Unmanned aerial vehicle intelligent control method and system based on machine learning
By integrating a phased intelligent control strategy with machine learning algorithms, the UAV can quickly adapt to and precisely optimize in complex environments, solving the problems of poor adaptability and low control precision in existing technologies and improving the control capability of UAVs in complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUANENG RENEWABLES CORP LTD HEBEI BRANCH
- Filing Date
- 2026-02-25
- Publication Date
- 2026-05-12
AI Technical Summary
Existing UAV control methods have poor adaptability in complex environments, low control accuracy, and weak anti-interference capabilities, making it difficult to achieve real-time and precise control and limiting their application in demanding scenarios.
The system employs a deep integration of phased intelligent control strategies and machine learning algorithms, including open-loop control in the initial adaptation phase, semi-closed-loop control in the dynamic adjustment phase, and closed-loop optimization control in the stable optimization phase. Through environmental adaptation models, dynamic adjustment models, and reinforcement learning agents, the system enables the UAV to adapt quickly and optimize precisely.
It improves the timeliness and accuracy of UAV control in complex environments, enhances anti-interference capabilities and autonomous decision-making capabilities, reduces flight deviation and energy consumption, and improves mission reliability and endurance.
Smart Images
Figure CN122018298A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and more specifically to a machine learning-based intelligent control method and system for UAVs. Background Technology
[0002] With the rapid development of drone technology, its application scenarios have expanded from simple aerial photography and mapping to fields such as material transportation, power line inspection, and emergency rescue in complex environments. In complex flight scenarios, such as mountainous terrain, strong winds, and low visibility conditions, drones face challenges such as rapid environmental changes, numerous interference factors, and high flight accuracy requirements.
[0003] Existing UAV control methods mostly employ traditional PID control algorithms, achieving flight control through preset fixed parameters or simple parameter self-tuning strategies. However, traditional control methods have the following drawbacks: First, they have poor adaptability to complex dynamic environments; fixed control parameters struggle to cope with sudden changes in wind speed, terrain undulations, and other unexpected situations, easily leading to unstable flight attitude. Second, their control precision is limited; they cannot accurately adjust parameters based on dynamic changes in the environment and mission, resulting in significant mission execution deviations. Third, they lack autonomous learning capabilities; they cannot optimize control strategies using historical flight data, and the control effect depends on initial parameter settings, resulting in weak anti-interference capabilities. Furthermore, while some existing UAV technologies incorporate simple intelligent algorithms, they suffer from poor data processing timeliness and insufficient model generalization ability, making it difficult to achieve real-time, precise control in complex scenarios and limiting the application of UAVs in demanding environments.
[0004] Therefore, how to propose a machine learning-based intelligent control method and system for unmanned aerial vehicles (UAVs) to overcome the shortcomings of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] In view of this, the present invention provides a machine learning-based intelligent control method and system for unmanned aerial vehicles (UAVs). Through the deep integration of a phased intelligent control strategy and machine learning algorithms, it achieves precise and stable control of UAVs in complex environments. This addresses the problems of low control accuracy, poor adaptability, and weak anti-interference capability in existing technologies. To achieve the above objectives, the present invention adopts the following technical solution: A machine learning-based intelligent control method for unmanned aerial vehicles (UAVs) includes: Determine if the user has entered a drone flight mission command. If the user has entered a flight mission command, collect flight environment parameters and determine if the drone is in a complex flight scenario. If it is in a complex flight scenario, control the drone to enter the initial adaptation phase of the complex flight mode. Determine whether the flight attitude stability of the drone is greater than a predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, control the drone to enter the dynamic adjustment stage of complex flight mode. Determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, then control the UAV to enter the stable optimization phase of the complex flight mode. The system determines whether the drone's mission execution deviation is less than a predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the system controls the drone to exit the complex flight mode and enter the normal flight control mode.
[0006] Optionally, the initial adaptation phase is used to execute a first control strategy, which includes open-loop control of UAV flight parameters based on a pre-trained environment adaptation model. The dynamic adjustment phase is used to execute the second control strategy, which includes semi-closed-loop control of flight parameters based on the dynamic output of real-time flight data and machine learning models. The stabilization optimization phase is used to execute a third control strategy, which includes closed-loop optimization control of flight parameters based on a reinforcement learning algorithm.
