Dynamic unmanned aerial vehicle cluster adaptive learning rate adjustment method and system

Abstract

This invention relates to a method and system for adaptive learning rate adjustment in dynamic unmanned aerial vehicle (UAV) swarms. The method includes: each UAV, through an environmental perception module, real-time monitoring and collection of environmental parameters and flight status data, transmitting this data to an adaptive learning module to calculate the adaptability of the current learning rate; adjusting the initial learning rate in real-time within a learning rate threshold range based on the adaptability of the current learning rate to obtain a target learning rate; and a central control unit dynamically adjusting flight mission data based on a distributed decision-making algorithm, deciding the flight path and actions of each UAV according to the target flight mission data. By real-time collection of environmental parameters and flight status data for adaptive learning rate adjustment, each UAV operates independently, while the central control unit performs collaborative optimization based on a distributed decision-making algorithm. This avoids the single point of failure problem of centralized control, improves robustness and flexibility, and ensures the efficient and stable operation of the UAV swarm in dynamic environments.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm control technology, and in particular to a dynamic UAV swarm adaptive learning rate adjustment method and system. Background Technology

[0002] Drone swarms offer powerful collaborative capabilities. Compared to individual drones, drone swarms are more efficient, robust, flexible in task allocation and scheduling, and more adaptable. Drone swarm control technologies primarily include centralized control, distributed control, and fixed-learning-rate adaptive control. Centralized control typically uses a central controller to coordinate the actions of all drones, enabling precise swarm control and facilitating management and monitoring. In distributed control, each drone operates independently but collaborates through communication, improving system robustness and flexibility and avoiding single points of failure. Fixed-learning-rate adaptive control employs machine learning and adaptive control algorithms, such as reinforcement learning and neural networks, to optimize drone control strategies.

[0003] However, centralized control methods result in the central controller becoming a single point of failure. If the central controller malfunctions, the entire drone swarm may lose control. Furthermore, centralized control has a high computational and communication load, which can easily lead to system latency and performance bottlenecks. Distributed control typically has low coordination efficiency, especially when there are a large number of drones, with high communication overhead and a tendency to cause communication congestion. Adaptive control methods with fixed learning rates usually use a fixed learning rate. In complex and dynamic environments, a fixed learning rate may lead to slow convergence of the learning process or an inability to adapt to environmental changes. For example, when the environment changes drastically, a fixed learning rate may not be able to quickly adjust the behavior of the drones, resulting in a decline in swarm performance.

[0004] Therefore, traditional UAV swarm control methods suffer from poor control performance and weak dynamic adaptability. Summary of the Invention

[0005] Based on this, in order to solve the above-mentioned technical problems, a method and system for adjusting the adaptive learning rate of a dynamic UAV swarm is provided, which can improve the dynamic adaptability of the UAV swarm and enhance its control performance.

[0006] A dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment method, the method comprising:

[0007] Each drone in the drone swarm acquires a pre-set initial learning rate and learning rate threshold range, and obtains flight mission data.

[0008] Each of the aforementioned UAVs executes flight missions based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through the environmental perception module, and transmits them to the adaptive learning module;

[0009] Each of the aforementioned UAVs calculates the adaptability of the current learning rate based on the environmental parameters and the flight status data through the adaptive learning module, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate, and sends the target learning rate to the central control unit.

[0010] The central control unit, based on a distributed decision-making algorithm, dynamically adjusts the flight mission data according to the target learning rate to obtain target flight mission data, and then determines the flight path and actions of each UAV based on the target flight mission data.

[0011] In one embodiment, the method further includes:

[0012] The central control unit sends task allocation instructions to each of the drones;

[0013] Each of the aforementioned UAVs executes a flight mission according to the task allocation instructions, and collects current environmental parameters and current flight status data in real time and sends them to the central control unit;

[0014] The central control unit continuously monitors each of the UAVs based on the current environmental parameters and current flight status data.

[0015] In one embodiment, each of the drones is equipped with a decision-making mechanism module; the method further includes:

[0016] Each of the aforementioned drones collects drone status information and transmits the drone status and environmental parameters to each other.

[0017] Each of the decision-making mechanism modules calculates the decision information for each of the drones based on the drone's status and the environmental parameters.

[0018] Each of the aforementioned drones periodically exchanges its drone status and decision information.

[0019] In one embodiment, the various drones periodically exchange drone status and decision information, including:

[0020] Each of the aforementioned drones randomly searches for neighboring drone nodes and exchanges drone status and decision information with the neighboring drone nodes to obtain exchanged data.

[0021] Each of the aforementioned drones performs information fusion based on the exchanged data and updates its status according to the fused information.

[0022] In one embodiment, the method further includes:

[0023] Each of the aforementioned UAVs continuously collects environmental parameters and flight status data through the environmental perception module, and performs data preprocessing to obtain a dataset;

[0024] A deep neural network is trained based on the dataset, and the parameters are updated and optimized using gradient descent to obtain an optimized machine learning model.

