed condition. The mean value of each type of indicator is used as a dimension for evaluating the system resilience, and the curve formed by these indicator mean values constitutes the standard for evaluating the system resilience.

[0100] Step 500: Select combination test cases from multiple combination test cases according to the resilience test requirements to form a test case set.

[0101] Step 600: In the unmanned system, execute each combination test case in the test case set in sequence to obtain a test value of an elasticity metric performance parameter, and determine the elasticity value of the unmanned system according to the test value and the standard value.

[0102] Wherein, step 600 specifically includes:

[0103] Step 601: Execute the test case set suti formed in step 500 in sequence, and calculate and generate the elasticity curve of each performance parameter.

[0104] Step 602: Compare the elasticity curve generated in step 601 with the system elasticity evaluation standard constructed based on the undisturbed system parameters in step 3022 to construct a cluster system elasticity evaluation index;

[0105] The "elasticity test requirements" refer to: according to the different requirements of the tester for elasticity testing at different levels of the unmanned system, some test cases in the orthogonal table can be selected to generate a test case set; for example, if you are more concerned about the impact of disturbances at the physical layer and the coordination layer on the entire unmanned system, you can choose to generate a test case set for more test cases at these two layers;

[0106] The specific method of "constructing a cluster system resilience evaluation index" is as follows: when the unmanned system injects test cases within the time period [t0, t1] according to the orthogonal table entries, the cluster performance will change within the time period, and this change will be reflected in the performance curve of the unmanned system. The change in the area enclosed by the performance curve after the disturbance is injected and the normal performance curve can quantify this change well. The smaller the area, the more stable the unmanned system is; therefore, in this embodiment, the unmanned system resilience evaluation index can be represented by the area enclosed by the performance curve of the unmanned system and the curve in the time period under normal performance after the test case i is injected within the time period [t0, t1]. The larger the area, the worse the resilience of the dimension when the unmanned aerial vehicle system faces the corresponding disturbance.

[0107] Step 700: Optimize the structure of the cluster (unmanned system) according to the elasticity evaluation result (elasticity value of the unmanned system).

[0108] The specific method of optimizing the cluster structure according to the elasticity evaluation results described in step 700 is as follows: through the elasticity calculation results of step 600, according to the dimensions of the cluster that the tester is concerned about, select the indicator with the strongest disturbance, and determine what kind of disturbance caused the drastic change in the indicator; taking the drone cluster as an example, a relatively strong disturbance to drone communications will cause the average size of the drone cluster team to drop significantly during the disturbance time. If you want the cluster to always maintain a large team size, you need to strengthen the ability to protect communications.

[0109] Example 2

[0110] In this embodiment, the unmanned system takes a drone swarm as an example to illustrate a method for determining the resilience of an unmanned system based on combinatorial testing. Specifically, this drone swarm consists of 50 drones and 4 transportation task distribution centers, and the swarm needs to complete the task of transporting specified targets. Taking this swarm as an example, test cases are generated for testing, and its data is collected to calculate the resilience index, so as to realize the evaluation of the system resilience.

[0111] A method for determining the resilience of an unmanned system based on combinatorial testing in this embodiment specifically includes the following steps.

[0112] Step A: Construct a hypernetwork model of the drone system.

[0113] Step B: Construct a perturbation model of the drone system.

[0114] Step C: Generate combinatorial test cases by perturbing the encoding.

[0115] Step D: Determine the resilience metric of the drone system.

[0116] Step E: Execute the test cases to evaluate the resilience of the drone system.

[0117] Step F: Optimize the swarm structure according to the resilience evaluation results.

[0118] Among them, the specific method of "constructing a hypernetwork model of the drone system" described in Step A is as follows:

[0119] Step A1: Model the physical layer.

[0120] The specific method is as follows: In this drone swarm system, the physical layer nodes are individual drones, and 50 drone flights form 50 nodes of the physical layer. The physical layer edges are the communications between drones;

[0121] Step A2: Model the cooperation layer.

