CN116520887A - Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles - Google Patents
Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles Download PDFInfo
- Publication number
- CN116520887A CN116520887A CN202310640306.2A CN202310640306A CN116520887A CN 116520887 A CN116520887 A CN 116520887A CN 202310640306 A CN202310640306 A CN 202310640306A CN 116520887 A CN116520887 A CN 116520887A
- Authority
- CN
- China
- Prior art keywords
- unmanned aerial
- aerial vehicle
- cluster
- heterogeneous
- communication
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000003044 adaptive effect Effects 0.000 title claims description 15
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 230000002787 reinforcement Effects 0.000 claims abstract description 10
- 238000004891 communication Methods 0.000 claims description 99
- 239000013598 vector Substances 0.000 claims description 29
- 230000006870 function Effects 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 16
- 230000009471 action Effects 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 230000008447 perception Effects 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 5
- 238000005265 energy consumption Methods 0.000 claims description 5
- 230000007246 mechanism Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 239000003795 chemical substances by application Substances 0.000 claims description 4
- 238000012423 maintenance Methods 0.000 claims description 3
- 230000001174 ascending effect Effects 0.000 claims description 2
- 230000004927 fusion Effects 0.000 abstract description 8
- 241000854291 Dianthus carthusianorum Species 0.000 description 12
- 238000004364 calculation method Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 210000001503 joint Anatomy 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
- G05D1/104—Simultaneous control of position or course in three dimensions specially adapted for aircraft involving a plurality of aircrafts, e.g. formation flying
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Aviation & Aerospace Engineering (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides a self-adaptive adjustment method for a cluster structure of a hybrid multi-unmanned aerial vehicle, which is characterized in that by designing a special type of state encoder, the attribute and the capability of different dimensions of heterogeneous unmanned aerial vehicles are converged to the same dimension, the dimension of a knowledge space is reduced, then multidimensional heterogeneous data which is processed by fusion of a cluster clustering algorithm based on reinforcement learning is used, a global adjustment strategy is learned, and the cluster structure and a reconstruction network are adjusted. The technical core of the state encoder is that a set of neural network is independently trained for each type of unmanned aerial vehicle, and special treatment is carried out on the attribute and the capability of different dimensions of each type of unmanned aerial vehicle, so that the difference among heterogeneous unmanned aerial vehicles is reflected better. According to the method, the characteristics of different dimensions of each type of unmanned aerial vehicle can be converged under the mixed situation, the problem of multi-dimensional heterogeneous data fusion is effectively solved, and the stability and the efficiency of the mixed multi-unmanned aerial vehicle cluster are improved.
Description
Technical Field
The invention relates to the field of unmanned aerial vehicles, in particular to a self-adaptive adjustment method for a cluster structure of a hybrid multi-unmanned aerial vehicle.
Background
The heterogeneous nature of multidimensional under the cluster situation of hybrid many unmanned aerial vehicles makes heterogeneous unmanned aerial vehicles have very big difference in motion state and communication ability, and these differences make the inside topological structure of cluster more likely dynamic variation, lead to the relative position and the communication cost to change between unmanned aerial vehicles, have caused cluster structure's unstability. The unstable cluster structure refers to frequent changes of the cluster structure caused by factors such as unmanned aerial vehicle variation or communication connection disconnection of the cluster. When the unstable cluster structure state occurs in the multi-unmanned aerial vehicle cluster system, the system needs to be subjected to necessary cluster structure adjustment to ensure the stability of the cluster structure due to the consideration of the optimization of the operation efficiency of the whole system. In the case of hybrid multi-unmanned aerial vehicle cluster situation, the cluster structure adjustment method needs to solve the problem of multi-dimensional heterogeneous data fusion, but most of existing multi-unmanned aerial vehicle cluster structure adjustment models are not considered (for example, when calculating the similarity of the motion states of heterogeneous unmanned aerial vehicles, only the current speed and direction of the unmanned aerial vehicles are considered, and the influence caused by the attribute of other dimensions such as the acceleration of the variable speed unmanned aerial vehicle is ignored), so that the state extraction is inaccurate and the local optimal solution is easily trapped. If the attributes of all different dimensions of the heterogeneous unmanned aerial vehicle are directly considered, the dimensions of all unmanned aerial vehicle attributes are expanded to include all different types of attributes (such as acceleration attribute and maximum speed attribute with an added value of 0 for a fixed-speed unmanned aerial vehicle), the problem that the problem solution space dimension is greatly increased can occur. Therefore, the problem of multi-dimensional heterogeneous data fusion under the situation of the hybrid multi-unmanned aerial vehicle cluster has extremely important influence on the operation efficiency optimization of the whole system.
The fusion of multidimensional heterogeneous data under the cluster situation of the hybrid multiple unmanned aerial vehicles is realized, so that the flexibility of a cluster structure adjustment model is improved, the dimension of a solution space is reduced, a better optimization effect is obtained, and the operation efficiency of the whole system is maximized. In order to achieve the optimization objective, a state encoder can be independently designed for each type of heterogeneous unmanned aerial vehicle in the system to gather the attribute with different dimensions. Each state encoder uses a specially trained neural network, combines the historical states of the heterogeneous unmanned aerial vehicle, and the states of the heterogeneous unmanned aerial vehicle such as attributes and capabilities of different dimensions are processed by the corresponding type of state encoder and are converged into unmanned aerial vehicle observation features of the same dimensions (for example, certain unmanned aerial vehicle has two-dimensional residual energy attributes, namely residual flight energy and residual communication energy), the state encoder of the unmanned aerial vehicle converges the two types of energy according to the consumption speeds of the two types of energy in the historical states of the unmanned aerial vehicle, namely, the energy attributes of the two dimensions are weighted and summed according to proper weights to obtain the energy attribute in the unmanned aerial vehicle observation features).
In order to integrate the processed multidimensional isomerism characteristics with the network structure, a cluster clustering algorithm based on reinforcement learning is introduced into the method, the algorithm builds a graph attention network based on the network structure of the system, the self-attention mechanism is used for integrating the characteristics of all isomerism unmanned aerial vehicles, the quality of different cluster structure adjustment strategies is learned through an Actor-Critic architecture model, and finally the optimal cluster structure adjustment strategy under the current situation is provided.