[0025] The central control unit, based on the machine learning model, determines the flight path and actions of each UAV according to the target flight mission data.

[0026] In one embodiment, the adaptive adjustment of the initial learning rate based on the current learning rate within the learning rate threshold range includes:

[0027] Determine the time step, and calculate the gradient vector based on the time step;

[0028] The adaptive learning rate is calculated based on the first and second moments of the gradient vector and the initial learning rate.

[0029] The initial learning rate is adjusted in real time within the learning rate threshold range to become the adaptive learning rate.

[0030] In one embodiment, the method further includes:

[0031] Calculate the exponential moving average of the first moment and the exponential moving average of the second moment of the gradient vector, and correct the deviation of the first moment and the second moment.

[0032] Based on the bias-corrected value and the adaptive learning rate, the parameter vector corresponding to the time step is updated.

[0033] In one embodiment, the method further includes:

[0034] Calculate the energy consumption rate of the UAV based on the environmental parameters and flight status data;

[0035] Calculate the influence coefficient of the target learning rate on the energy consumption rate, and calculate the energy consumption optimization target of the UAV based on the influence coefficient.

[0036] A dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment system, the system comprising: a central control unit located at the UAV swarm, several UAVs, a control module, an environmental perception module, and an adaptive learning module mounted on each of the UAVs; wherein:

[0037] The central control unit is used to set the initial learning rate and learning rate threshold range for each of the UAVs, and to acquire flight mission data.

[0038] The control module in each of the aforementioned UAVs controls the UAVs to perform flight missions based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through the environmental perception module, and transmits them to the adaptive learning module;

[0039] Each of the aforementioned UAVs calculates the adaptability of the current learning rate based on the environmental parameters and the flight status data through the adaptive learning module, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate, and sends the target learning rate to the central control unit.

[0040] The central control unit, based on a distributed decision-making algorithm, dynamically adjusts the flight mission data according to the target learning rate to obtain target flight mission data, and then determines the flight path and actions of each UAV based on the target flight mission data.

[0041] The aforementioned dynamic UAV swarm adaptive learning rate adjustment method and system optimizes the control strategy of the UAV swarm and enhances dynamic adaptability by collecting environmental parameters and flight status data in real time and adjusting the adaptive learning rate. Each UAV operates independently, and the central control unit performs collaborative optimization based on a distributed decision algorithm, which avoids the single point of failure problem of centralized control, while improving robustness and flexibility, and ensuring the efficient and stable operation of the UAV swarm in dynamic environments. Attached Figure Description

[0042] Figure 1 This is a diagram illustrating the application environment of a dynamic UAV swarm adaptive learning rate adjustment method in one embodiment.

[0043] Figure 2 This is a flowchart illustrating a dynamic UAV swarm adaptive learning rate adjustment method in one embodiment;

[0044] Figure 3 This is a block diagram of a dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment system in one embodiment.

[0045] Figure 4 This is a diagram of the internal structure of the central control unit in one embodiment. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0047] The dynamic UAV swarm adaptive learning rate adjustment method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown. For example... Figure 1 As shown, the application environment includes a central control unit 100 and various drones 200. The central control unit 100 is connected and communicates with each drone 200, and the drones 200 are interconnected. Each drone 200 in the drone swarm acquires a pre-set initial learning rate and learning rate threshold range, and acquires flight mission data. Each drone 200 executes flight missions based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through an environmental perception module, transmitting this data to an adaptive learning module. Each drone 200 calculates the adaptability of its current learning rate based on the environmental parameters and flight status data through the adaptive learning module, and adjusts its initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate, obtaining a target learning rate which is then sent to the central control unit 100. The central control unit 100 dynamically adjusts the flight mission data based on the target learning rate using a distributed decision-making algorithm, obtaining target flight mission data, and decides the flight path and actions of each drone based on the target flight mission data. The central control unit 100 can be, but is not limited to, various personal computers, laptops, smartphones, robots, tablets, and other devices.

[0048] In one embodiment, such as Figure 2 As shown, a dynamic UAV swarm adaptive learning rate adjustment method is provided, including the following steps:

[0049] Step 202: Each drone in the drone cluster acquires the set initial learning rate and learning rate threshold range, and acquires flight mission data.

[0050] Each drone in the drone swarm is equipped with a control module, an adaptive learning module, and an environmental perception module. After initializing each drone, an initial learning rate and a learning rate threshold range can be set on each drone, and then the adaptive learning module can be activated.

[0051] Step 204: Each UAV executes its flight mission based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through the environmental perception module, and transmits them to the adaptive learning module.

[0052] After initialization, each drone acquires flight mission data transmitted from the central control unit and then executes the flight mission based on the flight mission data.

[0053] During flight missions, each drone can monitor environmental parameters in real time through its environmental perception module and transmit the data to the adaptive learning module. Specifically, when transmitting environmental parameters, the drone's real-time flight status data is also transmitted to the adaptive learning module.

