Coverage formation control method of team-based heterogeneous multi-agent system in channel type area

Abstract

The invention relates to the field of coverage formation control, and discloses a coverage formation control method of a team-based heterogeneous multi-agent system in a channel-type region, and the method comprises the following steps: S1, dividing a target region: dividing the target region at a team level and an agent level; s2, modeling the sensors and the intelligent agents, and defining a corresponding sensing function for each intelligent agent to model a relationship between the health condition of the sensors and the sensing data of the intelligent agents; s3, designing an intelligent agent motion controller, enabling the intelligent agents to keep a specific formation based on the designed motion controller, and deploying the intelligent agents to a local optimal position of the target area; and S4, designing an intelligent body weight self-adaptive controller, dynamically adjusting the coverage area of the intelligent body based on the designed weight self-adaptive controller, deploying the intelligent body to the local optimal position of the task area by the controller designed by the invention, and dynamically adjusting the formation of the multi-intelligent body system according to the task demand.

Description

Technical Field

[0001] The present invention relates to the field of coverage formation control, and in particular to a coverage formation control method of a team-based heterogeneous multi-agent system in a channel-type area. Background Art

[0002] Coverage control is a core issue in the study of multi-agent systems, which is widely used in surveillance, military and environmental protection. The core of this problem is how to deploy the agent system to the local optimal position in the target area. Current research mainly focuses on coverage control in convex areas. However, in real applications, many areas are non-convex areas, and existing technologies have many shortcomings in coverage control in such areas.

[0003] Channel-type areas (such as rivers and military buffer zones) are typical non-convex areas. Continuous monitoring of these areas is not only crucial in environmental protection, but also has significant military value. However, there are few studies on the deployment of multi-agent systems in channel-type areas. The existing technology proposes a coverage control method based on Voronoi segmentation, which divides the convex area into Voronoi regions and assigns task areas to each agent. However, this method has limited processing of non-convex areas. Some improved methods attempt to extend the traditional Voronoi segmentation through mapping techniques or using geodesic distances, but these methods have high computational complexity and ignore the performance differences of agents. In addition, most of the existing mobile coverage control methods are only applicable to homogeneous agent systems, and there is no effective solution for the collaborative coverage between heterogeneous agents.

[0004] The existing technology has the following defects when dealing with channel-type areas: 1) It cannot fully consider the performance differences between intelligent agents, resulting in unbalanced load of the agents; 2) The existing partitioning method has high computational complexity when facing complex non-convex areas; 3) There is a lack of effective methods for multi-agent formation control in channel-type areas, and it is impossible to flexibly adjust the formation according to task requirements. Summary of the invention

[0005] The purpose of the present invention is to provide a coverage formation control method for a team-based heterogeneous multi-agent system in a channel-type area to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a coverage formation control method of a team-based heterogeneous multi-agent system in a channel-type area, comprising the following steps:

[0007] Step S1, target area division, dividing the target area at the team level and the agent level respectively;

[0008] Step S2, modeling of sensors and agents, defining a corresponding perception function for each agent to model the relationship between the health status of the sensor and the agent's perception data;

[0009] Step S3, designing a motion controller for the intelligent agent, based on the designed motion controller, making the intelligent agent maintain a specific formation and deploy it to a local optimal position in the target area;

[0010] Step S4, designing a weight adaptive controller for the intelligent agent, and dynamically adjusting the coverage area of ​​the intelligent agent based on the designed weight adaptive controller.

[0011] As a preferred embodiment, the specific process of the team-level target area division in step S1 is as follows: the front leader agent and the rear leader agent are respectively denoted as l F and l R , the two leader agents follow the center line of the target area channel If it moves, the moving coverage area Q determined by the Voronoi segmentation is defined as:

[0012] Q={q∈S|||ql||≤||ql i ||,i=F,R}. (1.3)

[0013] Where l represents the collective center of mass of the agent, S is the channel-type area where n heterogeneous agents are deployed, and the area consists of two regular curves. and Definition.

[0014] As a preferred embodiment: the agent level division in step S1 adopts a power diagram to define the agent positions and weight sets V(W,P)={V1,V2,...,V n}:

[0015] V t ={q∈Q||qp t || 2 -ω t ≤||qp s || 2 -ω s}, (1.4)

[0016] Where s=1,2,...,n,s≠t, W=(ω1,ω2,...,ω n ) represents the weights of n agents, where is the weight of the t-th agent, t = 1, 2, ..., n, and the agent includes an internal group and a boundary group, and the position is represented by p i and p b Represents, where i=1,2,...,n i ,ni represents the number of agents in the internal group, b=1,2,...,n b ,n b Represents the number of agents in the boundary group.

