Sea-air cross-domain unmanned platform formation method based on topology inference

Abstract

The invention discloses a sea-air cross-domain unmanned platform formation method based on topology inference, which belongs to a cross-domain unmanned system, and comprises the following steps: deploying an unmanned aerial vehicle-surface vessel-submersible vehicle cooperative sea cross-domain unmanned system to collect local state information, designing a topology inference estimator based on inverse reinforcement learning, optimizing formation topology, and obtaining formation information of the formation topology; and designing a formation controller based on inverse reinforcement learning, and carrying out iterative updating on the topological information until a target point is reached. According to the method, the optimal estimation of the heterogeneous network topology is realized, the real-time knowing of the networking topology is ensured, and the communication energy consumption is reduced to the greatest extent while the communication robustness of the heterogeneous networking topology is ensured.

Description

Technical Field

[0001] The present invention relates to a cross-domain unmanned system, and in particular to a sea-air cross-domain unmanned platform formation method based on topology inference. Background Art

[0002] The sea-air cross-domain unmanned platform formation refers to a technical and tactical deployment of multiple unmanned platforms to conduct cross-domain collaborative operations or perform tasks in the marine environment. This formation can include different types of unmanned platforms such as aerial drones, surface unmanned boats, and underwater unmanned submersibles, which work together to complete complex maritime tasks. How to effectively coordinate and control drones, surface boats, and submersibles is the prerequisite and basic guarantee for the sea-air cross-domain unmanned platform to perform complex dynamic tasks.

[0003] Patent document CN114967730A discloses a cross-domain collaborative control system for sea and air unmanned clusters. The document builds a sea and air cross-domain collaborative control system for UAVs and surface vessels, in which the UAVs and surface vessels are positioned through ultra-short baselines, and the speed and expected position of the UAVs and surface vessels at the next moment are estimated based on the positioning information received by the ground station. At the same time, a cross-domain collaborative cluster control algorithm is designed to improve the joint action capability of UAVs and surface vessels. However, this method assumes that the topology of the cluster is stable and unchanged, ignoring the dynamic changes in the formation topology caused by the complex sea and air environment, and cannot guarantee the robustness of the formation topology.

[0004] Patent document CN113037395A discloses a method for collaborative operation of cross-domain unmanned equipment on the water and underwater. The patent constructs a cross-domain unmanned system including drones, surface boats, submersibles, and command and control stations, wherein drone and surface boat sub-networks are constructed through water radio communications, and surface boat and submersible sub-networks are constructed through underwater acoustic communications, so that cross-domain unmanned equipment forms a collaborative network and performs tasks collaboratively under the condition of real-time interaction with the ground command and control center. This method uses neighborhood rules to construct topology, in which each drone, surface boat and submersible needs to communicate with its neighbors. Although the topological connectivity is guaranteed, the energy consumption of the communication network is increased. In summary, facing the complex dynamic environment of sea and air, how to construct a cross-domain unmanned platform formation method with connectivity robustness and energy efficiency under the condition of dynamic topological changes is still a problem that needs to be solved urgently. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a sea-air cross-domain unmanned platform formation method based on topological inference, so as to ensure that the topological robustness and energy efficiency of the cross-domain unmanned platform formation process are improved in the sea-air environment.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a sea-air cross-domain unmanned platform formation method based on topological inference, comprising the following steps:

[0007] Step 1: In the mission target monitoring area, deploy N1 drones in the air; deploy N2 surface boats on the surface; and deploy N3 submersibles underwater to form a maritime cross-domain unmanned system that coordinates drones, surface boats, and submersibles;

[0008] Step 2: In T k moment, where k∈{1,2,…} is the index of the time series. The UAV uses radar to collect the position and speed information of other UAVs within the perception range. At the same time, the UAV uses the camera mounted below to collect the position and speed information of surface vessels within its observation range. The surface vessel uses the camera to collect the position and speed information of surface vessels within the observation range. At the same time, the sonar is used to collect the speed and state information of the submersible within the perception range. The submersible uses the sonar to collect the position and speed information of surface vessels and submersibles within its perception range.

[0009] Step 3: The UAVs, surface vessels, and submersibles use the distance changes between them to map the topological links, and build topological inference optimization models based on the collected local state information. A topological inference estimator based on inverse reinforcement learning is designed to perform topological estimation on the UAV formation, surface vessel formation, and submersible formation respectively.

