A networked drone identification and control method and system

Abstract

The present application provides a method and system for identifying and controlling networked drones, wherein real-time flight activity information is received from heterogeneous data sources such as ground stations and satellite links; based on multi-agent system theory and distributed computing resources, this information is processed in parallel to dynamically establish a virtual isolation zone; deep recursive neural networks and long short-term memory networks are combined to evaluate the flight intentions of drones, and anomaly detection algorithms are used to identify abnormal behaviors and generate warning signals; based on the warning signals and flight intentions, flight path planning is formulated using geo-fencing technology, time window control strategies, and multi-objective optimization path planning algorithms, and sent to drones via quantum encrypted communication links; a closed-loop management process is formed based on real-time feedback information and Bayesian update mechanisms. The present application improves the safe operation capabilities of drones in complex airspace environments and the accuracy of abnormal behavior warnings.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of networked drones, and in particular to a method and system for identifying and controlling networked drones. Background Art

[0002] With the widespread application of drones in logistics distribution, agricultural monitoring, environmental monitoring, public safety and other fields, real-time monitoring and management of drone flight activities have become crucial. These application scenarios require the system to seamlessly receive and process real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly obtained from drones. In order to ensure the safety and efficient use of airspace, the system needs to process flight activity information in parallel based on multi-agent system theory and distributed computing resources to dynamically establish virtual isolation zones. In addition, the system also needs to use deep recurrent neural networks and long short-term memory networks to evaluate the flight intentions of each drone, identify potential abnormal behaviors through anomaly detection algorithms, and generate early warning signals. Based on these early warning signals, the system should be able to formulate reasonable flight path planning through geo-fencing technology, time window control strategy, and path planning algorithm based on multi-objective optimization, and send warning signals and path planning to the corresponding drones through quantum encryption communication links to achieve precise control. Finally, the system needs to combine real-time feedback information with the Bayesian update mechanism to form a closed-loop management process to ensure the safety and efficiency of drone operations.

[0003] Current drone flight management systems usually rely on centralized control systems that define flight areas and restrictions through geo-fencing technology and simple rule engines. To improve prediction accuracy, some systems have introduced machine learning models, such as deep recurrent neural networks and long short-term memory networks, to process historical flight records and real-time data to evaluate the flight intentions of drones. At the same time, in order to ensure the security of data transmission, quantum encryption communication links are used to ensure the security of instructions and key data during transmission. However, these existing solutions mainly rely on static settings and predefined rules, lack sufficient flexibility and adaptability, and are difficult to cope with complex and changing practical application needs.

[0004] Although the existing solutions have met the basic needs to a certain extent, there are still several significant defects, which are as follows: First, when faced with a large number of heterogeneous data sources, the data processing capabilities of traditional centralized systems are limited, and they cannot efficiently integrate and analyze real-time flight activity information from different channels; second, the existing virtual isolation zone settings are not flexible enough to respond quickly to emergencies or temporary mission requirements, resulting in poor performance in dynamic adjustment; third, the anomaly detection mechanism mainly relies on preset rules and lacks adaptive learning capabilities, resulting in inaccurate identification of abnormal flight behaviors and prone to false alarms or omissions; finally, in the entire data processing and communication process, especially in a distributed environment, how to effectively protect sensitive information from attacks remains a major challenge. Summary of the invention

[0005] The embodiments of the present application provide a method and system for identifying and controlling a networked drone, which is used to solve the problems in the prior art of low safe operation capability of drones in complex airspace environments and low accuracy of abnormal behavior warning.

[0006] In a first aspect, an embodiment of the present application provides a method for identifying and controlling a networked drone, including:

[0007] receiving real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly from the drone;

[0008] Based on multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone;

[0009] According to the virtual isolation zone and the real-time flight activity information, the flight intention of each drone is evaluated using a deep recurrent neural network and a long short-term memory network, and an anomaly detection algorithm is used to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the drone to obtain a warning signal, wherein the flight intention is to input the real-time flight activity information and the multi-source historical drone flight activity records of the drone into a machine learning model to generate a prediction of the expected behavior of the drone;

[0010] Based on the warning signal, the flight path is identified from the flight intention, and the flight path planning is obtained through the geo-fencing technology, the time window control strategy and the path planning algorithm based on multi-objective optimization. The warning signal and the flight path planning are sent to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information;

[0011] Based on the real-time control feedback information combined with the early warning signal, a networked drone identification and control process is generated through a Bayesian update mechanism.

