Low-altitude flight real-time monitoring method and system based on 5G communication

Abstract

The invention discloses a low-altitude flight real-time monitoring method and system based on 5G communication, and belongs to the technical field of low-altitude aircraft monitoring, and the method comprises the following steps: carrying out the classification and parameterization representation of low-altitude aircrafts, and constructing an aircraft model library; acquiring multi-source heterogeneous data of the aircraft based on the aircraft model library; performing feature extraction on the multi-source heterogeneous data, and constructing a three-dimensional model of a flight environment; based on the three-dimensional model of the flight environment, combining aircraft characteristics to construct a multi-aircraft motion model; on the basis of the multi-aircraft motion model, predicting a motion track of each aircraft; and when the motion trails of the multiple aircrafts have a conflict risk, monitoring and early warning are carried out, and an avoidance strategy is adopted to carry out avoidance. According to the method, the low-altitude flight environment perception capability and the risk assessment accuracy are remarkably improved, and powerful support is provided for guaranteeing the low-altitude flight safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of monitoring low-altitude aircraft, and in particular, relates to a real-time monitoring method and system for low-altitude flight based on 5G communication. Background Art

[0002] There are many technical difficulties and contradictions in the process of building a low-altitude intelligent network to realize real-time monitoring and management of low-altitude aircraft. First, there are many types of low-altitude aircraft, such as drones, helicopters, gliders, etc. Their flight altitudes, speeds, and routes are different, which puts high demands on the adaptability of the monitoring system. Secondly, the low-altitude flight environment is complex and changeable, and there are a large number of obstacles such as buildings, trees, and telephone poles, which seriously interfere with the transmission of wireless signals and lead to the appearance of monitoring blind spots. Furthermore, over the city, various types of low-altitude aircraft are densely intertwined, and the flight paths are difficult to predict, which poses a severe challenge to the real-time and accuracy of the monitoring system. At the same time, severe weather such as strong winds, heavy rains, and lightning will affect the normal operation of low-altitude aircraft, further increasing the difficulty of monitoring. Finally, in the process of processing and analyzing massive multi-source heterogeneous monitoring data, how to achieve cross-domain data fusion, improve computing efficiency, and ensure data security is also a key issue that needs to be solved urgently.

[0003] These technical contradictions are intertwined, making it difficult to build a low-altitude intelligent network with full coverage, multi-dimensional perception, and efficient collaboration. It is necessary to overcome a series of technical difficulties before we can ultimately achieve real-time monitoring and intelligent management of low-altitude aircraft. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a real-time monitoring method and system for low-altitude flight based on 5G communication to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a method for real-time monitoring of low-altitude flight based on 5G communication, comprising the following steps:

[0006] Classify and parameterize low-altitude aircraft and build an aircraft model library;

[0007] Based on the aircraft model library, acquiring multi-source heterogeneous data of the aircraft;

[0008] Extracting features from the multi-source heterogeneous data to construct a three-dimensional model of the flight environment;

[0009] Based on the three-dimensional model of the flight environment and in combination with the characteristics of the aircraft, a multi-aircraft motion model is constructed;

[0010] Based on the multi-aircraft motion model, predict the motion trajectory of each aircraft;

[0011] When there is a risk of conflict in the movement trajectories of multiple aircraft, monitoring and early warning will be carried out, and avoidance strategies will be adopted to avoid them.

[0012] Optionally, the process of classifying and parameterizing low-altitude aircraft and building an aircraft model library includes:

[0013] Establish aircraft classification rules in advance, classify low-altitude aircraft based on the aircraft classification rules, and obtain classification results; based on the classification results, use statistical learning methods to extract key parameters of low-altitude aircraft and obtain parameterized representations; build an initial model library based on the classification results and parameterized representations of low-altitude aircraft; preset model evaluation rules, evaluate the initial model library, and based on the model evaluation results, screen out high-quality models from the initial model library to build a standard aircraft model library; preset description specifications, describe the standard aircraft model library, and obtain standard aircraft description specifications; associate and store the standard aircraft model library and the standard aircraft description specifications to obtain a final aircraft model library and corresponding description specifications.

[0014] Optionally, the process of extracting features from the multi-source heterogeneous data and constructing a three-dimensional model of the flight environment includes:

[0015] The multi-source heterogeneous data are denoised and standardized, and features are extracted from the processed multi-source heterogeneous data. The extracted features are associated and matched and fused using a Bayesian inference method, and a three-dimensional model of the flight environment is constructed based on the fused features.

