A resilient distributed security monitoring method in a multi-sensor-observation network

Abstract

The present invention discloses a resilient distributed security monitoring method under a multi-sensor-observation network, which relates to the technical field of multi-agent security control technology. The key points of its technical solution are: first, a sensor-observation architecture and a corresponding communication topology channel are constructed according to the problem under study. Secondly, based on the measurement information of local sensors, each observer preliminarily constructs the observation information of the target system, and then exchanges information with neighboring observers through the corresponding communication topology. Finally, in view of the sparse characteristics of maliciously tampered information in the communication topology, an effective resilient information receiving and processing mechanism is analyzed and constructed for each local observer, so that all observers can still achieve safe and consistent state estimation of the monitored target under attack. Experimental results show that the method proposed in the present invention can effectively resist sparse malicious attacks in the observation communication topology, and ensure the safe and consistent monitoring of the true state of the target by the multi-sensor-observation network.

Description

Technical Field

[0001] The present invention relates to the field of multi-agent safety control technology, and more specifically, to a resilient distributed safety monitoring method under a multi-sensor-observation network. Background Art

[0002] In militarized warfare, in order to be familiar with the combat status of both sides in a harsh battlefield environment, battlefield intelligence is often obtained at great cost to improve one's own victory rate. Multi-agent systems, due to their distributed computing and strong damage resistance, provide strong support for highly intelligent, unmanned and autonomous combat environments, and have become an indispensable part of the military combat field. Therefore, in the field of battlefield intelligence monitoring and acquisition, the use of mutually unknown multi-sensor-observation networks to complete the distributed and secure acquisition of target states not only effectively reduces the risk loss caused by local sensor exposure, but also improves our ability to grasp battlefield intelligence. However, the heavy dependence of distributed sensor-observation network systems on communication networks has brought new entry points for enemy attackers and also caused serious obstacles to their practical application. Therefore, it is urgent to develop an algorithm that can effectively realize security status monitoring for large-scale multi-sensor-observation networks in malicious attack environments to avoid the limitations of the above methods.

[0003] Among the existing security state estimation algorithms, the literature (An L, Yang G H. Byzantine-resilient distributed state estimation: A min-switching approach. Automatica, 2021, 129: 109664.) uses the idea of ​​minimum switching to consider the distributed security estimation problem when Byzantine edges and Byzantine nodes exist in the observer network, transforming the estimation problem into a quadratic optimization problem, and using an event-triggered optimization algorithm to solve the security state estimation and reduce the search complexity of sparse malicious attack channels. The advantage of this scheme is that it can effectively achieve accurate estimation of the system state and eliminate attacks under general malicious communication attacks, but the disadvantage is that carefully designed attacks will cause the algorithm to fail, and problems such as state estimation delay cannot be ignored. The literature (A. Barboni, H. Rezaee, F. Boem and T. Parisini. Detection of covert cyber-attacks in interconnected systems: a distributed model-based approach. IEEE Transactions on Automatic Control, 2020, 65(9): 3728-3741.) constructs a distributed observer based on measurement output and a distributed Romberg observer that relies on unknown neighbor observation transmission information for each agent. The difference between the two different observers is used to locate the hidden malicious sensor attack to a certain subset, and then the specific attacked agent is determined from the subset through the residual threshold. The defect of this scheme is that it can only deal with the situation where there is only one attacked agent in the subsystem composed of the agent and its neighbors, and this method can only solve the situation where the hidden attack exists alone. The literature (Rezaee H, Parisini T, Polycarpou M M. Resiliency in dynamic leader–follower multiagent systems. Automatica, 2021, 125: 109384.) considers the consistent control problem of the dynamic leader-follower system under malicious attack environment. Through the coordinate clipping method, each follower discards the F neighbor information with the largest and smallest difference from its own state to achieve resistance to malicious attacks, and uses non-smooth saturation functions to compensate for the unknown dynamics and interference of the leader to achieve consistent tasks. The defect of this scheme is that it requires the dimension of each agent state to be one-dimensional, and the algorithm is difficult to break through the application of high-dimensional states.