[0012] Optionally, the flight intention of each UAV is evaluated using a deep recurrent neural network and a long short-term memory network according to the virtual isolation zone and the real-time flight activity information, and an anomaly detection algorithm is used to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV, and obtain a warning signal, including:

[0013] Using graph database technology, the real-time flight activity information and the data of the virtual isolation zone are integrated and processed to obtain a data set;

[0014] Collect historical drone flight activity records from multiple sources, and build a flight intention prediction model of deep recursive neural network and long short-term memory network based on the data set and the historical drone flight activity records, and obtain an optimized prediction model through transfer learning technology;

[0015] Predicting the flight intention, flight path and initial behavior pattern of the UAV from the optimized prediction model, simulating and analyzing the flight path and the initial behavior pattern using a reinforcement learning algorithm, obtaining a flight strategy, simulating the expected benefits under the flight strategy, evaluating the flight intention, and obtaining an optimal flight path and key behavior pattern;

[0016] Compare and analyze the optimal flight path and the key behavior pattern with the actual flight behavior of the UAV to obtain behavior differences, use an isolation forest combined with anomaly detection algorithm to identify the target behavior pattern from the behavior differences, introduce a local anomaly factor algorithm, and mark the behavior of the target behavior pattern as abnormal flight behavior;

[0017] The game theory model is used to predict conflict scenarios from the abnormal flight behavior, and the scenario simulation technology and Monte Carlo method are combined to estimate the probability of the conflict scenarios and generate early warning signals.

[0018] Optionally, the flight intention, flight path and initial behavior pattern of the UAV are predicted from the optimized prediction model, the flight path and the initial behavior pattern are simulated and analyzed by a reinforcement learning algorithm to obtain a flight strategy, and the expected benefits under the flight strategy are simulated to evaluate the flight intention to obtain an optimal flight path and key behavior pattern, including:

[0019] Based on the optimized prediction model combined with the attention mechanism and the introduction of a graph convolutional network, the flight intention, flight path and initial behavior pattern of each drone are predicted from the data set and the historical drone flight activity records;

[0020] Using a reinforcement learning algorithm, simulating and analyzing the flight path and the initial behavior pattern to obtain a flight strategy, introducing a multi-objective optimization framework to simulate the flight strategy, and obtaining a flight strategy with an expected benefit higher than a preset threshold;

[0021] Identify initial behavior performance from the initial behavior pattern, simulate the initial behavior performance and expected benefits of the drone in the flight strategy in combination with the virtual isolation zone and the preset airspace use rules, and generate a flight strategy evaluation report through scenario simulation technology and Monte Carlo method;

[0022] Based on the flight strategy evaluation report, the flight intention is evaluated to obtain an initial evaluation result, and a multimodal fusion technology is introduced in combination with a hierarchical clustering algorithm to classify the initial evaluation result to generate a comprehensive evaluation result;

[0023] Based on the comprehensive evaluation results, a collision detection algorithm is applied in combination with a genetic algorithm to determine the optimal flight path and key behavior patterns.

[0024] Optionally, the use of a reinforcement learning algorithm to simulate and analyze the flight path and the initial behavior pattern to obtain a flight strategy, and the introduction of a multi-objective optimization framework to simulate the flight strategy to obtain a flight strategy with an expected benefit higher than a preset threshold, includes:

[0025] Using a reinforcement learning algorithm, simulating and analyzing the flight path and the initial behavior pattern to obtain an initial flight strategy;

[0026] Using a multi-objective optimization framework, the initial flight strategy is simulated to generate a target flight strategy, and based on the target flight strategy, a Monte Carlo method combined with a Bayesian optimization algorithm is applied to generate a flight strategy evaluation report;

[0027] According to a pre-set condition that the expected benefit is higher than a preset threshold, based on the condition and in combination with a differential evolution algorithm, an optimal flight strategy is screened out from the flight strategy evaluation report.

[0028] Optionally, based on the warning signal, a flight path is identified from the flight intention, a flight path planning is obtained through geo-fencing technology, a time window control strategy and a path planning algorithm based on multi-objective optimization, and the warning signal and the flight path planning are sent to a corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information, including:

[0029] Based on the flight intention, analyzing the warning signal to identify a flight path related to the warning signal and obtain a preliminary flight path plan;

[0030] Applying geo-fencing technology to set a restricted area of ​​the virtual isolation zone, integrating information of the restricted area into the preliminary flight path plan to generate a flight path, and generating a flight path plan based on the flight path combined with a time window control strategy and a path planning algorithm based on multi-objective optimization;

[0031] By using a quantum encrypted communication link, the warning signal and the flight path planning are sent to the corresponding UAV to obtain a real-time adjusted flight status, and based on the flight status, real-time control feedback information is obtained.