[0017] As a preferred embodiment, the sensor and agent modeling process in step S2 uses a perception function specifically expressed as:

[0018] γ t (q,p t )=-(||qp t || 2 -h t ), (1.5)

[0019] Among them, h t represents the health index of the t-th agent, t=1,2,...,n.

[0020] As a preferred embodiment, in the design process of the intelligent body motion controller in step S3, a total coverage cost function is defined based on (1.3) and (1.4):

[0021]

[0022] in, represents the service factor of the t-th agent, represents the formation factor, t = 1, 2, ..., n. φ(q) is a distribution density function that represents the distribution of information in region S and applies the total cost function (1.6) to the agent p t The derivative of :

[0023]

[0024] Where N s Representing Boundaries The unit external normal vector at , according to equation (1.1), the first term on the right side of equation (1.8) can be rewritten as

[0025]

[0026] The second term on the right side of equation (1.8) can be rewritten as:

[0027]

[0028] Where N t Represents the agent p t The index set of agents that share a boundary, and the unit normal vectors of the shared boundary of adjacent agents are in opposite directions on both sides, satisfying the relationship: N t,s =-N s,t ,also, and represents the same shared boundary. According to the definition of power diagram (1.4), ||qp t || 2 -ω t =||qp s || 2 -ω s For any holds, so the first two terms on the right side of (1.10) are rewritten as:

[0029]

[0030] For an agent p in the boundary group b , the boundary shared with the leader agent indirectly depends on agent p t According to the chain rule, we get:

[0031]

[0032] in b=1,2,...,n b , and i∈{F,R}, so the last term on the right side of (1.10) is rewritten as:

[0033]

[0034] Substituting equations (1.11) and (1.13) into equation (1.10), we obtain:

[0035]

[0036] Finally, substituting equations (1.9) and (1.14) into equation (1.8), the total cost function (1.6) is given by t The derivative of is:

[0037]

[0038] in

[0039]

[0040]

[0041] A gradient descent-based controller is proposed, which is defined as follows:

[0042]

[0043] Where R t represents the positive gain of the agent's mobility, β t Defined by (1.16), η tDefined by equation (1.17), t = 1, 2, ..., n., the first term of equation (1.18) Drive the agent to the center of mass of its power map unit, the second term β of (1.18) t The third term η in (1.18) is related to the boundaries of its neighboring agents and the boundaries of the dynamic region Q. t Used to adjust the distance between the agent and the collective center of mass.

[0044] As a preferred embodiment: the design of the agent weight adaptive controller in step S4 uses the perception function (1.5) to represent the agent p t The weighted adaptive controller for the real-time detection performance at point q and the change in the agent's detection performance is defined as follows:

[0045]

[0046] Among them, k t is a positive gain scalar that determines the convergence speed, t=1,2,...,n., and is the neighboring agent p t and agent p s The power diagram cell boundaries shared between (1.19) The weights are adjusted by comparing the perception function values ​​of the shared boundaries. ‖qp t ‖ 2 -ω t =‖qp s ‖ 2 -ω s The adaptive controller (1.19) can be simplified to

[0047]

[0048] in, Indicates shared boundaries Length

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The controller designed by the present invention can not only deploy the agents to the local optimal position of the task area, but also dynamically adjust the formation of the multi-agent system according to the task requirements by adjusting the formation factor. In this process, each agent can dynamically adjust the size of the coverage area according to its detection capability, so that "the more capable, the more work" can be achieved.

