Multi-uav dynamic task allocation method based on proximity information correlation in emergency scenario

Abstract

The application discloses a multi-unmanned aerial vehicle dynamic task allocation method based on adjacent information correlation in an emergency scene, constructs a multi-unmanned aerial vehicle system model in the emergency scene; constructs a dynamic task model; constructs a multi-unmanned aerial vehicle dynamic task allocation problem as a maximum multi-unmanned aerial vehicle task coverage optimization problem and proposes a maximum task coverage objective function; represents the maximum unmanned aerial vehicle task coverage as a partially observable Markov decision process; utilizes a multi-unmanned aerial vehicle deep deterministic policy gradient algorithm based on adjacent information correlation, exchanges information among different unmanned aerial vehicles, forms a system state, and finds the maximum coverage of task points in the emergency scene within a limited time. The multi-unmanned aerial vehicle deep deterministic policy gradient collaborative method based on adjacent information correlation proposed in the application improves the efficiency of multi-unmanned aerial vehicle collaboration under dynamic tasks and meets the demand of maximizing task coverage of multi-unmanned aerial vehicles in the emergency scene within a limited time.

Description

Technical Field

[0001] This invention relates to a method and apparatus for dynamic task allocation of multiple unmanned aerial vehicles (UAVs) in emergency scenarios, which provides reasonable task planning for multiple UAVs in emergency scenarios. Specifically, in post-disaster emergency rescue, dynamic task allocation is performed for multiple UAVs, enabling them to complete more tasks within a limited time. This invention belongs to the field of multi-UAV task allocation technology. Background Technology

[0002] Multi-UAV task allocation refers to the process of effectively allocating and scheduling each UAV to complete a specific task when multiple UAVs (Unmanned Aerial Vehicles) are working collaboratively. It is an important issue in multiple UAV systems and is widely used in military, search and rescue, agricultural monitoring, environmental protection, and logistics distribution fields.

[0003] In the field of emergency rescue, traditional rescue methods are often hampered by factors such as geographical environment, traffic disruptions, and time constraints after a disaster, resulting in low rescue efficiency or inability to respond quickly. The emergence of drone technology provides a rapid, efficient, and safe solution for emergency rescue. Drones can quickly reach disaster areas to conduct real-time monitoring, search for trapped personnel, and drop relief supplies, thereby greatly improving the speed and accuracy of rescue efforts. For example, one group of drones can be responsible for aerial photography, while another group conducts ground searches, working together to complete a comprehensive disaster area assessment. After an earthquake, drones quickly fly to the disaster area to collect high-resolution images of the affected area, generating 3D topographic maps to help rescuers assess the disaster situation and plan rescue routes. In flooded areas, drones equipped with infrared imaging equipment can penetrate smoke and low-light conditions at night to search for the location of trapped personnel and transmit real-time images back to the command center to guide rescue operations. At the scene of large-scale fires such as forest fires, drones conduct real-time monitoring to obtain information on the spread of the fire and assist the command center in formulating firefighting strategies. For the coordinated task allocation of multiple drones, it can ensure the efficient execution of different tasks within the disaster area through reasonable scheduling and collaboration, avoiding conflicts between drones and maximizing resource utilization. Through intelligent task allocation, multiple drones can work together in complex disaster scenarios to complete more complex tasks. Therefore, by rationally planning the task allocation of multiple drones, we can collect more information in emergency situations and improve rescue efficiency.

