Non-cooperative target-oriented cluster interception cost rapid estimation method and system

Abstract

The invention discloses a non-cooperative target-oriented cluster interception cost rapid estimation method and system, and the method comprises the steps: designing a communication topology network and initializing the parameters of an intelligent agent through employing a navigation-following topology modeling and topology splitting method, and building a cluster multi-target expected state model; designing a distributed specified time target observer based on the cluster multi-target expected state model; establishing a non-cooperative target-oriented cluster interception cost estimation model based on a distributed specified time target observer; and designing an interception cost convex resolving method to solve the cluster interception cost estimation model, and resolving the interception cost. Compared with the prior art, situation awareness can be carried out on a plurality of non-cooperative targets within specified time, the cost value can be calculated on line, the overall calculation time is short, the method can be widely applied to the field of cluster task planning, and support is provided for planning decision making.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of swarm control, and particularly relates to a swarm interception cost fast estimation method and system for non-cooperative targets. BACKGROUND

[0002] The core value of the swarm lies in breaking through the single-body capability limit through a distributed architecture, and being capable of realizing swarm intelligence emergence in a communication limited environment, so that the swarm is widely applicable and has been widely used in national strategic fields such as space exploration, intelligent manufacturing and smart energy.

[0003] At present, the swarm situation awareness depends on radar and sensors on each intelligent agent to perceive and locate targets, and is capable of completing observation and perception of the target swarm within a certain error range. The current interception cost estimation methods include a forward BP neural network method and a wavelet neural network method, etc. These methods establish an estimation model of the missile interception cost through neural network approximation learning, and are capable of realizing preliminary estimation of the target interception cost under specific conditions, but still have the following limitations in the calculation of the swarm interception cost for non-cooperative targets:

[0004] Firstly, it is impossible to realize situation awareness of the non-cooperative targets in a short time. In actual application, the non-cooperative targets are unevenly distributed and the position change is complex, and the existing methods take a long time to calculate when observing the targets, so that the situation awareness result is difficult to adapt to the rapidly changing non-cooperative target battlefield environment.

[0005] Secondly, the calculation time of the cost estimation is relatively long. The existing interception cost estimation algorithms based on the BP neural network or the wavelet neural network method can only be used for offline calculation, and are difficult to meet the scene of online real-time calculation in flight. SUMMARY

[0006] In view of the limitations in the calculation of the swarm interception cost, the application provides a swarm interception cost fast estimation method and system for non-cooperative targets, which can realize situation awareness of multiple non-cooperative targets within a specified time, and can calculate the cost value online, so as to meet the scene of online real-time calculation in flight.

[0007] The specific content of the application is as follows:

[0008] A swarm interception cost fast estimation method for non-cooperative targets comprises the following steps:

[0009] Step 1: A leader-following topology modeling and topology splitting method is used to design a communication topology network, initialize intelligent agent parameters, and establish a swarm multi-target expected state model.

[0010] Step 2: A distributed specified time target observer is designed based on the swarm multi-target expected state model.

[0011] Step 3, based on the distributed specified time target observer, a cluster interception cost estimation model for non-cooperative targets is established;

[0012] Step 4, a convex interception cost solving method is designed to solve the cluster interception cost estimation model, and the interception cost is solved.

[0013] Further, the communication topology network uses an undirected graph The network of n followers is described using The graph composed of and the leader includes information of followers that can directly establish contact with the leader; the undirected graph is divided into several units, and the undirected edge topology of part of the followers in each unit is connected, if Each unit in the graph is connected to the leader through a directed edge, and the graph of this system is called is connected;

[0014] Let be the neighbor label set of the follower i at time t, be a matrix describing whether the followers can directly communicate with each other, indicates that the followers i and j are not adjacent; let be the label set of the followers adjacent to the leader at time t, be a matrix describing whether the leader and the followers can directly communicate with each other, indicates that the leader and the follower i cannot directly communicate, .

