Multi-unmanned aerial vehicle cooperative loading autonomous adjustment method under sudden fault

Abstract

The invention provides a multi-unmanned aerial vehicle cooperative loading autonomous adjustment method under a sudden fault. The method comprises the following steps: establishing an unmanned aerial vehicle and load physical model according to unmanned aerial vehicle information and load information; according to the unmanned aerial vehicle and load physical model, establishing a multi-unmanned aerial vehicle cooperative hanging system virtual model; carrying out iterative optimization on the virtual model of the multi-unmanned aerial vehicle cooperative suspension system through a depth deterministic strategy gradient network; dispatching the multi-unmanned aerial vehicle cooperative suspension system to execute a task, and controlling the multi-unmanned aerial vehicle cooperative suspension system based on the optimized virtual model of the multi-unmanned aerial vehicle cooperative suspension system; and carrying out real-time fault detection on the multi-unmanned aerial vehicle cooperative suspension system through the fault detection model. Compared with the prior art, the unmanned aerial vehicles in the multi-unmanned aerial vehicle cooperative suspension system are trained, so that each unmanned aerial vehicle realizes the optimal action through local information, and the robustness, the stability and the control precision of the multi-unmanned aerial vehicle cooperative suspension system are effectively improved.

Description

Technical Field

[0001] The present invention is applicable to the technical field of unmanned aerial vehicles, and in particular relates to a method for autonomously adjusting the coordinated loading of multiple unmanned aerial vehicles under sudden failure. Background Art

[0002] In recent years, drones have made some progress in the field of cargo transportation, especially the quadcopter suspension system, which has been proven to be able to complete logistics transportation tasks, including basic tasks such as vertical take-off and landing, hovering, etc., with certain maneuverability and safety. However, due to the low rated load capacity of a single quadcopter, its operating capacity is limited by the limited load capacity of a single quadcopter. The multi-drone collaborative suspension system can well solve the limitation of load capacity.

[0003] The stability of the formation control of the multi-UAV cooperative suspension system has a significant impact on the swing of the load. The instability of a single UAV can easily cause the entire suspension system to collapse. At present, the formation control algorithms of the multi-UAV cooperative suspension system include the pilot-follower method, the behavior-based method, the virtual structure method, the graph theory method, and the consistency-based method. Among them, the pilot-follower method is more widely used. The pilot in the pilot-follower method is usually directly controlled by a person or a computer and is responsible for determining the direction and speed of the entire formation, while the follower autonomously adjusts its position according to the relative position and direction information with the pilot to maintain the formation. However, this method relies on one or several navigators to guide the entire formation. If the navigator fails or is disturbed, the entire formation may lose direction, resulting in the collapse of the formation structure and the crash of the payload. The behavior-based method can provide distributed control and good adaptability, but the complexity of behavior fusion may lead to conflicts in control commands, making it difficult for the formation to maintain consistency. When facing dynamically changing environments or task requirements, behavioral conflicts may lead to reduced coordination among formation members and increase the risk of collision. The virtual structure method requires that the drones maintain a fixed relative position, which may not be practical when passing through narrow spaces or performing complex maneuvers. It lacks flexibility and adaptability when avoiding obstacles. Strict formation constraints will lead to frequent position adjustments when encountering obstacles or other drones, increase energy consumption and may cause actuator saturation, thereby affecting the stability of the formation. The graph theory method may find it difficult to quickly adapt to structural changes in large-scale or dynamically changing formations, resulting in control delays and decision errors. The consistency method relies on information consistency among formation members, but in practical applications, communication delays and information loss may lead to impaired consistency and difficulty in maintaining synchronization among formation members, thus affecting the overall performance and stability of the formation.

