A privacy protection and mean convergence control method for unmanned swarm systems
By adopting event-triggered privacy protection and mean convergence control methods in unmanned cluster systems, the problems of node information privacy leakage and DoS attacks are solved, and effective privacy protection and system stability are achieved.
Patent Information
- Application Number
- CN202211580974.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-09
- Publication Date
- 2025-04-22
- Estimated Expiration
- 2042-12-09
AI Technical Summary
Unmanned cluster systems face privacy leakage problems in node information when implementing mean convergence control, and are vulnerable to cyber attacks such as DoS attacks in open network environments.
The privacy protection and mean convergence control method of unmanned cluster systems based on event triggering is adopted. By establishing a node dynamic model and a DoS attack model, the Laplace distribution generates noise for differential privacy protection, and the event trigger control protocol is designed to resist DoS attacks.
It effectively protects the privacy of the initial state value of the node and the state value during the update process, and can still achieve mean convergence control in the presence of DoS attacks, resisting the impact of DoS attacks.
Smart Images
Figure CN115967550B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of coordinated control of unmanned cluster systems, and in particular to a privacy protection and mean convergence control method for unmanned cluster systems based on event triggering. Background Art
[0002] Unmanned swarm system is a self-organizing, highly stable distributed system in which multiple unmanned systems (intelligent agents or nodes) with common mission objectives are divided into tasks and, with the help of modern control and decision-making methods, maintain the unmanned systems in the form of a group to collaboratively complete complex tasks within a certain time and space. Clustering is not a simple connection and combination of multiple individuals, but rather enables these individuals to collaborate efficiently, expand the perception of environmental situations through group information sharing, realize collaborative task allocation and coordination, better complete the work that a single individual cannot complete, and effectively improve the ability of the group to complete complex tasks. In recent years, intelligent unmanned systems have developed rapidly, and a series of new products such as drones, unmanned vehicles, unmanned ships, and robots have emerged. Convergence control means that in a distributed system, each node is assigned a state value at the initial moment. It sends and receives information to neighboring nodes and iteratively updates its state value. Finally, it can calculate the state value of all nodes to reach a consensus. In recent years, the distributed convergence control problem of unmanned swarm systems has received much attention due to its wide application, such as intelligent transportation systems, sensor networks, and regional measurement, control, and search. The mean consensus control is to design a control method so that each node converges to the average of the initial state values of all nodes in the system. At the same time, the so-called privacy protection of unmanned cluster systems refers to the protection of the initial state value of the node and the state value during the update process in the process of achieving mean convergence, so that eavesdroppers in the network cannot accurately infer the initial state value of the node. In this case, the node is said to have obtained privacy protection.
[0003] At present, the distributed unmanned cluster system network faces two key problems: (1) The privacy leakage of node information. In order to achieve convergence of the distributed network, each node exchanges the real state value with its neighbors in the process of reaching convergence in the traditional convergence control algorithm. However, in some cases, the initial state of the node is private, so the node hopes to keep its sensitive information confidential. For example, compared with ordinary power grids, smart grids can read the measurement data of each meter through distributed sensors, thereby better achieving load balancing, but opponents can track the living conditions of residents by analyzing the data, posing a serious threat to the privacy of residents. (2) Unmanned cluster systems are generally exposed to an open network environment, so they are vulnerable to network attacks, such as common DoS (denial-of-service) attacks and deception attacks. Summary of the invention
[0004] The purpose of the present invention is to provide a privacy protection and mean convergence control method for an unmanned cluster system in view of the deficiencies in the prior art, which protects the initial state values of nodes and is suitable for a network environment where DoS attacks exist.
[0005] The method of the invention first establishes a node dynamic model and a DoS attack model, and then uses the node dynamic model and the DoS attack model to control the unmanned cluster system.
