Task allocation method and device for multiple unmanned aerial vehicles

Abstract

The invention provides a multi-unmanned aerial vehicle task allocation method and device. The method comprises the steps of determining a mixed integer linear programming problem model of multi-unmanned aerial vehicle task allocation in a wide area search attack scene; solving a mixed integer linear programming problem by using a branch variable selection strategy network to obtain a task executed by each unmanned aerial vehicle; wherein when the branch variable is used for selecting the strategy network, the original strategy or the special strategy is adaptively selected for branching according to the current state based on the adaptive strategy switching framework. According to the multi-unmanned aerial vehicle task allocation method and device provided by the invention, the multi-unmanned aerial vehicle task allocation is converted into a mixed integer linear programming problem, and in the solving process, the original strategy or the special strategy is adaptively selected for branching according to the current state through the adaptive strategy switching framework, so that the multi-unmanned aerial vehicle task allocation is realized. The generalization of unknown task allocation demand modification is improved, and the efficiency of multi-unmanned aerial vehicle task allocation is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multi-UAV task allocation method and device. Background Art

[0002] In a complex battlefield environment, unmanned system clusters face problems such as low autonomy, high uncertainty and poor coordination. In order to improve the combat effectiveness of the system, key technologies such as distributed situational awareness, adaptive intelligent navigation and cooperative control of unmanned systems must be solved. Among them, cooperative control technology needs to solve two key problems: including the design of cooperative control architecture of unmanned systems, formation control problems and cooperative decision-making problems of unmanned systems.

[0003] The problem of coordinated decision-making in formations mainly focuses on two aspects: one is the allocation of combat tasks and resource scheduling on the battlefield. This type of problem belongs to the campaign level. It thinks about the problem from the top level of the battlefield and gives the optimal combat resource allocation plan according to different combat objectives (such as: the shortest time to complete the task, the highest probability of annihilating the enemy, etc.). In solving this type of problem, we can start from the perspective of operations optimization; the second is the problem of impromptu decision-making in the actual combat process. This type of problem belongs to the tactical level, and common scenarios include air games between aircraft formations and the enemy. During the air confrontation between aircraft, decisions that need to be made include: when to launch missiles, in which direction to fly to avoid the opponent's missiles, when to turn on the radar, etc. To solve this type of problem, we can start from the perspectives of game theory and multi-agent reinforcement learning.

[0004] In the prior art, the multi-UAV task allocation problem is usually modeled as a mixed integer linear programming problem (MILP). To solve this type of problem, the widely used method is the branch-and-bound method (B&B). B&B branches the MILP problem into smaller sub-problems and uses a bounding algorithm to prune out the sub-problems that cannot contain the optimal solution. Therefore, when using B&B to solve the MILP problem, the two most basic decisions are the branching variable selection (BVS) and the node selection (Node Selection). These decisions will significantly affect the size of the B&B tree, and the reduction in tree size corresponds to a reduction in solution time.

[0005] Faced with the complex and ever-changing battlefield environment, the constraints of the multi-UAV task allocation problem are likely to change to adapt to the new requirements, which leads to the generation of distribution shift and heterogeneous MILP problems. The generalization performance of the BVS strategy in the existing technology is limited, resulting in technical problems such as long solution time and low efficiency for multi-UAV task allocation. Summary of the invention

[0006] The present invention provides a multi-UAV task allocation method and device, which are used to solve the technical problem of low efficiency of multi-UAV task allocation in the prior art.

[0007] The present invention provides a multi-UAV task allocation method, comprising: Determine the mixed integer linear programming problem model for multi-UAV task allocation in wide-area search and attack scenarios; The branch variable selection strategy network is used to solve the mixed integer linear programming problem and obtain the tasks to be performed by each UAV. Among them, when using the branch variable selection strategy network, the original strategy or special strategy is adaptively selected according to the current state for branching based on the adaptive strategy switching framework.

[0008] In some embodiments, the goal of the mixed integer linear programming problem model is to complete all tasks in the shortest time; The constraints of the mixed integer linear programming problem model include at least one of the following conditions: The logical relationships that need to be satisfied between different tasks; The time required to complete different tasks; The order of precedence between different tasks; The flight time of the drone.