[0015] Further, the cluster multi-target expected state model is an expected state model of each follower to the leader, that is:

[0016] ;

[0017] Wherein, m is the dimension of the position, velocity and control quantity, is a parameter related to the position relationship between the follower and the leader, is an expected unknown of the follower to the position relationship of the leader during the cluster flight, is the position of the leader, is the velocity of the leader, is the control quantity of the leader, and t is a variable representing time.

[0018] Further, the distributed specified time target observer is:

[0019] ;

[0020] wherein, , , , is a time-based regulator function, the switching signal is a piecewise function, is a diagonal matrix whose diagonal elements are , is the Laplacian matrix of n followers, in which the communication remains unchanged, is denoted by and , , are intermediate variables.

[0021] Further, the time-based regulator function satisfies:

[0022] (1) is a continuous, non-decreasing function;

[0023] (2) ;

[0024] (3) ;

[0025] wherein, is an initial time instant, is a specified end time instant.

[0026] Further, the time-based regulator function is:

[0027] .

[0028] Further, the cluster interception cost estimation model is:

[0029] ;

[0030] wherein, J represents an interception cost, represents a cost value related to terminal state deviation amount, represents a cost value related to energy consumption, represents a terminal constraint, represents an inequality constraint in flight, is all state quantities, is all control quantities, represents the number of cluster members.

[0031] Further, the step 4 specifically comprises:

[0032] Process 1): Initialize parameters: Use the standard trajectory of the agent as the initial parameters. If the difference between the initial point parameters and the standard trajectory is greater than the set threshold, that is, there is a fault in the previous flight segment, then directly use the current point as the initial point and the end point of the standard trajectory as the terminal point, and assign the initial value through linear interpolation.

[0033] Step 2): Set the relevant deviation tolerance parameter values: including Initial values ​​of penalty function coefficients and permissible scope and the requirements for dynamic convergence accuracy Terminal constraint accuracy requirements Maximum number of iterations ;

[0034] Process 3): Begin iterative solution, setting the iteration count k=0;

[0035] Step 4): Apply a two-step correction strategy to obtain the optimal solution. ;

[0036] Step 5): Determine if the linearization accuracy meets the requirements: If the linearization accuracy... Then let , If the result of this round of calculation is discarded, return to process 2); otherwise, update the optimization variables. , place ;

[0037] Step 6): Determine if the convergence accuracy is satisfied: If And the terminal constraint accuracy meets Then let , ;if If the j-th equality constraint is removed, then the relaxation factor corresponding to the j-th equality constraint is removed.

[0038] Process 7): If Then the problem can be solved directly by iteration;

[0039] Process 8): If the following convergence condition is met: Then the solution is complete, and the optimal solution is... ,if If the solution fails, the process will fail; otherwise, return to step 4) and start a new iteration.

[0040] A system for fast cost estimation of cluster interception against non-cooperative targets, comprising:

[0041] The cluster multi-objective desired state model building unit adopts the leader-follower topology modeling and topology splitting method to design the communication topology network and initialize the agent parameters, and establish the cluster multi-objective desired state model.

[0042] The design unit for a distributed time-specified target observer is based on a cluster multi-objective desired state model, which designs a distributed time-specified target observer.

[0043] The cluster interception cost estimation model building unit establishes a cluster interception cost estimation model for non-cooperative targets based on a distributed time-specified target observer.

[0044] The solution unit is designed to solve the cluster interception cost estimation model by using a convex solution method. The interception cost is then calculated.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] (1) A target observation method with a specified time is proposed, which can converge quickly within the set time, with short calculation time and strong real-time performance; (2) An online convex solution calculation method for interception cost is proposed, which can calculate the estimated interception cost online during the interception process and has strong applicability; (3) The overall calculation time is short and it can be widely applied in the field of cluster task planning, providing support for planning decisions. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the logical structure for fast estimation of cluster interception costs in this invention.

[0048] Figure 2 This is a flowchart of the method for rapid estimation of cluster interception costs in this invention.

[0049] Figure 3 This is a diagram of the cluster communication topology.