[0016] Optionally, based on the multi-aircraft motion model, the process of predicting the motion trajectory of each aircraft includes:

[0017] The multi-aircraft motion model is trained using a deep reinforcement learning algorithm to obtain a trained multi-aircraft motion model; the status data of the aircraft is acquired in real time, the status data of the aircraft is input into the trained multi-aircraft motion model, and the future motion trajectory of each aircraft is predicted in real time.

[0018] Optionally, the process of training the multi-aircraft motion model using a deep reinforcement learning algorithm includes:

[0019] The historical motion trajectory data of the aircraft are used as training samples to train the multi-aircraft motion model. During the model training process, the reward function is used to evaluate the difference between the predicted trajectory and the actual trajectory, and the model parameters are optimized through the policy gradient algorithm to finally obtain the trained multi-aircraft motion model.

[0020] Optionally, when there is a risk of conflict between the motion trajectories of multiple aircraft, the process of using an avoidance strategy to avoid conflict includes:

[0021] Based on the predicted future motion trajectories of each aircraft, the conflict risk between multiple aircraft is dynamically evaluated to determine whether there is a conflict risk. If it is assessed that there is a flight conflict risk, the optimal avoidance strategy and avoidance action for each aircraft is calculated based on the trained multi-aircraft motion model. The generated optimal avoidance strategy and avoidance action are sent to the corresponding aircraft to control the aircraft to perform avoidance action.

[0022] The present invention also provides a low-altitude flight real-time monitoring system based on 5G communication, which is used to implement a low-altitude flight real-time monitoring method based on 5G communication, including: a model library construction module, a multi-source data acquisition module, an environment model construction module, a motion model construction module and a motion trajectory prediction module;

[0023] The model library building module is used to classify and parameterize low-altitude aircraft and build an aircraft model library;

[0024] The multi-source data acquisition module is used to acquire multi-source heterogeneous data of the aircraft based on the aircraft model library;

[0025] The environment model building module is used to extract features from the multi-source heterogeneous data and build a three-dimensional model of the flight environment;

[0026] The motion model building module is used to build a multi-aircraft motion model based on the three-dimensional model of the flight environment and in combination with aircraft characteristics;

[0027] The motion trajectory prediction module is used to predict the motion trajectory of each aircraft based on the multi-aircraft motion model. When there is a risk of conflict between the motion trajectories of the multiple aircraft, monitoring and early warning are performed, and avoidance strategies are adopted to avoid the conflict.

[0028] The present invention also provides a computer device, comprising 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.

[0029] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0030] The present invention also provides a computer program product, comprising a computer program, which implements the steps of the method when executed by a processor.

[0031] Compared with the prior art, the present invention has the following advantages and technical effects:

[0032] The method disclosed in the present invention first constructs an aircraft model library, and based on the aircraft model library, obtains multi-source heterogeneous data of aircraft; extracts features from the multi-source heterogeneous data to construct a three-dimensional model of the flight environment; based on the three-dimensional model of the flight environment, combined with aircraft characteristics, constructs a multi-aircraft motion model; based on the multi-aircraft motion model, predicts the motion trajectory of each aircraft; when there is a risk of conflict in the motion trajectories of multiple aircraft, monitors and warns, and adopts an avoidance strategy to avoid. The present invention significantly improves the low-altitude flight environment perception capability and risk assessment accuracy, and provides strong support for ensuring low-altitude flight safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0034] Figure 1 It is a schematic diagram of an equation flow of an embodiment of the present invention;

[0035] Figure 2 The figure is a schematic diagram of the process of constructing a three-dimensional model of a flight environment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0037] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] Embodiment 1

[0039] like Figure 1 As shown, this embodiment provides a low-altitude flight real-time monitoring method based on 5G communication, including the following steps:

[0040] Classify and parameterize low-altitude aircraft and build an aircraft model library;

[0041] Based on the aircraft model library, acquiring multi-source heterogeneous data of the aircraft;

[0042] Extracting features from the multi-source heterogeneous data to construct a three-dimensional model of the flight environment;

[0043] Based on the three-dimensional model of the flight environment and in combination with the characteristics of the aircraft, a multi-aircraft motion model is constructed;

[0044] Based on the multi-aircraft motion model, predict the motion trajectory of each aircraft;

[0045] When there is a risk of conflict in the movement trajectories of multiple aircraft, monitoring and early warning will be carried out, and avoidance strategies will be adopted to avoid them.

