A resilient optimization method for multi-UAV cooperative search under global dynamic network attacks

Abstract

This invention discloses an elastic optimization method for multi-UAV cooperative search under global dynamic network attacks, comprising the following steps: S1, establishing a multi-UAV node network topology and a Byzantine network attack model; S2, using an elastic convex combination operator algorithm to ensure that each node receives information only from its unattacked neighbors in an elastic convex combination; S3, transforming the elastic convex combination operator problem into a linear programming problem; S4, introducing intermediate variables to transform the multi-UAV cooperative search problem into an optimization problem; S5, updating the state of each UAV node using gradient descent; S6, incorporating the elastic convex combination operator from step S3 into the distributed optimization algorithm from step S5, updating the state of each UAV, and determining whether consensus has been reached. This invention enables UAVs to resist Byzantine network attacks in multi-UAV cooperative search tasks under global dynamic network attacks, achieving distributed optimization for multi-UAV cooperative search.

Description

Technical Field

[0001] This invention relates to the field of robot intelligent algorithms, and in particular to a resilient optimization method for multi-UAV cooperative search under global dynamic network attacks. Background Technology

[0002] A multi-drone system is a multi-agent system based on drones, in which multiple drones work together to complete a task. With the continuous development and popularization of drone technology, the application background of distributed multi-drone systems is very wide, involving multiple fields, such as search and rescue, agriculture, building inspection, forest fire prevention, and urban management.

[0003] Multi-drone collaborative search technology is a technique that utilizes drones for search and rescue missions, widely applied in complex or hazardous environments such as earthquakes and volcanoes. In this technology, multiple drones can form a swarm, working collaboratively through distributed control. Multi-drone collaborative search offers the following advantages: efficiency, robustness, flexibility, safety, and versatility.

[0004] In distributed multi-UAV cooperative search, each UAV has its own cost function, reflecting the cost or benefit of completing the task. Because each UAV's cost function differs, coordination and planning of the UAV swarm are necessary to minimize the largest cost function within the swarm. To achieve this coordination and planning, distributed control methods are required. Distributed control means that each UAV has its own controller, and the UAVs exchange information and coordinate with each other through communication. In distributed control, each UAV can formulate its own action plan based on its observations and mission requirements. Simultaneously, the UAVs can coordinate and adjust their action plans to reach a consensus.

[0005] However, in distributed multi-drone systems, byzantine attacks can occur due to factors such as communication failures, software vulnerabilities, and hardware failures. Some nodes may propagate incorrect information, disrupting the system's coordination and causing errors and confusion in information transmission between drones, thus affecting the coordinated operation of the entire system.

[0006] Therefore, ensuring the security and reliability of distributed multi-UAV systems in the face of such global dynamic Byzantine network attacks has become an urgent problem to be solved. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to provide an elastic optimization method for multi-UAV cooperative search under global dynamic network attacks, aiming to fully utilize the advantages of elastic convex combination optimization among multiple UAVs to achieve the security and robustness of the entire UAV system under global dynamic network attacks.

[0008] Technical solution: The present invention provides an elastic optimization method for multi-UAV cooperative search under global dynamic network attacks, comprising the following steps:

[0009] S1, Establish a multi-drone node network topology and a Byzantine network attack model;

[0010] S2, through the elastic convex combination operator algorithm, enables each node to receive information only from its unattacked neighbors in an elastic convex combination.

[0011] S3 transforms the problem of elastic convex combination operators into a linear programming problem;

[0012] S4 introduces intermediate variables to transform the multi-UAV cooperative search problem into an optimization problem;

[0013] S5 updates the state of each drone node using gradient descent.

[0014] S6. Add the elastic convex combination operator from step S3 to the distributed optimization algorithm in step S5 to update the state of each drone. If each drone has reached a consensus, minimize the cost function of the drone with the maximum loss in the entire group, and the cooperative search task ends. Otherwise, continue to update the state of each drone.

[0015] Furthermore, in step S1, the network topology of the multi-UAV node network is as follows: Assume a time-invariant directed graph. Represents a set of N nodes. Let ε represent the set of all one-way communication channels between nodes, where (i,j)∈ε indicates that node i can obtain information from other nodes j. This is a time-invariant directed graph. There must exist a one-way path from every node i to every other node j in the graph;

[0016] The Byzantine network attack model is as follows: at each distance from the walk... The attacked node is represented as gather This represents nodes that have not been attacked; each node i cannot know which neighbors have been attacked, but can obtain an integer at each time step k. The upper bound representing the number of attacked neighbors.

[0017] Furthermore, in step S2, the algorithm for the elastic convex combination operator is as follows:

[0018] Construct set Each element is a subset of information received from neighboring nodes:

[0019]

[0020] in, This represents the state information of node j at step size k.