[0004] In emergency scenarios, the uncertainty of multi-drone task allocation leads to numerous new tasks arising during drone flights, increasing the difficulty of allocation. Furthermore, the sheer number of tasks means that the sheer number of drones cannot fully cover all of them. However, existing research primarily focuses on multi-drone task allocation under static conditions, assuming drones can cover all tasks. Most studies prioritize minimizing task completion time, which doesn't align with the limitations of multi-drone mission coverage in emergency scenarios. Summary of the Invention

[0005] Purpose of the Invention: To address the problems existing in the prior art, it is necessary to consider the challenges posed by the dynamic increase in tasks in emergency scenarios to the allocation of multiple UAV tasks. To address the increased computational complexity resulting from the large-scale increase in the number of tasks, this invention proposes a method and apparatus for dynamic multi-UAV task allocation in emergency scenarios based on proximity information association, solving the problems of increased dynamic tasks and computational complexity. Considering that UAV swarms cannot completely cover all tasks in emergency scenarios, maximizing UAV task coverage is taken as the optimization objective. This problem is represented as a partially observable Markov decision process, and a dynamic multi-UAV task allocation scheme based on proximity information association is proposed for emergency scenarios, enabling the UAV swarm to maximize task coverage and ensure maximum understanding of the latest situation in the disaster area.

[0006] Technical Solution: A method for dynamic task allocation of multiple unmanned aerial vehicles (UAVs) in emergency scenarios based on proximity information association, for multi-UAV systems in emergency scenarios, including the following steps:

[0007] (1) Construct a multi-UAV system model for emergency scenarios;

[0008] (2) Construct a dynamic task model;

[0009] (3) The problem of dynamic task allocation of multiple UAVs is transformed into an optimization problem of maximizing the task coverage of multiple UAVs, and an objective function for maximizing task coverage is constructed.

[0010] (4) Maximizing UAV mission coverage is represented as a partially observable Markov decision process;

[0011] (5) Treat multiple drones as different drones, and use the multi-drone deep deterministic strategy gradient algorithm based on proximity information association. Through information exchange between different drones (including current drone location information, current unfinished task information and newly added dynamic task information), and the drones' own current battery status and location, a system state is formed to find the task allocation scheme with the maximum coverage of multi-drone task points in the emergency scenario within a limited time.

[0012] (6) Randomly initialize the parameters of the Q network and the policy network, and use the system state information as the input for each UAV;

[0013] (7) During training, each drone receives its current reward r and its state o′ at the next moment based on the action it takes. Then, the drone obtains its current set of neighboring drones G = (G1, G2, ..., G...). N ), calculate the association information with each neighboring UAV, associate this neighbor information through the NIR module and generate a unified input vector neighbor information state φ(o)=(φ1(o)). k ),φ2(o k ),...,φ N (o k )) k∈G And the adjacent UAV actions φ(a)=(φ1(a) k ),φ2(a k ),...,φ N (a k )) k∈G ;

[0014] (8) The current state, action, neighbor association information data and reward (o,a,φ(o),φ(a),r,o′) are stored in the experience replay buffer. During each training session, a small batch of data is randomly drawn from the experience replay buffer to update the policy network and value network.

[0015] (9) Update the target network parameters;

[0016] (10) Repeat steps (6)-(9) until the iteration process ends, and find the task allocation scheme with the highest task coverage.

[0017] Furthermore, in step (1), a multi-UAV system model is constructed for an emergency scenario, comprising a UAV ensemble. There are N drones in total, which together form a multi-drone system, represented by a graph structure. Let ε represent the set of communication edges between multiple UAVs, representing direct communication and sensing between UAVs. The energy consumption of communication between UAVs is negligible.

[0018] The neighbor set of drone i is defined as follows: The neighbor set is dynamic and updates as the drone moves or the environment changes.

[0019] The Neighbor of Drone i Due to its communication radius R com Decision, satisfaction: in It is the Euclidean distance between drone i and drone i′ at time t. This represents the position of drone i at time t.

[0020] In step (2), a dynamic task model is constructed, and the task set is defined as follows: Task Collection It increases dynamically over time; the task j at the current time point t uses j. t This indicates that each task It includes the task coordinates, task priority, and task status.

[0021] in This indicates the state of the task at the current time t, pos = (x j ,y j The coordinates of the task are indicated by , and prio ∈ [0, 1] represents the priority of the task, where 0 represents a normal task and 1 represents an urgent task.

[0022] The new task is discovered by the drone during flight. Drone i discovers the new task j at time t. new It can share new mission information with neighboring drones within communication range.