[0051] The controller can effectively cope with the changes in dynamic task areas and respond in real time to the adjustments of the coverage area by the intelligent agent, while maintaining a specific formation. The existing technology cannot take these three points into account, resulting in low task efficiency in complex environments. The coverage formation controller of the present invention can control the distance between the intelligent agent and the collective center of mass by adjusting the formation factor, and when the formation factor increases, the distance between the intelligent agent and the collective center of mass decreases on average, thereby improving the flexibility of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flow chart of the present invention;

[0053] Figure 2 A coverage area division diagram in a channel type area;

[0054] Figure 3 It is the task area of ​​the embodiment of the present invention;

[0055] Figure 4 is the distribution density function;

[0056] Figure 5 is the weight of the agent in the embodiment of the present invention;

[0057] Figure 6 is the average distance between the intelligent agent and the collective center of mass in the embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Example

[0060] See also Figure 1 , a coverage formation control method of a team-based heterogeneous multi-agent system in a channel-type area, including the following steps:

[0061] Step S1, target area division, dividing the target area at the team level and the agent level respectively;

[0062] Step S2, modeling of sensors and agents, defining a corresponding perception function for each agent to model the relationship between the health status of the sensor and the agent's perception data;

[0063] Step S3, designing a motion controller for the intelligent agent, based on the designed motion controller, making the intelligent agent maintain a specific formation and deploy it to a local optimal position in the target area;

[0064] Step S4, designing a weight adaptive controller for the intelligent agent, and dynamically adjusting the coverage area of ​​the intelligent agent based on the designed weight adaptive controller.

[0065] First, in the present invention, represents n-dimensional Euclidean space, represents the set of positive real numbers, and the n×n identity matrix is ​​I n , ||·|| represents the Euclidean norm, and given the polygon area And including its interior, |Q| represents the Lebesgue measure of the region, and sets the distribution density function φ: It is bounded, measurable and absolutely continuous, reflecting the information distribution within the region Q.

[0066] Furthermore, V={V1,V2,...,V n} represents the division of region Q, which is a set of n subsets, and these subsets do not overlap each other. The position set of the agent is recorded as P = (p1, p2, ..., p n ), where p t represents the position of the t-th agent, and t=1,2,...,n., the generalized mass M in the region Q Q , centroid C Q and moment of inertia J Q,p They are defined as follows:

[0067]

[0068] The collective center of mass l of the agents is defined as:

[0069]

[0070] where p t represents the position of the tth agent.

[0071] n heterogeneous agents are deployed in the channel-shaped area S of the present invention, and the area consists of two regular curves and Define, the center line of the channel area is used as the reference line for deploying the agent, denoted as Then step S1 divides the target area from the team level and the agent level. At the team level, the front leader agent and the rear leader agent are respectively denoted as l F and l R , the two leader agents move along the center line of the channel Move and define a moving coverage area Q determined by the Voronoi segmentation:

[0072] Q={q∈S|||ql||≤||ql i ||,i=F,R}. (1.3)

[0073] Where l represents the collective center of mass of the agent; at the agent level, let W = (ω1, ω2, ..., ω n ) represents the weights of n agents, where is the weight of the t-th agent, t = 1, 2, ..., n., and defines a power graph V(W, P) = {V1, V2, ..., V n}:

[0074] V t ={q∈Q||qp t || 2 -ω t ≤||qp s || 2 -ω s}, (1.4)

[0075] Where s=1,2,...,n,s≠t. Considering that heterogeneous agents have different perception capabilities, standard Voronoi segmentation is not applicable to the context of the present invention. The generalized Voronoi segmentation adopted has a higher degree of freedom in assigning task areas to agents. Depending on the position of the agent, it may require multiple information sources to determine its coverage area. For example, the agent located inside the team only needs to obtain the position of the neighboring agent, while the agent at the boundary of the team needs to obtain the boundary of area S and the position of the leader agent. Therefore, the agents are divided into two groups: an internal group and a boundary group.

[0076] Internal group: Agents that only share boundaries with other agents in the team, with positions indicated by p i Represents, where i=1,2,…,n i ,n i Indicates the number of agents in the internal group.

[0077] Boundary group: Agents that share boundaries not only with agents in the team but also with region Q, with positions denoted by p b Indicates, where b = 1, 2, ..., n b ,n b Represents the number of agents in the boundary group.

[0078] The boundary of region Q is denoted by The boundary of the tth agent is denoted by The boundary is either an edge shared with a neighboring agent or an edge shared with the agent's dynamic region Q, t = 1, 2, ..., n. The set of agent indices that share a boundary with the tth agent is denoted by N t .Agent p t With neighboring agent p s The shared boundary of s∈N t , agent p t The shared parts with the front and rear boundaries of region S and dynamic region Q are expressed as and boundary The unit external normal vector is denoted by N t,s ,boundary The unit external normal vector is denoted by N i , where i∈{F,R}, Figure 2 Describes the power diagram unit V t The above boundary and its unit external normal vector, the red dashed line represents the boundary of the dynamic region Q. In (1.4), ω t It represents the weight of the t-th agent, reflecting its detection capability, t=1,2,...,n. Agents with stronger detection capability will be assigned larger task coverage areas. Based on the characteristics of weights, the agents can adjust the size of their coverage areas online according to their real-time performance.