[0004] The fault-tolerant control method of UAV formation refers to the technology and method of taking corresponding control strategies to maintain stable flight of the formation in the face of possible individual UAV failures during the UAV formation flight. The existing fault-tolerant control methods of UAV formation mainly include fault-tolerant algorithms based on distributed control, sliding mode control, and model predictive control. In the fault-tolerant algorithm based on distributed control, each UAV only needs to communicate and coordinate with adjacent UAVs without the participation of the global control center. When a UAV fails, other UAVs can adjust the formation through local information and continue to perform the task. However, when this method is applied to the suspension system, if some UAVs fail, the load swing may increase, resulting in increased control complexity; using the sliding mode control algorithm, the control strategy is quickly switched when the UAV fails to maintain the stability of the system, and the fault tolerance is achieved by online estimation of UAV failures and timely adjustment of control inputs. However, this method is prone to jitter when implemented, which affects the control accuracy. At the same time, the load control of the suspension system is weak, and it is difficult to cope with complex load dynamics and coordination of multiple UAVs; the fault-tolerant algorithm based on model predictive control uses the dynamic model of the system for prediction and makes the optimal control decision based on the current system state and future output trajectory. When a drone fails, the task is reallocated by adjusting the model prediction. This method is highly dependent on the system model. When the system model error is large, the control effect will be greatly reduced. Especially when the suspended load changes dramatically, this method has a large amount of calculation and poor real-time performance, which is prone to large delays and affects system stability.

[0005] The main problem of the multi-UAV cooperative suspension system lies in the control of the load swing and swing angle. Once some UAVs fail, the load swing of the suspension system may become difficult to control, resulting in the instability of the overall system. At the same time, the coordination and control difficulty of the multi-UAV cooperative suspension system increases. The failure of a single UAV will break the original formation balance, requiring other UAVs to quickly redistribute tasks. However, in the suspension system, the dynamic response of the load is more complex, which increases the difficulty of task redistribution. In addition, the suspension system has high requirements for real-time control. Any lag or error may cause load instability. The response speed of many fault-tolerant algorithms after the failure of the UAV may not be enough to cope with the dynamic changes in the suspension system. It can be seen that although the multi-UAV cooperative suspension system can increase the upper limit of the load weight, its safety also brings greater challenges. The failure of a single UAV can easily lead to the collapse of the entire suspension system.

[0006] Therefore, a new method for autonomous adjustment of multi-UAV collaborative loading under sudden failures is urgently needed to solve the above technical problems. Summary of the invention

[0007] The present invention provides an autonomous adjustment method for cooperative loading of multiple unmanned aerial vehicles under sudden failure, aiming at improving the robustness, stability and control accuracy of cooperative loading of multiple unmanned aerial vehicles.

[0008] The present invention provides a method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failure, comprising the following steps:

[0009] S1. Obtaining drone information and load information of a multi-drone cooperative suspension system, and establishing a drone and load physical model according to the drone information and load information;

[0010] S2. Establishing a virtual model of a multi-UAV cooperative suspension system according to the UAV and load physical models;

[0011] S3, establishing a deep deterministic policy gradient network, and iteratively optimizing the multi-UAV collaborative hanging system virtual model through the deep deterministic policy gradient network to obtain an optimized multi-UAV collaborative hanging system virtual model;

[0012] S4, dispatching the multi-UAV cooperative hanging system to perform a task, and controlling the multi-UAV cooperative hanging system based on the optimized multi-UAV cooperative hanging system virtual model;

[0013] S5. Establish a fault detection model based on the improved generalized linear model, and perform real-time fault detection on the multi-UAV collaborative suspension system through the fault detection model to determine whether there is a fault in any UAV in the multi-UAV collaborative suspension system. If so, adjust the position, load distribution and flight trajectory of the UAV in the multi-UAV collaborative suspension system according to the fault detection result.

[0014] Preferably, in step S1, the drone information includes drone position attitude change information, output thrust of the drone's motor and attitude adjustment torque information, and the load information includes load center of mass motion information, swing angle information and angular velocity information.

[0015] Preferably, step S2 includes the following sub-steps:

[0016] S21, setting the state space of the drone according to the physical model of the drone and the payload;

[0017] S22. Setting the action space of the drone according to the physical model of the drone and the payload to obtain the virtual model of the multi-drone cooperative suspension system.

[0018] Preferably, in step S21, the state space includes the position information, attitude information, load status, environmental information and fault status information of the UAV.

[0019] Preferably, in step S22, the action space includes thrust vector adjustment information, attitude adjustment information and speed adjustment information of the UAV, the thrust vector adjustment information is to adjust the thrust to change the position and attitude of the UAV, the attitude adjustment information is to adjust the pitch angle, roll angle and yaw angle of the UAV, and the speed adjustment information is to adjust the speed of the UAV.