[0021] subset Total Possibilities, therefore the set Represented as:

[0022]

[0023] Generate a set convex hull in The operator represents the solution for the convex hull of a set, where "∪" represents the union and "∩" represents the intersection, and it obtains the intersection of all convex hulls.

[0024] Choose intersection any point in An elastic convex combination of information received only by neighbors that have never been attacked.

[0025] Furthermore, in step S3, the following convex combination form is used to represent...

[0026]

[0027] in, coefficient satisfy

[0028] coefficient This is determined by solving the following linear programming problem:

[0029]

[0030] Each of them It includes all coefficients to be solved. column vectors, Aggregate collection In the context of all states, 1 represents a column vector where every element is 1. Let κ be the transpose of the vector, and κ be the variable to be optimized, representing the state. For each The first column in the table.

[0031] Furthermore, in step S4, the solution method for the original multi-UAV cooperative search optimization problem is as follows:

[0032] In a multi-UAV cooperative search task, the original optimization problem is:

[0033]

[0034] in, It is a private local cost function maintained by the i-th drone node, and this function is a convex function.

[0035] By introducing an auxiliary scalar variable The original optimization problem is transformed into the following form:

[0036]

[0037] Representing constraints as sets In addition, the two decision variables x and t are bound to a new variable. Based on this, the original optimization problem can be further expressed in the following form:

[0038]

[0039] in, It is the (p+1)th column of the identity matrix.

[0040] Furthermore, in step S5, the gradient descent method is used to update the state of each UAV node as follows: Each node i is updated at each time step... Maintain local state Iterative updates are performed using the following gradient descent method:

[0041]

[0042]

[0043]

[0044] Among them, the function Defined as h i (θ):=f i (x)-t, Represented as function h i The gradient of (·), The corresponding non-negative weights, α, when combining information from neighbors. k >0 and γ k >0 represents two step size types; operators Defined as [x] + := max{x,0} Indicates in set Euclidean projection on.

[0045] Furthermore, in step S6, the state of each UAV is updated using an elastic distributed min-max algorithm, and the steps are as follows:

[0046] S61 receives the states of all attacked and unattacked neighboring nodes, performs a resilient convex combination, and obtains...

[0047] S62, perform the gradient descent step to obtain intermediate results.

[0048] S63, based on intermediate results Update status

[0049] S64, make k→k+1, and continue the loop.

[0050] A device for multi-UAV cooperative search under a global dynamic network attack includes a memory and a processor, wherein:

[0051] Memory is used to store computer programs that can run on a processor;

[0052] The processor is configured to execute, while running the computer program, the elastic optimization method for multi-UAV cooperative search under any of the aforementioned global dynamic network attacks.

[0053] A storage medium storing a computer program, which, when executed by at least one processor, implements an elastic optimization method for multi-UAV cooperative search under any of the aforementioned global dynamic network attacks.

[0054] Compared with the prior art, the significant advantages of this invention are as follows:

[0055] 1. This invention establishes a multi-UAV topology network, enabling UAVs to communicate and exchange information with each other. This allows the UAV swarm to eliminate interference from erroneous information by only acquiring the upper bound of the attacked node among its neighbors, thereby achieving consensus on collaborative search tasks.

[0056] 2. By using the elastic convex combination algorithm, each node can achieve an elastic convex combination that only receives information from its unattacked neighbors. This can filter out erroneous information, enable the drone swarm to reach a consensus, and maximize search efficiency in collaborative search tasks. Attached Figure Description

[0057] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0058] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0059] This problem applies to any drone node in a collaborative search distributed system. Due to the large search area, multiple drones need to cooperate in the search. Each drone has its own cost function to measure its search efficiency and cost. The drone swarm needs to reach a consensus to minimize the cost function with the highest cost across the entire swarm. This is a typical distributed minimization-maximization problem, with the goal of collaboratively finding the minimum of the global function.

[0060]

[0061] in, It is a private local cost function maintained by the i-th drone node, and this function is a convex function. It is a global function that is subject to a convex feasible set. constraint; Let p represent a p-dimensional real space, where p is the dimension of the decision variable.

[0062] The distributed system of this invention includes N consensus nodes. During the collaborative search process, there is a global dynamic network attack. The attack target changes over time, and the attacked node will send erroneous information to the UAV network. However, each node can recover at least once within a certain time window and send the correct information. This method enables the UAV swarm to eliminate the interference of erroneous information and achieve consensus on the collaborative search task by only acquiring the upper bound of the attacked node among its neighbors.

[0063] like Figure 1 As shown, the elastic optimization method for multi-UAV cooperative search under global dynamic network attacks of the present invention includes the following steps:

[0064] Step 1: Establish a multi-UAV node network topology and a Byzantine network attack model;

[0065] Step 11: Establish the network topology of multiple unmanned aerial vehicle (UAV) nodes.