[0004] Therefore, the present invention aims to provide a resilient distributed security monitoring method under a multi-sensor-observation network to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and to provide a resilient distributed security monitoring method under a multi-sensor-observation network. The method uses multiple mutually unknown sensors to complete the measurement acquisition of the target system, and on the basis that the local observer can correctly exclude all transmission channels that are subject to malicious attacks, the interaction of communication topology is used to achieve consistent security state observation of the target, thereby providing a basis for the security decision-making and stable operation of the multi-sensor-observation network in a harsh network attack environment.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions: A resilient distributed security monitoring method under a multi-sensor-observation network, comprising the following steps:

[0007] S1: Provide the specific dynamic model of the monitored target system and the measurement output model of the corresponding sensor;

[0008] S2: Based on the upper limit of the specific number of malicious attacks, build a sensor measurement combination that meets the corresponding number of conditions, and construct an observation communication topology that meets the corresponding conditions to ensure the realization of secure and consistent state observation;

[0009] S3: Based on the characteristics of sparse malicious attacks and appropriate communication topology, each observer uses its own observation data to flexibly process the received neighbor information;

[0010] S4: Based on the dynamic model of the target system, a suitable observer dynamic framework is constructed in combination with the processed neighbor information to achieve safe and consistent state observation of the target.

[0011] Furthermore, the specific method of step S1 is:

[0012] S1-1: The dynamic model of the monitored target system and the corresponding sensor measurement model are described as follows:

[0013]

[0014] Among them, x(t),y i (t) are the n-dimensional state of the target system and the p of the i-th sensor. i dimensional measurement output, matrix A is the state parameter matrix of the system, C i is the measurement matrix of the i-th sensor, and C i is column rank deficient, but the overall measurement matrix The columns are full rank; an observer is configured for each sensor, and the information of neighboring observers is unknown to each observer; each observer can only complete the distributed security monitoring of the entire target system state through information exchange with its neighbors.

[0015] Furthermore, the specific design method of step S2 is:

[0016] S2-1: The sensor measurement architecture requirements of the multi-sensor-observation network are as follows:

[0017] For the entire measurement matrix Under the condition that the number of local malicious communication information in the observed communication topology does not exceed F, it is necessary to assume that C satisfies the 2F-sparse column full rank condition; that is, discard any 2F submatrices C from the matrix C i After that, its stacking matrix C still satisfies the column full rank property;

[0018] S2-2: The communication topology model of the multi-sensor-observation network is described as follows:

[0019] For a multi-sensor-observation network consisting of N observers, the communication topology between them is G = (V, E), where V = {1, 2, ..., N} represents the set of observation nodes. is the set of its communication edges;

[0020] If two observers can exchange information, then (i, j)∈E and a ij =1, otherwise, record And a ij =0;

[0021] Let the set of neighbors interacting with observer i be N i ={j∈V:(i,j)∈E}, and the number of its neighbors is N i =|N i |; Then, the set of information data manipulated and tampered by malicious attackers among all observer communication networks is recorded as E a (t);

[0022] S2-3: The observation communication topology of the multi-sensor-observation network meets the following requirements:

[0023] For any local observer i, it is assumed that at most F data among the information transmitted by its neighbors are tampered by malicious attackers, that is, |E a (t)∩{(j,i)|j∈N i}|≤F; then, the constructed observation communication network architecture is required to satisfy the 2F+1-robust property, that is, for any two non-empty node subsets that do not intersect There must be a subset between the two that satisfies: there is at least one node i∈S1 (or S2) that has more than 2F+1 neighbors outside the set S1 (or S2).