[0020] Preferably, step S3 includes the following sub-steps:

[0021] S31, setting a reward mechanism for the drone in the deep deterministic policy gradient network;

[0022] S32, establishing a strategy network and a value network for each drone in the multi-drone collaborative suspension system virtual model through the deep deterministic strategy network, wherein the strategy network is used to predict the optimal strategy, and the value network is used to evaluate the action value of the drone, and initialize the experience replay pool;

[0023] S33, training and iterating the virtual model of the multi-UAV cooperative suspension system according to the reward mechanism, and storing the data information of the UAVs in the virtual model of the multi-UAV cooperative suspension system after the training iteration into the experience playback pool; wherein the data information includes the current state, action, reward and next state of the UAVs;

[0024] S34, randomly selecting the data information from the experience replay pool, and updating the parameters of the policy network and the parameters of the value network according to the data information;

[0025] S35. Determine whether the current number of iterations is the maximum number of iterations. If so, output the current multi-UAV collaborative suspension system virtual model as the optimized multi-UAV collaborative suspension system virtual model; if not, return to step S33.

[0026] Preferably, the reward mechanism is set according to the relative distance between the drones, the flight trajectory of the drones, the obstacle avoidance capability of the drones, and the swing angle of the payload carried by the drones.

[0027] Compared with the prior art, the present invention obtains the drone information and load information of the multi-drone cooperative suspension system, establishes the drone and load physical model according to the drone information and load information; establishes the multi-drone cooperative suspension system virtual model according to the drone and load physical model; establishes a deep deterministic policy gradient network, iteratively optimizes the multi-drone cooperative suspension system virtual model through the deep deterministic policy gradient network, and obtains the optimized multi-drone cooperative suspension system virtual model; dispatches the multi-drone cooperative suspension system to perform tasks, and controls the multi-drone cooperative suspension system based on the optimized multi-drone cooperative suspension system virtual model; establishes a fault detection model based on an improved generalized linear model, and performs real-time fault detection on the multi-drone cooperative suspension system through the fault detection model, and determines whether the drone in the multi-drone cooperative suspension system has a fault. If so, the position, load distribution and flight trajectory of the drone in the multi-drone cooperative suspension system are adjusted according to the fault detection result. In this way, the present invention trains the drones in the multi-drone cooperative suspension system so that each drone can achieve the optimal action through local information, and detects faults in real time through the fault detection model, so as to realize the drone formation reorganization and load swing minimization, and effectively increase the robustness, stability and control accuracy of the multi-drone cooperative suspension system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention will be described in detail below in conjunction with the accompanying drawings. The above and other aspects of the present invention will become clearer and easier to understand through the detailed description made in conjunction with the following drawings. In the accompanying drawings:

[0029] Figure 1 It is a flowchart of a method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] Please refer to Figure 1 The present invention provides a method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failure, comprising the following steps:

[0032] S1. Obtain drone information and load information of a multi-drone cooperative suspension system, and establish a drone and load physical model based on the drone information and load information.

[0033] In an embodiment of the present invention, in step S1, the drone information includes the drone position attitude change information, the output thrust of the drone's motor and attitude adjustment torque information, and the load information includes the load's center of mass motion information, swing angle information and angular velocity information.

[0034] Specifically, the physical model of the drone and the load includes a kinematic model and a dynamic model of the drone and the load, wherein the dynamic model of the drone needs to include the change in the position and attitude of the drone and the thrust and attitude adjustment torque generated by the motor, and the dynamic model of the load needs to consider the swing of the load in three-dimensional space, including the movement of the center of mass, the swing angle and the angular velocity. In addition, the tension transmitted by the hanging line and the reaction force on the drone need to be considered. The hanging line can be modeled by a spring damping model to accurately describe the load swing caused by the change in tension. On this basis, the obstacle modeling adopts a rigid body model, and introduces a random distribution model of wind disturbances in the environment, and simulates the influence of wind on the system through the fluid mechanics perturbation equation.