[0007] Optionally, the open-loop control of the flight parameters includes controlling the UAV's power actuators, attitude actuators, and navigation actuators through the flight control unit to control the UAV's operation with a pre-generated initial set of flight parameters.
[0008] Optionally, the initial flight parameter set includes a flight speed reference value, an attitude angle reference value, a lift reference value, and a navigation path reference value.
[0009] Optionally, the initial flight parameter set is calculated by the flight control unit through an environment adaptation model. The environment adaptation model is a neural network model trained based on a deep learning algorithm. Its input parameters include wind speed, wind direction, air pressure, visibility, terrain slope, and obstacle density, and its output parameters are the corresponding initial flight parameter set.
[0010] Optionally, the training data for the environment adaptation model includes flight sample data under different complex environments, and the flight sample data includes environmental parameters, flight parameters, and flight performance evaluation values.
[0011] Optionally, the semi-closed-loop control of the flight parameters includes collecting UAV flight status data in real time through the flight control unit, inputting the flight status data into a dynamic adjustment model, and the dynamic adjustment model outputting flight parameter correction values based on the gradient descent algorithm to dynamically correct the initial flight parameter set.
[0012] Optionally, the flight status data includes actual flight speed, actual attitude angle, actual position coordinates, remaining battery power, and mission execution progress.
[0013] Optionally, the closed-loop optimization control of the flight parameters includes constructing a reinforcement learning agent through the flight control unit, using the maximization of task execution accuracy and the minimization of energy consumption as the reward function, and optimizing the flight parameters in real time based on the Markov decision process; the state space of the reinforcement learning agent includes environmental dynamic change parameters, UAV flight state parameters and task state parameters, and the action space includes a set of flight parameter adjustment amounts.
[0014] Optionally, a machine learning-based intelligent control system for unmanned aerial vehicles includes: Initial adaptation module: used to determine whether the user has entered the drone flight mission command. If the user has entered the flight mission command, the module collects flight environment parameters and determines whether the drone is in a complex flight scenario. If it is in a complex flight scenario, the module controls the drone to enter the initial adaptation stage of complex flight mode. Dynamic adjustment module: Used to determine whether the flight attitude stability of the UAV is greater than the predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, the UAV is controlled to enter the dynamic adjustment stage of complex flight mode. Stability optimization module: used to determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, the drone is controlled to enter the stability optimization phase of the complex flight mode. Normal flight module: Used to determine whether the mission execution deviation of the UAV is less than the predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the UAV is controlled to exit the complex flight mode and enter the normal flight control mode.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a machine learning-based intelligent control method and system for unmanned aerial vehicles, which has the following beneficial effects: This invention employs a phased control strategy. The initial adaptation phase utilizes a pre-trained model for rapid environment adaptation; the dynamic adjustment phase dynamically corrects parameters based on real-time data; and the stabilization optimization phase achieves globally optimal control through reinforcement learning. These stages are seamlessly integrated, effectively improving the control timeliness and accuracy of UAVs in complex environments. By integrating multiple machine learning algorithms, the environment adaptation model possesses powerful environmental parameter mapping capabilities, the dynamic adjustment model achieves rapid response, and the reinforcement learning agent enables autonomous optimization. This addresses the problems of poor adaptability and weak generalization ability in traditional control methods, significantly improving the UAV's anti-interference capability and autonomous decision-making level. By combining closed-loop and open-loop control, both rapid response in the initial stage and precise optimization in subsequent stages are ensured, balancing control speed and accuracy, reducing flight deviation and energy consumption, and improving the UAV's mission execution reliability and endurance. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of a machine learning-based intelligent control method for unmanned aerial vehicles (UAVs) provided by the present invention.