[0024] Furthermore, the specific method of step S3 is:

[0025] S3-1: Each observer i performs the following operations based on the received neighbor observation information and its own observation information:

[0026] 1) Each local observer transmits its state observation to its communication neighbors and receive information transmitted by its neighbors in Represents the communication information that may be tampered by malicious attackers, Γ ij (t)∈{1,0} indicates whether the information on the communication channel (j,i) has been tampered by a malicious attacker, and the malicious attack satisfies |{Γ ij =1|j∈N i}|≤F;

[0027] 2) After receiving the information transmitted by its neighbors, each observer Arranged in ascending order: in, Then, the consistency control input u of the local observer is designed i (t) are as follows:

[0028]

[0029]

[0030] Among them, the function Sign(e)=[sign(e1),...,sign(e n )] T ,|e(t)|=[|e1(t)|,...,|e n (t)|] T and Diag{μ i (t)} represents the vector μ i Each component of (t) is a diagonal matrix consisting of diagonal elements, ρ ij (t) = 0 or 1 indicates the trust of observer i in the information transmitted by channel (j, i);

[0031] S3-2: Attack detection function ρ for observer i ij (t), and its assignment standard is described as follows:

[0032]

[0033] Among them, ρij (t)=0 means that observer i determines that the information on the transmission channel (j, i) has been tampered by the attacker, otherwise, it is determined that the information on the transmission channel has not been attacked.

[0034] Furthermore, the specific method of step S4 is:

[0035] S4-1: For each observer i, construct its observation model dynamics as follows:

[0036]

[0037] in is the state estimate of the monitoring target by the i-th observer at time t, u i (t) is the consistency control input designed in the previous step, δ 1i (t) and δ 2i (t) is the designed time-varying control gain parameter;

[0038] S4-2: Control gain δ of observer i 1i (t) and δ 2i (t) is updated as follows:

[0039]

[0040]

[0041] Here, sat(·,1) represents a saturation function bounded by 1.

[0042] In summary, the present invention has the following beneficial effects:

[0043] 1. The present invention can ensure the security of the entire network and reconstruct the true state of the target system even if part of the transmission data in the observation communication channel is tampered by the attacker, thereby improving the safety factor of the monitoring system in a harsh network attack environment and providing a basis for actual combat;

[0044] 2. Compared with the existing security state estimation method, which converts the estimation problem into a high-dimensional optimization problem for solution, the present invention compresses the impact of malicious attacks within the tolerable limit of the observer by constructing a distributed state observer based on Romberg observation, and excludes the communication data tampered by malicious attacks over time, thereby ensuring that the entire observation network can safely and consistently estimate the true state of the target system, effectively curbing the impact of attackers on the monitoring system;

[0045] 3. The present invention completes the measurement of the target system through a combination of mutually unknown multiple sensors, and can achieve effective target status monitoring even if some sensors are discovered and destroyed by the enemy, thereby improving the survivability of the monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the steps of a method for elastic distributed safety status monitoring in a multi-sensor-observation network of the present invention;

[0047] Figure 2 It is a specific flow chart of a method for elastic distributed safety status monitoring in a multi-sensor-observation network of the present invention;

[0048] Figure 3 It is a structural diagram design of a multi-sensor-observation network provided by an example of the present invention to monitor a target system;

[0049] Figure 4 It is a schematic diagram of the actual position of the target system and the position estimated by the multi-sensor-observation network provided by the example of the present invention;

[0050] Figure 5 It is a schematic diagram of the actual speed of the target system and the speed estimated by the multi-sensor-observation network provided by the example of the present invention;

[0051] Figure 6 It is a schematic diagram of the actual acceleration of the target system and the acceleration estimated by the multi-sensor-observation network provided by the example of the present invention;

[0052] Figure 7 It is a schematic diagram of the channel indicators of the actual attack provided by the example of the present invention and the attack indicators identified by the observation network. DETAILED DESCRIPTION

[0053] The following is combined with Figure 1-7 The present invention is described in further detail.

[0054] Embodiment: A resilient distributed security monitoring method in a multi-sensor-observation network, such as Figure 1 As shown, the following steps are included:

[0055] Step 1: Based on the multi-sensor-observation network under study, construct a dynamic model for the monitored target system and the corresponding sensors;

[0056] In the present invention, this step is specifically:

[0057] Step 1-1: The dynamic model of the monitored target system and the corresponding sensor measurement model are described as follows:

[0058]

[0059] Among them, x(t),y i (t) are the n-dimensional state of the target system and the p of the i-th sensor. i Dimensional measurement output, matrix A, Ci are the system state parameter matrix and the measurement matrix respectively. And the overall measurement matrix Satisfies column full rank, but the local measurement matrix C i It is rank-deficient. Assume that each local sensor is equipped with a local observer, and the information of neighboring observers is unknown to each local observer. Then each observer can only complete the distributed security monitoring of the entire target state through mutual information exchange with its communication neighbors.