[0035] The multi-UAV cooperative suspension system consists of four quad-rotor drones and a carrier. The carrier is a rigid body. The quad-rotor drone and the carrier are connected by a flexible rope. The flexible rope is suspended at the center point of the bottom of the quad-rotor. The connection points between the flexible rope and the load are symmetrically distributed. The communication architecture of the system is a distributed architecture. The quad-rotor drones communicate directly point-to-point, and each quad-rotor drone makes autonomous decisions based on each other's status. A single quad-rotor drone is equipped with a force sensor, a positioning sensor, an inertial measurement unit, a lidar, and a visual sensor, which are used to measure the tension of the flexible rope on the load, the relative position between drones, the attitude, acceleration, angular velocity of the drone, detect the relative distance between surrounding obstacles and other drones, and identify the surrounding environment characteristics and the status of other drones through the visual system. An inertial measurement unit is also installed on the carrier, which is mainly used to measure the attitude and acceleration of the carrier.

[0036] S2. Establish a virtual model of a multi-UAV collaborative suspension system based on the UAV and payload physical models.

[0037] In the embodiment of the present invention, the physical model of the drone is linked to the physical model of the load. The drone and the load are coupled through the hanging wire, and the force transmission on the hanging wire needs to be modeled. The virtual model of the multi-drone collaborative hanging system needs to control the pulling force applied by each drone to ensure the stability of the load and prevent excessive swing or loss of control.

[0038] The core of the multi-UAV suspension coordination system is the collaborative work of multiple agents, so it is also necessary to model the collaboration of multiple UAVs. The communication between UAVs adopts a distributed control strategy, and each agent shares local information to achieve global goals. In order to deal with communication delays or data loss, the virtual model of the multi-UAV collaborative suspension system adopts a fault-tolerant control mechanism to ensure that the failure of a single UAV will not affect the stability of the entire formation. The collaborative control algorithm is based on a distributed consistency algorithm to achieve UAV tension distribution and load stability adjustment, ensuring that the virtual model of the multi-UAV collaborative suspension system can maintain collaborative consistency when facing complex tasks.

[0039] In this embodiment of the present invention, step S2 includes the following sub-steps:

[0040] S21, setting the state space of the drone according to the physical model of the drone and the payload;

[0041] S22. Setting the action space of the drone according to the physical model of the drone and the payload to obtain the virtual model of the multi-drone cooperative suspension system.

[0042] In an embodiment of the present invention, in step S21, the state space includes the position information, attitude information, load status, environmental information and fault status information of the drone. Specifically, the state space contains local information of each drone and relationship information with other drones, mainly including drone position information, attitude information, load status, environmental information and fault status. The drone position information includes its own spatial coordinates, the relative position between drones, the position of the formation center point and the distance from the target point. The attitude information includes the drone speed, attitude angle (pitch angle, yaw angle, roll angle), and motor output thrust state. The load status includes the center of mass position of the load, the swing angle and the tension of the hanging rope. The environmental information includes the distance from obstacles and the wind disturbance. The fault status includes the content of marking the faulty drone and its fault.

[0043] In an embodiment of the present invention, in step S22, the action space determines the control method of each drone, and the action space includes the thrust vector adjustment information, attitude adjustment information and speed adjustment information of the drone. The thrust vector adjustment information is to adjust the thrust to change the position and attitude of the drone, the attitude adjustment information is to adjust the pitch angle, roll angle and yaw angle of the drone, and the speed adjustment information is to adjust the speed of the drone.

[0044] S3. Establish a deep deterministic policy gradient network, and iteratively optimize the multi-UAV collaborative hanging system virtual model through the deep deterministic policy gradient network to obtain an optimized multi-UAV collaborative hanging system virtual model.

[0045] In the embodiment of the present invention, step S3 includes the following sub-steps:

[0046] S31, setting a reward mechanism for the drone in the deep deterministic policy gradient network;

[0047] In an embodiment of the present invention, the reward mechanism is set according to the relative distance between drones, the flight trajectory of the drones, the obstacle avoidance capability of the drones, and the swing angle of the load carried by the drones. Specifically, it includes minimizing the load swing, quantifying the swing amplitude according to the acceleration of the load and the swing amplitude of the rope; balancing the tension, ensuring that the tension distribution of each drone is uniform; reasonable formation position, avoiding collisions between drones or loss of control of the formation, and avoiding entanglement of different ropes; trajectory tracking effect, checking whether the system flies along the set flight trajectory; obstacle avoidance, when obstacles are detected by laser radar and visual sensors, timely adjusting the flight trajectory to avoid collision. Distributed control is achieved through autonomous learning of drones, and each drone learns the optimal cooperation strategy based on the interaction of local information with other drones. Even in the case of communication delay or local information loss, smoother and more flexible cooperation can be achieved between drones through autonomous learning. Its advantage is that it can dynamically adjust the formation and load distribution, so that it can adapt to various complex task requirements.