[0006] The node dynamic model is specifically:
[0007] n is the number of nodes in the unmanned cluster system network; x i (k) and x j (k) represents the i-th node v i and its neighbor node v j The state value at time k; ζ is a constant step between (0,1); For node v i The neighbor set of a ji For node v i With node v j The weight between them; if the node v i and node v j There is a communication channel between nodes v j For node v i The neighboring node of Then a ji > 0, otherwise a ji = 0. In addition, this method does not consider the node v i There is a case of own weight, that is, a ii =0.
[0008] The DoS attack model is specifically:
[0009] Set the attack frequency of DoS attack: For each communication link, there is β f ∈[0.5,1), so that Γ f (k,k+T) satisfies Γ f (k,k+T)≤β f ×T. Among them, β f is the frequency limiting constant of DoS attack. T=mk is an attack cycle. And in each attack period T, the DoS attack occurs independently. f (k, k+T) is the frequency of DoS attacks on each link in the time interval [k, k+T).
[0010] Set the duration of DoS attack: Each communication link has Γ f ≥0, so that Ψ d Satisfy d ×Γ f ≤T-1, where Ψ d is the duration of a DoS attack on each link within the time interval [k, k+T).
[0011] The specific control methods are:
[0012] Step (1) System initialization: Let k = 0, and assign an initial state value x to each node. i (0);
[0013] Step (2) Each node v i Select Laplace distribution, Laplace distribution Lap(k,μ,λ) is expressed as: μ is the mathematical expectation, parameter σ is the variance;
[0014] Step (3) Node v i Randomly generate a Laplace distributed variable η i (0), and randomly generate a pseudo-random value s i (0), For node v i The in-degree of the initial noise value θ i (0) = η i (0)+h*s i (0), the new state value after adding noise at the initial moment
[0015] Node v i Create a cache To store the state value after adding noise, Pass to neighbor node v j ;
[0016] Step (4) Status update:
[0017] Node v i After receiving the state value of the neighbor node, the state value is updated to:
[0018]
[0019] Step (5) Design control protocol:
[0020] Node v i Create cache area M i Store messages sent by neighboring nodes:
[0021] Represents node v i The triggering moment, if the node v i If the trigger condition is met, the node status value is updated according to the update rule, and then k is increased by 1 until convergence is reached; Represents node v i Neighbor nodes At the triggering moment The status value of Indicates the trigger time The cache collection;
[0022] Node v i The event trigger function is
[0023] Each node v i Generate a pseudo-random value s i (1) For node v i The out-degree of node v i After receiving the state value of the neighbor node, if the received value is the same as the state value of node v i If the state value required for the update is the same at all times, Stored in cache space M i Otherwise, the received value is discarded; if the node satisfies the event trigger function, node v i Update the status value and then clear M i , k=k+1, and then randomly generate a Laplace distribution variable to calculate the noise Where 0<ρ<1; adding noise to the true state value of the node, we have Then Stored in and will Broadcast to neighboring nodes; if no event is triggered, keep the node's state value unchanged;
[0024] Step (6) loop setting: if an event is triggered, repeat step (5), otherwise the actual state value of the node remains unchanged.
[0025] The beneficial effects of the method of the present invention include:
[0026] 1. The method of the present invention describes the corresponding DoS attack model based on the characteristic that DoS attacks can block the information interaction between nodes in an unmanned cluster system. Compared with the traditional differential privacy control algorithm, the method designed by the present invention can effectively resist DoS attacks.
[0027] 2. Since there may be curious or malicious information eavesdropping in the real environment, information leakage is inevitable, so it is very necessary to protect privacy. The method of the present invention can protect the privacy of the initial state value of the node and the privacy of the state value during the update process.
[0028] 3. The method of the present invention designs a distributed event triggering mechanism, and only nodes that meet the triggering function can be updated, thereby reducing the frequency of node updates and saving resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a directed topological structure diagram of an unmanned cluster system composed of five nodes established by the present invention.
[0030] Figure 2 This is the performance diagram of each node without adding pseudo-random values.
[0031] Figure 3 This is a graph showing the performance of each node with pseudo-random values added.
[0032] Figure 4 This is an example diagram of a DoS attack.