[0060] Step 2: Based on the known upper limit of the number of malicious attacks, build a number of sensor measuring devices that meet the corresponding conditions, and construct the corresponding observation communication topology to meet the following conditions;

[0061] In the present invention, this step is specifically:

[0062] Step 2-1: The sensor measurement architecture requirements of the multi-sensor-observation network are as follows:

[0063] For the entire measurement matrix Under the condition that the number of local malicious communication information in the observed communication topology does not exceed F, it is necessary to assume that C satisfies the 2F-sparse column full rank condition. That is, discard any 2F submatrices C from the matrix C. i Afterwards, its stacking matrix C still satisfies the column full rank property.

[0064] Step 2-2: The communication topology model of the multi-sensor-observation network is described as follows:

[0065] For a multi-sensor-observation network consisting of N observers, the communication topology between them is G = (V, E), where V = {1, 2, ..., N} represents the set of observation nodes. is the set of its communication edges. If two observers can exchange information, then (i, j)∈E and a ij =1, otherwise, record And a ij = 0. Let the set of neighbors that communicate and interact with observer i be N i ={j∈V:(i,j)∈E}, and the number of its neighbors is N i =|N i |. Further, the set of information data between all observer communication networks that is manipulated and tampered by malicious attackers is denoted as E a (t).

[0066] Step 2-3: The communication topology of the multi-sensor-observation network is constructed as follows:

[0067] For any local observer i, it is assumed that at most F data in the information transmitted by its neighbors are tampered by malicious attackers, that is, |Ea (t)∩{(j,i)|j∈N i}|≤F. To this end, the constructed observation communication network architecture is required to satisfy the 2F+1-robust property, that is, for any two non-empty node subsets that do not intersect, There must be a subset between the two that satisfies: there is at least one node i∈S1 (or S2) that has more than 2F+1 neighbors outside the set S1 (or S2).

[0068] Step 3: Based on the characteristics of malicious attacks and the communication topology, each observer performs elastic processing on the received neighbor information;

[0069] In the present invention, this step is specifically:

[0070] Step 3-1: Each observer i performs the following operations based on the received neighbor observation information and its own observation information:

[0071] Each local observer transmits its state observation to its communication neighbors. and receive information transmitted by its neighbors in Represents the communication information that may be tampered by malicious attackers, Γ ij (t)∈{1,0} indicates whether the information on the communication channel (j,i) has been tampered by a malicious attacker, and the malicious attack satisfies |{Γ ij =1|j∈N i}|≤F.

[0072] After receiving the information transmitted by its neighbors, each observer Arranged in ascending order: in Furthermore, we design the consistency control input u of the local observer i (t) are as follows:

[0073]

[0074]

[0075] where function Sign(e) = [sign(e1), ..., sign(e n )] T ,|e(t)|=[|e1(t)|,...,|e n (t)|] T and Diag{μ i (t)} represents the vector μ i Each component of (t) is a diagonal matrix consisting of diagonal elements, ρ ij(t) = 0 or 1 indicates the trust of observer i in the information transmitted by channel (j, i);

[0076] Step 3-2: Attack detection function ρ for observer i ij (t), and its assignment standard is described as follows:

[0077]

[0078] where ρ ij If (t) = 0, it means that observer i believes that the information sent by the transmission channel (j, i) may be tampered by the attacker, otherwise it is considered that the information on the transmission channel has not been attacked.

[0079] Step 4: Construct a suitable observation framework by considering the dynamic model of the target system combined with the processed neighbor information to achieve elastic and consistent safety state observation of the target;

[0080] In the embodiment of the present invention, the step is specifically as follows:

[0081] Step 4-1: For each observer i, construct its observation model dynamics as follows:

[0082]

[0083] in is the state estimate of the monitoring target by the i-th observer at time t, u i (t) is the consistency control input designed in the previous step, δ 1i (t) and δ 2i (t) is the designed time-varying control gain parameter.