[0046] As a specific implementation method, based on a method combining rules and statistical learning, the aircraft model is classified and parameterized, and the process of building an aircraft model library includes:

[0047] According to the pre-established aircraft classification rules, the aircraft are classified at multiple levels and in multiple granularities to obtain the classification results; based on the classification results, the key parameters of the aircraft are extracted using statistical learning methods to obtain parameterized representations; based on the classification results and parameterized representations, an aircraft model library is constructed to obtain an initial model library; according to the preset model evaluation rules, the initial model library is evaluated, and according to the model evaluation results, high-quality models are screened out from the initial model library to construct a standard aircraft model library; according to the preset description specifications, the standard aircraft model library is described to obtain a standard aircraft description specification; the standard aircraft model library and the description specification are associated and stored to obtain the final aircraft model library and description specification.

[0048] Specifically, aircraft classification is the basis for building a model library. A multi-level classification method can be used, such as military and civilian classification according to purpose; fixed-wing, rotary-wing, vertical take-off and landing according to flight mode; large, medium, small, etc. according to size. Multi-granular classification can be further refined, such as fixed-wing aircraft can be divided into fighters, bombers, transport aircraft, etc. This classification method helps to comprehensively and systematically cover various types of aircraft. Parametric representation is the extraction of key features of aircraft. Statistical learning methods such as principal component analysis can be used to extract the most representative parameters from a large amount of data. For example, for fixed-wing aircraft, key parameters such as maximum take-off weight, cruising speed, and range can be extracted. This method can accurately describe the main performance characteristics of the aircraft with a small number of parameters. When building the initial model library, the classification results can be combined with the parametric representation. For example, for a certain type of fighter, a model containing its classification information (such as military, fixed-wing, fighter) and key parameters (such as maximum speed, combat radius, bomb load, etc.) can be established. This method can comprehensively and concisely describe various types of aircraft. Model evaluation is a key step to ensure model quality. Evaluation rules can be set, such as parameter completeness, data accuracy, classification rationality, etc. If the evaluation score of a model is lower than the preset threshold, it needs to return to optimization. This iterative evaluation method can continuously improve the quality of the model. Screening high-quality models is an important part of building a standard model library. The model with the highest score can be selected based on the evaluation results. This method can ensure that each model in the model library is of high quality and representative. Formulating description specifications can unify the model expression method. It can include model naming rules, parameter unit unification, performance indicator expression method, etc. For example, it is stipulated that all speed units use Mach number and all weight units use kilograms. This unified description specification can improve the readability and comparability of the model. Associative storage is a method to achieve the integration of model and description specification. Associative database technology can be used to associate model data with description specifications for storage. For example, a model table and a specification table are established and associated through foreign keys. This method can ensure the consistency of model data with description specifications, which is convenient for subsequent query and maintenance. This series of steps forms a complete aircraft model library construction process. Starting from classification, through parameterized representation, model construction, evaluation optimization, screening standardization, specification formulation to final associative storage, each step is closely linked, jointly ensuring the comprehensiveness, accuracy and standardization of the final model library.

[0049] As a specific embodiment, based on the aircraft model library, the multi-sensor fusion technology is adopted, and the video, radar, and radio detection methods are comprehensively utilized to construct the all-round, multi-level environmental perception system to obtain multi-source heterogeneous data; the multi-source heterogeneous data of the aircraft includes video data, radar data, and radio data.

[0050] As a specific example, Figure 2As shown, the process of preprocessing, feature extraction and information fusion of the multi-source heterogeneous data, building a three-dimensional model of the flight environment in real time, and dynamically updating the environment map includes:

[0051] Acquire multi-source heterogeneous data, including visual images, lidar point clouds, inertial navigation data, etc., and standardize data from different sources and formats. Preprocess the standardized data to remove noise and redundant information and improve data quality. Use algorithms such as Kalman filtering to smooth and fuse the data. Extract key features such as edges, corners, planes, etc. from the preprocessed data to construct feature descriptors. Classify and merge features through clustering and segmentation algorithms. Fuse the extracted features, integrate feature information from different sensors, and generate a unified environmental representation. Use methods such as Bayesian inference to associate and match features. Based on the fused feature information, build a three-dimensional model of the flight environment in real time. Use data structures such as octrees and voxels to spatially divide and index the environment. Generate environmental maps based on the three-dimensional model, including static maps and dynamic maps. Static maps represent fixed objects in the environment, and dynamic maps represent the real-time positions of moving obstacles. As the drone continues to fly, new sensor data is continuously acquired and the environmental map is dynamically updated. Through inter-frame matching and loop detection, the map can be corrected and optimized in real time to maintain consistency with the real environment.