[0079] The agents performing the coverage task are equipped with professional sensors, and their performance is completely dependent on the quality of the sensors. In practical applications, the quality of the sensors may be affected by a variety of adverse factors. For example, when patrolling a river, fog or dust in the air can reduce the detection quality of the sensors. Even under optimal conditions, the aging of the sensors may cause their detection performance to deteriorate. For this reason, step S2 defines a perception function for each agent to model the relationship between the health status of the sensor and the agent's perception data, which is specifically expressed as:

[0080] γ t (q,p t )=-(||qp t || 2 -h t ), (1.5)

[0081] Among them, h t represents the health index of the t-th agent, t = 1, 2, ..., n. Equation (1.5) uses a quadratic approximation of the actual function value near the comparison point to model the impact of the sensor health status. Generally, γ t (q,p t ) can be obtained by measuring the sensor state of the agent. When different agents detect the same point, γ t(q,p t ) may take different values. Based on this feature, an adaptive controller can be designed to dynamically adjust the weight of the agent. In addition, the model effectively reflects the changes in sensor quality because when point q is far away from the agent, the agent's detection performance of the point will decrease.

[0082] Furthermore, the present invention defines an error factor, which quantifies the perceived performance deviation of the agent in the current task. t The difference between the health index and its weight is defined as the agent's error factor, that is, e t =h t -ω t , where t=1,2,…,n.

[0083] Executing the coverage formation control task in the channel-type area S requires independent area division at the team and agent levels. The coverage control problem of the present invention can be achieved by solving a two-layer optimization problem. However, when heterogeneous agents are deployed in the channel-type area S, the situation becomes more complicated. The agents not only need to maintain a specific formation, but also need to adjust the size of their coverage area and respond to changes in the dynamic area. In order to enhance the flexibility of the multi-agent system, the present invention considers the formation control problem of the multi-agent system in the channel-type area S. The goal of the multi-agent system is to achieve a specified formation within the mission area. The formation in the present invention requires that the agents maintain a specific distance from the collective center of mass. To achieve this goal, the present invention defines a total cost function to ensure that the agents maintain a specific formation during the mission. According to definitions (1.3) and (1.4), the total coverage cost function is defined as follows:

[0084]

[0085] in, represents the service factor of the t-th agent, represents the formation factor, t=1,2,…,n, φ(q) is a distribution density function, which represents the distribution of information in area S. By solving the minimum value of equation (1.6), the local optimal deployment of heterogeneous multi-agent system in the channel area can be achieved. When the multi-agent system is deployed to the local optimal position of the task area, the cost function is minimized. Different agents contribute differently to the cost function. Agents with stronger performance consume more energy when performing coverage tasks. Therefore, in the total cost function (1.6), agents with higher performance are given a larger service factor. The definition of service factor is as follows: The service factor of each agent is linearly related to its performance, that is, α t =αh t , where α is a positive real number, t=1,2,…,n.

[0086] At the same time, the present invention also adopts the following lemma: Lemma 1: If the agent p t and agent p s Shared borders Then for

[0087]

[0088] where t=1,2,…,n,s∈N t ; Lemma 2, let l represent the collective center of mass of the multi-agent system. The collective center of mass l is about the agent p t The partial derivative of

[0089]

[0090] in

[0091]

[0092] Lemma 3. If an agent p in the boundary group b With the leader agent i share some boundaries, then for

[0093]

[0094] Where b=1,2,…,n b ,i∈{F,R}, l represents the collective center of mass of the multi-agent system.