[0018] Figure 2 This invention provides a structural framework diagram of an unmanned aerial vehicle (UAV) intelligent control system based on machine learning. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0020] This invention discloses a machine learning-based intelligent control method for unmanned aerial vehicles (UAVs), such as... Figure 1 As shown, it includes: Determine if the user has entered a drone flight mission command. If the user has entered a flight mission command, collect flight environment parameters and determine if the drone is in a complex flight scenario. If it is in a complex flight scenario, control the drone to enter the initial adaptation phase of the complex flight mode. Determine whether the flight attitude stability of the drone is greater than a predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, control the drone to enter the dynamic adjustment stage of complex flight mode. Determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, then control the UAV to enter the stable optimization phase of the complex flight mode. The system determines whether the drone's mission execution deviation is less than a predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the system controls the drone to exit the complex flight mode and enter the normal flight control mode.
[0021] Furthermore, the initial adaptation phase is used to execute a first control strategy, which includes open-loop control of UAV flight parameters based on a pre-trained environment adaptation model. The dynamic adjustment phase is used to execute the second control strategy, which includes semi-closed-loop control of flight parameters based on the dynamic output of real-time flight data and machine learning models. The stabilization optimization phase is used to execute a third control strategy, which includes closed-loop optimization control of flight parameters based on a reinforcement learning algorithm.
[0022] Furthermore, the open-loop control of the flight parameters includes controlling the UAV's power actuators, attitude actuators, and navigation actuators via a flight control unit (FCU) to control the UAV's operation with a pre-generated initial set of flight parameters.
[0023] Furthermore, the initial flight parameter set includes a flight speed reference value, an attitude angle reference value, a lift reference value, and a navigation path reference value.
[0024] Furthermore, the initial flight parameter set is calculated by the flight control unit through an environment adaptation model. The environment adaptation model is a neural network model trained based on a deep learning algorithm. Its input parameters include wind speed, wind direction, air pressure, visibility, terrain slope, and obstacle density, and its output parameters are the corresponding initial flight parameter set.
[0025] Furthermore, the training data for the environment adaptation model includes flight sample data under different complex environments, and the flight sample data includes environmental parameters, flight parameters, and flight performance evaluation values.
[0026] Furthermore, the semi-closed-loop control of the flight parameters includes collecting UAV flight status data in real time through the flight control unit, inputting the flight status data into a dynamic adjustment model, and the dynamic adjustment model outputting flight parameter correction values based on the gradient descent algorithm to dynamically correct the initial flight parameter set.
[0027] Furthermore, the flight status data includes actual flight speed, actual attitude angle, actual position coordinates, remaining battery power, and mission execution progress.
[0028] Furthermore, the closed-loop optimization control of the flight parameters includes constructing a reinforcement learning agent through the flight control unit, using the maximization of task execution accuracy and the minimization of energy consumption as reward functions, and optimizing the flight parameters in real time based on the Markov decision process; the state space of the reinforcement learning agent includes environmental dynamic change parameters, UAV flight state parameters and task state parameters, and the action space includes a set of flight parameter adjustment quantities.