[0050] Figure 4 A convergence plot of the observer error for a specified time target.

[0051] Figure 5 This is a spatial trajectory diagram of a loitering rovers. Detailed Implementation

[0052] The technical solution of the present invention will be described in detail below with reference to specific sample simulations and accompanying drawings.

[0053] Example 1

[0054] Combination Figure 1 This implementation provides a fast method for estimating the cost of cluster interception for non-cooperative targets, including:

[0055] First, establish a multi-objective desired state model for the cluster. The original problem is the situational awareness of a cluster towards distributed multiple non-cooperative objectives, which requires topological decomposition into multiple systems consisting of one leader (non-cooperative objective) and n followers (cluster members). For example... Figure 3 Using undirected graphs Describe a network of n followers using Indicates by An undirected graph is a graph consisting of the leader and followers, and includes information about followers who can directly connect with the leader. It can be divided into several units, each unit containing a partial undirected edge topological connection of followers, if A system is called a graph if every element has at least one follower connected to the leader via a directed edge. They are connected.

[0056] In practical applications, the relationships between neighboring followers may change over time, and their interconnection topology may also change dynamically. Let... It is the set of neighbor tags of the follower i at time t. It is a matrix that describes whether followers can communicate directly. This indicates that followers i and j are not adjacent. Let... Let be the set of labels of the followers that are adjacent to the leader at time t. This is a matrix describing whether the leader and followers communicate directly. This indicates that the navigator and follower i cannot communicate directly.

[0057] Since the non-cooperative goal, i.e., the leader in the split topology, is in motion, its state changes continuously throughout the process, and its dynamics are described below:

[0058]

[0059] in, It is the position of the navigator. It's the speed of the leader.

[0060] During swarm flight, each follower expects the navigator to be in a specific position. With the Navigator status Directly related, suppose that the desired position of the navigator by the acquired followers satisfies the following conditions based on the navigator's state:

[0061]

[0062] in, Describe the relationship between the unknown expectations of followers regarding the navigator and the navigator's position during swarm flight. The parameters describing the positional relationship between the follower and the navigator are related to the topological relationship of the follower.

[0063] Based on the leader dynamics equations, a model of each follower's expected state of the leader when facing observations of a non-cooperative target is derived:

[0064]

[0065] Therefore, the original situational awareness problem of distributed multiple non-cooperative targets is decomposed into multiple situational awareness sub-problems targeting a single target, where m is the dimension of position, velocity, and control variables.

[0066] This invention employs a leader-follower topology modeling and topology decomposition method to break down the distributed multi-target situational awareness problem into multiple distributed single-target situational awareness problems.

[0067] Next, a distributed time-specified target observer is designed. Based on the above model, a time-specified target position observer is designed. However, during flight, the follower positions change in real time with the navigator's position, and it cannot be guaranteed that all followers can obtain the navigator's position in real time. Value, therefore, the follower must monitor the entire flight. Estimate the position of the single aircraft to obtain the ideal position. .

[0068] Assumption It is the follower i (i=1,...,n) pair Given the estimated value, the distributed time-specified target observer design is as follows:

[0069]

[0070] Where i = 1, ..., n, let It is the set of agent labels adjacent to follower i at time t. It is a matrix that describes whether followers are adjacent. This indicates that followers i and j are not adjacent. Let... Let be the set of labels of followers that are adjacent to the leader at time t. It is a matrix that describes whether the leader and followers are adjacent. This indicates that the navigator and the follower i are not adjacent. , It is a Time Base Generator (TBG) function, which must satisfy the following three conditions simultaneously:

[0071] (4) It is a continuous, non-decreasing function;

[0072] (5) ;

[0073] (6) .

[0074] This invention introduces a time-based regulator function to design a non-cooperative target state observer for a specified time, which can converge quickly within a preset time to achieve rapid observation of the non-cooperative target state, making it more suitable for practical application scenarios.

[0075] The dynamics of the observed variables are represented as follows:

[0076]

[0077] Where i = 1, ..., n, it is clear that the above two formulas only contain information about the product and its neighbors.