[0084] Step 4-2: Control gain δ of observer i 1i (t) and δ 2i (t) is updated as follows:

[0085]

[0086]

[0087] Here, sat(·,1) represents a saturation function bounded by 1.

[0088] Example 1

[0089] based on Figure 1 , Figure 2 The steps and flow chart shown are for executing the safety monitoring algorithm according to the flow chart steps in the following embodiments.

[0090] Step 1: For the target system consisting of two unmanned ground vehicles and the corresponding six measurement sensor systems, its dynamic model is as follows:

[0091]

[0092] in

[0093]

[0094]

[0095] The initial state parameters of the target system are selected as p1=-9, p2=-6, v1=15, v2=18, a1=-4, a2=5.

[0096] Step 2: From the construction of the measurement matrix Ci in step 1, it can be seen that it meets the 2-sparse column full rank requirement required in step 201. In addition, the communication topology between the six sensor-observation networks can be seen Figure 3 As shown, x represents the target system, Si represents the measurement sensor, and the dotted circle represents the local observer corresponding to each sensor; it can be verified that the communication topology between the observer networks satisfies the 3-robust property.

[0097] Step 3: Assuming that the initial state estimate of each observer is selected as the 0 vector, the attacker randomly selects an information transmission channel from the neighbor channel of each local observer every 5 seconds to inject bad data attack, and its bad data injection function is

[0098] Step 4: Set the initial value of the control gain to δ 1i (0) = δ 2i (0) = 1, and then each local observer uses the observation model given in step 4-1 and step 4-2 to perform distributed state estimation.

[0099] Figure 4-6 They are the true state values ​​of the target car regarding position, velocity, and acceleration and the estimated data corresponding to each local observer. It can be seen that after t>9s, when there are sparse malicious attackers in the communication network, the multi-sensor-observation network proposed in the present invention can still achieve safe and consistent state monitoring. This proves the effectiveness of the elastic distributed security state estimation algorithm under the multi-sensor-observation network proposed in the present invention.

[0100] Mark all the edges in the communication topology with the following indicators:

[0101]

[0102]

[0103] Then, Figure 7 That is, the indicators of attacks suffered by each communication transmission channel at each moment and the attack indicators identified by the observation network. It can be seen that after t>5s, the indicator identification of the communication attack by the observer node can successfully match the real attack indicator, indicating that sparse communication attacks can be successfully identified by the elastic distributed security algorithm under the multi-sensor-observation network proposed by the present invention.

[0104] In the above-mentioned embodiments of the present invention, the above-mentioned method of the present invention realizes that when part of the transmission data in the observation communication channel is tampered by the attacker, the entire network can still be ensured to reconstruct the true state of the target system, thereby improving the safety factor of the monitoring system in a harsh network attack environment and providing a basis for actual combat. On the one hand, the existing security state estimation method is to convert the estimation problem into a high-dimensional optimization problem for solution. The present invention compresses the impact of malicious attacks within the tolerable limit of the observer by constructing a distributed state observer based on Romberg observation, and excludes the communication data tampered by malicious attacks as time goes by, thereby ensuring that the entire observation network can safely and consistently estimate the true state of the target system, effectively curbing the impact of attackers on the monitoring system. On the other hand, the present invention completes the measurement of the target system through a combination of mutually unknown multiple sensors, so that even if some sensors are discovered and destroyed by the enemy, effective target state monitoring can be achieved, thereby improving the survivability of the monitoring system.