[0010] Step 4: Construct angle constraints for the UAV and the surface boat, and use the minimum angle rigidity to optimize the formation topology; construct distance constraints for the surface boat and the submersible, and use the optimal persistent rigidity to optimize the formation topology;

[0011] Step 5: According to the target situation information, the UAV, surface ship and submersible are respectively designed with inverse reinforcement learning-based formation controllers to achieve approaching the target point during the formation process;

[0012] Step 6: In T k+1 The topology of the UAV-surface vessel-submersible formation is re-inferred and updated at all times. If the topology changes, return to step 4; otherwise, the existing topology is maintained to approach the target point. When the target point is reached, the task ends.

[0013] A further improvement of the technical solution of the present invention is that the status information collected by the UAV, the surface vessel and the submersible in step 2 is respectively as follows:

[0014] The drone uses radar to collect the position and speed status of other drones within the sensing range, which is expressed as in Represents the drone i within the sensing range respectively UAV (i UAV ∈{1,2,...,N1}) in the X-axis, Y-axis and Z-axis directions, Indicates drone iUAV Velocity components on the X, Y, and Z axes;

[0015] The drone uses the camera mounted below to collect the position and speed information of the surface boat within the observation range, the surface boat uses the camera to collect the position and speed information of the surface boat within the observation range, and the submersible uses the sonar to collect the position and speed information of the surface boat within the perception range, which can be expressed as in are the surface vessels i within the observation range respectively. USV (i USV ∈{1,2,...,N2}) in the X-axis, Y-axis and Z-axis directions, Indicates surface boat USV Velocity components in the X, Y and Z directions;

[0016] Surface vessels and submersibles use the sonar they carry to collect information about the position and speed of submersibles within their sensing range, expressed as in For submersible AUV (i AUV ∈{1,2,...,N3}) in the X-axis, Y-axis and Z-axis directions, Submersible AUV The velocity components on the X, Y, and Z axes.

[0017] A further improvement of the technical solution of the present invention is that the topology inference optimization models of the UAV, surface vessel and submersible constructed in step 3 are:

[0018]

[0019] where i∈{i UAV ,i USV ,i AUV}, L i represents the Laplacian matrix, Represents its estimated value, S' i represents the state matrix, represents the velocity estimate, represents the offset vector estimator, for The jth row of I i The identity matrix, 1 i is the unit column vector, represents the Kronecker product.

[0020] A further improvement of the technical solution of the present invention is that the topology inference estimator based on inverse reinforcement learning designed in step 3 is:

[0021]

[0022] in, Expert strategies can be obtained by selecting the best performing strategy through the interaction between drones, surface vessels and submersibles and the environment; S' i is the velocity state matrix, g(L i ,u)=[Δ 1,i ,Δ 2,i ] T Q i,1 [Δ 1,i ,Δ 2,i ]+u i T R i,1 u i is the cost function, Q i,1 and R i,1 is a positive definite matrix, u i is the learning increment, by solving u i The optimal value of L i The best estimate of .

[0023] A further improvement of the technical solution of the present invention is that the method for optimizing the formation topology by using the minimum angle rigidity in step 4 is as follows:

[0024] For the water environment, the UAV and the surface boat first establish a neighbor list locally, and then generate the angle stiffness matrix based on it, that is,

[0025]

[0026] in, a,b,c∈{i UAV ,i USV}And a≠b≠c,i UAV ∈{1,2,...,N1} and i USV ∈{1,2,...,N2} represents the UAV and the surface ship respectively, α is the angle of ∠bac, d ab is the direction vector between a and b, l ab is the distance between a and b; and is the orthogonal projection operator; delete The extra rows in n a is the total number of a and its neighbors; the angles corresponding to the remaining rows are used as angle constraints; the angle constraints of UAVs and surface boats are recorded; the angle constraints of all UAVs and surface boats are obtained through the above calculations; these edges that constitute the angle constraints are used to form a new global angle rigid topology, namely the minimum angle rigid topology.