[0029] In a specific implementation, the closed-loop optimization control of the flight parameters specifically includes: constructing a closed-loop optimization framework based on reinforcement learning and the EMD-EGMGWO-LSTM model. With multiple objectives of maximizing UAV mission execution accuracy, minimizing energy consumption, and optimizing motor noise suppression, a reinforcement learning agent dynamically coordinates the motor noise state and flight parameters in real time. Based on a Markov decision process (MDP), the collaborative dynamic optimization of flight parameters and noise control strategies is achieved. The specific formulas and parameter definitions are as follows: (a) State space S The state space encompasses four dimensions: dynamic environmental changes, UAV flight state, mission state, and motor noise state. Its expression is: ; For parameters of dynamic environmental change: ; in, The ambient wind speed (m / s) ranges from [0, 20]. The wind direction angle (°) ranges from [0, 360]. The density of air (kg / m³) ranges from 1.10 to 1.30. The ambient temperature (°C) ranges from -10 to 45. For the drone's flight status parameters: ; in, The flight speed (m / s) ranges from [5, 30]. The pitch angle (°) ranges from -30 to 30. The roll angle (°) ranges from -30 to 30. The flight altitude (m) ranges from [10, 500]. The height change rate (m / s) ranges from [-5, 5]. The remaining mass of the drone (kg) ranges from [1.0 to 5.0]. The output power of the motor (kW) ranges from [0.3 to 3.0]. The motor operating current (A) has a range of [5, 30]. The motor speed (r / min) ranges from [3000 to 15000]. Task status parameters: ; in, Define the target coordinates (m) and set the range of values according to the task scenario; The current task execution error (m) ranges from [0, 10]. The remaining time for the task (s), with a value range of [0, 3600]. The remaining battery charge (%), with a value range of [10, 100]. Motor noise status parameters: ; in, The fundamental frequency (Hz) of the periodic noise from the motor cooling duct. ,in, This refers to the number of stator slots in the motor (values range from 12 to 24). The motor speed (r / min) is given. The diameter of the cooling air duct (m, ranging from 0.05 to 0.2). The noise level is the resonance noise (dB) of the cooling air duct. ,in, This is the duct damping coefficient (value ranges from 0.1 to 0.5). The cross-sectional area of the cooling air duct (m², valued at 0.002~0.03). The total volume of the air duct (m³, with a value of 0.0001~0.001). The length of the air duct (m, ranging from 0.2 to 1.0). The sound pressure level (dB) of the motor structure dynamic noise. ,in, For bearing friction, rotor imbalance, etc. The time-domain signal (Pa) of each noise source, where n is the number of noise sources (values range from 3 to 6). The noise sound pressure level (dB) of the motor gearbox. ,in, This is the time-domain signal (Pa) of gear meshing noise. The measurement time is 0.1 to 1.0 seconds. The sound pressure level (dB) is the noise level caused by airflow disturbance. ,in, The airflow noise figure (values range from 0.001 to 0.01). The relative velocity of the airflow (m / s) The swept area of the propeller (m², valued at 0.1~0.5). These are the n intrinsic mode function components of the noise signal after EMD decomposition. ,in, The original noise signal, For residual components, The number of IMF components (values range from 4 to 10).
[0030] (ii) Action Space A
[0031] The action space is the set of flight parameters and motor control parameter adjustments, expressed as: ; in, The flight speed adjustment (m / s) ranges from [-2, 2]. The pitch angle adjustment is in degrees (°), with a range of [-5, 5]. The roll angle adjustment amount (°) ranges from [-5, 5]. The height change rate adjustment (m / s) ranges from [-1, 1]. This represents the motor power adjustment (kW), with a value range of [-0.2, 0.2]. This represents the motor speed adjustment amount (r / min), with a value range of [-500, 500]. The adjustment factor for the EGMGWO algorithm is set to a value in the range of [-0.1, 0.1].
[0032] (iii) Reward function R(S, A)
[0033] Retain the multi-objective weighted logic, remove combustion noise-related terms, and optimize energy consumption reward to adapt to battery characteristics: ; in, , , For the weighting coefficients, satisfying + + =1, with default values of 0.4, 0.3, and 0.3; the sub-reward functions are defined as follows: Task execution accuracy reward : ; Energy consumption reward Adapted to battery power consumption characteristics: ; in, The remaining battery level (%) after the action is performed; Initial battery level (%, valued at 100); Minimum allowable power consumption (%, valued at 10%). Noise suppression reward Update the total noise calculation logic: ; in, Total noise sound pressure level; The maximum permissible noise level (dB, value 90). The target noise level is 55 dB.
[0034] (iv) Transition probabilities of Markov decision processes
[0035] Using the original product form, the motor parameter transfer probability is supplemented: ; The transition probabilities of motor parameters follow a Gaussian distribution: ; in, =0.05, power transfer standard deviation, and the transfer standard deviation of other parameters are similarly taken as 0.1~0.3.