[0105] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A resilient distributed security monitoring method in a multi-sensor-observation network, characterized by: The following steps are involved: S1: Provide a specific dynamic model of the monitored target system and a measurement model of the corresponding sensor; the specific method of step S1 is: S1-1: The dynamic model of the monitored target system and the measurement model of the corresponding sensor are described as follows: Among them, x(t),y i (t) are the n-dimensional state of the target system and the p of the i-th sensor. i dimensional measurement output, matrix A is the state parameter matrix of the system, C i is the measurement matrix of the i-th sensor, and C i is column rank deficient, but the overall measurement matrix The columns are full rank; an observer is configured for each sensor, and the information of neighboring observers is unknown to each observer; each observer can only complete the distributed security monitoring of the entire target system state through information interaction with its neighbors; S2: Based on the upper limit of the specific number of malicious attacks, a sensor measurement device combination that meets the corresponding number of conditions is built, and an observation communication topology that meets the corresponding conditions is constructed to ensure the realization of safe and consistent state observation; the specific design method of step S2 is: S2-1: The sensor measurement architecture requirements of the multi-sensor-observation network are as follows: For the entire measurement matrix Under the condition that the number of local malicious communication information in the observed communication topology does not exceed F, it is necessary to assume that C satisfies the 2F-sparse column full rank condition; that is, discard any 2F submatrices C from the matrix C i After that, its stacking matrix C still satisfies the column full rank property; S2-2: The observation communication topology model of the multi-sensor-observation network is described as follows: For a multi-sensor-observation network consisting of N observers, the observation communication topology between them is G = (V, E), where V = {1, 2, ..., N} represents the set of observation nodes. is the set of its communication edges; If two observers can exchange information, then (i, j)∈E and a ij =1, otherwise, record And a ij =0; Let the set of neighbors interacting with observer i be N i ={j∈V:(i,j)∈E}, and the number of its neighbors is N i =|N i |; Then, the set of information data manipulated and tampered by malicious attackers among all observer communication networks is recorded as E a (t); S2-3: The observation communication topology of the multi-sensor-observation network meets the following requirements: For any local observer i, it is assumed that at most F data among the information transmitted by its neighbors are tampered by malicious attackers, that is, |E a (t)∩{(j,i)|j∈N i }|≤F; then, the constructed observation communication network architecture is required to satisfy the 2F+1-robust property, that is, for any two non-empty node subsets S1, There must be a subset between the two that satisfies: there is at least one node i∈S1, which has more than 2F+1 neighbors outside the set S1; or there is at least one node i∈S2, which has more than 2F+1 neighbors outside the set S2; S3: Based on the characteristics of sparse malicious attacks and a suitable observation communication topology, each observer uses its own observation data to flexibly process the received neighbor information; the specific method of step S3 is: S3-1: Each observer i performs the following operations based on the received neighbor observation information and its own observation information: 1) Each local observer transmits its state observation to its communication neighbors and receive information transmitted by its neighbors in Represents the communication information that may be tampered by malicious attackers, Γ ij (t)∈{1,0} indicates whether the information on the communication channel (j,i) has been tampered by a malicious attacker, and the malicious attack satisfies |{Γ ij =1|j∈N i }|≤F; 2) After receiving the information transmitted by its neighbors, each observer Arranged in ascending order: δ1(t)≤δ2(t)≤…≤δ Ni (t), where Then, the consistency control input u of the local observer is designed i (t) are as follows: Among them, the function Sign(e)=[sign(e1),...,sign(e n )] T ,|e(t)|=[e1(t)|,...,|e n (t)] T and Diag{μ i (t)} represents the vector μ i Each component of (t) is a diagonal matrix consisting of diagonal elements, ρ ij (t) = 0 or 1, indicating the trust of observer i in the information transmitted by channel (j, i); S3-2: Attack detection function ρ for observer i ij (t), and its assignment standard is described as follows: Among them, ρ ij (t) = 0 means that observer i determines that the information on the communication channel (j, i) has been tampered by the attacker, otherwise, it is determined that the information on the communication channel has not been attacked; S4: Based on the dynamic model of the target system, a suitable observer dynamic framework is constructed in combination with the processed neighbor information to achieve safe and consistent state observation of the target. The specific method of step S4 is: S4-1: For each observer i, construct its observation model dynamics as follows: in is the state estimate of the monitoring target by the i-th observer at time t, u i (t) is the consistency control input designed in the previous step, δ 1i (t) and δ 2i (t) is the designed time-varying control gain parameter; S4-2: Control gain δ of observer i 1i (t) and δ 2i (t) is updated as follows: Here, sat(·,1) represents a saturation function bounded by 1.

Images