[0032] Optionally, the application of geo-fencing technology to set a restricted area of ​​the virtual isolation zone, integrating information of the restricted area into the preliminary flight path plan to generate a flight path, and generating a flight path plan based on the flight path combined with a time window control strategy and a path planning algorithm based on multi-objective optimization, including:

[0033] Applying geo-fencing technology to set a restricted area of ​​the virtual isolation zone;

[0034] Based on the restricted area, the preliminary flight path plan is integrated and processed, and information of the restricted area is input into the preliminary flight path plan to generate a flight path;

[0035] According to the flight path, combined with the predicted airspace use rules, a time window control strategy is introduced to obtain an optimized time window arrangement;

[0036] Based on the optimized time window arrangement, the flight path is optimized using a path planning algorithm based on multi-objective optimization to generate a flight path plan.

[0037] Optionally, based on multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone, including:

[0038] Using multi-agent system theory, the flight activity information is distributed to multiple agents in the multi-agent system theory to obtain flight activity information fragments;

[0039] Preprocessing the flight activity information fragments to generate preprocessed flight activity information;

[0040] Identifying the area requiring dynamic adjustment from the preprocessed flight activity information to obtain an evaluation result;

[0041] According to the evaluation results, a virtual isolation zone is constructed by combining distributed algorithms, utilizing space partitioning technology and geographic information system tools.

[0042] In a second aspect, the embodiment of the present application provides a networked drone identification and control system, including:

[0043] A receiving module for receiving real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly obtained from drones;

[0044] Establishing a module for parallel processing of the flight activity information based on multi-agent system theory and distributed computing resources to establish a virtual isolation zone;

[0045] An evaluation module, for evaluating the flight intention of each UAV using a deep recurrent neural network and a long short-term memory network according to the virtual isolation zone and the real-time flight activity information, and using an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV and obtain a warning signal, wherein the flight intention is to input the real-time flight activity information and the multi-source historical UAV flight activity records of the UAV into a machine learning model to generate a prediction of the expected behavior of the UAV;

[0046] An identification module is used to identify a flight path from the flight intention based on the warning signal, obtain a flight path plan through geo-fencing technology, a time window control strategy, and a path planning algorithm based on multi-objective optimization, and send the warning signal and the flight path plan to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information;

[0047] A generation module is used to generate a networked drone identification and control process based on the real-time control feedback information combined with the early warning signal through a Bayesian update mechanism.

[0048] In a third aspect, an embodiment of the present application provides a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a networked drone identification and control method as described in any one of the first aspects.

[0049] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a method for identifying and controlling a networked drone as described in any one of the first aspects.

[0050] In an embodiment of the present application, real-time flight activity information is received from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly obtained from drones; based on multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone; based on the virtual isolation zone and the real-time flight activity information, a deep recurrent neural network and a long short-term memory network are used to evaluate the flight intention of each drone, and an anomaly detection algorithm is used to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the drone, and obtain a warning signal, and the flight The intention is to input the real-time flight activity information and the multi-source historical UAV flight activity records of the UAV into the machine learning model to generate a prediction of the expected behavior of the UAV; based on the early warning signal, identify the flight path from the flight intention, and obtain the flight path planning through geographic fencing technology, time window control strategy and path planning algorithm based on multi-objective optimization, and use the quantum encryption communication link to send the early warning signal and the flight path planning to the corresponding UAV to obtain real-time control feedback information; based on the real-time control feedback information combined with the early warning signal, a networked UAV identification and control process is generated through the Bayesian update mechanism.

[0051] The technical solution of this application has the following beneficial effects:

[0052] This method not only achieves accurate monitoring and management of drone flight activities, but also effectively predicts potential risks through intelligent algorithms, improving flight safety and efficiency. At the same time, quantum encryption communication ensures the security of data transmission, improves the reliability and response speed of the networked drone system as a whole, and provides solid technical support for the large-scale application of drones.

[0053] Further,

[0054] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 A flowchart of a method for identifying and controlling a networked drone provided in an embodiment of the present application;

[0057] Figure 2A schematic diagram of the structure of a networked drone identification and control system provided in an embodiment of the present application;

[0058] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0061] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0062] Figure 1 A flowchart of a method for identifying and controlling a networked drone is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0063] Step 101: receiving real-time flight activity information from multiple heterogeneous data sources;

[0064] In this step, multiple heterogeneous data sources including ground stations, satellite links, mobile networks, and data directly obtained from drones provide information about the flight status of drones, such as location, speed, altitude, etc., for comprehensive monitoring and management of drone behavior.

[0065] In actual operation, the system receives real-time information about the flight activities of drones by integrating multiple different types of sensors and communication links. The ground station provides high-precision location information, the satellite link covers a wide area, the mobile network ensures the stability of the connection of drones in urban environments, and the data directly obtained from the drone contains more detailed flight parameters.