[0054] The environmental perception module includes sensors that can collect environmental data and flight status data, and transmit these data to the adaptive learning module of the adaptive learning layer.

[0055] Step 206: Each UAV calculates the adaptability of the current learning rate based on environmental parameters and flight status data through the adaptive learning module, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate, and sends the target learning rate to the central control unit.

[0056] The adaptive learning module runs a multi-dimensional environment perception algorithm, which performs comprehensive analysis based on multi-dimensional environmental data (such as weather conditions, obstacle distribution, etc.) and adjusts the learning rate parameter in real time.

[0057] The adaptive learning modules in each UAV can calculate the adaptability of the learning rate based on environmental changes and task execution feedback. Based on the adaptability of the current learning rate, the adaptive learning module can adjust the initial learning rate in real time within a learning rate threshold range to obtain the target learning rate. Specifically, if the environment changes drastically or the task execution is unsatisfactory, the adaptive learning module increases the learning rate to optimize the control strategy; if the environment changes smoothly and the task execution is good, the adaptive learning module decreases the learning rate to stabilize the learning process.

[0058] Step 208: The central control unit dynamically adjusts the flight mission data based on the target learning rate using a distributed decision-making algorithm to obtain the target flight mission data, and then decides the flight path and actions of each UAV based on the target flight mission data.

[0059] Distributed decision-making algorithms achieve autonomous decision-making and coordination by distributing the decision-making process among various drone nodes. Specifically, each drone node in the drone swarm possesses sensing and processing capabilities, enabling it to collect its own status and information about the surrounding environment. The status representation includes multiple parameters such as position, speed, remaining battery power, and task load, denoted as a vector: s i =[pi,v] i e i , l i ].

[0060] Each drone node independently calculates its decision based on its perceived state and information received from other nodes, aiming to maximize long-term gains. Let Q(s, a) be the value function of taking action a in state s: Q(s t a t )←Q(s t a t )+α[r t +γmax a Q(s t+1 ,a)-Q(s t a t )]; where α is the learning rate, γ is the discount factor, and r t For instant rewards, s t+1 To perform action a t The new state after that.

[0061] To ensure coordination within the drone swarm, each drone node needs to periodically exchange status and decision information. Specifically, a distributed approach using protocols can be employed, including information propagation and information fusion. During information propagation, drone node i randomly selects a neighboring drone node j and transmits its status and Q-value to node j. During information fusion, upon receiving the information, each drone node merges it with its local information to update its local status and decision. The information fusion process uses a weighted average method, with the formula: Among them, w i Let be the weight of node i, which is usually related to the trust level or importance of the node.

[0062] The central control unit is responsible for mission planning and command distribution for the entire drone swarm, communicating with the control module of each drone via a wireless network. Based on the learning performance data fed back by the drones, the central control unit dynamically adjusts the learning rate parameters of each drone, ensuring overall swarm coordination and mission completion efficiency. By introducing a swarm collaborative optimization mechanism, the individual learning rates are further optimized by analyzing the collaborative work effects of multiple drones, thereby improving the overall performance of the swarm.

[0063] The central control unit can analyze the overall collaborative effect based on feedback data from all drones in the drone swarm, and adjust the learning rate parameters of individual drones to ensure the efficient completion of swarm tasks.

[0064] Specifically, the central control unit can adjust the drone's movements and paths based on the optimized control strategy to better adapt to the current environment. During the execution of tasks by each drone, the drone continuously monitors environmental changes and provides task execution feedback to the central control unit, so as to adjust the learning rate and control strategy in real time. Based on the adjusted learning rate, the control strategy is optimized, enabling the drones to better adapt to the current environment and execute tasks, ensuring that the drone swarm can complete tasks efficiently and stably in dynamic environments.

[0065] In this embodiment, each UAV can integrate a multi-level control layer structure, including a basic control layer, a task execution layer, an adaptive learning layer, and an environmental perception layer. The basic control layer is responsible for the basic flight control and stability maintenance of the UAV, ensuring basic flight safety. The task execution layer is used for task planning and execution according to task instructions assigned by the central control unit. The adaptive learning layer dynamically adjusts the learning rate and optimizes the control strategy based on real-time environmental data and flight status data. It employs an innovative multi-dimensional environmental perception algorithm, enabling comprehensive perception of changes in the surrounding environment and rapid response. The environmental perception layer perceives environmental changes in real time through sensors and communication devices and feeds the information back to the adaptive learning module. Furthermore, during UAV swarm control, power consumption algorithms can be adjusted, significantly reducing power consumption through adaptive learning rate adjustments in the UAV swarm.

[0066] In this embodiment, the learning rate of the UAV is continuously optimized during mission execution through dynamic adjustment, resulting in a significant reduction in overall energy consumption. When the UAV performs long-term missions, energy consumption is evenly distributed to avoid rapid energy depletion. During large-scale patrols, the UAV's endurance is maximized and mission completion efficiency is improved by reasonably adjusting the learning rate.