[0066] Consider a time-invariant directed graph. Let N be the set of N drone nodes (hereinafter referred to as nodes). Let ε represent the set of all one-way communication channels between nodes, where (i,j)∈ε indicates that node i can obtain information from other nodes j. This is a time-invariant directed graph. It is strongly connected, meaning that there must exist a unidirectional path from every node i to every other node j in the graph.

[0067] Step 12, Establish a Byzantine network attack model

[0068] In a multi-UAV cooperative search system, each UAV node maintains its own local cost function f. iIn the collaborative target search process, drone nodes acquire cost function information sent by neighboring nodes, thereby enabling the drone swarm to reach a consensus. During information acquisition, the local information maintained by each drone is reliable; network attacks only occur within the communication channel, where individual drones may send incorrect information, affecting the swarm's consensus. The target of the attacked drone changes over time, but each attacked drone can recover at least once within a certain time window, allowing the attacked target to transmit correct local information to its neighboring nodes. At each distance k, the attacked node is represented as... gather This represents nodes that have not been attacked. Each node i cannot know which neighbors have been attacked, but it can obtain an integer at each time step k. The upper bound representing the number of attacked neighbors. It is the set of positive integers.

[0069] Step 2: Using the elastic convex combination operator algorithm, each node achieves an elastic convex combination that only receives information from its unattacked neighbors.

[0070] Will This represents the local state of the cost function maintained by each node, where each node i can obtain the number d of its neighboring nodes. i and the upper bound of the number of attacked neighbors. for All satisfied The elastic convex combination operator can eliminate the impact of erroneous information on the achievement of consensus in a drone swarm even without knowing which neighboring nodes have been attacked. The solution is as follows:

[0071] Construct set Each element is a subset of information received from neighboring nodes:

[0072]

[0073] in, This represents the state information of node j at step size k, a subset. Total Possibilities, therefore the set It can be represented as:

[0074]

[0075] Generate a set convex hull in The operator represents the solution for the convex hull of a set, where "∪" represents the union and "∩" represents the intersection, and it obtains the intersection of all convex hulls.

[0076]

[0077] Choose intersection any point in As a flexible convex combination of information received only by neighbors that have never been attacked, it enables the filtering of error messages.

[0078] Step 3: Transform the elastic convex combination operator problem into a linear programming problem;

[0079] Based on step two, it can be represented by the following convex combination form.

[0080]

[0081] in, coefficient satisfy

[0082] coefficient This can be determined by solving the following linear programming problem:

[0083]

[0084] Each of them It includes all coefficients to be solved. column vectors, Aggregate collection In the context of all states, 1 represents a column vector where every element is 1. Let κ be the transpose of the vector, and κ be the variable to be optimized, representing the state. For each The first column in the table.

[0085] Elastic convex combination operator The algorithm is shown in Table 1.

[0086] Table 1 Elastic Convex Combination Operators algorithm

[0087]

[0088] Step four: Introduce intermediate variables to transform the multi-UAV cooperative search problem into an optimization problem;

[0089] In a multi-UAV cooperative search task, the original optimization problem is:

[0090]

[0091] in, It is a private local cost function maintained by the i-th drone node, and this function is a convex function.

[0092] By introducing an auxiliary scalar variable The original optimization problem can be transformed into the following form:

[0093]

[0094] Representing constraints as sets In addition, the two decision variables x and t are bound to a new variable. Based on this, the original optimization problem can be further formulated into the following compact form:

[0095]

[0096] in, is the (p+1)th column of the identity matrix, and "T" indicates matrix transpose.

[0097] Step 5: Update the state of each drone node using gradient descent.

[0098] Make each node i at each time step Maintain local state Iterative updates are performed using the following gradient descent method:

[0099]

[0100] Among them, the function Defined as h i (θ):=f i (x)-t, Represented as function h i The gradient of (·), This is the (p+1)th column of the identity matrix. The corresponding non-negative weights, α, when combining information from neighbors. k >0 and γ k >0 represents two step size types, and the operator... Defined as [x] + := max{x,0} Indicates in set Euclidean projection on.

[0101] Step six: Add the elastic convex combination operator from step three to the distributed optimization algorithm from step five to update the state of each drone: if each drone has reached a consensus, minimize the cost function of the drone with the maximum loss in the entire group, and the cooperative search task ends; otherwise, continue to update the state of each drone.

[0102] During the iteration in step five, the only process involving communication is Each node needs to combine (through a weighted average) the information received from all its neighbors. To handle potential communication interruptions, the resilient convex combination operator from step three is used. Instead of the pure weighted average, the following is the elastic distributed min-max algorithm, as shown in Table 2.