[0023] Step (3) transforms the multi-UAV dynamic task allocation problem into a multi-UAV task coverage optimization problem and proposes a task coverage maximization objective function. In emergency rescue scenarios, multi-UAV task allocation involves M tasks, denoted as a set. The drones are arbitrarily distributed across known locations in the emergency scenario. Assume that the drones and tasks under discussion are of the same type, and that each task can be completed with only one drone. Based on these conditions, the problem of assigning M tasks to N drones can be expressed as the following optimization formula:

[0024]

[0025]

[0026] Where, ξ ij ξ represents the completion status of the corresponding task by the drone. If drone i completes task j, then ξ ij =1, otherwise ξ ij =0. Formula (1a) ensures that the number of tasks performed by the UAV does not exceed the maximum number of tasks to be performed. Formula (1b) ensures that each task is either not completed or is completed by only one UAV. Formula (1c) is the maximum flight time constraint, which means that the time taken for the UAV to perform all tasks cannot exceed the UAV's own maximum flight time. Formula (1d) is the safe distance constraint that must be met between UAVs. Formula (1e) is the constraint that ensures that the UAV cannot fly over the obstacle area.

[0027] In step (4), maximizing UAV mission coverage is represented as a partially observable Markov decision process, where the state space, action space, and reward function are specifically represented as follows:

[0028] A. State Space: The state space observed by the UAV at time t should contain the current position information of the UAV. Battery Information Nearby drone cluster The nearest drone set includes a set of locations. Battery Task Information and the current list of tasks. Therefore, the state space of the UAV at time t is represented as: The state space set of the drone is

[0029] B. Action Space: The actions performed by the drone at time t It should include the mission point, flight direction, and flight speed, as shown in the formula:

[0030] C. Reward Function: The goal of the reward function is to encourage drone swarms to improve mission completion efficiency, avoid collisions, dodge obstacles, and reduce mission completion time. Therefore, reward functions can be divided into mission completion rewards, collision rewards, and maximum mission completion time rewards.

[0031] The reward for completing the task is Including mission point rewards R close,i (t), and and Differentiate rewards for regular missions and emergency missions;

[0032] Collision rewards include collision rewards between drones and obstacle collision rewards, respectively and Among them, R collide,ij It's a collision reward between drones, R obstacle,i It is an obstacle collision reward between the drone and the obstacle. The distance between the drone and the obstacle is represented by λ1 and λ2, which are collision penalty coefficients. collide ,d obstacle It is the obstacle collision threshold.

[0033] Maximum task completion time reward: Among them, T minκ1 is the preset minimum time, κ2 is a constant that controls the strength of the effect of time on the reward and determines the rate of reward decay, and κ2 is a constant that controls the decay speed. MaxCompletionTime represents the maximum time to complete all tasks.

[0034] The overall reward function is:

[0035] In step (7), this method constructs a neighborhood information association module. In the neighborhood information association module, each UAV only needs to focus on the set of neighborhood information that is directly related to itself. There is no need to focus on global information. It contains two parts: This represents the set of drones and mission points within the current drone's distance range (i). This represents the set of actions within a set distance for the current drone i. For drone i, the associated proximity information is calculated using the following formula, and observations are aggregated: in It is the observation value of the neighboring related UAV k.

[0036] Action set: in This refers to the actions of neighboring robots k. The reciprocal of the distance between drones is chosen as the weight: Among them, based on the distance between drones Calculate the weighting coefficients Furthermore, a parameter β was added to adjust the weights. This design prioritizes weight allocation based on proximity, making it particularly suitable for emergency scenarios.

[0037] The proximity information association module φ function effectively integrates the collective knowledge and actions of nearby drones. This module not only ensures consistency in the input dimensions but also captures the complexity of environmental dynamics. By integrating information from observations and actions of neighboring drones, φ function allows drones to make more informed decisions under broader environmental conditions, reflecting the real-time dynamics of the environment and the synchronized behavior among surrounding robots.