[0067] In one embodiment, the provided dynamic UAV swarm adaptive learning rate adjustment method may further include a process of continuous monitoring of the UAVs. The specific process includes: the central control unit sending task allocation instructions to each UAV; each UAV executing flight tasks according to the task allocation instructions and collecting current environmental parameters and current flight status data in real time and sending them to the central control unit; and the central control unit continuously monitoring each UAV based on the current environmental parameters and current flight status data.

[0068] In one embodiment, each UAV is equipped with a decision-making mechanism module; the provided dynamic UAV swarm adaptive learning rate adjustment method may further include a UAV decision-making process, specifically including: each UAV collects UAV status information and transmits UAV status and environmental parameters to each other; each decision-making mechanism module calculates the decision information of each UAV based on the UAV status and environmental parameters; and each UAV periodically exchanges UAV status and decision information.

[0069] In one embodiment, the provided dynamic UAV swarm adaptive learning rate adjustment method may further include a process of UAVs exchanging information. The specific process includes: each UAV randomly searching for neighboring UAV nodes and exchanging UAV status and decision information with the neighboring UAV nodes to obtain exchange data; each UAV performing information fusion based on the exchange data and updating its status according to the fused information.

[0070] In one embodiment, the provided dynamic UAV swarm adaptive learning rate adjustment method may further include a model training process, specifically including: each UAV continuously collects environmental parameters and flight status data through an environmental perception module, and performs data preprocessing to obtain a dataset; a deep neural network is trained based on the dataset, and parameters are updated and optimized using gradient descent to obtain an optimized machine learning model; the central control unit, based on the machine learning model, decides the flight path and actions of each UAV according to the target flight mission data.

[0071] By optimizing the decision-making of drone swarms through well-trained machine learning models, machine learning models can provide effective assistance in handling complex tasks and responding to emergencies.

[0072] Specifically, during mission execution, the drone swarm continuously collects environmental data and status information. Environmental data includes weather conditions, terrain features, obstacle information, and the location of the mission target. To ensure data accuracy and consistency, preprocessing is required, mainly including data cleaning, normalization, and feature extraction.

[0073] In the model training and optimization process, deep neural networks (DNNs) and gradient descent are used to optimize the decision-making of the drone swarm. The model training process is as follows:

[0074] The basic unit of a neural network is the neuron, and its output can be represented as: Where l represents the number of layers, when It is the weight matrix of the l-th layer. It is the activation value of the previous layer. f is the bias term, and f is the activation function.

[0075] loss function Mean squared error (MSE) is usually used: Among them, y i This is the actual value. is the predicted value, and N is the number of samples.

[0076] To minimize the loss function, gradient descent is used for parameter updates. The gradient descent update is as follows: Where η is the learning rate. Let θ be the gradient of the loss function with respect to the parameter θ.

[0077] The optimization process of the model includes:

[0078] The model is trained using a deep neural network with a multi-layer perceptron (MLP) structure. The gradient of the loss function can be calculated using the backpropagation algorithm.

[0079] During the forward propagation, the activation value of each layer is calculated: Where L is the output layer. The loss function is calculated as follows: Among them, y i This is the actual value. This is a predicted value.

[0080] During backpropagation, the gradient of the loss function with respect to each parameter is calculated: for the output layer, the gradient is calculated using... For hidden layers, δ is used. [l] =(W [l+1] ) T δ [l+1] ☉f′(z [l] ); where ⊙ represents element-wise multiplication (Hadamard product).

[0081] The gradient calculation formula is: The parameter update process is as follows:

[0082] In one embodiment, a dynamic UAV swarm adaptive learning rate adjustment method may further include a process of adjusting the learning rate, specifically including: determining a time step, calculating a gradient vector based on the time step; calculating an adaptive learning rate based on the first moment, second moment, and initial learning rate of the gradient vector; and adjusting the initial learning rate to an adaptive learning rate in real time within a learning rate threshold range.

[0083] In one embodiment, the provided dynamic UAV swarm adaptive learning rate adjustment method may further include a parameter update process, specifically including: calculating the exponential moving average of the first moment and the exponential moving average of the second moment of the gradient vector, and correcting the deviation of the first moment and the second moment; updating the parameter vector corresponding to the time step based on the value after deviation correction and the adaptive learning rate.

[0084] During parameter updates, online learning models can be used. These models continuously receive new data and optimize themselves during task execution. Incremental updates are necessary; they update model parameters gradually, rather than waiting for all data to be collected before updating. The incremental update of the loss function is as follows: Where t represents the time step, η t Let t be the learning rate at time step t.

[0085] During online learning, dynamically adjusting the learning rate can improve model training performance. Adaptive learning rate adjustment, to further enhance online learning effectiveness, involves dynamically adjusting the learning rate. This method integrates multiple optimization strategies. Calculating the adaptive learning rate includes dynamically adjusting the learning rate based on the time step t and the initial learning rate η0. Where η0 is the initial learning rate; η t β1 is the adaptive learning rate at time step t; β2 is the decay rate of the first moment, usually taken as 0.9; β3 is the decay rate of the second moment, usually taken as 0.999.