[0036] (iv) Closed-loop optimization control process
[0037] (1) Initialization: Set the parameters of the reinforcement learning agent, initialize the EMD-EGMGWO-LSTM noise prediction model, adapt the new state parameters to 15 nodes in the input layer, the number of hidden layer nodes is 10~100, the output layer has 1 node, the number of GWO algorithm population is 50, and the maximum number of iterations is 500. (2) State awareness: Real-time collection of environmental, flight, mission and motor noise data, namely mechanical noise, duct noise and airflow noise, and IMF components are obtained through EMD decomposition to construct the current state S; (3) Action decision: Based on the current state and accumulated reward, the agent uses a greedy strategy to increase the exploration rate. Initially 0.9, linearly decaying to 0.1, select the optimal action A; (4) Execution and feedback: The execution action adjusts the flight parameters and motor control parameters, the EMD-EGMGWO-LSTM model predicts noise changes, and generates an anti-phase control signal to cancel the noise through the vibration suppression device; the new state S′ and the reward value R are collected; (5) Model update: Based on the TD(λ) algorithm, λ=0.8, update the state value function and iteratively optimize the action strategy; (6) Termination judgment: Task completed ( If the battery level is below the threshold (SOC<10%), output the optimal parameters and strategy; otherwise, return to step (2) to continue closed-loop optimization.
[0038] Specifically, the predetermined stability threshold ranges from 0.75 to 0.9, preferably 0.85; the predetermined adjustment duration ranges from 20 to 40 seconds, preferably 30 seconds; and the predetermined deviation threshold ranges from 0.5 to 2 meters, preferably 1 meter.
[0039] In a specific implementation, a machine learning-based intelligent control method for unmanned aerial vehicles (UAVs) includes the following steps: Flight Mission Initiation and Scene Determination: The system determines whether the user has input a drone flight mission command. The user can input commands containing information such as the mission objective and flight path via remote control, ground station system, etc. If a mission command is detected, flight environment parameters are collected in real time, including wind speed, wind direction, air pressure, visibility, terrain slope, and obstacle density. Based on preset complex scene determination rules and combined with environmental parameters, the system determines whether the drone is in a complex flight scene. If it is in a complex flight scene, the drone is controlled to enter the initial adaptation phase of the complex flight mode; if it is in a normal flight scene, it directly enters the normal flight control mode.
[0040] Initial Adaptation Phase Switching: During the initial adaptation phase, the UAV executes the first control strategy, outputting an initial flight parameter set through a pre-trained environment adaptation model, and quickly adapts to the current environment using an open-loop control method. Simultaneously, the inertial measurement unit (IMU) collects the UAV's flight attitude data in real time and calculates the flight attitude stability. It is determined whether the flight attitude stability is greater than a predetermined stability threshold (range 0.75-0.9, preferably 0.85). If it is greater than the predetermined stability threshold, it indicates that the UAV has initially adapted to the environment, and the control system enters the dynamic adjustment phase of complex flight modes; if not, the control strategy of the initial adaptation phase continues to be executed.
[0041] Dynamic Adjustment Phase Switching: During the dynamic adjustment phase, the second control strategy is executed, dynamically correcting flight parameters through a semi-closed-loop control method. Real-time flight status data of the UAV, including actual flight speed, actual attitude angles, actual position coordinates, remaining battery power, and mission progress, is collected and input into the dynamic adjustment model. The dynamic adjustment model analyzes and processes the real-time data based on the gradient descent algorithm, outputting flight parameter correction values to continuously adjust the initial flight parameter set. It is determined whether the duration of the dynamic adjustment phase exceeds the predetermined adjustment duration (range 20-40 seconds, preferably 30 seconds). If it does, the UAV is controlled to enter the stable optimization phase of the complex flight mode; otherwise, the control strategy of the dynamic adjustment phase continues.
[0042] Stable Optimization Phase and Mode Exit: During the stable optimization phase, the third control strategy is executed to achieve high-precision flight through closed-loop optimization control. The reinforcement learning agent constructed by the FCU uses maximizing task execution accuracy and minimizing energy consumption as the reward function. Based on the Markov decision process, it combines dynamic environmental parameters, UAV flight state parameters, and task state parameters to optimize flight parameters in real time. The task execution deviation is calculated in real time, and it is determined whether the task execution deviation is less than a predetermined deviation threshold (ranging from 0.5-2m, preferably 1m). If it is less than the predetermined deviation threshold, it indicates that the UAV has stably completed the core aspects of the task or has adapted to the environment and can accurately execute the task. In this case, the UAV is controlled to exit the complex flight mode and enter the normal flight control mode. If the threshold is not reached, the control strategy of the stable optimization phase continues to be executed.