[0103] Table 2. Elastic Distributed Mini-Max Algorithm

[0104]

[0105] In summary, this invention addresses the challenge of maximizing search efficiency in collaborative search tasks when drones, unaware of which neighboring nodes are under attack in a globally dynamic network attack, can achieve consensus through an elastic convex combination algorithm that filters out erroneous information.

[0106] A device for multi-UAV cooperative search under a global dynamic network attack includes a memory and a processor, wherein:

[0107] Memory is used to store computer programs that can run on a processor;

[0108] The processor is configured to, while running the computer program, execute any of the aforementioned resilient optimization methods for multi-UAV cooperative search under global dynamic network attacks.

[0109] A storage medium storing a computer program, which, when executed by at least one processor, implements an elastic optimization method for multi-UAV cooperative search under any of the aforementioned global dynamic network attacks.

Claims

1. A resilient optimization method for multi-UAV cooperative search under a global dynamic network attack, characterized in that, The steps include the following: S1, Establish a multi-drone node network topology and a Byzantine network attack model; S2, through the elastic convex combination operator algorithm, enables each node to receive information only from its unattacked neighbors in an elastic convex combination. S3 transforms the problem of elastic convex combination operators into a linear programming problem; S4 introduces intermediate variables to transform the multi-UAV cooperative search problem into an optimization problem; S5 updates the state of each drone node using gradient descent. S6, add the elastic convex combination operator from step S3 to the distributed optimization algorithm in step S5, and update the state of each drone; if each drone has reached a consensus, minimize the cost function of the drone with the largest loss in the entire group, and the cooperative search task ends. Otherwise, continue updating the status of each drone; In step S5, the gradient descent method is used to update the state of each UAV node as follows: Each node... At each time step Maintain local state Iteratively update according to the following gradient descent method: ， Among them, the function Defined as , Represented as a function gradient, The corresponding non-negative weights when combining information from neighbors and Two step size types; operators Defined as , Indicates in set Euclidean projection on; It is the set of positive integers; express 3D real space, where For decision variables; In step S1, the multi-UAV node network topology is as follows: Assume a time-invariant directed graph. , express A set of nodes, This represents the set of all one-way communication channels for all nodes. Represents a node Able to obtain other nodes Information, the time-invariant directed graph. The graph must contain elements from each node. to each other node A one-way path; The Byzantine network attack model is as follows: At each distance from the walk... The attacked node is represented as ,gather This represents nodes that have not been attacked; each node It's impossible to know which neighbors are under attack, but it's possible at each time step. Get integer , An upper bound representing the number of attacked neighbors; In step S2, the algorithm for the elastic convex combination operator is as follows: Construct set Each element is a subset of information received from neighboring nodes: ， in, Indicates the step size Time node Status information, subset Total a possibility, For nodes The number of neighboring nodes; therefore, the set Represented as: ， Generate a set convex hull ,in This represents the operator for finding the convex hull of a set. "represents the union of sets," " represents the intersection, and obtains the intersection of all convex hulls." , ; Choose intersection any point in As an elastic convex combination of information received only by neighbors that have never been attacked; In step S3, the following convex combination form is used to represent... : ， in, ,coefficient satisfy , ; coefficient This is determined by solving the following linear programming problem: ， Each of them It includes all coefficients to be solved. column vectors, Aggregates collections All states in This is represented as a column vector where every element is 1. This is the transpose of the vector. The variable to be optimized is the state. For each The first column in; In step S4, the solution method for the original multi-UAV cooperative search optimization problem is as follows: In a multi-UAV cooperative search task, the original optimization problem is: ， in, It is the first Each drone node maintains a private local cost function, which is a convex function. ; By introducing an auxiliary scalar variable The original optimization problem is transformed into the following form: ， Representing constraints as sets , Furthermore, the two decision variables... and Bind to a new variable Based on this, the original optimization problem can be further expressed in the following form: ， in, The first unit of the identity matrix List.

2. The elastic optimization method for multi-UAV cooperative search under global dynamic network attacks as described in claim 1, characterized in that, In step S6, the state of each UAV is updated using an elastic distributed min-max algorithm, and the steps are as follows: S61 receives the states of all attacked and unattacked neighboring nodes, performs a resilient convex combination, and obtains... ; S62, perform the gradient descent step to obtain intermediate results. ; S63, based on intermediate results Update status ; S64, makes The cycle continues.

3. A device for multi-UAV cooperative search under a global dynamic network attack, characterized in that, Includes memory and processor, wherein: Memory is used to store computer programs that can run on a processor; A processor, configured to, while running the computer program, execute the elastic optimization method for multi-UAV cooperative search under global dynamic network attacks as described in any one of claims 1-2.

4. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by at least one processor, implements the elastic optimization method for multi-UAV cooperative search under global dynamic network attacks as described in any one of claims 1-2.

Images