[0038] The method designs a centralized value function G. The corresponding shared value network G uses the neighboring observations and actions of UAV i. and adjacent related information and As input, the value G of drone i i The formula is as follows:

[0039] Target Value Network G i The formula for ′ is as follows:

[0040] pass To evaluate the expected Q-value, the shared value network G is trained by minimizing the generalized TD error: in, The μ parameter of the shared policy network is optimized by maximizing the expected reward of the drone, and its gradient formula is:

[0041] Using the soft update method θ′ g ←ηθ g +(1-η)θ′ g , θ′ a ←ηθ a +(1-η)θ′ a Update the weights of the target network until training terminates to obtain a task allocation scheme that maximizes task coverage.

[0042] A dynamic task allocation device for multiple unmanned aerial vehicles (UAVs) based on proximity information association for emergency scenarios includes:

[0043] The first module constructs a multi-UAV system model and a dynamic mission model for emergency scenarios.

[0044] The second module constructs the objective function that maximizes task coverage.

[0045] The third module represents maximizing UAV mission coverage as a partially observable Markov decision process;

[0046] The fourth module treats multiple drones as different drones and uses a multi-drone deep deterministic strategy gradient algorithm based on proximity information association. Through information exchange between different drones (including current drone location information, current unfinished task information, and newly added dynamic task information), as well as the drones' own current battery status and location, a system state is formed to find a task allocation scheme that maximizes the coverage of multi-drone task points in an emergency scenario within a limited time.

[0047] The parameters of the Q network and policy network are randomly initialized, and the system state information is used as the input for each UAV.

[0048] The drone will obtain the current set of neighboring drones G = (G1, G2, ..., G...). N ), calculate the association information with each neighboring UAV, associate this neighbor information through the NIR module and generate a unified input vector φ(o) = (φ1(o)). k ),φ2(o k ),...,φ N (o k )) k∈G , φ(a)=(φ1(a) k),φ2(a k ),...,φ N (a k )) k∈G During training, each drone receives a current reward r and a state o′ for the next moment based on the actions it takes.

[0049] Next, the drone will obtain the current set of neighboring drones G = (G1, G2, ..., G...). N ), calculate the association information with each neighboring UAV, associate this neighbor information through the NIR module and generate a unified input vector φ(o) = (φ1(o)). k ),φ2(o k ),...,φ N (o k )) k∈G , φ(a)=(φ1(a) k ),φ2(a k ),...,φ N (a k )) k∈G The current state, actions, neighbor association information, and rewards (o,a,φ(o),φ(a),r,o′) are stored in the experience replay buffer. During each training iteration, a small batch of data is randomly drawn from the experience replay buffer to update the policy network and the value network. The target network parameters are also updated. The process continues until the end of the iteration process, at which point the task allocation scheme with the highest task coverage is found.

[0050] The implementation process and method of the device are the same, and will not be repeated here.

[0051] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the dynamic task allocation method for multiple unmanned aerial vehicles (UAVs) based on proximity information association in emergency scenarios, as described above.

[0052] A computer-readable storage medium storing a computer program that executes the multi-UAV dynamic task allocation method based on proximity information association in emergency scenarios as described above.

[0053] In post-disaster emergency scenarios, the unpredictability of the situation leads to a large number of new tasks, and the number of drones cannot cover all tasks. If traditional methods are still used, it is difficult to meet the requirements of rationality and effectiveness in multi-drone task allocation in emergency scenarios. To address the challenge of dynamic multi-drone task allocation in post-disaster emergency scenarios, this paper proposes a method for dynamic multi-drone task allocation based on proximity information association. First, the problem of maximizing drone task coverage is represented as a partially observable Markov decision process (POMDP). Second, the task coverage of the drone swarm is improved by using a proximity information association module to associate the state and task information of other drones. Finally, the optimal task allocation strategy is solved by an algorithm based on the multi-drone deep deterministic policy gradient (NIR_MADDPG) of proximity information association.