[0078] The above formula and the distributed time-specified target observer design function are transformed into:

[0079]

[0080] Represented in a compact form as follows:

[0081]

[0082] in, , Switch signal It is a piecewise function. It is A diagonal matrix, where the diagonal elements are (1 when the follower can communicate directly with the navigator, 0 otherwise) It is the Laplace matrix of n followers, in When the communication remains unchanged, use and This is shown in the diagram. The asymptotic convergence of the observer for the specified time target is verified by establishing a Control Lyapunov Function (CLF) on the compact form system function. The error convergence plot is shown in the diagram. Figure 4 As shown.

[0083] Next, a cluster interception cost estimation model for non-cooperative targets is established.

[0084] The dynamic equations of the intelligent agent are established as follows:

[0085]

[0086] in, For all state variables, For all control variables.

[0087] The constraints are as follows:

[0088] (1) Initial constraints

[0089] Considering the initial state of multiple agents, the corresponding constraints are given:

[0090]

[0091] in, The initial time is referred to as 0" to represent the initial state.

[0092] (2) Input constraints

[0093] Inequality constraints introduced during flight:

[0094]

[0095] in, Enter the number of constraints.

[0096] (3) Terminal constraints

[0097]

[0098] in, Let be the terminal constraint function matrix of the agent. When the agent reaches a certain range of the target, the target attack process is considered to have ended.

[0099] The cost of interception is as follows:

[0100]

[0101] in, This represents the cost value related to the deviation in terminal status. This represents the cost value related to energy consumption.

[0102] In summary, the kinematics-based swarm interception cost estimation model can be expressed as:

[0103]

[0104] Finally, a convex solution method for interception cost is designed. The basic idea is to transform the original trajectory planning problem into a series of second-order cone programming subproblems through discretization and linearization, and then approximate the optimal solution of the original problem through iterative solving. The integral term in the interception cost can be calculated using Radau integration. The overall structure of this method is as follows: Figure 1 As shown, the operation flow of this method is as follows: Figure 2 As shown. The specific solution process of the interception cost convex solution calculation method is as follows:

[0105] Process 1): Initialize parameters: Use the standard trajectory of the agent as the initial parameters. If the initial point parameters differ significantly from the standard trajectory (i.e., there is a fault in the previous flight segment), the current point can be used as the initial point and the end point of the standard trajectory strike can be used as the terminal point. Initial values ​​can be quickly assigned through linear interpolation.

[0106] Step 2): Set the relevant deviation tolerance parameter values: including Initial values ​​of penalty function coefficients and permissible scope and the requirements for dynamic convergence accuracy Terminal constraint accuracy requirements Maximum number of iterations ;

[0107] Process 3): Begin iterative solution, setting the iteration count k=0;

[0108] Step 4): Apply a two-step correction strategy to obtain the optimal solution. ;

[0109] Step 5): Determine if the linearization accuracy meets the requirements: If the linearization accuracy... Then let , If the result of this round of calculation is discarded, return to process 2); otherwise, update the optimization variables. , place ;

[0110] Step 6): Determine if the convergence accuracy is satisfied: If And the terminal constraint accuracy meets Then let , .if If the j-th equality constraint is removed, then the relaxation factor corresponding to the j-th equality constraint is removed.

[0111] Process 7): If If the problem has basically converged at this point, then we can directly iterate to solve the problem and achieve fast convergence.

[0112] Process 8): If the following convergence condition is met: Then the solution is complete, and the optimal solution is... ,if If the solution fails, the process will fail; otherwise, return to step 4) and start a new iteration.

[0113] After obtaining the state observation value of the target, this invention uses a convex solution algorithm to quickly calculate the interception cost online, achieving a two-stage acceleration from fast prediction to fast estimation, resulting in a shorter overall solution time and online solution capability.