[0086] Adaptive learning rate, updating parameter θ t The steps include:

[0087] The first step is to calculate the gradient: calculate the gradient vector g at the current time step t. t :

[0088] Step 2: Exponential moving average of the first moment: Calculate the exponential moving average m of the first moment of the gradient. t ;m t =β1m t-1 +(1-β1)g t ;

[0089] Step 3: Exponential moving average of the second moment: Calculate the exponential moving average v of the second moment of the gradient. t ;v t =β2v t-1 +(1-β2)g t ⊙g t ;

[0090] Step 4: Maximum Second Moment Estimate: Ensure the second moment estimate is monotonically non-decreasing, and obtain...

[0091] Step 5: Deviation Correction: Correct the first and second moments for deviations, and obtain... and

[0092] Step 6: Adaptive Learning Rate Adjustment: Dynamically adjust the learning rate.

[0093] Step 7: Update parameters. Based on the bias-corrected value and the adaptive learning rate, update parameter θ. t :

[0094] In this embodiment, θ t A parameter vector representing time step t; The loss function representing time step t; g t The gradient vector at time step t; m t The exponential moving average of the first moment of the gradient; v t The exponential moving average of the second moment of the gradient; m represents the deviation correction t ; v represents the deviation correction t β1 represents the decay rate of the first moment, usually taken as 0.9; β2 represents the decay rate of the second moment, usually taken as 0.999; η represents the global learning rate.

[0095] By using a dynamic drone swarm adaptive learning rate adjustment method, drones can achieve more efficient and stable online learning and self-optimization.

[0096] In one embodiment, a dynamic UAV swarm adaptive learning rate adjustment method may further include an energy consumption optimization process, which specifically includes: calculating the UAV's energy consumption rate based on environmental parameters and flight status data; calculating the influence coefficient of the target learning rate on the energy consumption rate; and calculating the UAV's energy consumption optimization target based on the influence coefficient.

[0097] Here, we assume that the energy consumption rate of a single UAV is E(t), which is determined by the flight speed v(t), flight altitude h(t), payload L(t), and environmental factors ∈(t). The energy consumption rate can be expressed as: E(t) = αv(t) 2 +βh(t)+γL(t)+δ∈(t); where α,β,γ,δ are constants related to the characteristics of the UAV and the environment.

[0098] Assuming the learning rate is η(t), its adjustment affects the UAV's decision-making and path planning, thus indirectly affecting energy consumption. An energy optimization coefficient κ(t) is introduced to characterize the impact of the learning rate on energy consumption: κ(t) = f(η(t)); where f(η(t)) is a function of the learning rate, usually assumed to be a monotonically decreasing function to ensure that a higher learning rate leads to higher energy consumption. Specifically, it can take the form: Where λ and η0 are adjustment parameters that control the shape of the curve.

[0099] To minimize energy consumption throughout the entire task cycle, the energy optimization objective is defined as follows: Where T is the total task time.

[0100] In practical applications, drones need to perform different tasks (such as patrol, reconnaissance, and rescue), each with different energy consumption requirements. Let's assume the task set is... Each task The corresponding energy consumption is E m (t), then the comprehensive optimization objective can be expressed as: Among them, T m Let m be the execution time of task m.

[0101] At each time t, the learning rate η(t) is dynamically adjusted based on the current environmental state ∈(t) and task requirements to optimize energy consumption. The specific adjustment strategy is as follows: Where μ is the learning rate adjustment step size.

[0102] Specifically, in this embodiment, the energy consumption model of the UAV in a certain task m is assumed to be: E m (t)=0.01v(t) 2 +0.05h(t)+0.1L(t)+0.02∈(t); The effect function of the learning rate on energy consumption is: The optimization objective is:

[0103]

[0104] Dynamically adjust the learning rate using gradient descent:

[0105]

[0106] In this embodiment, the learning rate of the UAV is continuously optimized during task execution through dynamic adjustment, resulting in a significant reduction in overall energy consumption. When performing long-duration tasks, energy consumption is evenly distributed, preventing rapid energy depletion. During large-scale patrols, the UAV's endurance is maximized and task completion efficiency is improved by rationally adjusting the learning rate. Through a complex energy consumption model and learning rate adjustment strategy, optimal energy utilization of the UAV is ensured across various tasks, providing strong support for the efficient execution of long-duration, complex tasks.

[0107] In this embodiment, each UAV or the central control unit located in the UAV cluster is equipped with control software and learning algorithm software, which are used to realize the basic flight control, path planning and communication functions of the UAV; and to realize the adaptive learning rate adjustment algorithm, including initial parameter setting, environmental perception, feedback collection and strategy optimization.