[0054] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0055] This invention addresses a dynamic task allocation method for multiple unmanned aerial vehicles (UAVs) based on proximity information in emergency scenarios. Considering the need for emergency command centers to obtain as much disaster information as possible when a large number of new tasks are added, the invention takes maximizing UAV task coverage as the optimization objective. Representing this problem as a partially observable Markov decision process, the invention proposes a dynamic task allocation method for multiple UAVs based on proximity information in emergency scenarios. This invention has significant advantages in multi-UAV collaborative task allocation, improving task completion rate by approximately 5%-20% compared to traditional methods (Greedy, MAAC, MAPPO). When the task scale increases significantly, NIR_MADDPG (the method of this invention) also significantly outperforms the exponential or higher-order growth trend of baseline methods in terms of task coverage time. This indicates that the method reduces the communication and computational complexity of multi-UAV collaboration through proximity information aggregation, and has the ability to handle high-dimensional, large-scale task scheduling. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the dynamic task allocation process for multiple drones in an emergency scenario, as described in an embodiment of the present invention. Detailed Implementation

[0057] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0058] In post-disaster emergency scenarios, drone swarms initially fly according to a preset task allocation scheme. Upon detecting new tasks, this invention proposes a dynamic task allocation method for multiple drones based on proximity information in emergency scenarios. Figure 1 The process and method implementation includes the following steps:

[0059] (1) Construct a multi-UAV system model for emergency scenarios;

[0060] (2) Construct a dynamic task model;

[0061] (3) The problem of dynamic task allocation of multiple UAVs is transformed into an optimization problem of maximizing the task coverage of multiple UAVs, and an objective function for maximizing task coverage is constructed.

[0062] (4) Maximizing UAV mission coverage is represented as a partially observable Markov decision process;

[0063] (5) Treat multiple drones as different drones, and use the multi-drone deep deterministic strategy gradient algorithm based on proximity information association. Through information exchange between different drones (including current drone location information, current unfinished task information and newly added dynamic task information), and the drones' own current battery status and location, a system state is formed to find the task allocation scheme with the maximum coverage of multi-drone task points in the emergency scenario within a limited time.

[0064] (6) Randomly initialize the parameters of the Q network and the policy network, and use the system state information as the input for each UAV;

[0065] (7) During training, each drone receives its current reward r and its state o′ at the next moment based on the action it takes. Then, the drone obtains its current set of neighboring drones G = (G1, G2, ..., G...). N ), calculate the association information with each neighboring UAV, associate this neighbor information through the NIR module and generate a unified input vector φ(o) = (φ1(o)). k ),φ2(o k ),...,φ N (o k )) k∈G , φ(a)=(φ1(a) k ),φ2(a k ),...,φ N (a k )) k∈G ;

[0066] (8) The current state, action, neighbor association information data and reward (o,a,φ(o),φ(a),r,o′) are stored in the experience replay buffer. During each training session, a small batch of data is randomly drawn from the experience replay buffer to update the policy network and value network.

[0067] (9) Update the target network parameters;

[0068] (10) Repeat steps 6-10 until the iteration process ends, and find the task allocation scheme with the highest task coverage.

[0069] In step (1), a multi-UAV system model is constructed for an emergency scenario, and the UAV set is... There are N drones in total, which together form a multi-drone system, represented by a graph structure. Let ε represent the set of communication edges between multiple UAVs, representing direct communication and sensing between UAVs. The energy consumption of communication between UAVs is negligible in this chapter, as communication energy consumption accounts for only 1% or less of the total energy consumption in high-speed movement scenarios during disaster emergency rescue.

[0070] The neighbor set of drone i is defined as follows: The neighbor set is dynamic and updates as the drone moves or the environment changes.

[0071] The Neighbor of Drone i Due to its communication radius R com Decision, satisfaction: in The Euclidean distance between drone i and drone i′ at time t.

[0072] In step (2), a dynamic task model was constructed, and the task set was defined as follows: Task Collection It increases dynamically over time, for each task. It includes the task coordinates, task priority, and task status.