Disclosure of Invention
Technical problems:
the invention aims to provide a cluster structure adjusting method under a hybrid multi-unmanned aerial vehicle cluster, which mainly solves the problem that multidimensional heterogeneous data are difficult to fuse under a hybrid situation, and the problem can cause the conditions of overlarge algorithm solution space, frequent cluster structure change and the like. The core technology of the invention is that the heterogeneous unmanned aerial vehicle is converged to a single state attribute by utilizing the multi-dimensional information of the attribute and the capability of the heterogeneous unmanned aerial vehicle, the dimension of a solution space is reduced, the cluster clustering algorithm based on reinforcement learning is used for fusing the data such as the characteristic of the heterogeneous unmanned aerial vehicle and the cluster structure, the global cluster structure adjustment strategy is learned, the cluster structure is adjusted according to the strategy, and the communication network is reconstructed, so that the global cluster structure adjustment is realized, and the stability of the cluster structure and the efficiency of the system are further improved.
The technical scheme is as follows:
in a hybrid multi-unmanned aerial vehicle cluster system, each unmanned aerial vehicle bears different numbers of tasks, and meanwhile, unmanned aerial vehicles in the same cluster often have close relations. In order to establish a stable communication network in a Cluster at low cost, a communication unmanned aerial vehicle Cluster Head (CH) of the Cluster needs to be selected in each Cluster, and all member unmanned aerial vehicles Cluster (CMs) of the Cluster need to establish communication connection with the CH.
In the initial stage, a given situation is a hybrid multi-unmanned aerial vehicle cluster system with uniformly distributed unmanned aerial vehicles, U heterogeneous unmanned aerial vehicles with multidimensional isomerism are designed, and the types of unmanned aerial vehicles in each cluster are uniformly distributed. Aiming at the multi-dimensional isomerism of the isomerism unmanned aerial vehicle characteristics in each cluster, the method uses the specific type of state encoder aiming at the isomerism unmanned aerial vehicle to respectively process the state characteristics of different types of unmanned aerial vehicles, trains a set of neural network for each type of unmanned aerial vehicle to process the attribute and the capability state of different dimensionalities, and finally gathers the attribute and the capability characteristics of different dimensionalities through a full-connection layer in the neural network to obtain the unmanned aerial vehicle observation with the same dimensionality.
Each type of heterogeneous unmanned aerial vehicle is provided with a set of special neural network, the special neural network is obtained by training the historical state of the type of unmanned aerial vehicle and is used for combining the characteristics of the unmanned aerial vehicle in the past for a period of time and processing the characteristics of the unmanned aerial vehicle with different current dimensions so as to generate the observation characteristics with the same dimensions, the observation characteristics learn a global cluster structure adjustment strategy through a reinforcement learning method based on an Actor-Critic architecture, and the cluster structure is adjusted according to the learned strategy and the communication network is reconstructed.
Specifically, the neural network of each class of unmanned aerial vehicle uses a reset gate and an update gate to update its current characteristics in combination with the historical state and the current characteristics, while using a fully connected layer to aggregate the attributes and capabilities of different dimensions to the observed characteristics of the same dimension. These observation features are fed into the graph attention network, and the observation of all unmanned aerial vehicles in the system is processed through a self-attention mechanism to obtain the observation features representing the whole system. The Actor network will give a policy pi according to the observation characteristics of the system θ The probability of taking each global adjustment policy in the current system state is represented.
According to pi output by the Actor network θ And selecting a global adjustment strategy, weighting and summing all state factors of the heterogeneous unmanned aerial vehicle by using the utility weight in the adjustment strategy to obtain the utility of the unmanned aerial vehicle, clustering according to the utility and a communication range, and reconstructing a network structure of the hybrid multi-unmanned aerial vehicle cluster by clustering two layers of clusters and cross-clusters to finish primary cluster structure adjustment. Meanwhile, the Critic value network evaluates the optimization effect of the cluster structure adjustment, and calculates the value functions of the current state and the next state. And updating the Critic state value function and the Actor strategy parameter theta through the strategy gradient function, and training the model.
The method solves the problem of multi-dimensional isomerism data fusion under the situation of a hybrid multi-unmanned aerial vehicle cluster, gathers unmanned aerial vehicle states with multi-dimensional isomerism to unmanned aerial vehicle observation features with the same dimension, and reduces the dimension of the knowledge space. On the basis, an optimal cluster structure adjustment strategy is learned by using a reinforcement learning method, so that the best optimization effect is achieved.
The beneficial effects are that:
(1) The adaptive capacity of the hybrid multi-unmanned aerial vehicle cluster application scene is enhanced, the problem of fusion of multidimensional isomerism data under the hybrid multi-unmanned aerial vehicle cluster is solved, and the influence caused by the attribute and the capacity of different dimensionalities of the isomerism unmanned aerial vehicle under the hybrid scene is fully considered. The method can solve the problems that the conventional cluster structure adjustment method which does not consider multidimensional isomerism is poor in adaptability and easy to fall into a local optimal solution.
(2) The dimension of the knowledge space is reduced by using a state encoder aiming at the type of the heterogeneous unmanned aerial vehicle to carry out special processing on each type of unmanned aerial vehicle, and converging the attribute and the capability state of different dimensions of the state encoder to the dimension unified by all unmanned aerial vehicles in the system, so that the dimension of the knowledge space is greatly reduced.
(3) The self-adaptive cluster structure adjustment method for improving the stability of the multi-unmanned aerial vehicle cluster structure aiming at the hybrid situation can be used for fusing the attribute and the capability of different dimensions of the heterogeneous unmanned aerial vehicle with multidimensional heterogeneous data such as each cluster structure, providing an optimal cluster structure adjustment strategy, realizing the dynamic stability of the cluster structure and improving the operation efficiency of the whole system.
Drawings
Fig. 1 is a schematic structural diagram of a heterogeneous multi-drone cluster system.
FIG. 2 is an example diagram of a partial cluster structure.