[0027] A further improvement of the technical solution of the present invention is that the method for optimizing the formation topology by using the optimal persistent rigidity in step 4 is as follows:

[0028] For underwater environment, the optimal persistent rigid topology is used to optimize the formation topology. First, the distance stiffness matrix is ​​constructed as follows:

[0029]

[0030] where f,g∈{i USV ,i AUV}, i USV ∈{1,2,...,N2} and i AUV ∈{1,2,...,N3} represent surface vessels and submersibles respectively; The rows in the topology are sorted from small to large according to the corresponding edge lengths in the topology, and redundant rows are deleted to ensure Based on the remaining edges, a local distance rigid topology of the surface ship and the submersible is constructed; the above process is repeated to form a new global rigid topology, i.e., the optimal rigid topology, using the edges that are reused in these topologies; the optimal persistent topology can be obtained by directing the undirected edges in the optimal rigid topology; the optimal persistent topology needs to satisfy the condition that each point has no more than 3 directions pointing to its neighbors, i.e., the out-degree of each point Satisfies less than or equal to 3.

[0031] A further improvement of the technical solution of the present invention is that the design of the formation controller in step 5 is specifically: setting the cost function of the UAV, surface ship and submersible to:

[0032]

[0033] where i∈{i UAV ,i USV ,i AUV}, X i is the state vector; τ i is the control input; e i are target tracking error and formation error; and are the state weight matrix and the control weight matrix; a heterogeneous network formation control method based on inverse reinforcement learning is designed to seek the optimal control strategy, namely:

[0034]

[0035] in, It is an expert control strategy. It can be obtained by selecting the best strategy among the interactions between drones, surface vessels and submersibles and the environment. The neural network is used to approximate τ i ,Right now:

[0036]

[0037] Among them, φ i (t) The cost functions are and control strategy τ i The basis function of and are the critic network vector and the actor network respectively, by finding the optimal strategy Can raise the control level of drones, surface vessels and submersibles to the same level as expert strategies Same level.

[0038] A further improvement of the technical solution of the present invention is that: in step 6, the Laplace matrix L i , target tracking and formation error e i Indicators are provided, which can be expressed as:

[0039]

[0040] in, is the local Laplace matrix under the preset topology; ε1 is the threshold at which the topology does not change; ε2 is the threshold at which the formation task is completed.

[0041] Due to the adoption of the above technical scheme, the technical progress achieved by the present invention is: considering the dynamic change of the topological structure of the heterogeneous network of unmanned aerial vehicles, surface vessels and submersibles in the cross-domain environment of sea and air, the optimal estimation of the heterogeneous network topology is achieved by designing a topological inference estimator based on inverse reinforcement learning, and the real-time knowledge of the networking topology is ensured. The different characteristics of information transmission of different media in cross-domain scenarios are fully considered, and the network topology of unmanned aerial vehicles, unmanned aerial vehicles and surface vessels is optimized by adopting the minimum angle rigid topology, and the network topology of surface vessels, submersibles and submersibles is optimized by adopting the optimal persistent topology under distance rigidity, so as to ensure the robustness of the connectivity of the heterogeneous network topology while minimizing the communication energy consumption. By adopting a heterogeneous network formation controller based on inverse reinforcement learning, it is suitable for unmanned aerial vehicles, surface vessels and submersibles with different models, and the collaborative control of heterogeneous formations is realized. While ensuring the learning of the optimal control strategy, the control effect is improved to the same level as the expert strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 It is a flow chart of a sea-air cross-domain unmanned platform formation method based on topological inference of the present invention;

[0044] Figure 2 It is the overall block diagram of topology inference, optimization and formation control in the present invention;

[0045] Figure 3 This is a schematic diagram of the UAV-surface vessel-submersible network formation of the present invention; DETAILED DESCRIPTION

[0046] The present invention is further described in detail below in conjunction with embodiments:

[0047] like Figure 1 As shown, it is a flow chart of a method for sea-air cross-domain unmanned platform formation based on topological inference, which specifically includes the following steps:

[0048] Step 1: Figure 3 As shown in the figure, in the mission target monitoring area, N1 drones are deployed in the air; N2 surface boats are deployed on the surface; and N3 submersibles are deployed underwater, forming a maritime cross-domain unmanned system of drone-surface boat-submersible coordination. UAV-UAV, UAV-surface boat, and surface boat-surface boat communicate using electromagnetic waves as carriers, while surface boat-submersible and submersible-submersible communicate using sound waves.