[0095] In order to design the motion controller of the intelligent agent, we first need to obtain the local optimal position of the intelligent agent at each moment. Due to the dynamic changes of the leader intelligent agent, the boundaries of the intelligent agent will be affected, making the position optimization problem more complicated. Applying the generalized multivariable Leibniz law, the total cost function (1.6) is about the intelligent agent p t The derivative of :

[0096]

[0097] Where N s Representing Boundaries The unit external normal vector at , according to equation (1.1), the first term on the right side of equation (1.8) can be rewritten as

[0098]

[0099] The second term on the right side of equation (1.8) can be rewritten as:

[0100]

[0101] Where N tRepresents the agent p t The index set of agents that share a boundary, and the unit normal vectors of the shared boundary of adjacent agents are in opposite directions on both sides, satisfying the relationship: N t,s =-N s,t ,also, and represents the same shared boundary. According to the definition of power diagram (1.4), ||qp t || 2 -ω t =||qp s || 2 -ω s For any holds, so the first two terms on the right side of (1.10) are rewritten as:

[0102]

[0103] For an agent p in the boundary group b , the boundary shared with the leader agent indirectly depends on agent p t .

[0104] According to the chain rule, we have:

[0105]

[0106] in b=1,2,...,n b , and i∈{F,R}, so the last term on the right side of (1.10) is rewritten as:

[0107]

[0108] Substituting equations (1.11) and (1.13) into equation (1.10), we obtain:

[0109]

[0110] Finally, substituting equations (1.9) and (1.14) into equation (1.8), the total cost function (1.6) is given by t The derivative of is:

[0111]

[0112] in

[0113]

[0114]

[0115] A gradient descent-based controller is proposed, which is defined as follows:

[0116]

[0117] Where R t represents the positive gain of the agent's mobility, β t Defined by (1.16), η t Defined by equation (1.17), t = 1, 2, ..., n., the first term of equation (1.18) Drive the agent to the center of mass of its power map unit, the second term β of (1.18) t The third term η in (1.18) is related to the boundaries of its neighboring agents and the boundaries of the dynamic region Q. t Used to adjust the distance between the agent and the collective center of mass.

[0118] The leader agent applies specific dynamics according to its coverage task, affecting the boundary dynamics of the coverage area Q. In the present invention, the movement trajectory of the leader agent is Its horizontal moving speed is as follows:

[0119]

[0120] Where K F and K R is a positive constant. Before the coverage task, the initial positions of the two leader agents must be specified to ensure that the agent is located between them and the initial coverage area is not empty. The moving speed of the agent must be faster than the leader to ensure that it can still converge to the local optimal position as the coverage area changes. The information sent by the agent of the present invention also includes its position, weight, service factor and health index. Since the channel-type area S is non-convex, the coverage area of ​​the boundary group agent may also be non-convex. If the agent p in the boundary group b The centroid of the coverage area is located at its power map cell V b In addition, when the agent reaches the boundary, it may continue to move outward, b = 1, 2, ..., n b In this case, the agent needs to move along the projection direction of the boundary tangent to ensure that it is still in region V b Inside.

[0121] The weight ω assigned to each agent t represents its detection capability. The detection value for the same target can directly reflect the difference in detection capability between different agents. In order to describe this performance difference, the present invention uses the perception function (1.5) to represent the agent p t For the real-time detection performance at point q, the weighted adaptive controller for the agent’s detection performance change is defined as follows:

[0122]

[0123] Among them, k t is a positive gain scalar that determines the convergence speed, t=1,2,...,n, and is the neighboring agent p t and agent p s The power diagram cell boundaries shared between (1.19) The weights are adjusted by comparing the perception function values ​​of the shared boundaries. ||qp t || 2 -ω t =||qp s || 2 -ω s holds, so the adaptive controller (1.19) can be simplified to

[0124]

[0125] in, Indicates shared boundaries Length, it is worth noting that equation (1.19) and equation (1.20) are mathematically equivalent.

[0126] The present invention also adopts the following lemma, Lemma 4, the weight ω obtained by the adaptive controller (1.20) t is bounded, and there exists a constant c such that,

[0127]

[0128] Where τ represents the continuous time index. For n heterogeneous agents performing coverage tasks in a channel-type region S, the following Theorem 1 proves its convergence.

[0129] Theorem 1: Driven by the formation controller (1.18) and the weight adaptive controller (1.19), all agents will eventually converge asymptotically to the local optimal position in the task area to minimize the total cost function (1.6) while maintaining a specific formation:

[0130]

[0131] where β t Defined by (1.16), η t Defined by equation (1.17), t = 1, 2, ..., n., the formation factor δ tIt can be adjusted according to the task requirements to change the relative distance between the agent and the collective center of mass. Based on Theorem 1, Theorem 2 is obtained: Driven by the formation controller (1.18) and the weight adaptive controller (1.19), the collective center of mass of the agent will eventually converge asymptotically to the local optimal position in the task area to minimize the total cost function (1.6) while maintaining the predetermined formation.