[0073] in This indicates the state of the task at the current time t, pos = (x j ,y j The coordinates of the task are indicated by , and prio ∈ [0, 1] represents the priority of the task, where 0 represents a normal task and 1 represents an urgent task.

[0074] The new task is discovered by the drone during flight. Drone i discovers the new task j at time t. new It can share new mission information with neighboring drones within communication range.

[0075] Step (3) transforms the multi-UAV dynamic task allocation problem into a multi-UAV task coverage optimization problem and proposes a task coverage maximization objective function. In the emergency rescue scenario, the multi-UAV task allocation involves M tasks, denoted as a set. The drones are arbitrarily distributed across known locations in the emergency scenario. This chapter assumes that the drones and tasks discussed are of the same type, and that each task can be completed with only one drone. Based on these conditions, the problem of assigning M tasks to N drones can be expressed as the following optimization formula:

[0076]

[0077] Where, ξ ij ξ represents the completion status of the corresponding task by the drone. If drone i completes task j, then ξ ij =1, otherwise ξ ij =0. Formula (1a) ensures that the number of tasks performed by the UAV does not exceed the maximum number of tasks to be performed. Formula (1b) ensures that each task is either not completed or is completed by only one UAV. Formula (1c) is the maximum flight time constraint, which means that the time taken for the UAV to perform all tasks cannot exceed the UAV's own maximum flight time. Formula (1d) is the safe distance constraint that must be met between UAVs. Formula (1e) is the constraint that ensures that the UAV cannot fly over the obstacle area.

[0078] In step (4), maximizing UAV mission coverage is represented as a partially observable Markov decision process. The state space, action space, and reward function in the decision process are specifically represented as follows:

[0079] A. State Space: The state space observed by the UAV at time t should contain the current position information of the UAV. Battery Information Nearby drone swarm It contains a set of locations. Battery Task Information and the current list of tasks. Therefore, the state space of the UAV at time t is represented as: The state space set of the drone is

[0080] B. Action Space: The actions performed by the drone at time i. It should include the mission point, flight direction, and flight speed, as shown in the formula:

[0081] C. Reward Function: The goal of the reward function is to encourage the drone swarm to improve task completion efficiency, avoid collisions, dodge obstacles, and reduce task completion time. Therefore, the reward function can be divided into the following parts: Task completion reward. Including mission point rewards R close,i (t) and R normal,i and R emergency,iThe rewards for regular and emergency missions are differentiated, including collision rewards (comprising collision rewards between drones and obstacle collision rewards, respectively). and ) and the maximum task completion time reward (R time Where λ1 and λ2 are collision penalty coefficients, and d collide ,d obstacle It is the obstacle collision threshold.

[0082] Maximum task completion time reward: Among them, T min κ1 is a preset minimum time, which is a constant that controls the intensity of the effect of time on the reward and determines the rate of reward decay. κ2 is a constant that controls the decay rate.

[0083] The overall reward function is:

[0084] In step (7), this method constructs a neighborhood information association module. In the neighborhood information association module, each UAV only needs to focus on the set of neighborhood information that is directly related to itself. There is no need to focus on global information. It contains two parts: It is the set of drones and mission points that are currently closest to drone i; This is the set of actions that are closest to drone i. For drone i, the associated proximity information is calculated using the following formula, and observations are aggregated: in It is the observation value of the neighboring related UAV k.

[0085] Action set: in This refers to the actions of neighboring robots k. This chapter uses the reciprocal of the distance between drones as the weight: Among them, based on the distance between drones Calculate the weighting coefficients Furthermore, a parameter β was added to adjust the weights. This design, which prioritizes weight allocation based on proximity, is particularly suitable for the research scenario in this chapter.

[0086] The φ function, a neighborhood information association module, effectively integrates the collective knowledge and actions of nearby robots. This aggregation function not only ensures consistency in the input dimensions but also captures the complexity of environmental dynamics. By integrating information from observations and actions of neighboring robots, the φ function allows robots to make more informed decisions under broader environmental conditions, reflecting the real-time dynamics of the environment and the synchronized behavior among surrounding robots.