[0033] Figure 5 This is a diagram showing the status updates of each node under DoS attack with differential privacy added but without event-triggered control method.
[0034] Figure 6 This is a state update representation diagram of each node with differential privacy and event-triggered control method added under DoS attack.
[0035] Figure 7 It is a diagram showing the event triggering time of each node. DETAILED DESCRIPTION
[0036] The technical solution of the present invention is further explained below through specific embodiments.
[0037] This embodiment provides a privacy protection and mean convergence control method for an unmanned cluster system. First, a node dynamic model and a DoS attack model are established, and then the node dynamic model and the DoS attack model are used to control the unmanned cluster system.
[0038] The node dynamic model is specifically:
[0039] The dynamic model of the i-th node (node) in the unmanned cluster system network is:
[0040] n is the number of nodes in the unmanned cluster system network; x i (k) and x j (k) represents the i-th node vi and its neighbor node v j The state value at time k; ζ is a constant step between (0,1); For node v i The neighbor set of a ji For node v i With node v j The weight between them; if the node v i and node v j There is a communication channel between nodes v j For node v i The neighboring node of Then a ji > 0, otherwise a ji = 0. In addition, this method does not consider the node v i There is a case of own weight, that is, a ii =0.
[0041] DoS attack is a common type of existing malicious attack, and the attack occurs on the communication link between nodes in the unmanned cluster system. The DoS attack discussed in the present invention is a situation where the link information interaction between each node in the unmanned cluster system fails. During the active period of the DoS attack, that is, when a DoS attack occurs on the communication link between two nodes, the link between the nodes is cut off, and the information interaction is blocked, resulting in the system being unable to achieve accurate mean convergence. Moreover, from the perspective of resource constraints, the energy of a DoS attack is limited, and the attack needs to sleep for a period of time to accumulate energy for the next action. On different communication links, whether or not they are attacked is independent of each other.
[0042] The DoS attack model is specifically:
[0043] Set the attack frequency of DoS attack: For each communication link, there is β f ∈[0.5,1), so that Γ f (k,k+T) satisfies Γ f (k,k+T)≤β f ×T. Among them, β f is the frequency limiting constant of DoS attack. T=mk is an attack cycle. And in each attack period T, the DoS attack occurs independently. f (k, k+T) is the frequency of DoS attacks on each link in the time interval [k, k+T).
[0044] Set the duration of DoS attack: Each communication link has Γ f ≥0, so that Ψ d Satisfy d ×Γ f ≤T-1, where Ψd is the duration of a DoS attack on each link within the time interval [k, k+T).
[0045] The specific method of controlling the unmanned cluster system using the node dynamic model and DoS attack model is:
[0046] Step (1) System initialization: Let k = 0, and assign an initial state value x to each node. i (0).
[0047] Step (2) Select distribution: The present invention selects differential privacy method to achieve privacy protection. Since Laplace distribution can achieve different degrees of privacy for each node, while other distributions cannot, the present invention uses Laplace distribution to generate noise. Each node v i Select Laplace distribution, Laplace distribution Lap(k,μ,λ) is expressed as: μ is the mathematical expectation, parameter σ is the variance.
[0048] Step (3) Node v i Randomly generate a Laplace distributed variable η i (0), there may be honest but curious eavesdroppers in the unmanned cluster system. They will honestly follow the system's update rules to update their own state values, but they will collect information about themselves and their neighboring nodes during the information interaction between nodes, and information leakage is inevitable. Therefore, in order to avoid the leakage of information by honest but curious eavesdroppers and protect the privacy of the node's initial state value, the method of the present invention randomly generates and adds a pseudo-random value s to the noise i (0), For node v i The in-degree of the initial noise value θ i (0) = η i (0)+h*s i (0), the new state value after adding noise at the initial moment
[0049] Node v i Create a cache To store the state value after adding noise, Pass to neighbor node v j .