[0122] The specific method is as follows: In this drone swarm system, the drones cooperate with each other according to the instructions given by the reinforcement learning algorithm. In this drone swarm system, the cooperation layer nodes are the drone squads formed after the drones are grouped according to the algorithm instructions. The edges are the distribution and scheduling instructions of the reinforcement learning algorithm.

[0123] Step A3: Model the task layer.

[0124] The specific method is as follows: In this drone swarm system, in this drone swarm system, the task layer nodes are 4 transportation task distribution centers, and the edges are the information flows for task target allocation after the transportation task distribution centers disassemble the total transportation task.

[0125] Among them, for the "construction of the UAV system perturbation model" mentioned in step B, the specific method is as follows:

[0126] Step B1: Physical layer perturbation modeling; the specific steps are as follows:

[0127] Step B11: Analyze the behavior of physical layer objects to determine the perturbation category.

[0128] The specific method is as follows: The specific content of the "physical layer perturbation category" is: 1. The UAV fails and cannot participate in the instruction scheduling of the reinforcement learning algorithm.

[0129] Step B12: Perform perturbation modeling for different perturbation categories in the physical layer.

[0130] The specific method is as follows: Model the following perturbation for the perturbation category 1 in step B11: Perturbation D p1 : Make the UAV paralyzed with a probability of X1. The specific method of applying the perturbation is to generate a random number p in the range of 0 to 1 before the UAV moves. If p < X1, then let the UAV skip this round of actions. The above perturbation injection object is the node in the physical layer, and the domain R of X1 p1 is 0 - 100%.

[0131] Step B2: Cooperative layer perturbation modeling; the specific steps are as follows:

[0132] Step B21: Analyze the behavior of cooperative layer objects to determine the perturbation category.

[0133] The specific method is as follows: The specific content of the "cooperative layer perturbation category" is: 1. The reinforcement learning algorithm cannot give reasonable action instructions; 2. The UAV team cannot correctly execute the instructions given by the reinforcement learning algorithm; 3. Communication cannot be established between any two UAVs.

[0134] Step B22: Perform perturbation modeling for different perturbation categories in the cooperative layer;

[0135] The specific method is as follows: Model the following perturbation for the perturbation category 1 in step B21: Perturbation D c1 : For the key elements affecting the generation of instructions by the reinforcement learning algorithm (observations of the environment such as the distance to the target, the number of surrounding UAVs, etc.), make them deviate by X2 with a probability of X2. The specific method of applying the perturbation is to generate a random number p in the range of 0 - 1. If p < X2, then expand or shrink the original observation value X by X2 times. The above perturbation injection object is the reinforcement learning algorithm, and the domain R of X2 c1 is 0 - 100%.

[0136] Perturbation D c2:For the instructions (action) generated by the reinforcement learning algorithm, make it become other instructions with a probability of X3. The injection object is the reinforcement learning algorithm, and the domain of X3 is R c2 It is 0 - 100%.

[0137] Perturbation D c3 :For the trend (policy) of the instructions generated by the reinforcement learning algorithm, make it have a deviation with a fluctuation degree of X4 with a probability of X4. The injection object is the reinforcement learning algorithm, and the domain of X4 is R c3 It is 0 - 100%.

[0138] Model the following perturbation for the perturbation category 2 in step B21: Perturbation D c4 :For the action instructions sent by the reinforcement learning algorithm to the UAV team, make the UAV team execute incorrectly with a probability of X5. The specific method of applying the perturbation is to generate a random number p in the range of 0 - 1. If p < X5, the UAV does not execute the instructions given by the algorithm and acts randomly. The above perturbation injection object is the UAV team, and the domain of X5 is R c4 It is 0 - 100%.

[0139] Model the following perturbation for the perturbation category 3 in step B21: Perturbation D c5 :Make the communication between UAVs interrupted with a probability of X6. The specific method of applying the perturbation is to generate a random number p in the range of 0 - 1 when the UAVs are about to communicate. If p < X6, set all the transmitted data values to 0. The injection object is the link at the physical layer, and the domain of X6 is R c5 It is 0 - 100%.