[0132] In order to demonstrate the effectiveness of the coverage control strategy proposed in the invention, the present invention will illustrate the effectiveness of the invention through mathematical proof and numerical simulation respectively.

[0133] Mathematical Proof

[0134] Introducing the differentiable Lyapunov function

[0135]

[0136] This function has the same form as the total cost function (1.6), so the Lyapunov function V is positive definite. According to Lemma 4, we can get V with respect to ω t The partial derivative of :

[0137]

[0138] The derivative of the function V with respect to time is:

[0139]

[0140] Substituting the motion controller (1.18) and the weight adaptive controller (1.20) into equation (1.21), we get

[0141]

[0142] Due to R t is a positive constant, and the derivative of V with respect to time is semi-negative definite, that is, Therefore, it can be concluded that the Lyapunov function V is a non-increasing function with a lower bound of zero. In order to obtain We need to ensure the continuity of The boundedness of , by taking the derivative of (1.22) with respect to time, we can obtain:

[0143]

[0144] Due to R t and α t are all positive constants and bounded, and and p t Both In, therefore and p t is also bounded. According to formula (1.1), is also bounded. According to Reynolds' transport theorem, we can get and is bounded, represents the speed of the t-th agent, which is naturally bounded, from which we can conclude is bounded, which means is uniformly continuous. According to Barbalat's lemma, we finally get This also means:

[0145]

[0146] Where t = 1, 2, ..., n. Therefore, the group of agents gradually converges to the local optimal position in the task area.

[0147] Numerical simulation

[0148] The present invention has been numerically simulated on a river section in the Yellow River Basin. Figure 3 Shown is the designated mission area, which is located in the Inner Mongolia Autonomous Region with coordinates of 39.9019°N, 111.4359°E. Since the trajectory of the leader agent is known, its horizontal moving speed is set to 2.53m / s, and the moving speed of each agent is limited to 7.60m / s to ensure that they can respond quickly to changes in dynamic boundaries.

[0149] In this simulation, the Figure 4 The distribution density function shown is in the form of

[0150]

[0151] where [x1 x2 x3] = [150350560], [y1 y2 y3] = [50200300] and

[0152] The multi-agent system is assigned two sets of health factors and weights to reflect the differences in their detection capabilities. The health factors remain unchanged throughout the deployment process, while the weights need to be adaptively adjusted. The initial health factors and weights of the agents are set to [h1 h2 h3 h4 h5] = [30 20 30 45 40] and [ω1 ω2 ω3 ω4 ω5] = [45.5 30.5 55.5 37.5 40.5], respectively. The positive real number α is specified as 10 -5 , and the positive gain in the weight adaptive controller is set to k1=k2=k3=k4=k5=10 3 ,like Figure 5As shown in the figure, the weight of the agent gradually converges to a fixed value, and finally its weight converges to [ω1ω2ω3ω4ω5]=[36.26 26.26 36.26 51.26 46.26], and the service factor of the agent is [α1 α2 α3 α4α5]=10 -5 ×[30 20 30 45 40].

[0153] The formation factor assigned to each agent determines its distance from the collective center of mass. In the first round of simulation, the formation factor is set to [η1 η2 η3 η4 η5] = [0.9 0.7 0.8 1 0.6], and in the second round of simulation, the formation factor is set to [η1 η2 η3 η4 η5] = [42 30 37 40 36]. Figure 6 The average distance between the agents and the collective center of mass is recorded when different formation factors are assigned to the agents. By observing Figure 6, it can be concluded that the larger the average formation factor of the agents, the smaller the average distance between the agents and the collective center of mass.

[0154] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0155] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A coverage formation control method for a team-based heterogeneous multi-agent system in a channel-type area, characterized in that: The steps include: Step S1, target area division, dividing the target area at the team level and the agent level respectively; Step S2, modeling of sensors and agents, defining a corresponding perception function for each agent to model the relationship between the health status of the sensor and the agent's perception data; Step S3, designing a motion controller for the intelligent agent, based on the designed motion controller, making the intelligent agent maintain a specific formation and deploy it to a local optimal position in the target area; Step S4, designing a weight adaptive controller for the intelligent agent, and dynamically adjusting the coverage area of ​​the intelligent agent based on the designed weight adaptive controller.