Fig. 3 is a principal sketch of the method of the invention.
Fig. 4 is a network structure diagram of the method of the present invention.
The different shapes in fig. 1,2 represent different drone types. In fig. 2, the CH communication drone selected from the cluster is responsible for communication with all other CMs member drones in the cluster. The communication between isomorphic unmanned aerial vehicles is solid line connection, and the communication between heterogeneous unmanned aerial vehicles is broken line connection.
Detailed Description
The present invention is further illustrated in the following drawings and detailed description, which are to be understood as being merely illustrative of the invention and not limiting the scope of the invention.
As shown in the figure, the invention provides a self-adaptive adjustment method for a cluster structure of a hybrid multi-unmanned aerial vehicle.
1. Initial stage
In the initial stage, sensing heterogeneous unmanned aerial vehicles, clusters and hybrid multi-unmanned aerial vehicle clusters: the heterogeneous unmanned aerial vehicle perception process analyzes the type, attribute, capability, state and other characteristics of each unmanned aerial vehicle; the cluster sensing process needs to sense the number of clusters in the current multi-unmanned aerial vehicle cluster system and the composition of heterogeneous unmanned aerial vehicles in each cluster; the hybrid network structure sensing process needs to sense the network structure formed by communication connection between unmanned aerial vehicles in clusters and across clusters, and meanwhile, the communication cost which is constantly changed between heterogeneous unmanned aerial vehicles is calculated. In the initial stage, we describe the three perception objects and the perception process in turn:
1.1 heterogeneous unmanned aerial vehicle perception process: a specific analysis was performed for each drone. Kth cluster C k Any unmanned aerial vehicle in (a)
Can be represented by a triplet:Wherein->Represents the kth cluster heterogeneous unmanned plane +.>Is a type of unmanned aerial vehicle;Represents heterogeneous unmanned plane->The attribute and the capability characteristics (the multidimensional isomerism is reflected, the dimension of the characteristic vector P is different between the isomerism unmanned aerial vehicles), such as the residual energy, the speed and the like;Representing heterogeneous unmanned plane->The dimensions of the state feature vector v are all the same, irrespective of the type of drone), such as drone number, current position coordinates, direction, character, etc.
1.2 Cluster awareness process: in the initial stage, a specific analysis of the composition of the unmanned aerial vehicle cluster is first required. The multi-unmanned aerial vehicle cluster system is composed of a plurality of heterogeneous unmanned aerial vehicle clusters C k Constitute, each unmanned aerial vehicle cluster C k From heterogeneous unmanned aerial vehicleThe composition of the composite material comprises the components,
wherein N is k Is the number of drones in the cluster.
1.3 hybrid network structure awareness process: in the hybrid multi-unmanned aerial vehicle cluster system aimed at by the patent, two connection relations exist between heterogeneous unmanned aerial vehicles, one is communication connection inside a cluster, and the other is communication connection crossing the cluster. Communication connection between heterogeneous unmanned aerial vehicles constitutes the network structure that mixes in the system. Both connections are dynamic and will change with cluster structure adjustments. Intra-cluster and inter-cluster connections in a comprehensive heterogeneous multi-unmanned aerial vehicle system, a hybrid multi-unmanned aerial vehicle cluster system (MAS) with N clusters is ultimately described as mas= (N) 1 ,n 2 ,…,n N E), where n k ＝<C k ,CH k >
Representing the complete perception of the kth cluster, C k Heterogeneous unmanned aerial vehicle set being kth cluster, CH k Representing the current communication unmanned aerial vehicle of the kth cluster, E is an adjacency matrix representing the network structure in the hybrid multi-unmanned aerial vehicle cluster system. Using two tuples
Representing a communication connection in the system. All communication connections are constructed and the communication costs spent maintaining the connection at each subsequent time are expressed as +.>Representing cluster C k Heterogeneous unmanned aerial vehicle->And cluster C l Heterogeneous unmanned aerial vehicle->Where l is possibly equal to k.Function c will be according to the droneAnd->The type of (c) and the distance between them calculates the communication cost.
2. Model construction and problem formulation
In order to adjust the cluster structure through a clustering algorithm and a routing algorithm in the hybrid multi-unmanned aerial vehicle cluster system, the states of all unmanned aerial vehicles need to be evaluated, so that the original input obtained in a perception stage is needed to be used for calculating the utility of each unmanned aerial vehicle, and the states of the unmanned aerial vehicles are represented through the calculated utility values. The model designs six factors for the utility of the unmanned aerial vehicle: the remaining energy percentage, the degree, the average distance of the neighbors, the average motion state similarity of the neighbors, the average holding time of the communication connection and the average communication cost, and the utility of the unmanned aerial vehicle is obtained by weighted summation of the six factors.
The weight of each factor can influence the calculation of the utility, and further influence the accuracy of unmanned aerial vehicle state evaluation. Along with the movement of the heterogeneous unmanned aerial vehicle, the cluster structure is changed continuously, and the optimal utility calculation weight capable of accurately reflecting the state of the unmanned aerial vehicle in the current situation is changed. The cluster structure adjustment method under the cluster situation of the hybrid multi-unmanned aerial vehicle needs to consider multidimensional isomerism, fuses multidimensional isomerism data such as characteristics of different dimensions of the heterogeneous unmanned aerial vehicle and cluster structures, and gives out an optimal cluster structure adjustment strategy under the current situation, namely optimal utility calculation weight.
The optimization objective of the problem is that the given cluster structure adjustment strategy can achieve balance of cluster structure stability and system communication cost, and optimal system operation efficiency is achieved. By determining cluster structure adjustment policies, we intend to maximize system operating efficiency over a long period of time. The whole system is used as an agent, and the strategy is adjusted according to the cluster structure to be in butt joint with the environment where the system is located. The markov decision model of the present problem will be given below:
state: s (t) represents the cluster structure state observation of the whole system at the time interval t, and comprises the states, distribution conditions, communication network structures and CH communication unmanned aerial vehicles of the heterogeneous unmanned aerial vehicles in each cluster. First, at the beginning of each time interval t, the composition and state S of the heterogeneous unmanned aerial vehicle in the kth cluster are defined k (t) is:
wherein S is k (t) is the heterogeneous drone in the kth clusterObservation +.>Is set of (N) k And (t) is the number of heterogeneous unmanned aerial vehicles in the kth cluster at the time t. Definitions->The method comprises the following steps:
in the course of the time interval t,is unmanned plane->The type of (2);The attribute and the capability characteristics of the heterogeneous unmanned aerial vehicle comprise residual energy, speed and the like;The state characteristics of the heterogeneous unmanned aerial vehicle comprise unmanned aerial vehicle numbers, position coordinates, directions, roles and the like.