[0049] Step 2: In T k At the moment, k∈{1,2,…} is the index of the time series. The drone uses radar to collect the position and speed status of other drones within the sensing range, which is expressed as in Represents the drone i within the sensing range respectively UAV (i UAV ∈{1,2,...,N1}) in the X-axis, Y-axis and Z-axis directions, Indicates drone i UAV The velocity components on the X, Y, and Z axes.

[0050] The drone uses the camera mounted below to collect the position and speed information of the surface boat within the observation range, the surface boat uses the camera to collect the position and speed information of the surface boat within the observation range, and the submersible uses the sonar to collect the position and speed information of the surface boat within the perception range, which can be expressed as in are the surface vessels i within the observation range respectively. USV (i USV ∈{1,2,...,N2}) in the X-axis, Y-axis and Z-axis directions, Indicates surface boat USVThe components of velocity in the X, Y, and Z directions.

[0051] Surface vessels and submersibles use the sonar they carry to collect information about the position and speed of submersibles within their sensing range, expressed as in For submersible AUV (i AUV ∈{1,2,...,N3}) in the X-axis, Y-axis and Z-axis directions, Submersible AUV The velocity components on the X, Y, and Z axes.

[0052] Step 3: Based on the collected information, the topological link relationship is mapped by using the distance changes between the drones in the drone formation, between the surface boats in the surface boat formation, and between the submarines in the submarine formation. The topological inference optimization models of the drone formation, surface boat formation, and submarine formation are designed respectively. The three models are consistent in form:

[0053]

[0054] where i∈{i UAV ,i USV ,i AUV}, L i represents the Laplacian matrix, Represents its estimated value. S' i represents the state matrix, represents the velocity estimate, represents the offset vector estimator, for The jth row of I i The identity matrix, 1 i is the unit column vector, Represents the Kronecker product. Based on this, a topology estimator based on inverse reinforcement learning is given:

[0055]

[0056] in, It is an expert strategy that can be obtained by interacting with the environment through drones, surface vessels and submersibles, and selecting the strategy with the best performance. S' i is the velocity state matrix, g(L i ,u)=[Δ 1,i ,Δ 2,i ] T Q i,1 [Δ 1,i ,Δ 2,i ]+u iT R i,1 u i is the cost function, Q i,1 and R i,1 is a positive definite matrix, u i is the learning increment. By solving u i The optimal value of L i The topology inference estimator is used to perform topology estimation on the UAV formation, surface ship formation and submersible formation respectively.

[0057] Step 4: For the water environment, the UAV and the surface boat first establish a neighbor list locally, and generate an angle stiffness matrix based on it, that is,

[0058]

[0059] in, a,b,c∈{i UAV ,i USV}And a≠b≠c,i UAV ∈{1,2,...N1} and i USV ∈{1,2,...N2} represent the UAV and the surface boat respectively, α is the angle of ∠bac, d ab is the direction vector between a and b, l ab is the distance between a and b. and is the orthogonal projection operator. Delete The extra rows in n a is the total number of a and its neighbors. The angles corresponding to the remaining rows are taken as angle constraints. The angle constraints of the UAV and the surface ship are recorded. By analogy, the angle constraints of all UAVs and surface ships can be obtained. Using these edges that constitute the angle constraints, a new global angle rigid topology is formed, that is, the minimum angle rigid topology.

[0060] For underwater environment, the optimal persistent rigid topology is used to optimize the formation topology. First, the distance rigidity matrix is ​​constructed as:

[0061]

[0062] where f,g∈{i USV ,i AUV}, i USV ∈{1,2,...N2} and i AUV ∈{1,2,...N3} represent surface vessels and submersibles respectively. The rows in the topology are sorted from small to large according to the corresponding edge lengths in the topology, and redundant rows are deleted to ensure Based on the remaining edges, a local distance rigid topology for the surface ship and the submersible is constructed. Repeat the above process and use the edges that are reused in these topologies to form a new global rigid topology, that is, the optimal rigid topology. By directing the undirected edges in the optimal rigid topology, the optimal persistent topology can be obtained. The optimal persistent topology needs to meet the condition that each point has no more than 3 directions pointing to its neighbors, that is, the out-degree of each point Satisfies less than or equal to 3.