[0114] This embodiment also provides a system for rapidly estimating the cost of cluster interception for non-cooperative targets, including:

[0115] The cluster multi-objective desired state model building unit adopts the leader-follower topology modeling and topology splitting method to design the communication topology network and initialize the agent parameters, and establish the cluster multi-objective desired state model.

[0116] The design unit for a distributed time-specified target observer is based on a cluster multi-objective desired state model, which designs a distributed time-specified target observer.

[0117] The cluster interception cost estimation model building unit establishes a cluster interception cost estimation model for non-cooperative targets based on a distributed time-specified target observer.

[0118] The solution unit is designed to solve the cluster interception cost estimation model by using a convex solution method. The interception cost is then calculated.

[0119] Example 2

[0120] This embodiment takes the interception of non-cooperative targets on the water surface by a loitering rovers with overload control as an example. The implementation process of a method for rapid cost estimation of swarm interception for non-cooperative targets includes the following four specific steps.

[0121] Step 1: Establish a multi-objective desired state model for the cluster.

[0122] This step is the same as the step in establishing the clustered multi-target observation model in Example 1, and will not be repeated here.

[0123] Step 2: Establish a distributed time-specified target observer model.

[0124] The typical time-base adjustment function selected in the steps of this embodiment is:

[0125]

[0126] in, It is continuously differentiable of order n and satisfies , , , It can be used for the design of time controllers for high-order integral systems.

[0127] Everything else is the same as in Example 1.

[0128] Step 3: Establish a cluster interception cost estimation model for non-cooperative targets.

[0129] A simplified model of the i-th three-degree-of-freedom loitering glider's center of mass motion in the ballistic coordinate system is established:

[0130]

[0131] Where i = 1, 2, ..., n is the number of loitering drones.

[0132] The state space X and control space u of the above dynamic model are defined in vector form as follows:

[0133]

[0134] The meanings of the remaining variables are: velocity Ballistic inclination angle Ballistic deflection range ,high Lateral deviation Normal overload Lateral overload Gravitational acceleration .

[0135] The constraints are as follows:

[0136] (1) Initial constraints

[0137] Considering the initial state of multiple agents, the corresponding constraints are given:

[0138] , , , ,

[0139] Hereinafter, 0" represents the initial state.

[0140] (2) Overload constraint

[0141] To meet the structural stiffness constraints of the intelligent agent, normal overload... Lateral overload Apply constraints:

[0142]

[0143] The subscript "max" indicates the maximum value. This represents the maximum overload capacity of the intelligent agent.

[0144] (3) Terminal constraints

[0145]

[0146] The target engagement process is considered complete when the loitering rovers reach a certain range of the target.

[0147] The estimated interception cost of the model is as follows:

[0148]

[0149] In the formula, At the initial moment, For the terminal time, , , These are the coordinates of the loitering drone i at time t. , , These are the coordinates of the target at time t. and These are the normal overload and lateral overload of the i-th loitering pod, respectively.

[0150] In summary, the kinematics-based swarm interception cost estimation model can be expressed as:

[0151]

[0152] Step 4: Design a convex solution calculation method for interception costs.

[0153] The integral term in the interception cost can be calculated using Radau integration. The specific solution process for the convex solution of the interception cost is the same as in Example 1. The final iteratively solved spatial trajectory of one of the loitering gliders is as follows: Figure 5 As shown, the online calculation time is 0.74 seconds. Simulation results demonstrate that this invention can calculate the cost value online with a short overall calculation time, and can be widely applied in the field of cluster task planning, providing support for planning decisions.

Claims

1. A method for fast estimation of the cost of swarm interception for non-cooperative targets, characterized in that, The method comprises the following steps: Step 1, a communication topology network is designed and the parameters of intelligent agents are initialized by using a leader-follower topology modeling and topology splitting method, and a cluster multi-target desired state model is established; Step 2, a distributed specified time target observer is designed based on the cluster multi-target desired state model; Step 3, a cluster interception cost estimation model for non-cooperative targets is established based on the distributed specified time target observer; Step 4, an interception cost convex solution method is designed to solve the cluster interception cost estimation model, and the interception cost is solved.