[0087] The method designs a centralized value function G. The corresponding shared value network G uses the neighboring observations and actions of UAV i. and adjacent related information and As input, the value G of drone i i The formula is as follows:

[0088] Target Value Network G i The formula for ′ is as follows:

[0089] pass To evaluate the expected Q-value, the shared value network G is trained by minimizing the generalized TD error: in, The μ parameter of the shared policy network is optimized by maximizing the expected reward of the drone, and its gradient formula is:

[0090] Using the soft update method θ′ g ←ηθ g +(1-η)θ′ g , θ′ a ←ηθ a +(1-η)θ′ a The weights of the target network are updated, and the above training process will continue to loop until training terminates and a task allocation scheme that maximizes task coverage is obtained.

[0091] Obviously, those skilled in the art should understand that the steps of the multi-UAV dynamic task allocation method or the modules of the multi-UAV dynamic task allocation system with proximity information association in emergency scenarios described in the above embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by the computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

Claims

1. A method for dynamic task allocation of multiple unmanned aerial vehicles (UAVs) in emergency scenarios based on proximity information association, characterized in that, A multi-UAV system for emergency scenarios includes the following steps: (1) Construct a multi-UAV system model for emergency scenarios; (2) Construct a dynamic task model; (3) The problem of dynamic task allocation of multiple UAVs is transformed into an optimization problem of maximizing the task coverage of multiple UAVs, and an objective function for maximizing task coverage is constructed; (4) Maximizing UAV mission coverage is represented as a partially observable Markov decision process; (5) Treat multiple drones as different drones, and use the multi-drone deep deterministic strategy gradient algorithm based on proximity information association. Through information exchange between different drones and the drones' own current battery status and location, a system state is formed, and a task allocation scheme with the maximum coverage of multi-drone task points in an emergency scenario within a limited time is found. (6) Random initialization The parameters of the network and policy network are used, and the system status information is used as input for each UAV; (7) During training, each drone receives a reward based on the action it takes. And the state at the next moment; then the drone will obtain the current status of the nearby drone cluster. It calculates the association information with each neighboring UAV, associates this neighbor information through the NIR module, and generates a unified input vector neighbor information state. and nearby drone actions ; (8) Current status, actions, neighboring information, and rewards. Data is stored in the experience replay buffer, and during each training iteration, data is randomly drawn from the experience replay buffer to update the policy network and the value network. (9) Update the target network parameters; (10) Repeat steps (6)-(9) until the iteration process ends, and find the task allocation scheme with the highest task coverage; In step (7), a neighbor information association module is constructed. In the neighbor information association module, each UAV only needs to focus on the set of neighbor information that is directly related to itself. It is not necessary to focus on global information; It contains two parts: Representing current drones A collection of drones and mission points within a defined distance range; Representing current drones A set of actions within a set distance; for drones The associated proximity information is calculated using the following formula: Observation aggregation: ,in It is the observation value of the neighboring related UAV k; Action set: ,in The actions of neighboring robots k are considered; the reciprocal of the distance between drones is selected as the weight. Among them, based on the distance between drones Calculate the weighting coefficients And parameters were added. To adjust the weights.

2. The method for dynamic task allocation of multiple UAVs in emergency scenarios based on proximity information association as described in claim 1, characterized in that, In step (1), a multi-UAV system model is constructed for an emergency scenario, and the UAV set... ,common These drones together form a multi-drone system, using a graphical structure. To indicate; among which This represents the set of communication edges between multiple drones, signifying direct communication and sensing between drones. Indicates drone The nearest neighbor set is defined as The neighbor set is dynamic and updates as the drone moves or the environment changes; drones Neighbors Due to its communication radius Decision, satisfaction: ,in It is a drone and drones exist Euclidean distance over time , This represents the position of drone i at time t.