[0050] Step (4) Status update:
[0051] Node v i After receiving the state value of the neighbor node, the state value is updated to:
[0052]
[0053] Step (5) Design control protocol:
[0054] In order to enable the unmanned cluster system to achieve accurate mean convergence control even under DoS attacks, the present invention adopts an event-triggered control method.
[0055] Node v i Create cache area M i Store messages sent by neighboring nodes:
[0056] Represents node v i The triggering moment, if the node v i If the trigger condition is met, the node status value is updated according to the update rule, and then k is increased by 1 until convergence is reached; Represents node v i Neighbor nodes At the triggering moment The status value of Indicates the trigger time The cache collection;
[0057] Node v i The event trigger function is
[0058] Each node v i Generate a pseudo-random value s i (1) For node v i The out-degree of node v i After receiving the state value of the neighbor node, if the received value is the same as the state value of node v i If the state value required for the update is the same at all times, Stored in cache space M i Otherwise, the received value is discarded; if the node satisfies the event trigger function, node v i Update the status value and then clear M i , k=k+1, and then randomly generate a Laplace distribution variable to calculate the noise Where 0<ρ<1; adding noise to the true state value of the node, we have Then Stored in and will Broadcast to neighboring nodes; if no event is triggered, the node's state value remains unchanged.
[0059] Step (6) loop setting: if an event is triggered, repeat step (5), otherwise the actual state value of the node remains unchanged.
[0060] In order to verify the effectiveness of the control method designed by the present invention, the following implementation examples are used for verification.
[0061] Consider an unmanned cluster system consisting of five nodes. Its directed communication topology diagram is shown in Figure 1 The initial state value of each node is x(0) = [33, 11, 22, 36, 15], and the average value of the initial state of all nodes is 23.4. The system weight A = [a ij ]The values are as follows:
[0062]
[0063] Choose ρ = 0.8, ζ = 0.76, We also consider the case where there is a curious eavesdropper in the unmanned cluster system trying to guess the initial state value of node i. We can establish an observer (curious eavesdropper) to estimate the true state value of node 1, and its update rule is as follows:
[0064] The initial state value of the observer is set to the initial state value of node 1 with noise added, that is, The update of the observer is based on the accumulation of the difference between the sent state value and the estimated state value of node 1, that is,
[0065] In the absence of DoS attacks and without adding pseudo-random values, the node status trajectory of the system can be found in Figure 2 As can be seen in the figure, the solid line represents the state trajectory change of nodes 1-5, and the solid line marked with * represents the observer's estimated value of node 1. Each node in the system can achieve accurate mean convergence, but the observer can clearly observe the initial state value of node 1.
[0066] In the absence of DoS attacks, adding pseudo-random values to differential privacy, Figure 3 It represents the change trajectory of the state value of each node. As can be seen from the figure, the differential privacy method with pseudo-random values can not only achieve accurate mean convergence, but also protect the initial state value of node 1 and avoid privacy leakage.
[0067] Next, consider the case of DoS attack and select the period of DoS attack T=5. Figure 4Indicates the DoS attack in the system (the shaded part in the figure is the attack time). At time k = 1, no link is attacked by DoS, and all nodes communicate normally; at time k = 2, links (1, 3), (2, 1), (2, 5) and (5, 4) are attacked by DoS, information transmission is blocked, and other links communicate normally; at time k = 3, links (2, 1), (2, 5), (4, 2) and (3, 5) are attacked by DoS, and links that are not attacked can communicate normally. Under DoS attack, Figure 5 The following figure shows the case where only the differential privacy method is used without adding the event-triggered control method. It can be seen that the nodes cannot reach a consensus during the update process, so the general method cannot resist DoS attacks.
[0068] Figure 6 The following is the case with differential privacy and event-triggered control methods. It can be clearly seen that under DoS attacks, each node in the system can achieve accurate mean convergence, and the observer does not estimate the initial state value of node 1. The method completely protects the initial state of node 1 and avoids privacy leakage by curious nodes. At the same time, the triggering time of each node can be found in Figure 7 When a node satisfies the given trigger function, the node updates its state value; if the trigger function is not satisfied, the node's state value remains unchanged.