[0009] In some embodiments, the adaptive strategy switching framework includes a domain generalization strategy training module and a discriminator; The domain generalization strategy training module is used to train an original strategy and a special strategy for out-of-distribution data; The discriminator is used to adaptively switch the BVS strategy according to the state in the target domain.

[0010] In some embodiments, the training of a special strategy for out-of-distribution data includes: Generate enhanced data based on adversarial enhancement network; A meta-learning approach is adopted to train ad hoc policies by combining source domain features and enhanced domain features.

[0011] In some embodiments, adaptively switching the BVS strategy according to the state in the target domain includes: Using the discriminator in the adversarial enhancement network to determine the similarity between the features of the current state and the features of the source domain; When the similarity between the feature of the current state and the feature of the source domain is greater than a preset value, the original strategy is used; when the similarity between the feature of the current state and the feature of the source domain is less than or equal to the preset value, the special strategy is used.

[0012] In some embodiments, the tasks to be performed for a specific target are identification, attack, and verification in order; The mission performed by each drone includes at least one of identification, attack and verification.

[0013] The present invention also provides a multi-UAV task allocation device, comprising: A determination module is used to determine a mixed integer linear programming problem model for multi-UAV task allocation in a wide-area search and attack scenario; The allocation module is used to solve the mixed integer linear programming problem using the branch variable selection strategy network to obtain the tasks to be performed by each UAV; Among them, when using the BVS policy network, the original strategy or special strategy is adaptively selected for branching according to the current state based on the adaptive policy switching framework.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a multi-UAV task allocation method as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the multi-UAV task allocation method as described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the multi-UAV task allocation method as described in any one of the above-mentioned methods is implemented.

[0017] The multi-UAV task allocation method and device provided by the present invention convert the multi-UAV task allocation into a mixed integer linear programming problem. During the solution process, an adaptive strategy switching framework is used to adaptively select the original strategy or special strategy for branching according to the current state, thereby improving the generalization of modification in the face of unknown task allocation requirements and improving the efficiency of multi-UAV task allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is one of the flow charts of the multi-UAV task allocation method provided by the present invention.

[0020] Figure 2 This is the second flow chart of the multi-UAV task allocation method provided by the present invention.

[0021] Figure 3 It is a schematic diagram of the adaptive strategy switching framework provided by the present invention.

[0022] Figure 4 It is a schematic diagram comparing the dimensionality reduction results of candidate features and tree features on the benchmark data set provided by the present invention.

[0023] Figure 5 It is a structural schematic diagram of the multi-UAV task allocation device provided by the present invention.

[0024] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] The multi-UAV task allocation problem is an important part of the top-level decision-making of unmanned system formations. In order to solve the multi-UAV task allocation problem, it can be established as a mixed integer linear programming (MILP) problem and solved. For this type of MILP problem, more and more researchers are focusing on using deep learning (DL) to improve the branch and bound method to accelerate the solution, especially through learning branches. However, the effectiveness of deep learning models is based on the assumption that the test set and training set are evenly distributed. Faced with a complex and changing battlefield environment, the task allocation requirements are likely to change, which will lead to distribution shifts and heterogeneous MILP problems. The generalization performance of the BVS strategy in the existing technology is limited, resulting in technical problems such as long time and low efficiency in solving multi-UAV task allocation.

[0026] Based on the above technical problems, the present invention provides a multi-UAV task allocation method and device, which converts the multi-UAV task allocation into a mixed integer linear programming problem. During the solution process, an adaptive strategy switching framework (APS) is used to adaptively select the original strategy or special strategy for branching according to the current state, so as to avoid the problem of decreased solution performance caused by the failure of the BVS strategy when encountering out-of-distribution (OOD) problems in the target domain, thereby improving the generalization of modifications to unknown task allocation requirements and thus improving the efficiency of multi-UAV task allocation.

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Figure 1 : is a flow chart of the multi-UAV task allocation method provided by the present invention, such as Figure 1As shown, the method includes the following: Step 101: Determine a mixed integer linear programming problem model for multi-UAV task allocation in a wide-area search and attack scenario.