2. According to claim 1, a coverage formation control method for a team-based heterogeneous multi-agent system in a channel-type area is characterized by: The specific process of the team-level target area division in step S1 is as follows: the front leader agent and the rear leader agent are respectively denoted as l F and l R , the two leader agents follow the center line of the target area channel If it moves, the moving coverage area Q determined by the Voronoi segmentation is defined as: Q={q∈S|||q-l||≤||q-l i ||,i=F,R}. (1.3) Where l represents the collective center of mass of the agent, S is the channel-type area where n heterogeneous agents are deployed, and the area consists of two regular curves. and Definition.

3. According to claim 2, a coverage formation control method for a team-based heterogeneous multi-agent system in a channel-type area is characterized by: The step S1 agent level division uses a power diagram to define the agent positions and weight sets V(W,P) = {V1,V2,...,V n }: V t ={q∈Q∣||q-p t || 2 -ω t ≤||q-p s || 2 -ω s }, (1.4) Where s=1,2,...,n,s≠t, W=(ω1,ω2,...,ω n ) represents the weights of n agents, where is the weight of the t-th agent, t = 1, 2, ..., n, and the agent includes an internal group and a boundary group, and the position is represented by p i and p b Represents, where i=1,2,...,n i ,n i represents the number of agents in the internal group, b=1,2,...,n b ,n b Represents the number of agents in the boundary group.

4. According to claim 3, a coverage formation control method for a team-based heterogeneous multi-agent system in a channel-type area is characterized by: The sensor and agent modeling process in step S2 uses the perception function specifically expressed as: γ t (q,p t )=-(||q-p t || 2 -h t ), (1.5) Among them, h t represents the health index of the t-th agent, t=1,2,...,n.

5. According to claim 4, a method for controlling a coverage formation of a team-based heterogeneous multi-agent system in a channel-type area is characterized by: In the design process of the intelligent motion controller in step S3, the total coverage cost function is defined based on (1.3) and (1.4): in, represents the service factor of the t-th agent, represents the formation factor, t = 1, 2, ..., n. φ(q) is a distribution density function that represents the distribution of information in region S and applies the total cost function (1.6) to the agent p t The derivative of : Where N s Representing Boundaries The unit external normal vector at , according to equation (1.1), the first term on the right side of equation (1.8) can be rewritten as, The second term on the right side of equation (1.8) can be rewritten as: Where N t Represents the agent p t The index set of agents that share a boundary, and the unit normal vectors of the shared boundary of adjacent agents are in opposite directions on both sides, satisfying the relationship: N t,s =-N s,t ,also, and represents the same shared boundary. According to the definition of power diagram (1.4), ||qp t || 2 -ω t =||qp s || 2 -ω s For any holds, so the first two terms on the right side of (1.10) are rewritten as: For an agent p in the boundary group b , the boundary shared with the leader agent indirectly depends on agent p t According to the chain rule, we get: in b=1,2,...,n b , and i∈{F,R}, so the last term on the right side of (1.10) is rewritten as: Substituting equations (1.11) and (1.13) into equation (1.10), we obtain: Finally, substituting equations (1.9) and (1.14) into equation (1.8), the total cost function (1.6) is given by t The derivative of is: in A gradient descent-based controller is proposed, which is defined as follows: Where R t represents the positive gain of the agent's mobility, β t Defined by (1.16), η t Defined by equation (1.17), t = 1, 2, ..., n., the first term of equation (1.18) Drive the agent to the center of mass of its power map unit, the second term β of (1.18) t The third term η in (1.18) is related to the boundaries of its neighboring agents and the boundaries of the dynamic region Q. t Used to adjust the distance between the agent and the collective center of mass.

6. The method for controlling a coverage formation of a team-based heterogeneous multi-agent system in a channel-type area according to claim 5, characterized in that: The design of the agent weight adaptive controller in step S4 uses the perception function (1.5) to represent the agent p t The weighted adaptive controller for the real-time detection performance at point q and the change in the agent's detection performance is defined as follows: Among them, k t is a positive gain scalar that determines the convergence speed, t=1,2,...,n., and is the neighboring agent p t and agent p s The power diagram cell boundaries shared between (1.19) The weights are adjusted by comparing the perception function values ​​of the shared boundaries. ||qp t || 2 -ω t =||qp s || 2 -ω s The adaptive controller (1.19) can be simplified to in, Indicates shared boundaries Length.

Images