[0140] Step B3: Task layer perturbation modeling; the specific steps are as follows:

[0141] Step B31: Analyze the behavior of task layer objects to determine the perturbation category.

[0142] The specific method is as follows: The specific content of "task layer perturbation category" is: 1. The transportation task distribution center fails and cannot disassemble the transportation task; 2. There is a communication failure between the transportation task distribution center and the UAV team, and the task assignment is lost;

[0143] Step B32: Perform perturbation modeling for different perturbation categories in the task layer.

[0144] Model the following perturbation for the perturbation category 1 in step B31: Perturbation D a1 :Paralyze the transportation task distribution center with a probability of X7. The injection object is the node in the task layer, and the domain of X7 is R a1 It is 0 - 100%; Model the following perturbation for the perturbation category 2 in step B31: Perturbation D a2For the information flow of task objective allocation after task layer decomposition, there is an information flow loss with a probability of X8, and the injection object is the connection edge of the task layer. The domain R of X8 is a2 is 0 - 100%.

[0145] Among them, the specific content of "generating combined test cases through perturbation coding" described in step C is as follows:

[0146] Step C1: For the perturbation D in step B ai , D pi , D ci is encoded within its domain R.

[0147] The specific approach is as follows: For the perturbation D p1 , within its domain R p1 , three values of 0, 50%, and 90% are selected as test inputs, and after integration, the encoding shown in Table 3 is formed.

[0148] Table 3 The third encoding table

[0149]

[0150]

[0151] For the perturbation D c1 , D c2 , D c3 , D c4 , D c5 , D a1 , D a2 : Similarly, three values of 0, 50%, and 90% are selected as test inputs, and after integration, test inputs such as Input2, Input3, Input4, Input5, Input6, Input7, and Input8 are formed; at the same time, according to the test requirements, perturbations are applied from 0 to 25 seconds and from 75 to 100 seconds after the start of the task, and the time range is used as the test input Inputt. After encoding, Inputt is shown in Table 4.

[0152] Table 4 The fourth encoding table

[0153] Number Optional test input Encoding corresponding to the test input Inputt [0,25]，[75,100] 1

[0154] Step C2: Design an orthogonal array according to the number of perturbations.

[0155] The specific content is as follows:

[0156] Step C21: Determine the number of test factors.

[0157] The specific content is: According to the test inputs determined in step C1, the number of test factors can be obtained as 8.

[0158] Step C22: Determine the number of factor levels.

[0159] Specifically, according to the selectable test inputs determined in Step C1, the number of levels for each factor is determined to be 3.

[0160] Step C23: Design an orthogonal array based on the number of test factors and levels.

[0161] Specifically, according to the number of test factors 8 in Steps C21 and C22, and the number of factor levels 3 in Step C22, an L27(3**8) orthogonal array is designed. The generated orthogonal array is shown in Table 5.

[0162] Table 5 Orthogonal Array

[0163]

[0164]

[0165] Step D: Determine the resilience metric for the UAV system.

[0166] Specifically:

[0167] Step D1: Identify the performance parameters for resilience assessment at each level of the UAV system.

[0168] Specifically, the performance parameter for resilience assessment at the physical layer is: single aircraft failure rate; the performance parameters for resilience assessment at the coordination layer are: UAV team size, number of UAV teams, communication reach rate; the performance parameter for resilience assessment at the mission layer is: delivery mission completion rate.

[0169] Step D2: Construct the resilience assessment criteria for the unmanned system.

[0170] Specifically:

[0171] Step D21: Collect the values of the performance parameters determined in Step D1 for the UAV system in the normal operating state multiple times.

[0172] Step D22: Construct the normal performance curve based on the system parameter values in D21.