[0029] Specifically, in the wide-area search and attack scenario of a drone formation, the drone formation needs to perform search, identification, attack, and verification tasks. Among these four types of tasks, the search task is not a task performed on a specific target. In the search task, the UAV (Unmanned Aerial Vehicle, UAV) will fly along a preset route to cover the specified area. When the UAV receives an order to perform other tasks, it will leave its search path to perform other tasks. After completing other tasks, the UAV will return to the location where it left the search task and continue to perform the search task.

[0030] For a specific target, the tasks to be performed are identification, attack, and verification. Before performing the identification task, it is assumed that we already know the location of the target. We must first perform the identification task on the target to obtain more information about the target; then, we attack the target; finally, in order to determine whether the target is successfully destroyed, we need to perform the verification task. The tasks performed by each drone include at least one of identification, attack, and verification.

[0031] First, define the relevant parameters of the wide area search attack scenario as follows: (1) There are a total of Our UAV and targets, UAV and target are respectively and express.

[0032] (2) These UAVs and targets can be defined nodes, using Indicates. When , it indicates the target node; when , it indicates the UAV source node.

[0033] (3) In addition to this In addition to the nodes, a search node is also defined ,Entering the search node indicates that the UAV has completed all assigned tasks and continues to execute the previously interrupted search task.

[0034] (4) For each enemy target , need to be executed in sequence tasks, use In the wide-area search attack scenario, , .

[0035] When building the model, the decision variables used are shown in Table 1.

[0036] Table 1 Decision variable table When establishing decision variables, some variables will be eliminated in the logical relationship: , (1) When When only the variable exists because As a starting point, only the aircraft Located at this starting point. (2) When , the task to be performed is the recognition task, because the recognition task is the first task, the starting point and goals Cannot be the same, that is ;when At this time, the verification mission needs to be performed. Since the attack and verification missions of the same target need to be performed by different UAVs, ; and for , target identification and attack can be performed by the same UAV, allowing .

[0037] In summary, when When, for Three situations, Variables shared when When the starting point is the UAV source node, it will definitely satisfy , Total So, The number of variables is .

[0038] In some embodiments, the goal of the mixed integer linear programming problem model is to complete all tasks in the shortest time.

[0039] Specifically, in the embodiment of the present invention, the objective function can be established with the goal of completing all tasks in the shortest time, and the expression is as follows: in, Time to complete all tasks, For aircraft Slave Node To target Execute the task The time spent, is the total number of tasks, is the total number of aircraft, The total number of enemy targets.

[0040] In some embodiments, the constraints of the mixed integer linear programming problem model include at least one of the following conditions: The logical relationships that need to be satisfied between different tasks; The time required to complete different tasks; The order of precedence between different tasks; The flight time of the drone.

[0041] Specifically, considering the logical relationship that needs to be satisfied between different tasks in task allocation, the constraints are established as follows: (1) For each enemy target, the three types of tasks only need to be completed once. The constraint conditions are expressed as follows: (2) Each UAV visits the target node at most once from the outside (eliminating the possibility of loops), the constraint expressions are as follows: (3) Each UAV can only enter the sink node once (eliminating the possibility of loops). The constraint condition is expressed as follows: (4) Each UAV leaves the node At most once (eliminating the possibility of loops), the constraint is expressed as follows: (5) If a UAV is assigned to verify a target, it cannot be assigned to attack a target. The constraint is expressed as follows: (6) If the UAV enters the target To perform task 3, the UAV must leave the node , the constraint expression is as follows: (7) If the UAV enters the target To execute task 1, the UAV must leave or attack the node. , the constraint expression is as follows: (8) If the UAV enters the target Execute Task 2, then the UAV cannot leave the node anymore , the constraint expression is as follows: (9) If the UAV is not assigned to visit a node, then it cannot be assigned to fly out of the node. The constraint is expressed as follows: (10) All UAVs must leave the source node, even if they are flying to the sink node, to continue the search mission. The constraint condition is expressed as follows: (11) Unless the UAV enters the target to classify, otherwise the UAV cannot target As the starting point, also towards the goal To launch an attack, the constraint condition is expressed as follows: In addition, For the maximum endurance of the UAV, considering time and precedence, the following constraints are established: (12) Time constraints (linear), is the maximum endurance time in the UAV, and the constraint condition is expressed as follows: in, .

[0042] (13) For ,Right now right To identify and attack, additional constraints are required. The constraint conditions are expressed as follows: in, .