The network structure E (t) of the system at time t is then defined as an adjacency matrix.
The state S (t) of the final overall system can be expressed as:
S(t)＝(S 1 (t),S 2 (t),…,S k (t),…,S N(t) (t),E(t))
wherein N (t) is the number of heterogeneous unmanned aerial vehicle clusters in the t-moment system.
Action:a cluster structure adjustment policy of the system is defined. Is provided with->The policy to be taken by the system at time interval t is defined as:
wherein lambda is i (i∈[1,6]) Utility weight, lambda, representing the ith factor of the drone i The following constraints are satisfied.
λ 1 +λ 2 +λ 3 +λ 4 +λ 5 +λ 6 ＝1
Reward function: r (t), the system adopts a strategy at the moment of t and the state S (t)Obtained by performing cluster structure adjustment
The resulting global rewards may be expressed as:
lambda is the weight, r s Awarding for cluster structure stability, r c Negative rewards for energy consumption.
r s ＝α 1 r ATT +α 2 r CTD +α 3 r SMD +α 4 r CTC
r c ＝-(β 1 r build +β 2 r maintance +β 3 r isolated +β 4 r fly )
Wherein alpha and beta are weights, r ATT Representing an average time that the drone remains within the current cluster; r is (r) CTD The compactness of the network structure among unmanned aerial vehicles in each cluster is represented; r is (r) SMD Representing clustersSimilarity of motion states among the inner unmanned aerial vehicles; r is (r) CTC Representing the average time for maintaining the communication connection between the unmanned aerial vehicles in each cluster; r is (r) build Representing communication costs of new communication connection construction caused by cluster structure adjustment; r is (r) maintance Representing the maintenance costs spent by the communication network in the system at the previous time to maintain to that time; r is (r) isolated Representing the additional communication loss caused by isolated clusters in all clusters; r is (r) fly And the flight energy consumption of all unmanned aerial vehicles in the system at the current moment is represented.
Where N is the number of clusters, N i For cluster C i Number of inner unmanned aerial vehicles, mt i,j For cluster C i Middle jth unmanned aerial vehicleThe time at the current cluster is maintained.
Wherein the method comprises the steps ofFor cluster C i Unmanned aerial vehicle->Communication unmanned aerial vehicle CH in cluster i Distance between them.
Wherein the method comprises the steps ofFor cluster C i Unmanned aerial vehicle->Communication unmanned aerial vehicle CH in cluster i Similarity of motion states, speed is the current speed of the unmanned aerial vehicle, and theta is the direction of the unmanned aerial vehicle.
Wherein the method comprises the steps ofFor cluster C i Unmanned aerial vehicle->Communication unmanned aerial vehicle CH in cluster i The time during which the communication connection is maintained.
Wherein the method comprises the steps ofCommunication costs built for the communication connection calculated in 1.4, I t For the newly constructed communication connection at the time t, the formula is as follows:
I t ＝E(t)-(E(t)∩E(t-1))
the communication cost required by the communication connection newly constructed by the hybrid multi-unmanned aerial vehicle cluster at the time t is calculated.
The communication maintenance cost of all communication connections in the hybrid multi-unmanned aerial vehicle cluster at the last moment is calculated.
r isolated ＝num isolated
Wherein num is isolated The number of orphaned clusters in the system, across which no communication connection exists with any other cluster, i.e. the number of clusters in which no communication connection exists in E. The isolated clusters cannot directly communicate with other clusters, and the actions and states of the other clusters can be acquired through the system base station, so that each isolated cluster can cause additional communication loss.
Wherein the method comprises the steps ofFor cluster C i The flight energy consumption of the jth unmanned aerial vehicle at the current moment is influenced by the type and the speed of the unmanned aerial vehicle.
On this basis, time steps t=1, 2, …, T according to the decision model, and the strategy of the systemIndicating that in state S (t) the system takes action +.>Is a probability of (2).
The strategy of each cluster should meet the optimization condition, i.e. the strategy of the system should maximize its reward function r (t), according to the definition of the markov decision, by the formula:
wherein the method comprises the steps ofThe action space that can be taken for the system is the set of all possible utility weights that can be taken.
3. Cluster structure adjustment algorithm based on reinforcement learning
The present patent uses reinforcement learning algorithms to solve the above-specified problems by first building a neural network for a Markov decision model, then training through an Actor-Critic framework, and updating the value network parameters and policy network parameters. In order to enable the neural network to fuse the multidimensional heterogeneous data dynamically changing in the hybrid multi-unmanned aerial vehicle cluster, the construction of the neural network in the algorithm is divided into two layers: a special state encoder for heterogeneous unmanned aerial vehicle types, a hybrid multi-unmanned aerial vehicle cluster structure.
3.1 Special State encoder for heterogeneous unmanned aerial vehicle types
Firstly, the heterogeneous unmanned aerial vehicles need to be processed in different attribute and capability characteristics of the dimensions, n types of heterogeneous unmanned aerial vehicles exist in the system, and a network is independently trained for each unmanned aerial vehicle to process the characteristics. Ith unmanned aerial vehicle in kth cluster at time tSystem state S (t) = (S) according to original input 1 (t),S 2 (t),…,S k (t),…,S N(t) (t), E (t)) S k Observation of the respective unmanned aerial vehicle contained in (t)>Extracting->Wherein->Representing the type number of the heterogeneous drone.