[0048] S32, establishing a strategy network and a value network for each drone in the multi-drone collaborative suspension system virtual model through the deep deterministic strategy network, wherein the strategy network is used to predict the optimal strategy, and the value network is used to evaluate the action value of the drone, and initialize the experience replay pool;

[0049] S33, training and iterating the virtual model of the multi-UAV cooperative suspension system according to the reward mechanism, and storing the data information of the UAVs in the virtual model of the multi-UAV cooperative suspension system after the training iteration into the experience playback pool; wherein the data information includes the current state, action, reward and next state of the UAVs;

[0050] S34, randomly selecting the data information from the experience replay pool, and updating the parameters of the policy network and the parameters of the value network according to the data information;

[0051] In the embodiment of the present invention, the TD error is used to minimize the gap between the predicted and actual rewards and update the network. In addition, the policy network is optimized by gradient ascent to find a strategy that can maximize the long-term cumulative reward. An entropy regularization term is added during the policy update to encourage the exploratory nature of the strategy and avoid falling into the local optimal solution. The data in the experience replay pool is selected by the priority replay mechanism to ensure that important state transitions are updated first.

[0052] At the same time, the policy network and value network are updated every few iterations. The training process terminates when the set number of training steps is reached, or the swing angle of the payload and the error of the drone formation reach a certain threshold. All drones learn in parallel during the training process, sharing experience and rewards. Using a shared experience replay pool can improve learning efficiency and promote collaborative behavior among all drones.

[0053] S35. Determine whether the current number of iterations is the maximum number of iterations. If so, output the current multi-UAV collaborative suspension system virtual model as the optimized multi-UAV collaborative suspension system virtual model; if not, return to step S33.

[0054] Specifically, the optimal formation position is found through the learning strategy of the deep deterministic policy gradient network, which can effectively deal with complex nonlinear control problems and UAV formation fault tolerance problems, and can optimize the strategy in high-dimensional continuous action space. The deep deterministic policy gradient network has the following three characteristics: (1) The optimal strategy obtained through learning can give the optimal action only by using local information when applied. (2) It does not need to know the dynamic model of the environment and special communication requirements. (3) The network can be used not only in cooperative environments but also in competitive environments. The framework of the network is centralized training and decentralized execution. Each UAV has an independent policy network and value network to predict the optimal strategy and evaluate the value of the action. The basic framework of reinforcement learning training consists of an actor, a state, and an action. During the training process, the actor first selects an action based on the current state. Then, the critic calculates a Q value based on this state-action. This Q value is an evaluation feedback of the action selected by the actor. The evaluator compares the estimated Q value with the actual Q value to train itself, and the executor adjusts and updates its strategy based on the evaluator's feedback. During training, some additional information can be added to the evaluator stage to obtain a more accurate Q value, such as the status and actions of other drones, that is, each drone evaluates the value of the current action not only based on its own situation, but also considering the behavior of other drones. Decentralized execution means that after each drone has been fully trained, each executor can independently take appropriate actions based on the current status, and no longer needs the status or action information of other drones.

[0055] S4, dispatching the multi-UAV cooperative hanging system to perform a task, and controlling the multi-UAV cooperative hanging system based on the optimized multi-UAV cooperative hanging system virtual model;

[0056] S5. Establish a fault detection model based on the improved generalized linear model, and perform real-time fault detection on the multi-UAV collaborative suspension system through the fault detection model to determine whether there is a fault in any UAV in the multi-UAV collaborative suspension system. If so, adjust the position, load distribution and flight trajectory of the UAV in the multi-UAV collaborative suspension system according to the fault detection result.