[0043] (14) The time constraint for executing the first task is expressed as follows: in, .

[0044] (15) Task order, the constraint condition is expressed as follows: in, .

[0045] Step 102: Use the branch variable selection strategy network to solve the mixed integer linear programming problem to obtain the tasks performed by each drone. When using the branch variable selection strategy network, the original strategy or the special strategy is adaptively selected according to the current state based on the adaptive strategy switching framework to branch.

[0046] Specifically, after establishing the MILP problem model, the present invention generates a large number of MILP instances of wide-area search attacks and solves them based on the branch-and-bound method using strong branches as the BVS strategy. In the branch-and-bound tree, the features of the current state and the selected branch variables are matched to establish a data set.

[0047] In some embodiments, the adaptive strategy switching framework includes a domain generalization strategy training module and a discriminator; The domain generalization strategy training module is used to train an original strategy and a special strategy for out-of-distribution data; The discriminator is used to adaptively switch the BVS strategy according to the state in the target domain.

[0048] Specifically, Figure 3 is a schematic diagram of the adaptive strategy switching framework provided by the present invention, such as Figure 3 As shown in Figure 2, the domain generalization strategy training process includes training an original strategy and a special strategy for out-of-distribution (OOD) data.

[0049] In some embodiments, the training of a special strategy for out-of-distribution data includes: Generate enhanced data based on adversarial enhancement network; A meta-learning approach is adopted to train ad hoc policies by combining source domain features and enhanced domain features.

[0050] Specifically, in the state-of-the-art (SOTA) methods for MILP problems such as multi-UAV task allocation, two types of features are used for the BVS strategy: candidate features And tree characteristics ,in represents the number of candidate variables. All features are related to The dynamic process of the tree is related. Candidate features captures the state of the candidate variable, while the tree feature Capture the state of the search tree. From this perspective, all MILP instances can be processed in the same feature space through neural networks.

[0051] However, although all instances share the same feature space, the features of different instances may be significantly different, especially the candidate features. Therefore, OOD data may exist in the test set (i.e., the target domain). The present invention applies principal component analysis (PCA) to reduce the high-dimensional features of the training set and the test set to two dimensions respectively. Figure 4 is a schematic diagram comparing the dimensionality reduction results of candidate features and tree features on the benchmark data set provided by the present invention, such as Figure 4As shown in the figure, it can be seen that from the perspective of candidate features, there are obviously some OOD data in the target domain compared to the source domain, while from the perspective of tree features, the distribution of the source domain can basically cover the target domain. Therefore, in order to further improve the generalization ability of the BVS strategy on OOD data that may exist in the target domain, it is necessary to expand the candidate features of the source domain.

[0052] In order to expand the candidate features of the training domain under the condition of limited source domain data, the present invention generates data that causes increased model loss based on the adversarial enhancement idea, and incorporates these data into the training process of the model.

[0053] In order to expand the candidate feature distribution under the condition of limited source domain data, the generated augmented data should be both diverse and different from the original data, but still keep the same label. Here, it is assumed that the generated candidate features can compensate for those OOD features in the target domain and correspond to the same variable selection as the original candidate features.

[0054] In the data generation step, define the overall objective function as follows: in, and For control and exist The proportion in . is the BVS strategy network parameter, is the candidate feature, is the loss function of the original BVS strategy, which is used to ensure that the additional data will increase the loss of the original strategy. is the hidden state before the pooling layer of the BVS strategy network, For the autoencoder pre-trained on the source domain Parameters.

[0055] Keep the selected variables unchanged by minimizing the change in hidden state after data augmentation, The expression is as follows: in, is the hidden state of the augmented data.

[0056] Enhance candidate features by increasing the reconstruction error of the autoencoder Diversity, The expression is as follows: in, is the enhanced candidate feature, express After autoencoder Candidate features after reconstruction.

[0057] Based on the overall objective function , the augmented data is generated through the following iterative process: in, represents the enhanced candidate features after t+1 iterations, Indicates The enhanced candidate features after iterations, is the learning rate, Indicates The hidden state of the augmented data after iterations.