This patent uses a GRU to design such a special state encoder for heterogeneous drone types. The input of the encoder is the type of the heterogeneous unmanned aerial vehicleAttributes and capabilities of different dimensions of heterogeneous unmanned aerial vehicleSyndrome of->Feature vector of unified dimension is output>The following description is->Is a specific calculation process of (a).
Unmanned aerial vehicle with input recording functionType number->Unmanned aerial vehicle stored in GRU for type +.>Historical (hidden) state of (a)Is a vector for representing the unmanned plane +.>Historical features at the last moment. In the process->At this time, we can add +.>Added to the input data. We use two gating for type unmanned aerial vehicle: reset gate r type And update gate u type Reset gate is used to control whether to ignore the hidden state of GRU +.>The calculation formula is as follows:
wherein type is unmanned aerial vehicleUnmanned aerial vehicle type number, +.>And->Input feature and hidden state pair r in GRU network for processing type unmanned plane state type Weight matrix of>Is the bias vector and σ is the sigmoid function.
Update gate u type For controlling GRU update hidden statusTo a degree of (3). The calculation formula is as follows:
and->Is a pair u of input features and hidden states in a GRU network that handles type unmanned plane states type Weight matrix of>Is the bias vector and σ is the sigmoid function.
According to r type And u type Calculating candidate features at the current time
Wherein the method comprises the steps ofEach element representing a reset gate vector and a hidden state vector is multiplied, +.>Andis the input and hidden state pair +.>Weight matrix of>Is a bias vector, tanh represents a hyperbolic tangent function, mapping the input value to the interval [ -1,1]And (3) inner part.
Finally, the processed candidate features are processed through a full connection layerMapping to a unified dimension:
where H is a nonlinear activation function,is a special dimension for the unmanned aerial vehicleThe degree mapping matrix is a weight matrix of Z rows and D columns, wherein Z is +.>D is->Dimension b of (b) z Is the bias vector. Obtain the heterogeneous unmanned aerial vehicle->Feature vector processed by state encoder of specific type +.>According to u type Update->As the hidden state at the next moment.
By implementing this state encoder, we have completed the processing of the properties and capability features of different dimensions of the heterogeneous drone. Subsequently we will be unmanned aerial vehicle typeFeature vectors of the same dimension after processing>Status feature of unmanned aerial vehicleSpliced to obtain the isomerism unmanned aerial vehicle +.>Observation features of the same dimension->
3.2 modeling of cluster structure of hybrid multiple unmanned aerial vehicles
And each cluster is internally provided with a CH communication unmanned aerial vehicle which establishes communication connection with all other CMs member unmanned aerial vehicles, and communication connection can also exist among unmanned aerial vehicles of different clusters, wherein the communication connection is formed by communication networks in a hybrid multi-unmanned aerial vehicle cluster and is represented by an adjacency matrix E obtained by a sensing process. To train the system as an intelligent agent, the cluster structure of the hybrid multiple unmanned aerial vehicles needs to be modeled, and therefore the problem of fusion of data such as heterogeneous unmanned aerial vehicles, feature vectors with the same dimension after processing, state features of unmanned aerial vehicles, and communication network structures in the system is solved.
This patent uses the graphic and ideological network (GAT) to design a hybrid multi-unmanned aerial vehicle cluster structure. The GAT input is the observation characteristics of the same dimension after the treatment of each heterogeneous unmanned aerial vehicle obtained in 3.1And outputting the communication network structural characteristics E (t) of the hybrid multi-unmanned aerial vehicle cluster system as the observation O (t) of the whole hybrid multi-unmanned aerial vehicle cluster system after being processed by the graph annotation force network.
At the position ofThe number l and the number x of the unmanned aerial vehicle are obtained l (t) i.e. heterogeneous unmanned aerial vehicle with the number l->Is of (1)
Taking a communication network structure E (t) of the system as an edge (namely network structure characteristics) of the graph, and using a self-attention mechanism to carry out weighted average convergence on the observations of all heterogeneous unmanned aerial vehicles so as to obtain the observations of the whole system.
The following specific implementation process and formula are as follows:
according to the communication network structure E (t) of the system at the moment t (namely the adjacency matrix representing the communication network structure of the system), E in the adjacency matrix E (t) ij (t) as edge of graph attention network, if e ij And (t) is 1, the unmanned aerial vehicle with the number i and the unmanned aerial vehicle with the number j are in communication connection, and otherwise, the unmanned aerial vehicle with the number i and the unmanned aerial vehicle with the number j are 0. Specially, e ii (t) is 1.
The attention weight between each unmanned aerial vehicle in the system is then calculated, and the formula is:
wherein N (t) is the number of non-isolated unmanned aerial vehicles in the cluster structure at the moment t. W is the matrix of shared weights and,for the feedforward neural network, performing dot product with the spliced feature vector to obtain a real number, and finally normalizing by a LeakyReLU activation function to obtain the attention weight alpha at the moment t ij (t) represents the relative importance of the heterogeneous drone numbered j to the heterogeneous drone numbered i.
And calculating a new observation feature vector according to the attention weight, wherein the formula is as follows:
m is the number of heterogeneous unmanned aerial vehicles in the system. Weighted summation is carried out to obtain a new observation feature vector X of the heterogeneous unmanned aerial vehicle with the number of i i And (t), finally splicing the observation characteristics of all unmanned aerial vehicles after being processed by the self-attention mechanism to obtain the observation X (t) of the system
X(t)＝(X 1 (t)||X 2 (t)||…||X i (t)||…||X m (t))
Where m is the total number of drones in the system. X (t) is processed by a full-connection layer, and final observation characteristics O (t) of the hybrid multi-unmanned aerial vehicle cluster system are obtained through aggregation
3.3 model training
And training a reinforcement learning model by using an Actor-Critic architecture, regarding the hybrid multi-unmanned aerial vehicle cluster system as an intelligent agent, deploying a trained Actor strategy network for the hybrid multi-unmanned aerial vehicle cluster system, wherein the Actor network gives actions to be taken by the system at the current moment, namely a cluster structure adjustment strategy. Training is performed using a Critic value network, and a value estimate of the action is calculated.