[0108] In one embodiment, a dynamic drone swarm adaptive learning rate adjustment method may include:

[0109] System initialization: Set an initial learning rate (e.g., 0.01) and a learning rate threshold range (e.g., 0.001 to 0.1) for each UAV, start the adaptive learning module, and prepare to receive environmental data and feedback information;

[0110] Mission execution: All UAVs begin executing their predetermined missions (such as patrolling, search and rescue, etc.) according to the initial control strategy; the environmental perception module monitors environmental parameters in real time and transmits the data to the adaptive learning module;

[0111] Learning rate adjustment and strategy optimization: The adaptive learning module calculates the adaptability of the current learning rate based on environmental changes and task execution feedback. If the environment changes drastically or the task execution is unsatisfactory, the adaptive learning module increases the learning rate (e.g., from 0.01 to 0.05) to accelerate strategy optimization. If the environment changes smoothly and the task execution is good, the adaptive learning module decreases the learning rate (e.g., from 0.01 to 0.005) to stabilize the learning process. The control module adjusts the UAV's flight path and actions in real time according to the optimized control strategy.

[0112] Continuous monitoring and adjustment: During mission execution, continuously monitor environmental changes and mission execution feedback, and adjust the learning rate and control strategy in real time; ensure that the drone swarm can complete the mission efficiently and stably in a dynamic environment.

[0113] In one embodiment, the dynamic UAV swarm adaptive learning rate adjustment method of this application is applied to post-disaster search and rescue missions in complex urban environments. The UAVs need to patrol, avoid obstacles, identify targets, and locate trapped personnel in complex urban environments. The specific implementation process is as follows:

[0114] System initialization: Set the initial learning rate to 0.01 for each UAV participating in the search and rescue mission, with a learning rate threshold range of 0.001 to 0.1; start the adaptive learning module to prepare for receiving environmental data and feedback information;

[0115] Mission execution: The UAV begins to execute the search and rescue mission according to the initial control strategy, patrolling and searching for targets within the predetermined area; the environmental perception module monitors the surrounding environmental parameters in real time, including building structure, debris distribution, wind speed, etc., and transmits the data to the adaptive learning module;

[0116] Learning rate adjustment and strategy optimization: The adaptive learning module calculates the adaptability of the current learning rate based on environmental changes and task execution feedback (such as path deviation, energy consumption, target recognition accuracy, etc.). For example, when the drone enters an area with dense collapsed buildings, where the environment changes drastically, the adaptive learning module increases the learning rate (e.g., from 0.01 to 0.05) to accelerate strategy optimization and better avoid obstacles. In a more open area with less environmental change, the adaptive learning module decreases the learning rate (e.g., from 0.01 to 0.005) to stabilize the learning process and ensure stable drone flight and effective patrolling.

[0117] Continuous monitoring and adjustment: Throughout the search and rescue mission, environmental changes and mission execution feedback are continuously monitored, and the learning rate and control strategy are adjusted in real time to ensure that the drone swarm can complete the search and rescue mission efficiently and stably, accurately identify and locate trapped personnel, and provide reliable data support for the rescue operation;

[0118] Application Results: By employing a dynamic UAV swarm adaptive learning rate adjustment method, UAV swarms can quickly adapt to environmental changes in complex urban environments, optimize flight control strategies, and achieve efficient post-disaster search and rescue missions. Compared to traditional fixed learning rate methods, this significantly improves the adaptability, mission execution efficiency, and system stability of UAV swarms, providing a new solution for the application of UAVs in dynamic and complex environments.

[0119] In one embodiment, experimental data is used to verify the effectiveness of the dynamic UAV swarm adaptive learning rate adjustment method in this application:

[0120] Experimental Design: Conduct search and rescue missions in a simulated complex urban environment to compare the performance of UAV swarms using the technology described in this application and traditional fixed learning rate methods.

[0121] Experimental data:

[0122]

[0123] Results analysis:

[0124] Task completion time: The task completion time of the UAV swarm using the dynamic UAV swarm adaptive learning rate adjustment method in this application is reduced by about 33%, indicating a significant improvement in task execution efficiency.

[0125] Target recognition accuracy improved by about 10%, indicating that the optimization of the control strategy was effective.

[0126] Energy consumption: Energy consumption was reduced by about 20%, indicating that energy efficiency was improved through optimized control strategies.

[0127] Environment adaptation time: The environment adaptation time was shortened by about 67%, indicating that the adaptive learning rate adjustment enabled the drone to adapt to environmental changes more quickly.

[0128] Task success rate: The task success rate increased by approximately 15%, indicating a significant improvement in the overall performance of the system and the reliability of task execution.