[0058] At the same time, every time a new enhanced domain is generated, the data of this domain is used to retrain : Specifically, the autoencoder uses an autoencoder that uses Wasserstein distance as a distribution difference metric, referred to as WAE (Wasserstein Auto-Encoder). This WAE includes an encoder and a decoder ,in Indicates input, represents bottleneck embedding, Using NoTree structure, Similar to the reverse NoTree structure. When , we can adopt the training idea of ​​GANs and define a parameter The discriminator defined Assumptions The prior distribution of and sample from , while from Sampling By maximizing the objective function Update the discriminator , The expression is as follows: in, , represents the regularization coefficient; Indicates the number of samples; Indicates input from The sampled Data , the output of the discriminator; Indicates input from The sampled Data , the output of the discriminator. The above formula means that the discriminator Should be from The sampled data is classified as positive, and the encoder The sampled data is classified as negative class. The update is done by minimizing the objective function accomplish, The expression is as follows: in, Indicates the number of samples; express and The cost function of the difference between represents the first samples; Indicates input from The sampled Data, decoder Output: Indicates input from The sampled Data , the output of the discriminator. Its meaning is to let the decoder The output of the encoder is as close as possible to the training set, and the data sampled from the encoder is judged as positive by the discriminator as much as possible. After the pre-training is completed, a WAE can be obtained. and a discriminator , thus calculating .

[0059] In some embodiments, adaptively switching the BVS strategy according to the state in the target domain includes: Using the discriminator in the adversarial enhancement network to determine the similarity between the features of the current state and the features of the source domain; When the similarity between the feature of the current state and the feature of the source domain is greater than a preset value, the original strategy is used; when the similarity between the feature of the current state and the feature of the source domain is less than or equal to the preset value, the special strategy is used.

[0060] Specifically, after data enhancement, a meta-learning method is used to combine source domain features and enhanced domain features To train a special strategy. In order to objectively compare the improvement effect of the proposed method in the verification experiment, the same neural network as the SOTA method TreeGate is used as the domain generalization strategy training network in this step. Meta-training on the augmented domain Meta-testing is performed on the unknown target domain to enable the BVS strategy to achieve better generalization ability on the unknown target domain. The specific steps are as follows: In meta-training, the source domain is used The data is calculated by the parameters The loss function of the defined model , and update it as follows get : in, is the learning rate. The above process obtains the model parameters after one step of gradient descent training on the source domain.

[0061] Then in the meta-test, for each enhanced domain (in , is the number of enhanced domains) This step evaluates the model’s ability to handle the augmented domain.

[0062] Finally, in the meta-update, the loss functions of meta-training and meta-testing are combined and the BVS policy parameters are updated as follows : When actually dealing with unknown instance problems on the target domain, the present invention uses an adaptive strategy switching judgment process to select whether to use the original strategy or the special strategy obtained in the domain generalization strategy training process: that is, the original strategy is used when the characteristics of the current state are similar to the source domain, and the special strategy is used when the characteristics of the current state exhibit OOD characteristics.

[0063] Among them, the discriminator obtained during the WAE training process is used As a metric function. Since the training data of WAE pre-training comes from the source domain, the trained discriminator Data similar to the source domain can be classified as positive. , the present invention can adaptively switch strategies according to the characteristics of the current state.

[0064] Experiments show that APS outperforms the SOTA methods in related technologies and reduces the number of nodes by 18% in the unknown task allocation problem.

[0065] The multi-UAV task allocation method provided by the present invention converts the multi-UAV task allocation into a mixed integer linear programming problem. During the solution process, an adaptive strategy switching framework is used to adaptively select the original strategy or the special strategy for branching according to the current state, thereby improving the generalization of modification in the face of unknown task allocation requirements and improving the efficiency of multi-UAV task allocation.

[0066] The multi-UAV task allocation device provided by the present invention is described below. The multi-UAV task allocation device described below and the multi-UAV task allocation method described above can be referred to each other.

[0067] Figure 5 : is a schematic diagram of the structure of the multi-UAV task allocation device provided by the present invention, such as Figure 5 As shown, the present invention provides a multi-UAV task allocation device, comprising: The determination module 501 is used to determine a mixed integer linear programming problem model for multi-UAV task allocation in a wide area search and attack scenario; The allocation module 502 is used to solve the mixed integer linear programming problem using a branch variable selection strategy network to obtain the tasks to be performed by each drone; Among them, when using the BVS policy network, the original strategy or special strategy is adaptively selected for branching according to the current state based on the adaptive policy switching framework.