3. The method for dynamic task allocation of multiple UAVs in emergency scenarios based on proximity information association as described in claim 1, characterized in that, In step (2), a dynamic task model was constructed, and the task set was defined as follows: Task set It increases dynamically over time; the task j at the current time point t uses j. t This indicates that each task Includes task coordinates, task priority, and task status; in Indicates the current time The status of the task. Indicate the task coordinates, This represents the priority level of the task; 0 represents a normal task, and 1 represents an urgent task. The new mission was discovered by the drone during flight. In time Discover new tasks all the time It can share new mission information with neighboring drones within communication range.

4. The method for dynamic task allocation of multiple UAVs in emergency scenarios based on proximity information association as described in claim 1, characterized in that, Step (3) transforms the multi-UAV dynamic task allocation problem into a multi-UAV task coverage optimization problem and proposes a task coverage maximization objective function. Multi-UAV task allocation in emergency rescue scenarios includes... A set of tasks The drones are arbitrarily distributed at various known locations in the emergency scenario; assuming that the drones and tasks discussed are of the same type, and that a task can be completed with only one drone, then... Tasks are assigned to The problem involving a single drone can be expressed by the following optimization formula: （1） （1a） （1b） （1c） （1d） （1d） in, This represents the status of the drone completing the corresponding task. Mission accomplished ,but ,otherwise Formula (1a) ensures that the number of tasks performed by the UAV does not exceed the maximum number of tasks to be performed. Formula (1b) means that each task is either not completed or is completed by only one UAV. Formula (1c) is the maximum flight time constraint, which means that the time for the UAV to perform all tasks cannot exceed the maximum flight time of the UAV itself. Formula (1d) is the safe distance constraint that must be met between UAVs. Formula (1e) is the constraint that ensures that the UAV cannot fly over the obstacle area.

5. The method for dynamic task allocation of multiple UAVs in emergency scenarios based on proximity information association as described in claim 1, characterized in that, In step (4), maximizing UAV mission coverage is represented as a partially observable Markov decision process. The state space, action space, and reward function in the decision process are specifically represented as follows: A. State Space: The state space is in the [missing information]... The system space observed by the drone should include the drone's current location information. Battery information Nearby drone clusters The nearest drone set includes a set of locations. Battery Task Information and the current list of tasks. ,so The state space representation of the drone at time t is as follows The state space set of the UAV is ; B. Action Space: The drone in the... Actions made at all times It should include the mission point, flight direction, and flight speed; C. Reward Function: The goal of the reward function is to encourage drone swarms to improve mission completion efficiency, avoid collisions, avoid obstacles, and reduce mission completion time. The reward function is divided into mission completion reward, collision reward, and maximum mission completion time reward. The reward for completing the task is Including mission point rewards ,as well as and Differentiate rewards for regular and emergency missions; Collision rewards include collision rewards between drones and obstacle collision rewards, respectively and ;in, It's a collision reward between drones. It is an obstacle collision reward between the drone and the obstacle. Indicates the distance between the drone and the obstacle. It is the collision penalty coefficient. It is the obstacle collision threshold; Maximum task completion time reward: ,in, It is the preset shortest time. It is a constant that controls the strength of the effect of time on the reward, and determines the rate at which the reward decays. It is a constant that controls the decay rate, and MaxCompletionTime represents the maximum time to complete all tasks; The overall reward function is: .

6. The method for dynamic task allocation of multiple UAVs in emergency scenarios based on proximity information association as described in claim 1, characterized in that, A centralized value function was designed. The corresponding shared value network With drones Nearest observations and actions and neighboring related information and As input, drones value The formula is as follows: ; Target Value Network The formula is as follows: pass To evaluate Value expectations, shared value network Training is performed by minimizing the generalized TD error: ,in, Shared policy network The parameters are optimized by maximizing the expected reward of the drone, and their gradient formula is: ; Using the soft update method , Update the weights of the target network until training terminates to obtain a task allocation scheme that maximizes task coverage.

7. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the multi-UAV dynamic task allocation method based on proximity information association in emergency scenarios as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that executes the multi-UAV dynamic task allocation method based on proximity information association in emergency scenarios as described in any one of claims 1-6.

Images