[0043] Furthermore, the environment adaptation model is trained based on deep learning algorithms, such as a hybrid model combining convolutional neural networks and recurrent neural networks. The training process includes: A large amount of flight sample data under different complex environments is collected. The flight sample data includes environmental parameters (input features), flight parameters (intermediate variables), and flight performance evaluation values (labels). The sample data is preprocessed by normalization, outlier handling, and other operations. The preprocessed sample data is divided into training set, validation set, and test set for model training, hyperparameter tuning, and performance verification. The model parameters are iteratively updated through the backpropagation algorithm until the model's prediction error meets the preset requirements.
[0044] Furthermore, the dynamic adjustment model employs a gradient descent algorithm based on the least squares method. By calculating the error between the predicted flight effect and the actual flight effect in real time, it dynamically adjusts the model weights and outputs the optimal flight parameter correction amount, ensuring that the UAV can respond and adjust quickly when the environment changes.
[0045] Furthermore, the training process of the reinforcement learning agent includes: defining the state space, action space, and reward function; initializing agent parameters; generating experience data through interaction with the environment, storing and sampling experience data using an experience replay mechanism; updating the agent's value function and policy function based on a temporal difference learning algorithm, continuously optimizing the control policy, and achieving a balance between task execution accuracy and energy consumption.
[0046] In a specific implementation, a machine learning-based intelligent control device for unmanned aerial vehicles (UAVs) includes a data acquisition module, a flight control unit (FCU), an actuation module, and a mission management module. These modules work together to achieve intelligent control of the UAV.
[0047] The data acquisition module consists of multiple sensors used to comprehensively collect environmental and UAV status data during flight. These include wind speed, wind direction, and barometric pressure sensors for collecting environmental meteorological parameters; visual sensors and obstacle detection sensors for collecting information on surrounding obstacles and terrain features; a GPS positioning module for obtaining the UAV's real-time position coordinates; an inertial measurement unit (IMU) for collecting flight status data such as attitude angles, angular velocity, and acceleration; and a battery level sensor for monitoring remaining battery power. Data from these sensors is transmitted to the FCU in real time, providing data support for control decisions.
[0048] Flight Control Unit (FCU): As the core control unit of the system, it integrates a scene judgment module, a mode switching module, and a machine learning processing module. The scene judgment module analyzes environmental parameters based on preset rules to determine the flight scene type; the mode switching module smoothly switches between different control modes according to the UAV's flight status (such as attitude stability, runtime, mission deviation, etc.); the machine learning processing module has built-in environment adaptation models, dynamic adjustment models, and reinforcement learning agents, and is responsible for calculating, correcting, and optimizing flight parameters.
[0049] The actuation module includes a power actuator, an attitude actuator, and a navigation actuator, corresponding to the UAV's power system, attitude control system, and navigation system, respectively. The power actuator controls the motor speed, adjusting flight speed and lift; the attitude actuator controls the servo motor movements, adjusting the UAV's pitch, roll, and yaw angles; the navigation actuator controls the navigation system, correcting the flight path. The actuation module receives control commands from the FCU and drives the corresponding actuators to achieve precise execution of flight parameters.
[0050] The mission management module receives flight mission commands sent by users via ground station or remote controller, parses information such as mission objectives, flight paths, and mission priorities; monitors the progress of UAV mission execution in real time, calculates mission execution deviations by combining GPS positioning data and mission objective coordinates; and provides mission adjustment suggestions based on remaining battery power and mission progress to ensure safe and efficient mission completion.
[0051] In a specific implementation, a machine learning-based intelligent control system for unmanned aerial vehicles (UAVs) is, for example... Figure 2 As shown, it includes: Initial adaptation module: used to determine whether the user has entered the drone flight mission command. If the user has entered the flight mission command, the module collects flight environment parameters and determines whether the drone is in a complex flight scenario. If it is in a complex flight scenario, the module controls the drone to enter the initial adaptation stage of complex flight mode. Dynamic adjustment module: Used to determine whether the flight attitude stability of the UAV is greater than the predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, the UAV is controlled to enter the dynamic adjustment stage of complex flight mode. Stability optimization module: used to determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, the drone is controlled to enter the stability optimization phase of the complex flight mode. Normal flight module: Used to determine whether the mission execution deviation of the UAV is less than the predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the UAV is controlled to exit the complex flight mode and enter the normal flight control mode.
[0052] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0053] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A machine learning-based intelligent control method for unmanned aerial vehicles (UAVs), characterized in that, include: Determine if the user has entered a drone flight mission command. If the user has entered a flight mission command, collect flight environment parameters and determine if the drone is in a complex flight scenario. If it is in a complex flight scenario, control the drone to enter the initial adaptation phase of the complex flight mode. Determine whether the flight attitude stability of the drone is greater than a predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, control the drone to enter the dynamic adjustment stage of complex flight mode. Determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, then control the UAV to enter the stable optimization phase of the complex flight mode. The system determines whether the drone's mission execution deviation is less than a predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the system controls the drone to exit the complex flight mode and enter the normal flight control mode.
2. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 1, characterized in that, The initial adaptation phase is used to execute a first control strategy, which includes open-loop control of UAV flight parameters based on a pre-trained environment adaptation model. The dynamic adjustment phase is used to execute the second control strategy, which includes semi-closed-loop control of flight parameters based on the dynamic output of real-time flight data and machine learning models. The stabilization optimization phase is used to execute a third control strategy, which includes closed-loop optimization control of flight parameters based on a reinforcement learning algorithm.
3. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 2, characterized in that, The open-loop control of the flight parameters includes controlling the UAV's power actuators, attitude actuators, and navigation actuators through the flight control unit to control the UAV's operation with a pre-generated initial set of flight parameters.
4. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 3, characterized in that, The initial flight parameter set includes reference values for flight speed, attitude angle, lift, and navigation path.
5. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 4, characterized in that, The initial flight parameter set is calculated by the flight control unit through an environment adaptation model. The environment adaptation model is a neural network model trained based on a deep learning algorithm. Its input parameters include wind speed, wind direction, air pressure, visibility, terrain slope, and obstacle density, and its output parameters are the corresponding initial flight parameter set.
6. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 5, characterized in that, The training data for the environment adaptation model includes flight sample data under different complex environments, and the flight sample data includes environmental parameters, flight parameters, and flight performance evaluation values.
7. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 2, characterized in that, The semi-closed-loop control of the flight parameters includes collecting UAV flight status data in real time through the flight control unit, inputting the flight status data into a dynamic adjustment model, and the dynamic adjustment model outputting flight parameter correction values based on the gradient descent algorithm to dynamically correct the initial flight parameter set.
8. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 7, characterized in that, The flight status data includes actual flight speed, actual attitude angle, actual position coordinates, remaining battery power, and mission execution progress.
9. The machine learning-based intelligent control method for unmanned aerial vehicles according to claim 2, characterized in that, The closed-loop optimization control of the flight parameters includes constructing a reinforcement learning agent through the flight control unit, using the maximization of task execution accuracy and the minimization of energy consumption as the reward function, and optimizing the flight parameters in real time based on the Markov decision process; the state space of the reinforcement learning agent includes environmental dynamic change parameters, UAV flight state parameters and task state parameters, and the action space includes a set of flight parameter adjustment variables.
10. A machine learning-based intelligent control system for unmanned aerial vehicles (UAVs), characterized in that, include: Initial adaptation module: used to determine whether the user has entered the drone flight mission command. If the user has entered the flight mission command, the module collects flight environment parameters and determines whether the drone is in a complex flight scenario. If it is in a complex flight scenario, the module controls the drone to enter the initial adaptation stage of complex flight mode. Dynamic adjustment module: Used to determine whether the flight attitude stability of the UAV is greater than the predetermined stability threshold. If the flight attitude stability is greater than the predetermined stability threshold, the UAV is controlled to enter the dynamic adjustment stage of complex flight mode. Stability optimization module: used to determine whether the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration. If the runtime of the dynamic adjustment phase is greater than the predetermined adjustment duration, the drone is controlled to enter the stability optimization phase of the complex flight mode. Normal flight module: Used to determine whether the mission execution deviation of the UAV is less than the predetermined deviation threshold. If the mission execution deviation is less than the predetermined deviation threshold, the UAV is controlled to exit the complex flight mode and enter the normal flight control mode.