[0052] Specifically, the acquisition of multi-source heterogeneous data is the basis of environmental perception. Taking drones as an example, their visual cameras can capture high-resolution images, lidar can obtain accurate three-dimensional point clouds, and inertial navigation units provide attitude and position information. These heterogeneous data need to be standardized, such as unifying the coordinate systems of different sensors to the body coordinate system. Data preprocessing is crucial to improving the quality of perception. For example, downsampling and filtering the lidar point cloud can remove noise points such as ground reflections. Applying Gaussian filtering to visual images can smooth images and suppress noise. Kalman filtering can fuse inertial navigation and GPS data to obtain more accurate pose estimates. Feature extraction is the key to constructing environmental representation. For visual images, SIFT or SURF feature points can be extracted; for point cloud data, geometric features such as normal vectors and curvature can be extracted. These features can be used for subsequent target detection and scene understanding. Clustering algorithms such as K-means can classify feature points, which helps to identify different objects. Feature fusion aims to make comprehensive use of multi-sensor information. For example, visual feature points can be combined with corresponding lidar depth information to generate visual features with three-dimensional information. Bayesian inference can be used to handle the uncertainty of different sensors, such as calculating the probability distribution of the current position based on historical observations. 3D reconstruction of the environment is a high-level task of perception. SLAM (simultaneous localization and mapping) can be achieved based on feature matching, and the environmental model can be continuously updated. The octree structure is suitable for representing large-scale outdoor scenes and can efficiently store and query spatial information. Voxel grids are suitable for representing dense indoor environments. Dynamic map construction requires distinguishing between static and dynamic objects. Moving targets can be detected by differentials between consecutive frames, and their trajectories can be tracked using Kalman filters. Static maps can be represented by probabilistic occupancy grids, and dynamic targets can be modeled using particle filters. Map optimization is the key to ensuring long-term navigation accuracy. Loop detection can identify repeatedly visited areas and correct accumulated positioning errors. Graph optimization methods can be used to construct pose estimates and observation data into factor graphs, and the entire trajectory can be optimized by minimizing the error function. This environmental perception system enables drones to adapt to complex and changing environments. For example, in urban environments, static and dynamic obstacles such as buildings, pedestrians, and vehicles can be accurately perceived, providing a reliable basis for path planning and obstacle avoidance. In post-disaster rescue, a 3D map of the disaster area can be quickly constructed to assist rescuers in making decisions. By continuously accumulating and optimizing environmental models, the autonomy and mission execution capabilities of drones will be significantly improved.

[0053] As a specific embodiment, the process of building the multi-aircraft motion model by combining the three-dimensional model of the flight environment and the characteristics of the aircraft, using the deep reinforcement learning algorithm to predict the future motion trajectory of the aircraft in real time, and dynamically evaluating the flight conflict risk to generate the optimal avoidance strategy includes:

[0054] The three-dimensional model data of the flight environment and the characteristic parameters of multiple aircraft are obtained, and the two are used as input to construct a multi-aircraft motion model; the multi-aircraft motion model is trained using a deep reinforcement learning algorithm to obtain a trained multi-aircraft motion model; in the process of the aircraft performing the task, the status data of the aircraft is obtained in real time and input into the trained multi-aircraft motion model; the trained multi-aircraft motion model is used to predict the motion trajectory of each aircraft in the future in real time according to the current aircraft status data; based on the predicted future motion trajectory of each aircraft, the conflict risk between multiple aircraft is dynamically evaluated to determine whether there is a conflict risk; if the assessment shows that there is a flight conflict risk, the trained multi-aircraft motion model is used to calculate and generate the optimal avoidance strategy and avoidance action for each aircraft; the generated avoidance strategy and avoidance action are sent to each aircraft, and the aircraft is controlled to perform the corresponding avoidance maneuver to eliminate the flight conflict risk and ensure flight safety.

[0055] Specifically, the core of the multi-aircraft cooperative obstacle avoidance system lies in building an accurate motion model and using deep reinforcement learning to make intelligent decisions. First, the system needs to obtain the three-dimensional model data of the flight environment, which can be achieved through laser radar scanning or photogrammetry. For example, in an urban environment, the point cloud data of static obstacles such as buildings and roads can be collected by vehicle-mounted laser radar, and then a detailed city model can be generated through a three-dimensional reconstruction algorithm. At the same time, the system also needs to obtain the characteristic parameters of each aircraft, such as maximum speed, acceleration, turning radius, etc. These parameters can be obtained through flight tests or simulations. For example, the maximum horizontal speed of a certain type of quadcopter drone is 20 meters per second and the maximum climb rate is 5 meters per second. Based on the environmental model and aircraft parameters, the system builds a multi-aircraft motion model. This model needs to take into account the mutual influence between aircraft, such as the wake effect. For example, when a large fixed-wing drone flies over, the wake it generates will cause disturbances to the small rotor drone behind it. The motion model also needs to include environmental factors, such as wind speed, air pressure, etc. These factors will affect the actual motion trajectory of the aircraft. The training process of the deep reinforcement learning algorithm is the key to the system. Algorithms such as dual Q networks or dominant actor-critic can be used. The training environment can be a simulation platform based on the above motion model. The design of the reward function is crucial and needs to balance multiple goals such as safety, efficiency and stability. For example, positive rewards can be given for successful obstacle avoidance and negative rewards can be given for approaching other aircraft or obstacles, while considering fuel consumption and flight time. During the training process, the scene complexity and the number of aircraft can be gradually increased to improve the generalization ability of the model. In actual flight, the system needs to obtain the status data of the aircraft in real time, including position, speed, attitude, etc. These data can be obtained through onboard sensors and transmitted to the central processing system through data links. For example, a drone reports its latitude and longitude, altitude, three-axis speed and Euler angle every 0.1 seconds. The trained multi-aircraft motion model predicts the future trajectory based on the current state. The selection of the prediction time window requires a trade-off between computational complexity and prediction accuracy. A prediction window of 5-10 seconds can usually be selected. The prediction result can be a series of discrete future position points or a continuous trajectory function. Based on the predicted trajectory, the system evaluates the risk of flight conflict. A probabilistic collision model can be used, taking into account prediction errors and aircraft size. For example, if the minimum distance between two drones in the next 5 seconds is less than 50 meters and the probability of collision is more than 1%, it is determined that there is a risk of collision. When a risk of collision is detected, the deep reinforcement learning model generates avoidance strategies for each aircraft. These strategies may include changing altitude, speed, or heading. For example, for two drones flying towards each other, the model may recommend that one rise 100 meters and the other turn right 15 degrees. The avoidance action needs to take into account aircraft performance limitations and airspace restrictions. Finally, the system sends avoidance instructions to each aircraft. This requires a reliable communication link and a standardized instruction format. After receiving the instruction, the aircraft needs to quickly perform the avoidance maneuver.The entire process from detection to execution should be completed within a few seconds to ensure safety. This system can effectively cope with complex multi-aircraft scenarios. For example, in urban air traffic management, the movement of dozens of manned aircraft and drones can be coordinated to avoid collisions between them and with buildings. In large-scale drone cluster operations, such as forest fire fighting or farmland spraying, the system can ensure that the aircraft can work together efficiently without collisions. Through continuous learning and optimization, the system can continuously improve its ability to handle various complex situations, laying the foundation for future air traffic safety.

[0056] Furthermore, the process of using the trained multi-aircraft motion model to predict the motion trajectory of each aircraft in the future in real time according to the current aircraft status data includes:

[0057] The current state data of each aircraft, including position, velocity, acceleration, attitude and other information, is obtained as the input of the multi-aircraft motion model. The historical motion trajectory data of the aircraft is used as training samples to train the multi-aircraft motion model so that the model can predict the future trajectory based on the current state. During the model training process, the reward function is used to evaluate the difference between the predicted trajectory and the actual trajectory, and the policy gradient algorithm is used to optimize the model parameters to improve the prediction accuracy. For the current state data of each aircraft, the trained multi-aircraft motion model is used to predict its motion trajectory in the future in real time. According to the predicted motion trajectory, it is judged whether there is a potential collision risk between different aircraft. If there is a collision risk, the motion parameters of the relevant aircraft are adjusted in time to avoid collision. By continuously obtaining the latest state data of the aircraft and constantly updating the input of the trained multi-aircraft motion model, the real-time prediction and dynamic adjustment of the aircraft motion trajectory are realized. The predicted future motion trajectory information of the aircraft is transmitted to the ground control center in real time for relevant personnel to monitor and make decisions to ensure the safe operation of the aircraft group.

[0058] As an additional implementation method, based on the three-dimensional model of the flight environment, a multi-factor comprehensive evaluation model of the flight environment is constructed, and the machine learning algorithm is used to evaluate and warn of flight risks under the severe weather conditions, formulate the emergency response plan, and generate real-time monitoring data.

[0059] Obtain three-dimensional model data of the flight environment, and build a comprehensive assessment model for multiple influencing factors of the flight environment; use machine learning algorithms such as support vector machines and random forests to assess flight risks under severe weather conditions; determine whether the warning threshold has been reached based on the flight risk assessment results, and trigger the warning mechanism if the threshold has been reached; after the warning is triggered, determine the corresponding disposal measures and processes based on the pre-established emergency response plan; obtain real-time monitoring data of the flight environment, including meteorological conditions, flight status, etc., as input for risk assessment; input real-time monitoring data into the flight risk assessment model to dynamically assess the current flight risk level; generate specific instructions and action plans based on the risk assessment results and emergency response plans to guide the adjustment of flight missions and emergency operations.

[0060] Specifically, to obtain the three-dimensional model data of the flight environment, it is first necessary to use advanced technologies such as laser radar and satellite remote sensing to perform high-precision scanning of the terrain, buildings, vegetation, etc. in the flight area to generate a refined three-dimensional model containing information such as height, slope, and material. For example, in a mountainous flight environment, the three-dimensional model should not only show the outline of the mountain, but also accurately mark dangerous terrain such as cliffs and steep slopes so that the aircraft can avoid these high-risk areas when planning the path. In view of the multiple influencing factors of the flight environment, when constructing a comprehensive evaluation model, it is necessary to comprehensively consider multi-dimensional factors such as meteorological conditions, air traffic flow, and electromagnetic interference. For example, when flying in coastal areas, in addition to terrain factors, it is also necessary to focus on meteorological data such as wind speed, wind direction, and humidity, as well as radar signal interference from ships on the sea. By incorporating these factors into the evaluation model, flight risks can be predicted more comprehensively. The use of machine learning algorithms such as support vector machines and random forests for flight risk assessment is based on the advantages of these algorithms in processing nonlinear and high-dimensional data. For example, support vector machines can effectively distinguish high-risk and low-risk flight conditions by finding the optimal hyperplane; while random forests can integrate multiple factors for risk assessment by constructing multiple decision trees. In severe weather conditions, such as severe storms, the model will assess the flight risk level, such as "high risk" or "medium risk", based on historical data and real-time meteorological data. Based on the flight risk assessment results, it is determined whether the warning threshold has been reached. The warning threshold is usually determined by expert experience combined with historical data analysis, such as wind speed exceeding 30 meters per second, visibility less than 500 meters, etc. If the assessment results show that the current flight environment reaches or exceeds the warning threshold, the warning mechanism is triggered and the system automatically sends an alarm message to the flight control center. After the warning is triggered, the corresponding disposal measures and processes are determined according to the pre-established emergency response plan. For example, if the warning is a severe storm, the plan may include immediately adjusting the flight path, lowering the flight altitude, and finding the nearest alternate airport. The formulation of the emergency response plan needs to comprehensively consider factors such as the performance of the aircraft and the importance of the flight mission to ensure that it can respond quickly and effectively in an emergency. Obtain real-time monitoring data of the flight environment, including meteorological conditions, flight status, etc., as input for risk assessment. For example, through sensors installed on the aircraft, data such as wind speed, air pressure, and temperature are collected in real time and transmitted to the ground control center through data links. These real-time data provide the basis for dynamic risk assessment. Input the real-time monitoring data into the flight risk assessment model to dynamically assess the current flight risk level. For example, during the flight, if the real-time meteorological data shows that the wind speed suddenly increases, the model will immediately re-evaluate the risk level and update the warning status. This dynamic assessment mechanism can ensure real-time monitoring and warning of flight risks. Based on the risk assessment results and emergency response plans, specific instructions and action plans are generated to guide the adjustment of flight missions and emergency operations.For example, if the risk assessment results show that the current flight path is high risk, the system will generate a new flight path plan and issue it to the aircraft through the command system to ensure that the aircraft can adjust its course in time to avoid high-risk areas. Through the above steps, not only can the risks of the flight environment be fully assessed, but also countermeasures can be taken quickly when risks occur, effectively improving the safety and reliability of the flight. This comprehensive risk assessment and early warning mechanism provides a strong guarantee for flight missions in complex environments.

[0061] As an additional implementation method, the big data processing architecture and distributed computing technology are used to build the efficient and scalable data processing platform, the monitoring data is cleaned, normalized and feature-engineered, and the incremental learning technology is combined to realize online updating of the model, and the online learning technology is used to adapt to the dynamic changes of the data distribution.

[0062] Use distributed computing frameworks such as Hadoop and Spark to build an efficient and scalable big data processing platform to support real-time processing of massive monitoring data. Design a data cleaning process for the original monitoring data, remove noise data through filtering, completion and other operations to improve data quality. Normalize the cleaned data to map features of different scales to a unified range to eliminate the dimensional effects between features. Use feature engineering methods such as feature selection and feature extraction to build a high-quality feature set to improve the performance of subsequent machine learning models. For dynamically changing data distribution, use incremental learning technology to regularly update existing models with newly arrived data so that the models can adapt to changes in data distribution. Obtain the latest monitoring data to determine whether it meets the preset data quality threshold. If it does, add it to the incremental training set to update the existing machine learning model. Use online learning algorithms such as online gradient descent to adjust model parameters based on incremental training data to obtain an updated model with better performance for subsequent monitoring data processing and analysis.

[0063] Specifically, distributed computing frameworks such as Hadoop and Spark provide efficient and scalable solutions for massive data processing. Taking meteorological monitoring as an example, meteorological stations across the country generate a large amount of observation data every minute, which needs to be processed in real time to support weather forecasts. By storing raw data through the Hadoop distributed file system and using MapReduce parallel processing, data cleaning and feature extraction can be completed quickly. Data cleaning is a key step to ensure data quality. Taking aircraft sensor data as an example, there may be problems such as missing values ​​and outliers. Filtering can be performed by setting reasonable thresholds, such as marking temperature data that exceeds the normal range as abnormal. For missing data, methods such as adjacent value interpolation can be used to complete it. These operations can significantly improve the accuracy of subsequent analysis. Normalization helps to eliminate the dimensional effects between different features. For example, in flight risk assessment, the numerical ranges of speed (m / s) and altitude (m) are quite different. Through min-max normalization, all features can be mapped to the 0-1 interval, so that different features have the same weight in the model. Feature engineering is crucial to the performance of machine learning models. In flight environment analysis, key features can be extracted through methods such as principal component analysis. For example, the main influencing factors such as temperature, humidity, and wind speed are extracted from the original meteorological data to reduce the data dimension and improve the efficiency of model training. Incremental learning technology can adapt to dynamically changing data distribution. Taking air traffic control as an example, flight patterns may change with the opening of new routes or seasonal changes. By regularly introducing new data to update the model, the timeliness and accuracy of the model can be maintained. In specific implementation, the model can be incrementally updated once a week or a month. Online learning algorithms such as online gradient descent can achieve real-time updates of the model. Taking the drone control system as an example, the control parameters can be continuously adjusted according to real-time flight data. When a new data sample is detected, the gradient of the loss function is immediately calculated, and the model parameters are adjusted slightly, so that the model can quickly adapt to environmental changes. The comprehensive application of these technologies can significantly improve the effect of flight environment monitoring and risk assessment. Through efficient data processing and model updating, potential risks can be predicted more accurately, providing strong protection for flight safety. For example, when bad weather is detected, the system can issue an early warning in time and recommend adjusting the route or delaying takeoff. At the same time, this method based on big data and machine learning also provides valuable data support for the long-term development of the aviation industry, helping to optimize route planning and improve operational efficiency.

[0064] As an additional implementation method, the ontology- and semantic network-based knowledge representation and reasoning technology is used to construct the low-altitude flight field knowledge graph, realize the semantic association and mapping of the multi-source heterogeneous data, and realize the integration and sharing of the cross-domain data through the rule-based reasoning and graph-based search algorithm.

[0065] Based on the knowledge of low-altitude flight, a domain ontology is constructed to define concepts, relationships and attributes, and form an ontology hierarchy and semantic associations. For multi-source heterogeneous data, natural language processing technology is used for semantic analysis to extract entities, relationships and attributes, map them to ontology concepts, and establish semantic indexes. Through knowledge graph representation technology, the ontology and semantic index are converted into graph data structures, with nodes representing concepts, edges representing relationships, and attributes as additional information for nodes and edges. For specific cross-domain data integration needs, a rule-based reasoning method is used to perform semantic reasoning on the knowledge graph, and new associated knowledge is generated based on the relationship between concepts. According to the topics of data sharing and user queries, graph-based search algorithms such as shortest path and subgraph matching are used to retrieve relevant nodes on the knowledge graph and obtain associated data. For search results, a sorting algorithm such as PageRank is used to sort the data based on factors such as node importance and relevance to improve the accuracy and coverage of queries. The knowledge graph and reasoning search results are visualized, and an interactive operation interface is provided to support users to conduct exploratory analysis of cross-domain data and discover new association patterns and insights.

[0066] This embodiment also provides a low-altitude flight real-time monitoring system based on 5G communication, which is used to implement a low-altitude flight real-time monitoring method based on 5G communication, including: a model library construction module, a multi-source data acquisition module, an environment model construction module, a motion model construction module and a motion trajectory prediction module;

[0067] The model library building module is used to classify and parameterize low-altitude aircraft and build an aircraft model library;

[0068] The multi-source data acquisition module is used to acquire multi-source heterogeneous data of the aircraft based on the aircraft model library;

[0069] The environment model building module is used to extract features from the multi-source heterogeneous data and build a three-dimensional model of the flight environment;

[0070] The motion model building module is used to build a multi-aircraft motion model based on the three-dimensional model of the flight environment and in combination with aircraft characteristics;

[0071] The motion trajectory prediction module is used to predict the motion trajectory of each aircraft based on the multi-aircraft motion model. When there is a risk of conflict between the motion trajectories of the multiple aircraft, monitoring and early warning are performed, and avoidance strategies are adopted to avoid the conflict.

[0072] Embodiment 2

[0073] This embodiment further provides a computer device, including 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.

[0074] Embodiment 3

[0075] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0076] Embodiment 4

[0077] This embodiment also provides a computer program product, including a computer program, which implements the steps of the method when executed by a processor.

[0078] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A real-time monitoring method for low-altitude flight based on 5G communication, characterized in that: The following steps are involved: Classify and parameterize low-altitude aircraft and build an aircraft model library; Based on the aircraft model library, acquiring multi-source heterogeneous data of the aircraft; Extracting features from the multi-source heterogeneous data to construct a three-dimensional model of the flight environment; Based on the three-dimensional model of the flight environment and in combination with the characteristics of the aircraft, a multi-aircraft motion model is constructed; Based on the multi-aircraft motion model, predict the motion trajectory of each aircraft; When there is a risk of conflict in the movement trajectories of multiple aircraft, monitoring and early warning will be carried out, and avoidance strategies will be adopted to avoid them.

2. The method according to claim 1, characterized in that The process of classifying and parameterizing low-altitude aircraft and building an aircraft model library includes: Establish aircraft classification rules in advance, classify low-altitude aircraft based on the aircraft classification rules, and obtain classification results; based on the classification results, use statistical learning methods to extract key parameters of low-altitude aircraft and obtain parameterized representations; build an initial model library based on the classification results and parameterized representations of low-altitude aircraft; preset model evaluation rules, evaluate the initial model library, and based on the model evaluation results, screen out high-quality models from the initial model library to build a standard aircraft model library; preset description specifications, describe the standard aircraft model library, and obtain standard aircraft description specifications; associate and store the standard aircraft model library and the standard aircraft description specifications to obtain a final aircraft model library and corresponding description specifications.

3. The method according to claim 1, characterized in that The process of extracting features from the multi-source heterogeneous data and constructing a three-dimensional model of the flight environment includes: The multi-source heterogeneous data are subjected to denoising and standardization processing, and features are extracted from the processed multi-source heterogeneous data. The extracted features are associated and matched and fused using a Bayesian inference method, and a three-dimensional model of the flight environment is constructed based on the fused features.

4. The method according to claim 1, characterized in that: Based on the multi-aircraft motion model, the process of predicting the motion trajectory of each aircraft includes: The multi-aircraft motion model is trained using a deep reinforcement learning algorithm to obtain a trained multi-aircraft motion model; the status data of the aircraft is acquired in real time, the status data of the aircraft is input into the trained multi-aircraft motion model, and the future motion trajectory of each aircraft is predicted in real time.

5. The method according to claim 4, characterized in that The process of training the multi-aircraft motion model using a deep reinforcement learning algorithm includes: The historical motion trajectory data of the aircraft are used as training samples to train the multi-aircraft motion model. During the model training process, the reward function is used to evaluate the difference between the predicted trajectory and the actual trajectory, and the model parameters are optimized through the policy gradient algorithm to finally obtain the trained multi-aircraft motion model.

6. The method according to claim 4, characterized in that When there is a risk of conflict in the motion trajectories of multiple aircraft, the process of using the avoidance strategy includes: Based on the predicted future motion trajectories of each aircraft, the conflict risk between multiple aircraft is dynamically evaluated to determine whether there is a conflict risk. If it is assessed that there is a flight conflict risk, the optimal avoidance strategy and avoidance action for each aircraft is calculated based on the trained multi-aircraft motion model. The generated optimal avoidance strategy and avoidance action are sent to the corresponding aircraft to control the aircraft to perform avoidance action.

7. A low-altitude flight real-time monitoring system based on 5G communication, characterized in that: A method for real-time monitoring of low-altitude flight based on 5G communication for implementing any one of claims 1 to 6, comprising: a model library construction module, a multi-source data acquisition module, an environment model construction module, a motion model construction module and a motion trajectory prediction module; The model library building module is used to classify and parameterize low-altitude aircraft and build an aircraft model library; The multi-source data acquisition module is used to acquire multi-source heterogeneous data of the aircraft based on the aircraft model library; The environment model building module is used to extract features from the multi-source heterogeneous data and build a three-dimensional model of the flight environment; The motion model building module is used to build a multi-aircraft motion model based on the three-dimensional model of the flight environment and in combination with aircraft characteristics; The motion trajectory prediction module is used to predict the motion trajectory of each aircraft based on the multi-aircraft motion model. When there is a risk of conflict between the motion trajectories of the multiple aircraft, monitoring and early warning are performed, and avoidance strategies are adopted to avoid the conflict.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images