[0173] The specific approach is as follows: For the data of each indicator collected under the condition of no disturbance, calculate the average value. The average value of each indicator serves as a dimension for evaluating the system resilience, and the curve formed by these indicator means constitutes the criteria for evaluating the system resilience.

[0174] Step E: Execute the test case set to evaluate the resilience of the UAV system.

[0175] Specifically:

[0176] Step E1: Generate a test case set according to the elastic test requirements of the drone.

[0177] The specific approach is as follows: Screen the test case set generated by the orthogonal table to generate a test case set suti for different requirements; for example, the test case set suti that focuses on the elasticity of the drone system task completion rate - composed of tests numbered 27, 13, and 2.

[0178] Step E2: Sequentially execute the test case set suti formed in Step E1, and calculate and generate the elastic curve of each performance parameter.

[0179] Step E3: Calculate the area enclosed by the curve of each index calculated in Step E2 and the corresponding normal performance curve within [0, 25] and [75, 100] seconds. The calculation method is as follows:

[0180]

[0181] Among them, Gi(t) represents the elastic curve generated by executing the test case, G(t) represents the index of the performance parameter corresponding to Gi(t) (the standard value for measuring the elastic performance parameter), and t represents time.

[0182] R i The larger it is, the worse the elasticity of the drone system in this dimension when facing the corresponding disturbance.

[0183] Step F: Optimize the cluster structure according to the elastic evaluation results.

[0184] The specific content is:

[0185] Step F1: According to the test requirements, select the task completion rate as the system index we are concerned about.

[0186] Step F2: Analyze the test cases that cause large fluctuations in the task completion rate and optimize the system structure.

[0187] The specific approach is as follows: Select the test cases that cause large fluctuations in the task completion rate, analyze the influence of different factors on the task completion rate, and obtain that when the intensities of both Input1 and Input3 factors are relatively strong, it will significantly affect the winning rate; according to the above results, in order to improve the task completion rate, the drone cluster should maintain a low unmanned failure rate during task execution and ensure the normal execution of the command algorithm.

[0188] Example 3

[0189] As Figure 2 shown, this example provides an unmanned system elasticity determination system based on combinatorial testing, including:

[0190] The hypernetwork model construction module 1 is used to construct a hypernetwork model for the unmanned system; the hypernetwork model includes a physical layer network, a collaborative layer network, and a task layer network.

[0191] The perturbation model construction module 2 is used to construct a physical perturbation model according to the physical layer of the unmanned system, construct a collaborative perturbation model according to the collaborative layer of the unmanned system, and construct a task perturbation model according to the task layer of the unmanned system. The physical perturbation model includes at least one physical layer perturbation category sub-model, the collaborative perturbation model includes at least one collaborative layer perturbation category sub-model, and the task perturbation model includes at least one task layer perturbation category sub-model.

[0192] The combined test case generation module 3 is used to encode each sub-model in the constructed physical perturbation model, the task perturbation model, and the collaborative perturbation model and generate a plurality of combined test cases.

[0193] The standard value determination module 4 of the elastic metric performance parameter is used to determine a plurality of elastic metric performance parameters of the unmanned system and the standard value of each elastic metric performance parameter.

[0194] The test case set construction module 5 is used to select combined test cases from a plurality of combined test cases according to the elastic test requirements to form a test case set.

[0195] The elastic value determination module 6 of the unmanned system is used to sequentially execute each combined test case in the test case set in the unmanned system, obtain the test value of the elastic metric performance parameter, and determine the elastic value of the unmanned system according to the test value and the standard value.