[0129] This application provides a dynamic UAV swarm adaptive learning rate adjustment method, which involves an adaptive learning rate adjustment algorithm, an environmental perception and feedback mechanism, distributed control and collaborative optimization, a real-time dynamic adjustment mechanism, initial learning rate and threshold setting, and an efficient energy management method. Specifically: an adaptive learning rate adjustment algorithm is introduced to dynamically adjust the learning rate based on environmental changes and task feedback, optimizing the control strategy; an integrated environmental perception module monitors environmental parameters (such as wind speed and obstacles) in real time and feeds them back to the learning module, enabling the UAVs to dynamically adapt to complex environments, providing real-time environmental data, supporting adaptive learning rate adjustment decisions, and enhancing the system's dynamic adaptability; a distributed control method is adopted, where each UAV operates independently and collaborates for optimization, avoiding... This approach eliminates the single point of failure associated with centralized control, while simultaneously enhancing system robustness and flexibility, thus improving cluster collaboration efficiency. During task execution, continuous monitoring of environmental changes and task performance, along with real-time adjustments to the learning rate and control strategy, ensures the efficient and stable operation of the UAV cluster in dynamic environments, serving as a crucial guarantee for dynamic adaptive adjustment. Setting initial learning rates and threshold ranges based on specific tasks and environments provides reasonable initial parameters for subsequent adaptive adjustments, ensuring the effectiveness and stability of the learning process. Through complex energy consumption models and learning rate adjustment strategies, it ensures optimal energy utilization for UAVs across various tasks, providing strong support for the efficient execution of long-duration, complex UAV missions.

[0130] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0131] In one embodiment, such as Figure 3 As shown, a dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment system is provided, comprising: a central control unit 100 located at the UAV swarm, several UAVs 200, a control module 210, an environmental perception module 220, and an adaptive learning module 230 located on each UAV 200; wherein:

[0132] The central control unit 100 is used to set the initial learning rate and learning rate threshold range of each UAV 200, and to acquire flight mission data;

[0133] The control module 210 in each UAV 200 controls the UAV to perform flight missions according to flight mission data, and the environmental perception module 220 monitors and collects environmental parameters and flight status data in real time and transmits them to the adaptive learning module 230.

[0134] Each UAV 200 calculates the adaptability of the current learning rate based on environmental parameters and flight status data through the adaptive learning module 230, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate, and sends the target learning rate to the central control unit 100.

[0135] The central control unit 100 dynamically adjusts the flight mission data based on the target learning rate according to the distributed decision-making algorithm to obtain the target flight mission data, and makes decisions on the flight path and actions of each UAV 200 based on the target flight mission data.

[0136] In one embodiment, the central control unit 100 is also used to send task allocation instructions to each UAV 200; each UAV 200 executes flight tasks according to the task allocation instructions and collects current environmental parameters and current flight status data in real time and sends them to the central control unit 100; the central control unit 100 continuously monitors each UAV 200 based on the current environmental parameters and current flight status data.

[0137] In one embodiment, each UAV 200 is equipped with a decision-making mechanism module; each UAV 200 collects UAV status information and transmits the UAV status and environmental parameters to each other; each decision-making mechanism module calculates the decision information of each UAV 200 based on the UAV status and environmental parameters; each UAV 200 periodically exchanges UAV status and decision information.

[0138] In one embodiment, each UAV 200 randomly searches for neighboring UAV nodes and exchanges UAV status and decision information with the neighboring UAV nodes to obtain exchange data; each UAV 200 performs information fusion based on the exchange data and updates its status according to the fused information.

[0139] In one embodiment, each UAV 200 continuously collects environmental parameters and flight status data through an environmental perception module, and performs data preprocessing to obtain a dataset; a deep neural network is trained based on the dataset, and parameters are updated and optimized using gradient descent to obtain an optimized machine learning model; the central control unit, based on the machine learning model, decides the flight path and actions of each UAV 200 according to the target flight mission data.

[0140] In one embodiment, the adaptive learning module 230 is further configured to determine a time step, calculate a gradient vector based on the time step, calculate an adaptive learning rate based on the first moment, second moment, and initial learning rate of the gradient vector, and adjust the initial learning rate to the adaptive learning rate in real time within a learning rate threshold range.

[0141] In one embodiment, the adaptive learning module 230 is further configured to calculate the exponential moving average of the first moment and the exponential moving average of the second moment of the gradient vector, and to perform bias correction on the first moment and the second moment; based on the bias-corrected value and the adaptive learning rate, update the parameter vector corresponding to the time step.

[0142] In one embodiment, the adaptive learning module 230 is further configured to calculate the energy consumption rate of the UAV based on environmental parameters and flight status data; calculate the influence coefficient of the target learning rate on the energy consumption rate; and calculate the energy consumption optimization target of the UAV based on the influence coefficient.

[0143] In one embodiment, a central control unit is provided at a drone swarm, the internal structure of which can be shown in the following diagram. Figure 4 As shown, the central control unit includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the central control unit is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a dynamic UAV swarm adaptive learning rate adjustment method. The display screen of the central control unit can be an LCD screen or an e-ink screen. The input devices of the central control unit can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the housing of the central control unit, or an external keyboard, touchpad, or mouse, etc.