[0063] Step 5: According to the target situation, design the formation controller based on inverse reinforcement learning for UAV, surface ship and submersible respectively. The three formation controllers have the same form. Specifically, the cost function of UAV, surface ship and submersible is the same as follows:

[0064]

[0065] where i∈{i UAV ,i USV ,i AUV}, X i is the state vector; τ i is the control input; e i are target tracking error and formation error; and is the state weight matrix and the control weight matrix. Furthermore, a heterogeneous network formation control method based on inverse reinforcement learning is designed to seek the optimal control strategy, that is,

[0066]

[0067] in, It is an expert control strategy, which can be obtained by selecting the best strategy among the interactions between drones, surface vessels and submersibles and the environment. i ,Right now

[0068]

[0069] Among them, φ i (t) The cost functions are and control strategy τ i The basis function of and are the critic network vector and the actor network respectively. By finding the optimal strategy Can raise the control level of drones, surface vessels and submersibles to the same level as expert strategies Same level.

[0070] Step 6: In T k+1 At this moment, the topology of the UAV-surface ship-submersible formation is inferred and the topology information is updated, where the Laplace matrix Li , target tracking and formation error e i Provides indicators, which can be expressed as

[0071]

[0072] in, is the local Laplace matrix under the preset topology; ε1 is the threshold value of the topology not changing; ε2 is the threshold value of the formation task completion. If the above indicators are not met, return to step 4.

[0073] The overall block diagram of the above topology inference, optimization and formation control is as follows: Figure 2 shown.

[0074] The embodiments described above are merely descriptions of preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A sea-air cross-domain unmanned platform formation method based on topological inference, characterized by: The steps include: Step 1: In the mission target monitoring area, deploy N1 drones in the air; deploy N2 surface boats on the surface; and deploy N3 submersibles underwater to form a maritime cross-domain unmanned system that coordinates drones, surface boats, and submersibles; Step 2: In T k At that moment, k∈{1,2,…} is the index of the time series. The UAV uses the radar to collect the position and speed information of other UAVs within the sensing range. At the same time, the UAV uses the camera mounted below to collect the position and speed information of the surface boats within its observation range. The surface vessel uses the camera to collect the position and speed information of the surface vessel within the observation range, and uses the sonar to collect the speed and status information of the submersible within the perception range; the submersible uses the sonar to collect the position and speed information of the surface vessel and the submersible within its perception range; Step 3: The UAVs, surface vessels, and submersibles use the distance changes between them to map the topological links, and build topological inference optimization models based on the collected local state information. A topological inference estimator based on inverse reinforcement learning is designed to perform topological estimation on the UAV formation, surface vessel formation, and submersible formation respectively. Step 4: Construct angle constraints for the UAV and the surface boat, and use the minimum angle rigidity to optimize the formation topology; construct distance constraints for the surface boat and the submersible, and use the optimal persistent rigidity to optimize the formation topology; Step 5: According to the target situation information, the UAV, surface ship and submersible are respectively designed with inverse reinforcement learning-based formation controllers to achieve approaching the target point during the formation process; Step 6: In T k+1 The topology of the UAV-surface vessel-submersible formation is re-inferred and updated at all times. If the topology changes, return to step 4; otherwise, the existing topology is maintained to approach the target point. When the target point is reached, the task ends.

2. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The status information collected by the drone, surface vessel and submersible in step 2 are as follows: The drone uses radar to collect the position and speed status of other drones within the sensing range, which is expressed as in Respectively represent the drone i within the sensing range UAV (i UAV ∈{1,2,...,N1}) in the X-axis, Y-axis and Z-axis directions, Indicates drone i UAV Velocity components on the X, Y, and Z axes; The drone uses the camera mounted below to collect the position and speed information of the surface boat within the observation range, the surface boat uses the camera to collect the position and speed information of the surface boat within the observation range, and the submersible uses the sonar to collect the position and speed information of the surface boat within the perception range, which can be expressed as in are the surface vessels i within the observation range respectively. USV (i USV ∈{1,2,...,N2}) in the X-axis, Y-axis and Z-axis directions, Indicates surface boat USV Velocity components in the X, Y and Z directions; Surface vessels and submersibles use the sonar they carry to collect information about the position and speed of submersibles within their sensing range, expressed as in For submersible AUV (i AUV ∈{1,2,...,N3}) in the X-axis, Y-axis and Z-axis directions, The submersible AUV The velocity components on the X, Y, and Z axes.

3. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The topology inference optimization models of the UAV, surface ship and submersible constructed in step 3 are: where i∈{i UAV ,i USV ,i AUV }, L i represents the Laplacian matrix, Represents its estimated value, S' i represents the state matrix, represents the velocity estimate, represents the offset vector estimator, for The jth row of I i The identity matrix, 1 i is the unit column vector, represents the Kronecker product.

4. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The topology inference estimator based on inverse reinforcement learning designed in step 3 is: in, Expert strategies can be obtained by selecting the best performing strategy through the interaction between drones, surface vessels and submersibles and the environment; S' i is the velocity state matrix, g(L i ,u)=[Δ 1,i ,Δ 2,i ] T Q i,1 [Δ 1,i ,Δ 2,i ]+u i T R i,1 u i is the cost function, Q i,1 and R i,1 is a positive definite matrix, u i is the learning increment, by solving u i The optimal value of L i The best estimate of .

5. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The method for optimizing the formation topology using minimum angle rigidity in step 4 is as follows: For the water environment, the UAV and the surface boat first establish a neighbor list locally, and then generate the angle stiffness matrix based on it, that is, in, a,b,c∈{i UAV ,i USV }And a≠b≠c, i UAV ∈{1,2,...,N1} and i USV ∈{1,2,...,N2} represents the UAV and the surface ship respectively, α is the angle of ∠bac, d ab is the direction vector between a and b, l ab is the distance between a and b; and is the orthogonal projection operator; delete The extra rows in n a is the total number of a and its neighbors; the angles corresponding to the remaining rows are used as angle constraints; the angle constraints of UAVs and surface boats are recorded; the angle constraints of all UAVs and surface boats are obtained through the above calculations; these edges that constitute the angle constraints are used to form a new global angle rigid topology, namely the minimum angle rigid topology.

6. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The method for optimizing the formation topology using the optimal persistent rigidity in step 4 is as follows: For underwater environment, the optimal persistent rigid topology is used to optimize the formation topology. First, the distance stiffness matrix is ​​constructed as follows: where f,g∈{i USV ,i AUV },i USV ∈{1,2,...,N2} and i AUV ∈{1,2,...,N3} represent surface vessels and submersibles respectively; The rows in the topology are sorted from small to large according to the corresponding edge lengths in the topology, and redundant rows are deleted to ensure Based on the remaining edges, a local distance rigid topology of the surface ship and the submersible is constructed; the above process is repeated to form a new global rigid topology, i.e., the optimal rigid topology, using the edges that are reused in these topologies; the optimal persistent topology can be obtained by directing the undirected edges in the optimal rigid topology; the optimal persistent topology needs to satisfy the condition that each point has no more than 3 directions pointing to its neighbors, i.e., the out-degree of each point Satisfies less than or equal to 3.

7. The method for sea-air cross-domain unmanned platform formation based on topological inference according to claim 1 is characterized by: The design of the formation controller in step 5 is as follows: the cost functions of the UAV, surface ship and submersible are set as: where i∈{i UAV ,i USV ,i AUV }, X i is the state vector; τ i is the control input; e i are target tracking error and formation error; and are the state weight matrix and the control weight matrix; a heterogeneous network formation control method based on inverse reinforcement learning is designed to seek the optimal control strategy, namely: in, It is an expert control strategy. It can be obtained by selecting the best strategy among the interactions between drones, surface vessels and submersibles and the environment. The neural network is used to approximate τ i ,Right now: Among them, φ i (t) The cost functions are and control strategy τ i The basis function of and They are the critic network vector and the actor network respectively, by finding the optimal strategy Can raise the control level of drones, surface vessels and submersibles to the same level as expert strategies Same level.

8. The method for sea-air cross-domain unmanned platform formation based on topology inference according to claim 1 is characterized by: The Laplace matrix L in step 6 i , target tracking and formation error e i Indicators are provided, which can be expressed as: in, is the local Laplace matrix under the preset topology; ε1 is the threshold at which the topology does not change; ε2 is the threshold at which the formation task is completed.

Images