[0069] Any description not involved in the specific implementation of the present invention belongs to the known technology in the field and can be implemented with reference to the known technology. The above specific implementation is a specific support for the technical idea of the privacy protection and mean convergence control method of an unmanned cluster system under a denial of service attack proposed by the present invention, and cannot be used to limit the protection scope of the present invention. Any equivalent changes or equivalent modifications made on the basis of the technical solution of the present invention in accordance with the technical idea proposed by the present invention still fall within the scope of protection of the technical solution of the present invention.
Claims
1. A privacy protection and mean convergence control method for an unmanned cluster system, characterized by: The steps include: Step (1) first establishes a node dynamic model and a DoS attack model; The DoS attack model is specifically: Set the attack frequency of DoS attack: For each communication link, there is β f ∈[0.5,1), so that Γ f (k,k+T) satisfies Γ f (k,k+T)≤β f ×T, where β f is the frequency limiting constant of DoS attack, T=mk is an attack cycle, And in each attack cycle T, DoS attacks occur independently, Γ f (k,k+T) is the frequency of DoS attacks on each link in the time interval [k,k+T), Set the duration of DoS attack: Each communication link has Γ f ≥0, so that Ψ d Satisfy d ×Γ f ≤T-1, where Ψ d is the duration of a DoS attack on each link within the time interval [k, k+T); Step (2) System initialization: Let k = 0, and assign an initial state value x to each node. i (0); Step (3) Each node v i Select Laplace distribution, Laplace distribution Lap(k,μ,λ) is expressed as: i=1,2,…,n;μ is the mathematical expectation, parameter σ is the variance; Step (4) Node v i Randomly generate a Laplace distributed variable η i (0), and randomly generate a pseudo-random value s i (0), For node v i The in-degree of For node v i The noise value θ at the initial moment i (0) = η i (0)+h*s i (0), the new state value after adding noise at the initial moment Node v i Create a cache To store the state value after adding noise, Pass to neighbor node v j ; Step (5) Status update: Node v i After receiving the state value of the neighbor node, the state value is updated to: Among them, ζ is a constant step between (0,1); x i (k) and x j (k) represents the i-th node v i and its neighbor node v j The state value at time k; a ji For node v i With node v j The weight between Step (6) designing a control protocol; The control protocol is designed as follows: Node v i Create cache area M i Store messages sent by neighboring nodes: Represents node v i The triggering moment, if the node v i If the trigger condition is met, the node status value is updated according to the update rule, and then k is increased by 1 until convergence is reached; Represents node v i Neighbor nodes At the triggering moment The status value of Indicates the trigger time The cache collection; Node v i The event trigger function is Each node v i Generate a pseudo-random value s i (1) For node v i The out-degree of node v i After receiving the state value of the neighbor node, if the received value is the same as the state value of node v i If the state value required for the update is the same at all times, Stored in cache space M i Otherwise, the received value is discarded; if the node satisfies the event trigger function, node v i Update the status value and then clear M i , k=k+1, and then randomly generate a Laplace distribution variable to calculate the noise Where 0<ρ<1; adding noise to the true state value of the node, we have Then Stored in and will Broadcast to neighboring nodes; if no event is triggered, keep the node's state value unchanged; Step (7) loop setting.
2. The privacy protection and mean convergence control method for an unmanned cluster system according to claim 1, characterized in that: The node dynamic model is specifically: i,j=1,2,…,n,i≠j,n is the number of nodes in the unmanned cluster system network; if node v i and node v j There is a communication channel between nodes v j For node v i The neighboring node of Then a ji > 0, otherwise a ji =0; In addition, this method does not consider the node v i There is a case of own weight, that is, a ii =0.
3. The privacy protection and mean convergence control method for an unmanned cluster system according to claim 2, characterized in that: The loop setting in step (7) is specifically as follows: if an event is triggered, step (6) is repeated, otherwise the real state value of the node remains unchanged.