[0068] Specifically, the above-mentioned multi-UAV task allocation device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be described in detail here.

[0069] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communications interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the multi-UAV task allocation method, which includes: Determine the mixed integer linear programming problem model for multi-UAV task allocation in wide-area search and attack scenarios; The branch variable selection strategy network is used to solve the mixed integer linear programming problem and obtain the tasks to be performed by each UAV. Among them, when using the branch variable selection strategy network, the original strategy or special strategy is adaptively selected according to the current state for branching based on the adaptive strategy switching framework.

[0070] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0071] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer can execute the multi-UAV task allocation method provided by the above methods, the method comprising: Determine the mixed integer linear programming problem model for multi-UAV task allocation in wide-area search and attack scenarios; The branch variable selection strategy network is used to solve the mixed integer linear programming problem and obtain the tasks to be performed by each UAV. Among them, when using the branch variable selection strategy network, the original strategy or special strategy is adaptively selected according to the current state for branching based on the adaptive strategy switching framework.

[0072] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the multi-UAV task allocation method provided by the above methods, the method comprising: Determine the mixed integer linear programming problem model for multi-UAV task allocation in wide-area search and attack scenarios; The branch variable selection strategy network is used to solve the mixed integer linear programming problem and obtain the tasks to be performed by each UAV. Among them, when using the branch variable selection strategy network, the original strategy or special strategy is adaptively selected according to the current state for branching based on the adaptive strategy switching framework.

[0073] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0074] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-UAV task allocation method, characterized in that: include: Determine the mixed integer linear programming problem model for multi-UAV task allocation in wide-area search and attack scenarios; The branch variable selection strategy network is used to solve the mixed integer linear programming problem and obtain the tasks to be performed by each UAV. Among them, when using the branch variable selection strategy network, the original strategy or special strategy is adaptively selected according to the current state for branching based on the adaptive strategy switching framework.

2. The multi-UAV task allocation method according to claim 1, characterized in that: The goal of the mixed integer linear programming problem model is to complete all tasks in the shortest time; The constraints of the mixed integer linear programming problem model include at least one of the following conditions: The logical relationships that need to be satisfied between different tasks; The time required to complete different tasks; The order of precedence between different tasks; The flight time of the drone.

3. The multi-UAV task allocation method according to claim 1, characterized in that: The adaptive strategy switching framework includes a domain generalization strategy training module and a discriminator; The domain generalization strategy training module is used to train an original strategy and a special strategy for out-of-distribution data; The discriminator is used to adaptively switch the BVS strategy according to the state in the target domain.

4. The multi-UAV task allocation method according to claim 3 is characterized in that: The training strategy for out-of-distribution data includes: Generate enhanced data based on adversarial enhancement network; A meta-learning approach is adopted to train ad hoc policies by combining source domain features and enhanced domain features.

5. The multi-UAV task allocation method according to claim 4, characterized in that: The adaptively switching the BVS strategy according to the state in the target domain includes: Using the discriminator in the adversarial enhancement network to determine the similarity between the features of the current state and the features of the source domain; When the similarity between the feature of the current state and the feature of the source domain is greater than a preset value, the original strategy is used; when the similarity between the feature of the current state and the feature of the source domain is less than or equal to the preset value, the special strategy is used.

6. The multi-UAV task allocation method according to any one of claims 1 to 5, characterized in that: The tasks to be performed for a specific target are identification, attack, and verification; The mission performed by each drone includes at least one of identification, attack and verification.

7. A multi-UAV task allocation device, characterized in that: include: A determination module is used to determine a mixed integer linear programming problem model for multi-UAV task allocation in a wide-area search and attack scenario; The allocation module is used to solve the mixed integer linear programming problem using the branch variable selection strategy network to obtain the tasks to be performed by each UAV; Among them, when using the BVS policy network, the original strategy or special strategy is adaptively selected for branching according to the current state based on the adaptive policy switching framework.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-UAV task allocation method as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-UAV task allocation method as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-UAV task allocation method as described in any one of claims 1 to 6 is implemented.

Images