[0057] In an embodiment of the present invention, when the fault detection model detects a fault, the faulty UAV in the multi-UAV collaborative suspension system will be marked, the state space will be expanded, the load tension distribution identifier will be added, the weight of the reward mechanism will be adjusted, the UAV formation will be reconstructed and the tension will be redistributed, etc.

[0058] Specifically, when the fault detection model finds that a certain drone fails, the state space of the multi-drone cooperative suspension system will automatically expand, adding fault identification and load distribution change information. The reward function is switched according to the fault state, emphasizing rewards in terms of minimizing load swing, reorganizing the formation, and drone safety. After the fault occurs, the other drones will automatically adjust the formation according to the state of the faulty drone and adopt new strategies to maintain the balance of the load. The multi-drone cooperative suspension system will replan the target position, posture, and tension distribution of each drone. The remaining drones share the load of the faulty drone by adjusting the thrust and position to ensure that the load does not swing violently due to the drone failure. Fault detection based on the fault detection model can timely identify the fault of a single drone in the multi-drone suspension system, and realize the strategy conversion under the fault state through the deep deterministic policy gradient network. Through the expansion of the state space and the adjustment of the reward function, the system can quickly respond to fault conditions, ensure the stability of the drone formation, reduce the swing of the load, and realize the fault-tolerant control of the system.

[0059] When the multi-UAV suspension system is detected to have non-fatal faults such as sensor failure, power system energy supply reduction, slight structural damage and other factors that cause the UAV's movement ability to be limited and its carrying capacity to be reduced, the state space needs to be expanded and adjusted to reflect the impact of the fault on the system. Add an identification variable to the state space to distinguish the faulty UAVs and their impact range. For example, identify a current UAV as a faulty UAV, and mark its control variables as unavailable. In addition, due to the exit of the faulty UAV, other UAVs need to bear additional loads. The state space needs to add an identification of the current load carried by each UAV. Due to UAV failure, the load balance may be broken, and the state space should also include information on the degree of load imbalance, such as the load swing angle change rate, load offset, etc.

[0060] When the multi-UAV suspension system enters a fault state, the reward function needs to be redesigned and different weights should be set for the multi-objective reward function. After the UAV fails, the reward function should focus on encouraging the remaining UAVs to quickly adjust the formation to minimize the load swing, so the reward weight of the load should be increased. The reward function should also encourage the remaining UAVs to take on more loads and complete the formation reorganization in the shortest time. At this time, the pulling force and load of each UAV need to be included in the reward function to encourage even distribution of pulling force. In addition, in the case of a fault, it is key to prevent other UAVs from failing due to overload or lack of coordination. The reward function should punish multiple UAV failures or other UAV overloads. In addition, the reward function should maintain system stability as much as possible. Its reward function should also include fault response time, minimization of load swing, stability of the new formation structure, and avoidance of rope entanglement.

[0061] In the multi-UAV cooperative suspension system, fault diagnosis is crucial. In order to ensure the stable and safe operation of the system, the fault diagnosis method based on the improved generalized linear model (BLS) can effectively identify and handle single UAV or system faults. BLS is an efficient neural network model that can quickly perform online learning and fault diagnosis. Compared with traditional deep learning models, BLS avoids complex hierarchical structures by directly constructing generalized feature maps and enhancing nodes, greatly improving learning efficiency. Based on the improved BLS model, faults in multi-UAV systems can be quickly learned and diagnosed, and the online learning mechanism can adapt to environmental changes. It is suitable for fault diagnosis of complex and dynamic multi-UAV systems.

[0062] Compared with the prior art, the present invention obtains the drone information and load information of the multi-drone cooperative suspension system, establishes the drone and load physical model according to the drone information and load information; establishes the multi-drone cooperative suspension system virtual model according to the drone and load physical model; establishes a deep deterministic policy gradient network, iteratively optimizes the multi-drone cooperative suspension system virtual model through the deep deterministic policy gradient network, and obtains the optimized multi-drone cooperative suspension system virtual model; dispatches the multi-drone cooperative suspension system to perform tasks, and controls the multi-drone cooperative suspension system based on the optimized multi-drone cooperative suspension system virtual model; establishes a fault detection model based on an improved generalized linear model, and performs real-time fault detection on the multi-drone cooperative suspension system through the fault detection model, and determines whether the drone in the multi-drone cooperative suspension system has a fault. If so, the position, load distribution and flight trajectory of the drone in the multi-drone cooperative suspension system are adjusted according to the fault detection result. In this way, the present invention trains the drones in the multi-drone cooperative suspension system so that each drone can achieve the optimal action through local information, and detects faults in real time through the fault detection model, so as to realize the drone formation reorganization and load swing minimization, and effectively increase the robustness, stability and control accuracy of the multi-drone cooperative suspension system.