Acotr policy network
The Actor network is used for selecting actions according to the current system state O (t)Outputting a policy->The parameter theta of the strategy can optimize the performance of the strategy through a gradient ascending algorithm, and a better cluster structure adjustment strategy is given.
Critic value network
Strategy for systematically observing O (t) and Acotr network output according to hybrid multi-unmanned aerial vehicle cluster output by neural networkCritic network calculates the value function of the current state, outputs the value function of the current state, and takes +.>Is to be used for reporting the expectations.
Policy gradient update
Training by adopting a PPO algorithm, and giving a strategy gradient formula:
wherein θ is the policySlightly parameters, T represents the total running time of the system, r t And (theta) setting epsilon limit probability ratio change for the probability ratio of the updated strategy after training to the old strategy, and ensuring that the strategy change is more gentle.And calculating a reward function and a value function which are given out during the construction of the decision model as the advantage function.
After model training is completed, the method can give out an optimal cluster structure adjustment strategy according to different circumstances, and balance the communication cost of the system, the state of the heterogeneous unmanned aerial vehicle and the stability of the network structure, so that the best optimization effect is achieved.
3.4 Cluster Structure adjustment and network reconstruction
And adjusting a strategy action (t) according to a cluster structure given by the trained model, and carrying out weighted summation on six factors of each unmanned aerial vehicle to obtain a utility value of each unmanned aerial vehicle. The detailed design of these six factors is given below:
percentage of remaining energy
Unmanned observation x from 3.1 i Extracting the remaining energy attribute in (t), divided by x i (0) Is the energy of the initial state.
Degree of drone in network architecture
The degree of unmanned aerial vehicle divided by the total number of other unmanned aerial vehicles in the system
Average distance between unmanned aerial vehicle and neighbor unmanned aerial vehicle (negative utility)
N i When (t) is tNumber set of neighbor unmanned aerial vehicle, d ij For the distance between two unmanned aerial vehicles, observing x by unmanned aerial vehicle i (t) and x j And (3) calculating the position coordinates of the unmanned aerial vehicle in (t).
Average motion state similarity of unmanned aerial vehicle and neighbor unmanned aerial vehicle
N i (t) is the number set of the neighbor unmanned aerial vehicle at the moment t, and speed is the unmanned aerial vehicle observation x obtained in 3.1 i (t) and x j The speed attribute of the unmanned aerial vehicle converged in (t), theta is the direction in the state characteristic in observation, V ij (t) is the degree of similarity of the motion states between the two unmanned aerial vehicles,for the average value of the similarity degree of the motion states of the neighbor unmanned aerial vehicle, solving the variance to obtain the average motion state similarity factor 4 。
Average hold time of communication connection
N i (t) is the number set of the neighbor unmanned aerial vehicle at the moment t, wherein ct ij And (t) is the time for maintaining the communication connection between the two unmanned aerial vehicles at the moment t.
Average communication costs of drone and neighbor drone
N i (t) is the number set of the neighbor unmanned aerial vehicle at the moment t, cost ij And (t) is a calculation method for communication cost in the perception process, the corresponding unmanned aerial vehicle is found through the numbers i and j of the heterogeneous unmanned aerial vehicles, and the communication cost between the two unmanned aerial vehicles is calculated according to the type and the distance of the heterogeneous unmanned aerial vehicles, so that the multidimensional isomerism of the communication capability is reflected.
Clustering is carried out according to utility values of unmanned aerial vehicles in a communication range by using a clustering algorithm, CH communication unmanned aerial vehicles are selected, and each cluster is divided. And then using a routing algorithm to construct a communication network in the cluster through the communication unmanned aerial vehicle, and sending out a signal with a cluster number by all unmanned aerial vehicles in the system and then receiving signals sent out by other unmanned aerial vehicles. From the received signals, the unmanned aerial vehicle knows the clusters where the signal sources are located, and for unmanned aerial vehicles which can receive unmanned aerial vehicle signals from other clusters, the unmanned aerial vehicles are used as candidates for cross-cluster communication connection, and finally the candidate with the lowest communication cost is selected as the cross-cluster communication connection between the two clusters and is constructed. Thus, the whole cluster structure adjustment process is completed.
Claims (8)
1. A self-adaptive adjustment method for a cluster structure of a hybrid multi-unmanned aerial vehicle is characterized in that a state encoder aiming at heterogeneous unmanned aerial vehicles is provided, unmanned aerial vehicles of different types are processed in a distinguishing mode, attributes and capabilities of different dimensions among the heterogeneous unmanned aerial vehicles under a converged hybrid situation are obtained, and characteristics of the heterogeneous unmanned aerial vehicles with the same dimensions are obtained; and then merging the processed multidimensional heterogeneous data by using a clustering algorithm based on reinforcement learning, and giving a global adjustment strategy to adjust the cluster structure and reconstruct the communication network.
2. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicles according to claim 1, wherein the adaptive adjustment method comprises the following steps: adopt unmanned aerial vehicle to distribute even many unmanned aerial vehicle cluster system of mixing, kth cluster C k Any unmanned aerial vehicle in (a)Represented by a triplet:Wherein->Represents the kth cluster heterogeneous unmanned plane +.>Is a type of unmanned aerial vehicle;Represents heterogeneous unmanned plane->Attributes and capability features of (a);Representing heterogeneous unmanned plane->Status features of (2);
hybrid multi-unmanned aerial vehicle cluster system is composed of a plurality of heterogeneous unmanned aerial vehicle clusters C k Constitute, each unmanned aerial vehicle cluster C k From heterogeneous unmanned aerial vehicleComposition (S)/(S)>Wherein N is k The number of unmanned aerial vehicles in the cluster; two connection relations exist between heterogeneous unmanned aerial vehicles in the system, wherein one is communication connection inside a cluster, and the other is communication connection crossing the cluster; communication connection between heterogeneous unmanned aerial vehicles forms a hybrid network structure in the system;both connections are dynamic and will change as cluster structure adjusts; intra-cluster connection and inter-cluster connection in a comprehensive heterogeneous multi-unmanned aerial vehicle system, a hybrid multi-unmanned aerial vehicle cluster system MAS with N clusters is ultimately described as mas= (N) 1 ,n 2 ,…,n N E), where n k ＝<C k ,CH k >Representing the complete perception of the kth cluster, C k Heterogeneous unmanned aerial vehicle set being kth cluster, CH k Representing the current communication unmanned aerial vehicle of the kth cluster, wherein E is an adjacency matrix representing the network structure in the hybrid multi-unmanned aerial vehicle cluster system; use of the binary group->Representing a communication connection in the system; all communication connections are constructed and the communication costs spent maintaining the connection at each subsequent time are expressed asRepresenting cluster C k Heterogeneous unmanned aerial vehicle->And cluster C l Heterogeneous unmanned aerial vehicle->Is a communication cost of (a);function c will be according to unmanned plane +.>And->The type of (c) and the distance between them calculates the communication cost.
3. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicles according to claim 2, wherein the adaptive adjustment method comprises the following steps:
the state S (t) represents the cluster structure state observation of the whole system at the time interval t, including the state, the distribution condition, the communication network structure and the CH communication unmanned aerial vehicle of the heterogeneous unmanned aerial vehicle in each cluster, and is represented as:
S(t)＝(S 1 (t),S 2 (t),…,S k (t),…,S N(t) (t),E(t))
wherein N (t) is the number of heterogeneous unmanned aerial vehicle clusters in the t moment system, E (t) is the communication network structure of the t moment system, S k (t) is the heterogeneous drone in the kth clusterObservation +.>Is a set of (3):
actionThe cluster structure adjustment strategy adopted by the system at the time t is expressed as:
wherein lambda is i (i∈[1,6]) Utility weight, lambda, representing the ith factor of the drone i The following constraints are satisfied;
λ 1 +λ 2 +λ 3 +λ 4 +λ 5 +λ 6 ＝1
rewarding r (t) to act when the system is in state S (t) at time tGlobal rewards obtained by making cluster structure adjustments are expressed as:
lambda is the weight, r s Awarding for cluster structure stability, r c Negative rewards for energy consumption;
r s ＝α 1 r ATT +α 2 r CTD +α 3 r SMD +α 4 r CTC
r c ＝-(β 1 r build +β 2 r maintance +β 3 r isolated +β 4 r fly )
wherein alpha and beta are weights, r ATT Representing an average time that the drone remains within the current cluster; r is (r) CTD The compactness of the network structure among unmanned aerial vehicles in each cluster is represented; r is (r) SMD Representing the similarity of the motion states of unmanned aerial vehicles in each cluster; r is (r) CTC Representing the average time for maintaining the communication connection between the unmanned aerial vehicles in each cluster; r is (r) build Representing communication costs of new communication connection construction caused by cluster structure adjustment; r is (r) maintance Representing the maintenance costs spent by the communication network in the system at the previous time to maintain to that time; r is (r) isolated Representing the additional communication loss caused by isolated clusters in all clusters; r is (r) fly Representing flight energy consumption of all unmanned aerial vehicles in the system at the current moment;
the strategy of each cluster should meet the optimization condition, i.e. the strategy of the system should maximize its reward function r (t), according to the definition of the markov decision, by the formula:
wherein the method comprises the steps ofThe action space that can be taken for the system is the set of all possible utility weights that can be taken.
4. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicles according to claim 1, wherein the adaptive adjustment method comprises the following steps: using GRU to design a state encoder aiming at heterogeneous unmanned aerial vehicle type, wherein the input of the state encoder is the heterogeneous unmanned aerial vehicle typeAttribute and capability features of different dimensions of a heterogeneous unmanned aerial vehicle +.>Feature vector of unified dimension is output>Unmanned plane with input recording function>Type number->Unmanned aerial vehicle stored in GRU for type +.>History state of->Is a vector for representing the unmanned plane +.>Historical features of the last moment; in the process->When in use, will->Adding to the input data;
reset gate r in GRU type Controlling whether to ignore hidden status of GRUThe formula is as follows:
wherein type is unmanned aerial vehicleUnmanned aerial vehicle type number, +.>And->Input feature and hidden state pair r in GRU network for processing type unmanned plane state type Weight matrix of>Is a bias vector, σ is a sigmoid function;
update gate u in GRU type For controlling GRU update hidden statusThe degree of (2) is as follows:
wherein the method comprises the steps ofAnd->Is a pair u of input features and hidden states in a GRU network that handles type unmanned plane states type Weight matrix of>Is the bias vector and σ is the sigmoid function.
5. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicle according to claim 4, wherein: according to r type And u type Calculating candidate features at the current time
Wherein the method comprises the steps ofEach element representing a reset gate vector and a hidden state vector is multiplied, +.>And->Is the input and hidden state pair +.>Weight matrix of>Is a bias vector, tanh represents a hyperbolic tangent function, mapping the input value to the interval [ -1,1]An inner part; the processed candidate feature is treated by a fully connected layer>Mapping to a unified dimension:
where H is a nonlinear activation function,is a dimension mapping matrix special for the unmanned aerial vehicle, is a weight matrix of Z rows and D columns, wherein Z is +.>D is->Dimension b of (b) z Is a bias vector; obtain the heterogeneous unmanned aerial vehicle->Feature vector processed by state encoder of specific type +.>Unmanned plane type->Feature vectors of the same dimension after processingStatus feature of unmanned aerial vehicle>Splicing to obtain the heterogeneous unmanned aerial vehicle +.>Observation features of the same dimension->
6. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicle according to claim 5, wherein the adaptive adjustment method comprises the following steps: according to the communication network structure E (t) of the system at the moment t, namely an adjacent matrix representing the communication network structure of the system, E in the adjacent matrix E (t) ij (t) as an edge of the graph attention network, calculating the attention weight among the unmanned aerial vehicles in the system, wherein the formula is as follows:
wherein N (t) is the number of non-isolated unmanned aerial vehicles in the cluster structure at the moment t, W is a shared weight matrix,for the feedforward neural network, performing dot product with the spliced feature vector to obtain a real number, and finally normalizing by a LeakyReLU activation function to obtain a t-timeAttention weight alpha of the score ij (t) representing the relative importance of the heterogeneous drone numbered j to the heterogeneous drone numbered i;
and calculating a new observation feature vector according to the attention weight, wherein the formula is as follows:
m is the number of heterogeneous unmanned aerial vehicles in the system; weighted summation is carried out to obtain a new observation feature vector X of the heterogeneous unmanned aerial vehicle with the number of i i And (t), finally splicing the observation characteristics of all unmanned aerial vehicles after being processed by the self-attention mechanism to obtain the observation X (t) of the system
X(t)＝(X 1 (t)||X 2 (t)||…||X i (t)||…||X m (t))
And (3) processing the X (t) through a full-connection layer, and converging to obtain the final observation feature O (t) of the hybrid multi-unmanned aerial vehicle cluster system.
7. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicles according to claim 6, wherein the adaptive adjustment method comprises the following steps: training a reinforcement learning model by using an Actor-Critic architecture, regarding the hybrid multi-unmanned aerial vehicle cluster system as an intelligent agent, deploying a trained Actor strategy network for the hybrid multi-unmanned aerial vehicle cluster system, wherein the Actor network is used for selecting actions according to the current system state O (t)Outputting a policy->The parameter theta of the strategy optimizes the performance of the strategy through a gradient ascending algorithm, and gives a better cluster structure adjustment strategy; strategy for systematically observing O (t) and Acotr network output according to hybrid multi-unmanned aerial vehicle cluster output by neural network>Critic netCalculating the value function of the current state, outputting the value function of the current state, and taking +.>And then calculate the dominance function update policy gradient and value function.
8. The adaptive adjustment method for the cluster structure of the hybrid multiple unmanned aerial vehicle according to claim 7, wherein: based on actions taken by the systemNamely, the utility weight of the cluster structure is adjusted, and the utility is calculated by weighting and summing each factor of the unmanned aerial vehicle; the utility factors of unmanned aerial vehicles have six kinds: the method comprises the steps of remaining energy percentage, the degree of the unmanned aerial vehicle in a network structure, the average distance between the unmanned aerial vehicle and the other unmanned aerial vehicle, the average motion state similarity of the unmanned aerial vehicle and the neighbor unmanned aerial vehicle, the average communication connection maintaining time and the average communication cost of the unmanned aerial vehicle and the neighbor unmanned aerial vehicle; clustering is carried out according to utility values of unmanned aerial vehicles in a communication range by using a clustering algorithm, CH communication unmanned aerial vehicles are selected, each cluster is divided, then a communication network in the cluster is built through the communication unmanned aerial vehicles CH by using a routing algorithm, and the cluster structure adjustment process is completed. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310640306.2A CN116520887A (en) | 2023-05-31 | 2023-05-31 | Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310640306.2A CN116520887A (en) | 2023-05-31 | 2023-05-31 | Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116520887A true CN116520887A (en) | 2023-08-01 |
Family
ID=87396016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310640306.2A Pending CN116520887A (en) | 2023-05-31 | 2023-05-31 | Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116520887A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117215331A (en) * | 2023-10-19 | 2023-12-12 | 浙江大学 | Decentralized heterogeneous unmanned aerial vehicle cluster motion planning method |
-
2023
- 2023-05-31 CN CN202310640306.2A patent/CN116520887A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117215331A (en) * | 2023-10-19 | 2023-12-12 | 浙江大学 | Decentralized heterogeneous unmanned aerial vehicle cluster motion planning method |
| CN117215331B (en) * | 2023-10-19 | 2024-09-13 | 浙江大学 | Decentralized heterogeneous unmanned aerial vehicle cluster motion planning method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111552314B (en) | Self-adaptive formation tracking control method for multiple unmanned aerial vehicles | |
| Ge et al. | Multi-agent transfer reinforcement learning with multi-view encoder for adaptive traffic signal control | |
| CN113741533A (en) | Unmanned aerial vehicle intelligent decision-making system based on simulation learning and reinforcement learning | |
| CN114489127B (en) | Unmanned plane cluster-oriented self-organizing formation method | |
| CN114815882B (en) | Unmanned aerial vehicle autonomous formation intelligent control method based on reinforcement learning | |
| CN112770256B (en) | Node track prediction method in unmanned aerial vehicle self-organizing network | |
| CN112464004A (en) | Multi-view depth generation image clustering method | |
| CN116520887A (en) | Adaptive adjustment method for cluster structure of hybrid multiple unmanned aerial vehicles | |
| CN115099133B (en) | TLMPA-BP-based cluster system reliability assessment method | |
| CN104809502A (en) | Dynamic adjusting method for nodes in hidden layers of restricted Boltzmann machines | |
| Xiao et al. | Graph attention mechanism based reinforcement learning for multi-agent flocking control in communication-restricted environment | |
| CN106647272A (en) | Robot route planning method by employing improved convolutional neural network based on K mean value | |
| Zhang et al. | Adaptive differential evolution-based distributed model predictive control for multi-UAV formation flight | |
| CN116582442A (en) | Multi-agent cooperation method based on hierarchical communication mechanism | |
| CN116108383A (en) | Ship track prediction method based on improved goblet sea squirt multi-output support vector | |
| CN115563490A (en) | Super-large-scale low-orbit satellite fault diagnosis method based on improved locust optimization algorithm | |
| CN113592079B (en) | Collaborative multi-agent communication method oriented to large-scale task space | |
| Yang et al. | FedDD: Federated double distillation in IoV | |
| Quan et al. | A Multi-Dimensional Dynamic Evaluation Method for the Intelligence of Unmanned Aerial Vehicle Swarm | |
| CN114859719A (en) | Graph neural network-based reinforcement learning cluster bee-congestion control method | |
| CN114564681A (en) | Method for analyzing inorganic salt content data of forest soil | |
| CN114997306A (en) | Target intention identification method based on dynamic Bayesian network | |
| Er et al. | A novel framework for automatic generation of fuzzy neural networks | |
| Wu et al. | An Attention Mechanism and Adaptive Accuracy Triple-Dependent MADDPG Formation Control Method for Hybrid UAVs | |
| Wang et al. | Multistrategy integrated marine predator algorithm applied to 3D surface WSN coverage optimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |