Artificial intelligence unmanned aerial vehicle ground station control method, system, device and medium

Abstract

The invention relates to an artificial intelligence unmanned aerial vehicle ground station control method, system and device and a medium, and belongs to the technical field of unmanned aerial vehicle control. Performing space-time alignment on the unmanned aerial vehicle sensor data, and performing normalization processing on a heterogeneous data format to generate a standardized space-time data packet; constructing a dynamic obstacle map based on the historical flight database and the standardized spatio-temporal data packet; inputting the standardized spatio-temporal data packet into a machine learning model, predicting the flight path deviation and the equipment fault risk of the unmanned aerial vehicle, and obtaining a prediction result set; based on a preset task target of the unmanned aerial vehicle, generating a corresponding control instruction set according to the dynamic obstacle map and the prediction result set; and converting the control instruction set into an unmanned aerial vehicle flight action sequence, and sending the unmanned aerial vehicle flight action sequence to the unmanned aerial vehicle for flight state adjustment. According to the invention, the safety and adaptability of the unmanned aerial vehicle system are improved.

Description

Technical Field

[0001] The present application relates to the field of drone control technology, and in particular to an artificial intelligence drone ground station control method, system, equipment and medium. Background Art

[0002] With the rapid development of drone technology, drones are increasingly being used in civil, commercial, and military fields, playing a particularly important role in agriculture, environmental monitoring, logistics distribution, disaster relief, and security surveillance. Autonomous flight capabilities are a key technology for drones to achieve intelligent operation. However, ensuring drone safety and effective mission execution in complex and dynamic flight environments relies on a highly integrated and intelligent ground station control system. As the core link in drone operation and control, ground stations handle multiple functions, including data processing, flight monitoring, mission planning, and command issuance. As drone missions become increasingly complex, traditional flight control methods are no longer able to meet the demand for efficient, precise, and intelligent control, especially in complex flight scenarios such as multi-missions, multiple obstacles, and dynamic environments.

[0003] Currently, drones are typically equipped with a variety of sensors to collect environmental information and motion data during flight. However, existing drone flight control systems rely solely on data from these sensors to plan a preset flight path, lacking real-time flight adjustments and intelligent decision-making capabilities. This often hinders timely flight plan adjustments when encountering obstacles, environmental factors, or equipment failures that change in real time during flight. This significantly reduces the adaptability and safety of drone systems during long missions. Summary of the Invention

[0004] In order to improve the safety and adaptability of drone systems, this application provides an artificial intelligence drone ground station control method, system, device and medium.

[0005] In the first aspect, the present application provides an artificial intelligence drone ground station control method, which adopts the following technical solutions: A method for controlling an artificial intelligence unmanned aerial vehicle ground station, the method comprising: Collecting drone sensor data; the drone sensor data includes positioning data, inertial measurement data, environmental perception data, and drone status data; Performing spatiotemporal alignment on the drone sensor data and normalizing heterogeneous data formats to generate standardized spatiotemporal data packets; constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package; Inputting the standardized spatiotemporal data packet into a machine learning model to predict the flight path deviation and equipment failure risk of the UAV, thereby obtaining a prediction result set; Based on the preset mission objectives of the UAV, a corresponding control instruction set is generated according to the dynamic obstacle map and the prediction result set; The control instruction set is converted into a UAV flight action sequence and sent to the UAV for flight state adjustment.

[0006] By implementing this technical solution, integrating close collaboration between the ground station and the drone, and through precise data collection, real-time environmental modeling and risk prediction, dynamic control command generation, and online optimization mechanisms, autonomous flight control of the drone in complex environments is achieved. By effectively identifying and responding to in-flight obstacles, path deviations, and equipment failure risks, the drone not only flies autonomously according to predetermined mission objectives but also promptly adjusts its flight strategy in the face of unexpected changes, significantly improving its safety and adaptability.

[0007] Optionally, the step of constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package includes: Collect obstacle coordinate records and sensor false alarm marks in the historical flight database, perform feature cleaning, and obtain a historical obstacle coordinate list and historical obstacle weight labels; Obtaining point cloud data and camera image streams based on the environmental perception data in the standardized spatiotemporal data packet; downsampling and ground segmentation processing are performed on the point cloud data to extract a candidate point set of non-ground obstacles; performing target re-identification on the non-ground obstacle candidate point set based on the camera image stream to obtain a fusion confidence score matrix; Separating static obstacles from dynamic obstacles based on the fused confidence score matrix and historical obstacle weight labels, and obtaining a static obstacle coordinate list and a dynamic obstacle tracking information table; Probabilistically fusing the historical obstacle coordinate list with the static obstacle coordinate list to construct a rasterized static obstacle map; Mapping the dynamic obstacle tracking information table to a dynamic layer to generate a risk heat map containing predicted trajectories; The static obstacle map, dynamic layer and risk heat map are encoded into a standard format and output at a preset frequency.

[0008] By integrating historical flight data with real-time sensor perception and building a dynamic obstacle map, the aforementioned technical solution provides drones with precise environmental awareness and flight path optimization, ensuring the system can accurately identify obstacles and adjust flight strategies in real time in complex and dynamic flight environments. Furthermore, standardized data output and the generation of risk heat maps enable the flight control system to make efficient and timely obstacle avoidance decisions, improving flight safety and mission reliability.

[0009] Optionally, the machine learning model includes a path deviation prediction model and an equipment failure warning model; the input of the path deviation prediction model is positioning data, and the output is a flight path deviation prediction result within a preset time in the future; the input of the equipment failure warning model is environmental perception data, and the output is an equipment failure risk probability prediction result.

[0010] By employing the aforementioned technical solutions, based on the combined application of a path deviation prediction model and an equipment failure warning model, the UAV system can monitor path deviations and equipment health status in real time during flight, proactively predicting potential risks. Path deviation prediction helps the flight control system adjust the flight trajectory based on real-time data to avoid deviations from the target, while the equipment failure warning model assesses equipment health in real time, proactively identifying potential battery, temperature, or engine failure risks and enabling timely intervention. These predictive capabilities significantly enhance the safety and reliability of the UAV system and improve mission execution efficiency.

[0011] Optionally, based on the preset mission objectives of the UAV, the step of generating a corresponding control instruction set according to the dynamic obstacle map and the prediction result set includes: Analyze the preset mission objectives of the UAV and generate quantitative track parameters; Based on the static obstacle distribution of the dynamic obstacle map, an initial waypoint list is generated and the spatiotemporal conflict areas are marked, and an initial track file with conflict markings is output; Combining the prediction result set and the initial track file, calculating the obstacle collision probability, equipment failure impact factor and environmental interference coefficient, and outputting a comprehensive risk score level; A corresponding control instruction set is generated according to the comprehensive risk score level and the quantified track parameters.

[0012] By employing this technical solution, based on quantified trajectory parameters and comprehensive risk assessment, the system can dynamically adjust the flight path and generate real-time control instructions based on the actual flight mission requirements. This solution not only improves flight control flexibility and safety, but also enables timely obstacle avoidance decisions in complex environments, ensuring that the drone can complete its mission safely and effectively.

[0013] Optionally, the step of generating a corresponding control instruction set according to the comprehensive risk score level and the quantified track parameters includes: If the comprehensive risk score is low, the quantified track parameters are called to generate heading angle and speed fine-tuning instructions; If the comprehensive risk score is medium or high, the global trajectory is reconstructed in combination with the flight path deviation prediction result to generate an obstacle avoidance trajectory or an emergency return instruction.

[0014] By adopting the above technical solution, an intelligent decision-making mechanism based on risk assessment is introduced into flight control, effectively addressing potential risks in different flight environments. In low-risk situations, the system fine-tunes flight parameters to improve mission execution accuracy and efficiency. In medium- and high-risk environments, the system ensures flight safety through real-time path deviation prediction and global trajectory reconstruction, enabling dynamic obstacle avoidance and emergency return command generation. This flexible control strategy enables the system to adapt to environmental changes in real time, avoid unforeseen risks, and improve the drone's autonomous flight capabilities and mission success rate in complex environments.

[0015] Optionally, after the step of converting the control instruction set into a UAV flight action sequence and sending the sequence to the UAV for flight state adjustment, the method further includes: monitoring the execution status of the UAV and generating control error feedback data; performing weight updating on the machine learning model based on the control error feedback data; The updated machine learning model is validated in the digital twin environment to obtain an optimized machine learning model.

[0016] By adopting the above technical solution, real-time monitoring of execution status and generation of control error feedback data helps identify deviations between the flight path and control instructions and make timely adjustments. By updating the weights of the machine learning model based on this feedback data, the system can continuously improve prediction accuracy and achieve precise control of flight paths and equipment status. At the same time, verification in the digital twin environment further ensures the effectiveness and robustness of the updated model in actual flight. This closed-loop self-optimization mechanism not only improves flight safety but also enhances the drone's adaptability in dynamic environments, greatly improving the success rate and reliability of flight missions.

[0017] In the second aspect, this application provides an artificial intelligence drone ground station control system, which adopts the following technical solutions: An artificial intelligence UAV ground station control system, the control system comprising: A data acquisition module is used to collect drone sensor data; the drone sensor data includes positioning data, inertial measurement data, environmental perception data and drone status data; A data standardization module is used to perform spatiotemporal alignment on the drone sensor data and normalize heterogeneous data formats to generate standardized spatiotemporal data packets; A dynamic obstacle map construction module, configured to construct a dynamic obstacle map based on a historical flight database and the standardized spatiotemporal data package; A prediction module, configured to input the standardized spatiotemporal data packet into a machine learning model to predict the flight path deviation and equipment failure risk of the UAV, and obtain a prediction result set; A control instruction generation module is used to generate a corresponding control instruction set based on the preset mission objectives of the UAV and the dynamic obstacle map and the prediction result set; The flight state adjustment module is used to convert the control instruction set into a UAV flight action sequence and send it to the UAV for flight state adjustment.

[0018] Optionally, the control system further includes: An execution monitoring module, configured to monitor the execution status of the UAV and generate control error feedback data; A weight updating module, configured to update the weight of the machine learning model based on the control error feedback data; The model verification module is used to verify the validity of the updated machine learning model in the digital twin environment to obtain the optimized machine learning model.

[0019] In a third aspect, the present application provides a computer device that adopts the following technical solution: A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to the first aspect.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium stores a computer program capable of being loaded by a processor and executing any one of the methods in the first aspect.

[0021] In summary, this application has at least one of the following beneficial technical effects: It comprehensively utilizes sensor data, historical flight databases, and machine learning models to construct a dynamic obstacle map in real time and predict flight path deviations and equipment failure risks, thereby providing precise control instructions for the drone. This technical solution, through seamless collaboration between the ground station and the drone, enables it to cope with various unpredictable flight challenges, ensuring the drone's safe obstacle avoidance in complex environments and responding to flight deviations and equipment failure risks in real time. This greatly enhances the drone's autonomous decision-making capabilities and the reliability of its mission execution, and has broad application prospects, such as in drone cruising, monitoring, and emergency response tasks in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a first flow chart of an artificial intelligence UAV ground station control method according to one of the embodiments of the present application.

[0023] Figure 2 This is a second flow chart of an artificial intelligence UAV ground station control method according to one of the embodiments of the present application.

[0024] Figure 3 This is a third flow chart of an artificial intelligence UAV ground station control method according to one of the embodiments of the present application.

[0025] Figure 4 This is a fourth flow chart of an artificial intelligence UAV ground station control method according to one of the embodiments of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-4 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0027] The embodiments of the present application disclose a method for controlling an artificial intelligence drone ground station.

[0028] Reference Figure 1 , a control method for an artificial intelligence UAV ground station, the control method comprising: Step S101, collecting drone sensor data; Drone sensor data includes positioning data, inertial measurement data, environmental perception data, and drone status data. Ground stations receive real-time sensor data from drones. Since drones are typically equipped with multiple sensors to collect different types of data, this data primarily includes positioning data (GPS), inertial measurement data (IMU), environmental perception data (such as LiDAR point cloud data and camera image streams), and drone status data. Each sensor has different functions and uses, but together they provide the drone with critical data about its flight status and surrounding environment.

[0029] Specifically, GPS provides the drone's location information, calculating its longitude, latitude, and altitude above the Earth's surface through the Global Positioning System. The inertial measurement unit (IMU) uses sensors like accelerometers and gyroscopes to measure the drone's acceleration, angular velocity, and orientation, helping to calculate its flight attitude and trajectory. LiDAR uses lasers to scan the surrounding environment, generating high-precision point cloud data and capturing obstacles on the ground or in the air, such as buildings and trees. Drone status data includes internal battery voltage, temperature, and engine status.

[0030] Step S102: performing spatiotemporal alignment on the drone sensor data and normalizing the heterogeneous data formats to generate standardized spatiotemporal data packets; While collected sensor data can provide rich information, it is often heterogeneous due to differences in sources and formats. Directly applying this data to subsequent analysis can lead to errors and inconsistencies. To ensure that this data can be effectively fused at the same time and in the same coordinate system, it is necessary to align them in time and space.

[0031] Specifically, spatiotemporal alignment consists of two key components: time synchronization and spatial coordinate alignment. Time synchronization involves matching and aligning the data collected by each sensor to the same timestamp, ensuring that the time windows of different data sources are consistent. For example, GPS and IMUs may have time delays, so time alignment must be achieved through interpolation or synchronization algorithms (such as Kalman filtering). Spatial coordinate alignment involves converting data from different sensors into a unified spatial coordinate system, typically based on the drone's own coordinate system, to ensure that all sensor data has a consistent reference frame in three-dimensional space.

[0032] Furthermore, data normalization standardizes the format and range of data output by different sensors so that data of different formats can be integrated and compared. For example, the point cloud data generated by LiDAR and the acceleration data from IMU may use different units and scales. The normalization step brings them into a consistent numerical range, facilitating subsequent calculations and analysis.

[0033] Step S103, constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package; The key to building a dynamic obstacle map lies in leveraging environmental data collected by sensors and historical flight data to update and represent obstacles in the flight environment in real time. A dynamic obstacle map not only includes the geographic location of static obstacles but also needs to reflect changes in obstacles over time and space, such as weather and lighting changes, or dynamic objects (such as birds and flying objects).

[0034] Specifically, a preliminary obstacle map is constructed using environmental perception data collected by sensors such as LiDAR. This data, through spatial parsing and clustering algorithms, identifies the location and form of objects and obstacles in the environment, such as roads, buildings, or trees. Furthermore, data from historical flight databases can support dynamic obstacle maps. This is particularly true in unknown or unstable environments, where historical data can help infer changing patterns of obstacles, particularly their potential impact on the flight path.

[0035] Step S104: Input the standardized spatiotemporal data packet into the machine learning model to predict the flight path deviation and equipment failure risk of the UAV, and obtain a prediction result set; Among them, the machine learning model includes a path prediction model (the input is the positioning data sequence, and the output is the path deviation within a preset time in the future) and a fault warning model (the input is drone status data such as battery voltage, temperature, engine status, etc., and the output is the fault probability).

[0036] Specifically, flight path deviation prediction analyzes the drone's historical flight data (such as position changes, speed, acceleration, etc.) and combines it with real-time sensor data (such as IMU and GPS). Using regression analysis, neural networks, or other machine learning models, it predicts the extent to which the drone is likely to deviate from its planned path in the next few minutes. This prediction helps the flight control system adjust the flight trajectory in a timely manner to avoid the danger of straying from the target. Equipment failure risk prediction analyzes sensor data, particularly data related to the drone's health (such as battery voltage, temperature, and engine status), to assess the current health status and predict the probability of failure. For example, by using a fault detection model, problems such as low battery or abnormal motor temperature can be predicted, prompting timely warnings or adjustments to the flight plan.

[0037] Step S105, based on the preset mission objectives of the UAV, a corresponding control instruction set is generated according to the dynamic obstacle map and the prediction result set; Based on the dynamic obstacle map and flight path prediction results, the system generates a corresponding set of control instructions according to the preset mission objectives. This solution combines multiple factors, including real-time environmental data, flight mission objectives, and equipment health status, to generate the optimal flight control instructions.

[0038] For example, if the mission objective is to perform a fixed-point cruise and the obstacle map shows that there are obstacles on the path, the system may adjust the flight path or control the speed based on the type and location of the obstacle. If the system predicts that a device is about to fail, the control instructions may include changing the flight direction or speed or issuing an emergency return command. These control instruction sets are designed to take into account all environmental factors and ensure that the drone can complete its mission safely and stably under various flight conditions.

[0039] Step S106: convert the control instruction set into a UAV flight action sequence and send it to the UAV for flight state adjustment.

[0040] Among them, the control instruction set includes multiple instructions, involving the UAV's flight parameters such as speed, heading, attitude, and tilt angle. The UAV needs to adjust its flight status according to these instructions to complete the flight mission.

[0041] Specifically, the UAV's flight control system converts control instructions into a sequence of flight maneuvers. Using flight control algorithms, these instructions are precisely translated into specific actions, such as steering control and motor speed adjustment. This process requires the control system to be extremely precise, capable of instantaneously adjusting flight parameters and responding to environmental changes in real time.

[0042] The above-mentioned implementation integrates close collaboration between the ground station and the drone, enabling autonomous flight control in complex environments through precise data collection, real-time environmental modeling and risk prediction, dynamic control command generation, and online optimization. By effectively identifying and addressing in-flight obstacles, path deviations, and equipment failure risks, the drone not only flies autonomously according to predetermined mission objectives but also promptly adjusts its flight strategy in the face of unexpected changes, significantly improving its safety and adaptability.

[0043] Reference Figure 2 As an implementation of step S103, the step of constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package includes: Step S201: Obstacle coordinate records and sensor false alarm marks are collected from the historical flight database, and feature cleaning is performed to obtain a historical obstacle coordinate list and historical obstacle weight labels; Obstacle coordinates and sensor false alarm flags are extracted from the historical flight database. The historical flight database contains obstacle data detected during previous flights. Each obstacle record may be accompanied by a sensor false alarm flag, which can interfere with obstacle map construction. To improve map accuracy, feature cleaning is required on this historical data to remove low-confidence or false alarm obstacle data, retaining only high-quality obstacle coordinate data.

[0044] For example, suppose some records in the historical database indicate that the GPS or IMU sensor mistakenly marked a dynamic object (such as a moving car) as a static obstacle. Through the feature cleaning process, the system identifies these false positives and removes them, ensuring that subsequent map construction relies only on accurate obstacle coordinates.

[0045] Step S202, obtaining point cloud data and camera image stream based on the environmental perception data in the standardized spatiotemporal data packet; The standardized spatiotemporal data package is a dataset that has undergone spatiotemporal alignment and data normalization, and includes real-time data from sensors such as LiDAR and cameras. By extracting point cloud data (LiDAR) and image stream data (camera), the system can obtain high-precision perception data about the current environment.

[0046] Specifically, LiDAR can provide a high-density point cloud dataset from real-time sensor data, representing the spatial structure of the surrounding environment. At the same time, the camera image stream provides visual information about the environment, such as the shape and color of objects. By combining these two, the system can form a comprehensive environmental perception input.

[0047] Step S203: downsampling and ground segmentation processing are performed on the point cloud data to extract a set of candidate points of non-ground obstacles; Point cloud data is often very large, so it needs to be downsampled to reduce the computational burden while ensuring that important information is not lost. Ground segmentation is used to identify and separate ground points from non-ground obstacle points. This allows the system to focus only on non-ground points that may be obstacles, such as buildings, trees, or vehicles.

[0048] For example, if the LiDAR point cloud data represents a city street environment, the ground segmentation processing will identify the ground (such as points on the road) and remove it, leaving the point cloud data of non-ground obstacles such as buildings and trees, which will serve as the obstacle candidate point set.

[0049] Step S204: re-identify the non-ground obstacle candidate point set based on the camera image stream to obtain a fusion confidence score matrix; The camera image stream provides additional visual information for object re-identification of non-ground obstacle candidates in the point cloud data. Using deep learning models such as YOLOv5, the system can identify specific objects (such as cars and pedestrians) in the point cloud and assign a confidence score to each obstacle candidate. These scores reflect the probability of each obstacle being correctly identified and are used for subsequent obstacle classification and processing.

[0050] For example, suppose that in the camera image stream, the YOLOv5 model identifies a moving car, and its corresponding point cloud data is also marked as a candidate point set for the car. The system assigns a high confidence score to these candidate points, indicating that they are more likely to be identified as obstacles.

[0051] Step S205: Separate static obstacles from dynamic obstacles based on the fused confidence score matrix and historical obstacle weight labels, and obtain a static obstacle coordinate list and a dynamic obstacle tracking information table; Based on the fused confidence score matrix, the system can determine whether each obstacle is static or dynamic based on historical obstacle weight labels. Dynamic obstacles are typically objects whose position changes over multiple time frames, while static obstacles remain constant. By separating static and dynamic obstacles, the system can further track dynamic obstacles and assign a unique tracking identifier to each dynamic obstacle.

[0052] For example, if an obstacle (such as a car) changes position in several consecutive frames, the system will mark it as a dynamic obstacle, assign it a unique tracking ID, calculate its speed and motion trajectory, and obtain dynamic obstacle tracking information; a building whose position remains unchanged will be marked as a static obstacle.

[0053] Step S206: Probabilistically fuse the historical obstacle coordinate list with the static obstacle coordinate list to construct a rasterized static obstacle map; Among them, the construction of the static obstacle map relies on the fusion of historical obstacle coordinates and real-time detected static obstacle coordinates. Through the probabilistic fusion method, combined with historical weight labels and real-time obstacle counting, the system can generate a static obstacle map in the gridded space. This map is used to represent the probability of obstacles in each grid area, thereby providing accurate obstacle avoidance information for the flight control system.

[0054] Specifically, in a rasterized map, the value of each grid cell represents the probability of the existence of an obstacle in that area. For example, a grid cell in one area may represent a high probability of a building, while a grid cell in another area may represent a low probability of a road.

[0055] Step S207: Map the dynamic obstacle tracking information table to the dynamic layer to generate a risk heat map containing the predicted trajectory; The dynamic obstacle tracking information table contains the trajectory and predicted path of each dynamic obstacle. This information is mapped onto a dynamic layer to generate a risk heat map. This heat map displays the risk level of different areas, helping the flight control system identify high-risk areas and implement obstacle avoidance or emergency measures. For example, the predicted trajectory of a dynamic obstacle (such as a moving car) is annotated on the heat map, showing the areas these objects may pass through. High-risk areas may appear red, and the system will adjust the flight strategy based on this information.

[0056] Step S208: Encode the static obstacle map, dynamic layer, and risk heat map into a standard format and output them at a preset frequency.

[0057] Specifically, all constructed map information, including static obstacle maps, dynamic layers, and risk heat maps, needs to be encoded into a standardized data format (such as JSON format) and output to the flight control system at a preset frequency, and the output frequency must be synchronized with the drone control instruction cycle.

[0058] In this implementation, integrating historical flight data with real-time sensor perception to construct a dynamic obstacle map provides the drone with precise environmental awareness and flight path optimization, ensuring the system can accurately identify obstacles and adjust flight strategies in real time in complex and dynamic flight environments. Furthermore, standardized data output and the generation of risk heat maps enable the flight control system to make efficient and timely obstacle avoidance decisions, improving flight safety and mission reliability.

[0059] As an implementation method of the machine learning model, the machine learning model includes a path deviation prediction model and an equipment failure warning model; the input of the path deviation prediction model is positioning data, and the output is the flight path deviation prediction result within a preset time in the future; the input of the equipment failure warning model is environmental perception data, and the output is the equipment failure risk probability prediction result.

[0060] As one implementation of the path deviation prediction model, its training process primarily relies on historical UAV flight data and real-time sensor data. First, a large amount of historical flight data must be collected, including the UAV's position, velocity, acceleration, and other data under different conditions. Next, the system compares this data with real-time sensor data (such as IMU and GPS) and uses machine learning methods such as regression analysis or neural networks to build a model to predict the UAV's likely path deviation over the next period of time. For example, the use of a long short-term memory (LSTM) network can effectively process time series data, learn the UAV's motion patterns, and thus predict future path changes. During training, the model is optimized based on the actual path deviations from historical flight data, and parameters are continuously adjusted until the model can accurately predict future path deviations.

[0061] For example, if the training data includes information about a drone's flight path, speed, and acceleration under different weather conditions, the model will learn how different environmental factors affect the flight path. Based on this data, the trained LSTM model can predict the drone's path deviation within the next five minutes, thereby helping the flight control system adjust the flight trajectory in a timely manner.

[0062] As one implementation of the equipment failure early warning model, its training relies on drone health data, including battery voltage, temperature, and engine status. This sensor data provides critical information about the drone's health. The training process begins by collecting historical data under different device states and annotating fault and non-fault conditions. During training, classification algorithms (such as random forests, support vector machines (SVMs), or neural networks) are used to predict the probability of device failure. By analyzing the relationship between sensor data such as battery voltage and temperature and the occurrence of faults, the model learns how these features affect the health of the device and predicts high-risk failures before they occur. The training dataset contains data under normal operating conditions and critical data when faults occur. The model extracts useful features from this data and outputs a probability value for device failure.

[0063] For example, assuming that the training data includes failure cases when the drone battery voltage is lower than a certain value, or failure cases when the engine temperature is abnormal, the trained model can identify these key data points and issue equipment failure warnings in a timely manner.

[0064] In the above implementation, the combined application of a path deviation prediction model and an equipment failure warning model enables the UAV system to monitor path deviation and equipment health status in real time during flight, predicting potential risks in advance. Path deviation prediction helps the flight control system adjust the flight trajectory based on real-time data to avoid deviation from the target, while the equipment failure warning model assesses equipment health in real time, identifying potential failure risks such as battery, temperature, or engine failures in advance and enabling timely intervention. These predictive capabilities significantly enhance the safety and reliability of the UAV system and improve the efficiency of flight mission execution.

[0065] Reference Figure 3 As an implementation of step S105, based on the preset mission objectives of the UAV, the step of generating a corresponding control instruction set according to the dynamic obstacle map and the prediction result set includes: Step S301, analyzing the preset mission objectives of the UAV and generating quantitative track parameters; A drone's pre-set mission objectives typically include the flight area, destination, and mission requirements (e.g., covering a specific area, avoiding specific obstacles, etc.). Based on these mission objectives, the system generates quantified track parameters, including a target coordinate set (indicating the target location the drone needs to reach), coverage density (indicating the degree of coverage of the mission area), and priority weights (indicating the relative importance of different objectives within the mission). These quantified track parameters provide a clear basis for subsequent flight path planning and decision-making.

[0066] For example, suppose a drone's mission is to cover a specific area while avoiding obstacles. The mission objectives might include multiple coordinate points. By parsing these objectives, the system generates a target coordinate set. If the area requires frequent surveillance, the coverage density will be higher. Priority weights might be set based on the urgency and importance of the mission (for example, prioritizing coverage of a specific area).

[0067] Step S302: Based on the static obstacle distribution of the dynamic obstacle map, an initial waypoint list is generated and the spatiotemporal conflict areas are marked, and an initial track file with conflict markings is output; Among them, the static obstacle distribution information extracted from the dynamic obstacle map is used to generate the drone's initial waypoint list. These waypoints are selected as key locations that the drone needs to pass through during flight. The system analyzes the path between the waypoints and marks possible spatiotemporal conflict areas, that is, areas that may collide with static obstacles. The final output initial track file contains these waypoints and conflict markers, providing a basis for subsequent path optimization and risk assessment. Assume that the initial track includes multiple waypoints from the starting point to the end point, and one section of the path may pass through a building area. Through static obstacle distribution and path analysis, the system will mark this path segment as a spatiotemporal conflict area to avoid overlap between the flight path and obstacles.

[0068] Step S303: combining the prediction result set and the initial track file, calculating the obstacle collision probability, equipment failure impact factor, and environmental interference coefficient, and outputting a comprehensive risk score level; The system combines path prediction and equipment failure prediction results with information from the initial track file to perform a risk assessment. The obstacle collision probability is calculated using an algorithm that detects the spatiotemporal overlap between the drone's predicted path and the trajectory of dynamic obstacles. The equipment failure impact factor is assessed by analyzing data such as battery, temperature, and engine status. The environmental interference factor considers external factors such as weather and signal interference. After integrating these factors, the system outputs a comprehensive risk score, which helps the flight control system assess the overall risk level during flight.

[0069] For example, suppose the flight path passes through an area containing a dynamic obstacle (e.g., a moving car) and the prediction model indicates that the battery charge is about to decrease. Combining this information, the system calculates a collision probability, equipment failure risk, and environmental interference factor, ultimately outputting an overall risk score (e.g., "medium risk").

[0070] Step S304: Generate a corresponding control instruction set based on the comprehensive risk score level and the quantified track parameters.

[0071] Based on the comprehensive risk score (e.g., low, medium, or high) and quantified trajectory parameters, the system generates a corresponding set of control instructions. For low-risk levels, the system generates heading and speed fine-tuning commands to ensure the drone stays on the intended path. For medium or high-risk levels, the system reconstructs the trajectory based on the predicted path deviations and generates obstacle avoidance trajectories or emergency return commands to ensure the drone safely avoids obstacles and returns in the event of a malfunction. Ultimately, these control instructions are sent to the flight control system to execute the corresponding flight operations.

[0072] In this implementation, based on quantified trajectory parameters and comprehensive risk assessment, the system dynamically adjusts the flight path and generates real-time control instructions based on the actual flight mission requirements. This solution not only improves flight control flexibility and safety, but also enables timely obstacle avoidance decisions in complex environments, ensuring the drone can complete its mission safely and efficiently.

[0073] As an implementation of step S304, the step of generating a corresponding control instruction set according to the comprehensive risk score level and the quantified track parameters includes: If the comprehensive risk score is low, the quantified track parameters are called to generate heading angle and speed fine-tuning instructions; When the system assesses a low comprehensive risk score, it indicates a high degree of flight path safety, with no significant obstacle threats or equipment failure risks in the flight environment. Therefore, the flight control system does not require extensive adjustments, instead fine-tuning the heading angle and flight speed based on quantitative track parameters (such as target coordinates and path accuracy). These fine-tuning commands refine flight trajectory tracking, improving flight path accuracy and mission execution efficiency.

[0074] For example, suppose a drone is performing a mission to cover a certain area. After risk assessment, the system determines that there are no obvious threats on the current path. At this time, it will generate fine-tuning heading angle instructions (for example, fine-tuning the flight angle by 1 degree) and speed fine-tuning instructions (for example, adjusting the flight speed to 95% of the original speed) based on the target position and current flight status to ensure that the drone continues to move along the optimal path during flight.

[0075] If the comprehensive risk score is medium or high, the global trajectory is reconstructed based on the flight path deviation prediction results to generate an obstacle avoidance trajectory or emergency return instructions.

[0076] Among them, when the comprehensive risk score is medium or high, it means that the potential risk in the flight environment has increased (such as approaching obstacles, possible equipment failure, severe environmental interference, etc.). At this time, the system needs to make a more urgent and precise response. First, the flight control system will combine the flight path deviation prediction results. These results are based on historical flight data, real-time sensor data, and the prediction of the path deviation model to analyze the possible deviations of the flight path. Based on these predictions, the system will reconstruct the global trajectory and recalculate the safest flight route to ensure that possible obstacles are avoided and high-risk areas are avoided. If the system predicts an unavoidable collision risk or other emergency situation, it will also generate an emergency return instruction to instruct the drone to immediately return to a safe location to avoid further risks.

[0077] For example, suppose during flight, the system detects a dynamic obstacle (such as a fast-moving car) approaching the drone's flight path, and the path deviation prediction indicates that continued flight will result in a collision with the obstacle. The flight control system will then replan the flight path based on the current risk level and the path deviation prediction, generating obstacle avoidance trajectory instructions (such as avoiding the current obstacle and yaw to the left) to avoid the collision. If the risk assessment indicates that equipment failure or other reasons make it impossible to continue the mission, the system may generate an emergency return instruction, directing the drone back to its starting point or a safe area.

[0078] The above-mentioned implementation incorporates an intelligent decision-making mechanism based on risk assessment into flight control, effectively addressing potential risks in diverse flight environments. In low-risk situations, the system fine-tunes flight parameters to improve mission execution accuracy and efficiency. In medium- and high-risk environments, the system ensures flight safety through real-time path deviation prediction and global trajectory reconstruction, enabling dynamic obstacle avoidance and emergency return command generation. This flexible control strategy enables the system to adapt to environmental changes in real time, avoid unforeseen risks, and enhance the drone's autonomous flight capabilities and mission success rate in complex environments.

[0079] Reference Figure 4 As a further embodiment of the control method, after converting the control instruction set into a UAV flight action sequence and sending it to the UAV for flight state adjustment, the method further includes: Step S401, monitoring the execution status of the UAV and generating control error feedback data; Monitoring the drone's execution status is a key step in ensuring the correct execution of flight commands and making appropriate adjustments. This monitoring primarily involves acquiring real-time flight data from the drone, such as position, speed, heading, acceleration, and the effectiveness of control commands. By comparing this real-time status with the target (i.e., the desired flight path or speed), control errors (such as position deviation and velocity error) are calculated. This error data serves as feedback, helping the system determine whether control commands are executing as expected and whether corrections are necessary.

[0080] For example, suppose a drone receives a command to adjust its flight path, but due to external disturbances (such as wind speed changes), the drone's actual flight path deviates. By monitoring the drone's state, the system can detect this deviation and generate corresponding control error data (for example, a 2-meter difference between the actual and desired positions).

[0081] Step S402, updating the weights of the machine learning model based on the control error feedback data; When an error occurs between the execution state monitored by the system and the target state, the machine learning model needs to be adjusted so that it can better adapt to changes in real-time flight. The core of this step is to update the weights in the machine learning model based on control error feedback data. Machine learning models are commonly used in flight control to predict flight path deviations, equipment status, etc., and continuously optimize the prediction results based on past data. By using real-time control error feedback as input, the system can use incremental learning methods (such as backpropagation and gradient descent) to adjust the model parameters, thereby improving prediction accuracy.

[0082] For example, suppose a machine learning model predicts that a drone will deviate by a certain distance during flight, but the observed error is actually greater than the predicted value. The system adjusts the model's weights based on this control error to reduce future prediction errors, allowing the model to more accurately reflect future flight deviations.

[0083] Step S403: Verify the validity of the updated machine learning model in the digital twin environment to obtain an optimized machine learning model.

[0084] Specifically, the updated machine learning model needs to be verified in a digital twin environment. Digital twin technology simulates various flight scenarios by creating a model consistent with the real-world drone and flight environment in a virtual environment. Through verification in a digital twin environment, the system can evaluate whether the updated machine learning model can accurately predict flight paths, equipment status, and other issues, and effectively respond to various complex situations that may occur during flight. This process ensures the robustness and accuracy of the model by repeatedly simulating different flight missions, various environmental interferences, equipment failures, and other situations.

[0085] For example, let's assume that an updated machine learning model is applied to a digital twin environment to simulate a flight mission under complex weather conditions. During this simulation, the system compares the simulated results with the actual flight mission to ensure that the model can accurately account for factors such as wind speed changes and equipment temperature fluctuations that may affect the flight.

[0086] It is understandable that by conducting validity verification in a digital twin environment, the system can evaluate and improve the machine learning model without actual risk, ensuring its feasibility and reliability in a real flight environment. This verification method can significantly improve the accuracy and stability of the model and avoid unnecessary deviations during actual flight.

[0087] In the above-mentioned implementation, real-time monitoring of the execution status and generation of control error feedback data help to detect deviations between the flight path and control instructions and make timely adjustments. By updating the weights of the machine learning model based on this feedback data, the system can continuously improve prediction accuracy and achieve precise control of the flight path and equipment status. At the same time, verification in the digital twin environment further ensures the effectiveness and robustness of the updated model in actual flight. This closed-loop self-optimization mechanism not only improves flight safety, but also enhances the adaptability of drones in dynamic environments, greatly improving the success rate and reliability of flight missions.

[0088] The embodiment of the present application also discloses an artificial intelligence UAV ground station control system.

[0089] An artificial intelligence UAV ground station control system, the control system includes: Data acquisition module, used to collect drone sensor data; drone sensor data includes positioning data, inertial measurement data, environmental perception data and drone status data; The data standardization module is used to align the UAV sensor data in time and space, normalize the heterogeneous data formats, and generate standardized time and space data packets; Dynamic obstacle map construction module, used to build dynamic obstacle maps based on historical flight database and standardized spatiotemporal data package; The prediction module is used to input the standardized spatiotemporal data packets into the machine learning model to predict the flight path deviation and equipment failure risk of the UAV and obtain a set of prediction results; The control instruction generation module is used to generate the corresponding control instruction set based on the preset mission objectives of the UAV, the dynamic obstacle map and the prediction result set; The flight state adjustment module is used to convert the control instruction set into a UAV flight action sequence and send it to the UAV for flight state adjustment.

[0090] As a further embodiment of the control system, it further includes: The execution monitoring module is used to monitor the execution status of the UAV and generate control error feedback data; The weight update module is used to update the weight of the machine learning model based on the control error feedback data; The model verification module is used to verify the validity of the updated machine learning model in the digital twin environment to obtain the optimized machine learning model.

[0091] An artificial intelligence UAV ground station control system of an embodiment of the present application can implement any of the above-mentioned control methods, and the specific working process of each module in the control system can refer to the corresponding process in the above-mentioned method embodiment.

[0092] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a module is merely a logical functional division, and in actual implementation, other division methods may be used, such as combining or integrating multiple modules into another system, or ignoring or not implementing certain features.

[0093] The embodiment of the present application also discloses a computer device.

[0094] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, an artificial intelligence UAV ground station control method as described above is implemented.

[0095] The embodiment of the present application also discloses a computer-readable storage medium.

[0096] A computer-readable storage medium stores a computer program that can be loaded by a processor and executed by any one of the above-mentioned artificial intelligence drone ground station control methods.

[0097] Among them, computer-readable storage media can be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0098] It should be noted that, in the above embodiments, the description of each embodiment has different emphases. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0099] The above are all preferred embodiments of the present application and are not intended to limit the scope of protection of this application. Unless otherwise specified, any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features. In other words, unless otherwise specified, each feature is merely an example of a series of equivalent or similar features.

Claims

1. An artificial intelligence UAV ground station control method, characterized in that: The control method includes: Collecting drone sensor data; the drone sensor data includes positioning data, inertial measurement data, environmental perception data, and drone status data; Performing spatiotemporal alignment on the drone sensor data and normalizing heterogeneous data formats to generate standardized spatiotemporal data packets; constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package; Inputting the standardized spatiotemporal data packet into a machine learning model to predict the flight path deviation and equipment failure risk of the UAV, thereby obtaining a prediction result set; Based on the preset mission objectives of the UAV, a corresponding control instruction set is generated according to the dynamic obstacle map and the prediction result set; The control instruction set is converted into a UAV flight action sequence and sent to the UAV for flight state adjustment.

2. The artificial intelligence UAV ground station control method according to claim 1, characterized in that: The steps of constructing a dynamic obstacle map based on the historical flight database and the standardized spatiotemporal data package include: Collect obstacle coordinate records and sensor false alarm marks in the historical flight database, perform feature cleaning, and obtain a historical obstacle coordinate list and historical obstacle weight labels; Obtaining point cloud data and camera image streams based on the environmental perception data in the standardized spatiotemporal data packet; Downsampling and ground segmentation processing are performed on the point cloud data to extract a candidate point set of non-ground obstacles; performing target re-identification on the non-ground obstacle candidate point set based on the camera image stream to obtain a fusion confidence score matrix; Separating static obstacles from dynamic obstacles based on the fused confidence score matrix and historical obstacle weight labels, and obtaining a static obstacle coordinate list and a dynamic obstacle tracking information table; Probabilistically fusing the historical obstacle coordinate list with the static obstacle coordinate list to construct a rasterized static obstacle map; Mapping the dynamic obstacle tracking information table to a dynamic layer to generate a risk heat map containing predicted trajectories; The static obstacle map, dynamic layer and risk heat map are encoded into a standard format and output at a preset frequency.

3. The artificial intelligence UAV ground station control method according to claim 1, characterized in that: The machine learning model includes a path deviation prediction model and an equipment failure warning model; the input of the path deviation prediction model is positioning data, and the output is the flight path deviation prediction result within a preset time in the future; the input of the equipment failure warning model is environmental perception data, and the output is the equipment failure risk probability prediction result.

4. The artificial intelligence UAV ground station control method according to claim 3, characterized in that: Based on the preset mission objectives of the UAV, the steps of generating a corresponding control instruction set according to the dynamic obstacle map and the prediction result set include: Analyze the preset mission objectives of the UAV and generate quantitative track parameters; Based on the static obstacle distribution of the dynamic obstacle map, an initial waypoint list is generated and the spatiotemporal conflict areas are marked, and an initial track file with conflict markings is output; Combining the prediction result set and the initial track file, calculating the obstacle collision probability, equipment failure impact factor and environmental interference coefficient, and outputting a comprehensive risk score level; A corresponding control instruction set is generated according to the comprehensive risk score level and the quantified track parameters.

5. The artificial intelligence UAV ground station control method according to claim 4, characterized in that: The step of generating a corresponding control instruction set according to the comprehensive risk score level and the quantified track parameters includes: If the comprehensive risk score is low, the quantified track parameters are called to generate heading angle and speed fine-tuning instructions; If the comprehensive risk score is medium or high, the global trajectory is reconstructed in combination with the flight path deviation prediction result to generate an obstacle avoidance trajectory or an emergency return instruction.

6. The artificial intelligence UAV ground station control method according to any one of claims 1 to 5, characterized in that: After the step of converting the control instruction set into a UAV flight action sequence and sending it to the UAV for flight state adjustment, the method further includes: monitoring the execution status of the UAV and generating control error feedback data; performing weight updating on the machine learning model based on the control error feedback data; The updated machine learning model is validated in the digital twin environment to obtain an optimized machine learning model.

7. An artificial intelligence UAV ground station control system, characterized in that: The control system includes: A data acquisition module is used to collect drone sensor data; the drone sensor data includes positioning data, inertial measurement data, environmental perception data and drone status data; A data standardization module is used to perform spatiotemporal alignment on the drone sensor data and normalize heterogeneous data formats to generate standardized spatiotemporal data packets; A dynamic obstacle map construction module, configured to construct a dynamic obstacle map based on a historical flight database and the standardized spatiotemporal data package; A prediction module, configured to input the standardized spatiotemporal data packet into a machine learning model to predict the flight path deviation and equipment failure risk of the UAV, and obtain a prediction result set; A control instruction generation module is used to generate a corresponding control instruction set based on the preset mission objectives of the UAV and the dynamic obstacle map and the prediction result set; The flight state adjustment module is used to convert the control instruction set into a UAV flight action sequence and send it to the UAV for flight state adjustment.

8. The artificial intelligence UAV ground station control system according to claim 7, characterized in that: The control system further comprises: An execution monitoring module, configured to monitor the execution status of the UAV and generate control error feedback data; A weight updating module, configured to update the weight of the machine learning model based on the control error feedback data; The model verification module is used to verify the validity of the updated machine learning model in the digital twin environment to obtain the optimized machine learning model.

9. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 6 when executing the program.

10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 6.

Images