[0063] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0064] The embodiments of the present invention are described above in conjunction with the accompanying drawings. What is disclosed is only the preferred embodiment of the present invention. However, the present invention is not limited to the above-mentioned specific implementation manner. The above-mentioned specific implementation manner is only illustrative rather than restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms and equivalent changes without departing from the scope of protection of the purpose of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failure, characterized in that: The following steps are involved: S1. Obtaining drone information and load information of a multi-drone cooperative suspension system, and establishing a drone and load physical model according to the drone information and load information; S2. Establishing a virtual model of a multi-UAV cooperative suspension system according to the UAV and load physical models; S3, establishing a deep deterministic policy gradient network, and iteratively optimizing the multi-UAV collaborative hanging system virtual model through the deep deterministic policy gradient network to obtain an optimized multi-UAV collaborative hanging system virtual model; S4, dispatching the multi-UAV cooperative hanging system to perform a task, and controlling the multi-UAV cooperative hanging system based on the optimized multi-UAV cooperative hanging system virtual model; S5. Establish a fault detection model based on the improved generalized linear model, and perform real-time fault detection on the multi-UAV collaborative suspension system through the fault detection model to determine whether there is a fault in any UAV in the multi-UAV collaborative suspension system. If so, adjust the position, load distribution and flight trajectory of the UAV in the multi-UAV collaborative suspension system according to the fault detection result.

2. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures according to claim 1 is characterized in that: In step S1, the drone information includes the drone position attitude change information, the output thrust of the drone motor and the attitude adjustment torque information, and the load information includes the load center of mass motion information, swing angle information and angular velocity information.

3. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures according to claim 1 is characterized in that: Step S2 includes the following sub-steps: S21, setting the state space of the drone according to the physical model of the drone and the payload; S22. Setting the action space of the drone according to the physical model of the drone and the payload to obtain the virtual model of the multi-drone cooperative suspension system.

4. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures according to claim 3 is characterized in that: In step S21, the state space includes the position information, attitude information, load status, environmental information and fault status information of the UAV.

5. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures as claimed in claim 3 is characterized in that: In step S22, the action space includes thrust vector adjustment information, attitude adjustment information and speed adjustment information of the UAV, the thrust vector adjustment information is to adjust the thrust to change the position and attitude of the UAV, the attitude adjustment information is to adjust the pitch angle, roll angle and yaw angle of the UAV, and the speed adjustment information is to adjust the speed of the UAV.

6. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures according to claim 1, characterized in that: Step S3 includes the following sub-steps: S31, setting a reward mechanism for the drone in the deep deterministic policy gradient network; S32, establishing a strategy network and a value network for each drone in the multi-drone collaborative suspension system virtual model through the deep deterministic strategy network, wherein the strategy network is used to predict the optimal strategy, and the value network is used to evaluate the action value of the drone, and initialize the experience replay pool; S33, training and iterating the virtual model of the multi-UAV cooperative suspension system according to the reward mechanism, and storing the data information of the UAVs in the virtual model of the multi-UAV cooperative suspension system after the training iteration into the experience playback pool; wherein the data information includes the current state, action, reward and next state of the UAVs; S34, randomly selecting the data information from the experience replay pool, and updating the parameters of the policy network and the parameters of the value network according to the data information; S35. Determine whether the current number of iterations is the maximum number of iterations. If so, output the current multi-UAV collaborative suspension system virtual model as the optimized multi-UAV collaborative suspension system virtual model; if not, return to step S33.

7. The method for autonomously adjusting the coordinated loading of multiple UAVs under sudden failures according to claim 6 is characterized in that: The reward mechanism is set according to the relative distance between the drones, the flight trajectory of the drones, the obstacle avoidance capability of the drones, and the swing angle of the payload carried by the drones.

Images