[0144] Those skilled in the art will understand that Figure 4The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the central control unit to which the present application is applied. The specific central control unit may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0145] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for adaptive learning rate adjustment in dynamic unmanned aerial vehicle (UAV) swarms, characterized in that, The method includes: Each drone in the drone swarm acquires a pre-set initial learning rate and learning rate threshold range, and obtains flight mission data. Each of the aforementioned UAVs executes flight missions based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through the environmental perception module, and transmits them to the adaptive learning module; Each of the aforementioned UAVs calculates the adaptability of the current learning rate based on the environmental parameters and flight status data through the adaptive learning module, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate. This includes: determining a time step; calculating a gradient vector based on the time step; calculating an adaptive learning rate based on the first and second moments of the gradient vector and the initial learning rate; adjusting the initial learning rate to the adaptive learning rate in real time within the learning rate threshold range to obtain a target learning rate, which is then sent to the central control unit; calculating the exponential moving average of the first and second moments of the gradient vector, and correcting for deviations in the first and second moments; and updating the parameter vector corresponding to the time step based on the corrected values ​​and the adaptive learning rate. The central control unit, based on a distributed decision-making algorithm, dynamically adjusts the flight mission data according to the target learning rate to obtain target flight mission data, and then determines the flight path and actions of each UAV based on the target flight mission data.

2. The dynamic UAV swarm adaptive learning rate adjustment method according to claim 1, characterized in that, The method further includes: The central control unit sends task allocation instructions to each of the drones; Each of the aforementioned UAVs executes a flight mission according to the task allocation instructions, and collects current environmental parameters and current flight status data in real time and sends them to the central control unit; The central control unit continuously monitors each of the UAVs based on the current environmental parameters and current flight status data.

3. The dynamic UAV swarm adaptive learning rate adjustment method according to claim 1, characterized in that, Each of the aforementioned drones is equipped with a decision-making mechanism module; the method further includes: Each of the aforementioned drones collects drone status information and transmits the drone status and environmental parameters to each other. Each of the decision-making mechanism modules calculates the decision information for each of the drones based on the drone's status and the environmental parameters. Each of the aforementioned drones periodically exchanges its drone status and decision information.

4. The dynamic UAV swarm adaptive learning rate adjustment method according to claim 3, characterized in that, Each of the aforementioned drones periodically exchanges its drone status and decision information, including: Each of the aforementioned drones randomly searches for neighboring drone nodes and exchanges drone status and decision information with the neighboring drone nodes to obtain exchanged data. Each of the aforementioned drones performs information fusion based on the exchanged data and updates its status according to the fused information.

5. The dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment method according to claim 1, characterized in that, The method further includes: Each of the aforementioned UAVs continuously collects environmental parameters and flight status data through the environmental perception module, and performs data preprocessing to obtain a dataset; A deep neural network is trained based on the dataset, and the parameters are updated and optimized using gradient descent to obtain an optimized machine learning model. The central control unit, based on the machine learning model, determines the flight path and actions of each UAV according to the target flight mission data.

6. The dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment method according to claim 1, characterized in that, The method further includes: Calculate the energy consumption rate of the UAV based on the environmental parameters and flight status data; Calculate the influence coefficient of the target learning rate on the energy consumption rate, and calculate the energy consumption optimization target of the UAV based on the influence coefficient.

7. A dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment system, characterized in that, The system includes: a central control unit located at the drone cluster, several drones, a control module, an environmental perception module, and an adaptive learning module installed on each drone; wherein: The central control unit is used to set the initial learning rate and learning rate threshold range for each of the UAVs, and to acquire flight mission data. The control module in each of the aforementioned UAVs controls the UAVs to perform flight missions based on the flight mission data, and monitors and collects environmental parameters and flight status data in real time through the environmental perception module, and transmits them to the adaptive learning module; Each of the aforementioned UAVs calculates the adaptability of the current learning rate based on the environmental parameters and flight status data through the adaptive learning module, and adjusts the initial learning rate in real time within the learning rate threshold range based on the adaptability of the current learning rate. This includes: determining a time step; calculating a gradient vector based on the time step; calculating an adaptive learning rate based on the first and second moments of the gradient vector and the initial learning rate; adjusting the initial learning rate to the adaptive learning rate in real time within the learning rate threshold range to obtain a target learning rate, which is then sent to the central control unit; calculating the exponential moving average of the first and second moments of the gradient vector, and correcting for deviations in the first and second moments; and updating the parameter vector corresponding to the time step based on the corrected values ​​and the adaptive learning rate. The central control unit, based on a distributed decision-making algorithm, dynamically adjusts the flight mission data according to the target learning rate to obtain target flight mission data, and then determines the flight path and actions of each UAV based on the target flight mission data.

8. The dynamic unmanned aerial vehicle (UAV) swarm adaptive learning rate adjustment system according to claim 7, characterized in that, The central control unit is also used to send task allocation instructions to each of the drones. Each of the aforementioned UAVs executes a flight mission according to the task allocation instructions, and collects current environmental parameters and current flight status data in real time and sends them to the central control unit; The central control unit continuously monitors each of the UAVs based on the current environmental parameters and current flight status data.

Images