[0196] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0197] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for determining the resilience of an unmanned system based on combinatorial testing, characterized in that, Including: Constructing a hypernetwork model of an unmanned system; the hypernetwork model includes a physical layer network, a collaboration layer network, and a task layer network; Constructing a physical perturbation model according to the physical layer of the unmanned system, constructing a collaboration perturbation model according to the collaboration layer of the unmanned system, and constructing a task perturbation model according to the task layer of the unmanned system. The physical perturbation model includes at least one physical layer perturbation category sub-model, the collaboration perturbation model includes at least one collaboration layer perturbation category sub-model, and the task perturbation model includes at least one task layer perturbation category sub-model; Encoding each sub-model in the physical perturbation model, the task perturbation model, and the collaboration perturbation model and generating multiple combined test cases; Determining multiple elastic metric performance parameters of the unmanned system and the standard values of each elastic metric performance parameter; the elastic metric performance parameters include physical layer performance parameters, collaboration layer performance parameters, and task layer performance parameters; Selecting combined test cases from multiple combined test cases according to the elastic test requirements to form a test case set; In the unmanned system, sequentially executing each combined test case in the test case set to obtain the test values of the elastic metric performance parameters, and determining the elasticity value of the unmanned system according to the test values and the standard values.

2. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 1, wherein The nodes in the physical layer network are unmanned execution entities, and the edges in the physical layer network are information interactions between each unmanned execution entity; the nodes in the collaboration layer network are physical entity unit groups, and the physical entity unit group includes multiple unmanned execution entities that execute set tasks. The edges in the collaboration layer network are the allocation and scheduling instructions of the collaboration layer of the unmanned system to the physical entity unit group; the nodes in the task layer network are sub-tasks obtained by disassembling the top-level task target by the task layer of the unmanned system, and the edges in the task layer network are the information flows of the disassembled sub-tasks.

3. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 2, wherein The physical entity unit group is determined by the collaboration layer using a collaboration layer algorithm model to dynamically integrate each unmanned execution entity based on the collected sensor parameters; the collaboration layer algorithm model includes a MADDPG algorithm model, and the sensor parameters include the speeds of each unmanned execution entity.

4. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 3, wherein The physical layer perturbation category sub-model includes that the unmanned execution entity cannot participate in the resource integration and scheduling of the collaboration layer.

5. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 3, wherein The collaboration layer perturbation category sub-models include that the collaboration layer algorithm cannot give action instructions, the physical entity unit group cannot correctly execute the instructions given by the collaboration layer algorithm, and there is no communication between any two physical entity unit groups.

6. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 3, wherein The task layer perturbation category sub-models include that the task layer cannot disassemble and allocate tasks and there are losses in task allocation due to communication failures between nodes in the task layer and the collaboration layer.

7. The method for determining the resilience of an unmanned system based on combinatorial testing according to claim 1, wherein Determining multiple elastic metric performance parameters of the unmanned system and the standard values of each elastic metric performance parameter specifically includes: For each elastic metric performance parameter, obtaining the mean curve of the elastic metric performance parameter in the normal operation state of the unmanned system, and using the mean curve as the standard value of the elastic metric performance parameter.

8. An unmanned system resilience determination system based on combinatorial testing, characterized in that, Including: A hypernetwork model construction module for constructing a hypernetwork model of an unmanned system; the hypernetwork model includes a physical layer network, a collaboration layer network, and a task layer network; A perturbation model construction module for constructing a physical perturbation model according to the physical layer of the unmanned system, constructing a collaboration perturbation model according to the collaboration layer of the unmanned system, and constructing a task perturbation model according to the task layer of the unmanned system. The physical perturbation model includes at least one physical layer perturbation category sub-model, the collaboration perturbation model includes at least one collaboration layer perturbation category sub-model, and the task perturbation model includes at least one task layer perturbation category sub-model; A combined test case generation module for encoding each sub-model in the physical perturbation model, the task perturbation model, and the collaboration perturbation model and generating a plurality of combined test cases; A standard value determination module for elastic metric performance parameters for determining a plurality of elastic metric performance parameters of the unmanned system and the standard value of each elastic metric performance parameter; A test case set construction module for selecting combined test cases from a plurality of combined test cases according to elastic test requirements to form a test case set; An elastic value determination module for the unmanned system, for sequentially executing each combined test case in the test case set in the unmanned system, obtaining a test value of the elastic metric performance parameter, and determining the elastic value of the unmanned system according to the test value and the standard value.

Images