2. The method of claim 1, wherein, The communication topology network uses an undirected graph A network of n followers is described using An undirected graph representing the information of the followers that can be directly contacted by the leader; the undirected graph and the leader, including the information of the followers that can be directly contacted by the leader; the undirected graph is divided into several units, and in each unit, the undirected edge topology of part of the followers is connected, and if each unit in the system has at least one follower connected to the leader through a directed edge, then the graph of this system is connected; Let is the set of follower i's neighbors at time t, is a matrix describing whether followers can directly communicate with each other, denotes that followers i and j are not neighbors; let is the set of followers that are neighbors to the leader at time t, is a matrix describing whether the leader can directly communicate with the followers, denotes that the leader cannot directly communicate with follower i, .

3. The method of claim 2, wherein, The cluster multi-target desired state model is a desired state model of each follower to the leader, that is: ； where m is the dimension of the position, velocity and control, expressing parameters related to the position relationship between the follower and the leader, describing the expected unknown of the follower to the position relationship of the leader in the process of cluster flight, is the position of the leader, is the velocity of the leader, is the control of the leader, and t is a variable representing time.

4. The method of claim 3, wherein, The distributed specified time target observer is: ； wherein , , , is a time base regulator function, the switching signal is a piecewise function, is a diagonal matrix with diagonal elements , is the Laplacian matrix of the n followers, in which the communication remains unchanged, is denoted by and , , are intermediate variables.

5. The method of claim 4, wherein, The time-based regulator function satisfies: (1) is a continuous, non-decreasing function; （2） ； （3） ； wherein is an initial time, is a specified end time.

6. The method of claim 5, wherein, The time-based regulator function is: 。 7. The method of claim 4, wherein, The cluster interception cost estimation model is: ; where J denotes the interception cost, denotes the cost value related to terminal state bias, denotes the cost value related to energy consumption, denotes the terminal constraints, denotes the inequality constraints during flight, for all state variables, for all control variables, denotes the number of swarm members.

8. The method of claim 7, wherein, Step 4 specifically comprises: Process 1): initialization of parameters: taking the standard trajectory of the intelligent agent as the initial parameter, if the initial point parameter and the standard trajectory differ by more than a set threshold, that is, there is a fault in the previous flight segment, then directly taking the current point as the initial point, the standard trajectory impact end point as the terminal point, and assigning initial values through linear interpolation; Process 2): Setting deviation allowance related parameter values: including , penalty function coefficient initial value and allowable range , and dynamic convergence accuracy requirement , terminal constraint accuracy requirement , maximum iteration number ; Process 3): start iteration solution, set iteration number k=0; Process 4): Apply two-step correction strategy to solve, get the optimal solution ; Process 5): Determine whether the linearization accuracy meets the requirement: if the linearization accuracy , then let , , and discard the current calculation result, return to process 2), otherwise update the optimization variable , and set ; Process 6): judge whether the convergence precision meets the condition: if and the terminal constraint precision meets , then let , ; if , then remove the relaxation factor corresponding to the jth equality constraint; Process 7): If then the problem is solved directly by iteration; Process 8): If the following convergence condition is satisfied: then the solution is complete, and the optimal solution is If then the solution fails, otherwise return to process 4) to start a new iteration.

9. A cluster-interception cost fast estimation system for implementing the method of any one of claims 1-8, characterized by The method comprises the following steps: A cluster multi-target desired state model construction unit is adopted to design a communication topology network and initialize the parameters of intelligent agents by using a leader-follower topology modeling and topology splitting method, and a cluster multi-target desired state model is established; A distributed specified time target observer design unit is adopted to design a distributed specified time target observer based on the cluster multi-target desired state model; A cluster interception cost estimation model establishment unit is adopted to establish a cluster interception cost estimation model for non-cooperative targets based on the distributed specified time target observer; A solving unit is adopted to design an interception cost convex solution method to solve the cluster interception cost estimation model, and the interception cost is solved.

Images