[0066] For example, in a smart agriculture monitoring project, drones are used to monitor the growth of large areas of farmland. The system first receives basic location information of the drone through a ground station and uses a satellite link for remote control and data transmission. At the same time, the mobile network ensures that the drone can maintain a stable connection even in remote areas, and the drone itself also uploads its sensor data (such as temperature, humidity, images, etc.) for further analysis.

[0067] Step 102: Based on multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone;

[0068] In this step, the virtual isolation zone refers to a safe area set in the airspace to prevent potential conflicts between drones. The multi-agent system theory combined with distributed computing resources can efficiently process a large amount of flight activity information and dynamically adjust the setting of the virtual isolation zone.

[0069] In actual operation, through the distributed computing architecture, the system can quickly process massive amounts of flight activity information and update the location and range of the virtual isolation zone in real time, thereby avoiding the risk of collision between drones. This method improves the system's response speed and processing capabilities, ensuring the safe flight of drones in complex environments.

[0070] For example, in the above-mentioned smart agriculture monitoring project, as more and more drones are deployed for different tasks (such as spraying pesticides, sowing, etc.), the system needs to dynamically create multiple virtual isolation zones to avoid collisions between drones. Through distributed computing resources, the system can quickly respond to changes in flight paths, ensure that all drones fly within the safe area, and flexibly adjust the size and location of the isolation zone according to actual needs.

[0071] Step 103: Based on the virtual isolation zone and the real-time flight activity information, the flight intention of each UAV is evaluated using a deep recurrent neural network and a long short-term memory network, and an anomaly detection algorithm is used to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV, and obtain a warning signal. The flight intention is to input the real-time flight activity information and the multi-source historical UAV flight activity records of the UAV into a machine learning model to generate a prediction of the expected behavior of the UAV;

[0072] In this step, the Deep Recursive Neural Network (DRNN) and Long Short-Term Memory (LSTM) are used in combination with anomaly detection algorithms to evaluate the UAV's flight intention and identify abnormal behaviors. Flight intention is to input real-time flight activity information and historical records into the machine learning model to generate expected behavior predictions.

[0073] In actual operation, the system first builds a flight intention prediction model based on historical flight records and real-time data, then compares the predicted intention with the actual flight behavior, identifies any behavior that deviates from the normal pattern, and issues a warning signal. This step improves the detection accuracy of abnormal behavior and reduces the false alarm rate.

[0074] For example, in the smart agriculture monitoring project, the system uses DRNN and LSTM models to analyze the drone's historical flight data and real-time flight information to predict its future flight path. For example, when a drone suddenly deviates from the planned path, the system will immediately issue an early warning to remind the operator to check whether there is a mechanical failure or other problem and take appropriate measures.

[0075] Step 104: Based on the warning signal, a flight path is identified from the flight intention, and a flight path planning is obtained through geo-fencing technology, a time window control strategy, and a path planning algorithm based on multi-objective optimization. The warning signal and the flight path planning are sent to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information;

[0076] In this step, based on the early warning signal, the flight path is identified from the flight intention, and the flight path planning is formulated through geo-fencing technology, time window control strategy and path planning algorithm based on multi-objective optimization. It is then sent to the corresponding drone through a quantum encrypted communication link to ensure the safe transmission of instructions.

[0077] In actual operation, the system optimizes the flight path of the drone based on early warning signals and flight intentions, combined with geo-fencing technology and time window control strategy to maximize efficiency and safety. The quantum encrypted communication link ensures the security of command transmission and prevents unauthorized access.

[0078] For example, in the smart agriculture monitoring project, when the system detects that a drone may enter a no-fly zone, it will re-plan the flight path of the drone to avoid the no-fly zone. The optimized path is sent to the drone via a quantum encrypted communication link to ensure the safe transmission and execution of instructions. In addition, the system will also consider the time window control strategy to ensure that the drone completes the task within a specific time period.

[0079] Step 105: Based on the real-time control feedback information and the early warning signal, a networked drone identification and control process is generated through a Bayesian update mechanism.

[0080] In this step, based on real-time control feedback information combined with early warning signals, the networked drone identification and control process is generated through the Bayesian update mechanism. The Bayesian update mechanism allows the system to continuously adjust its decision model based on new information and improve prediction accuracy.

[0081] In actual operation, the system continuously collects real-time feedback information from drones, combines it with early warning signals, and continuously optimizes its identification and control processes through the Bayesian update mechanism. This method improves the system's adaptive ability and response speed, ensuring the safe operation of drones in complex environments.

[0082] For example, in the smart agriculture monitoring project, as drones complete their missions and return to base, the system will adjust future flight path planning based on their real-time feedback information and previous warning signals. For example, if a drone frequently exhibits abnormal behavior in a certain area, the system will re-evaluate the risk of the area through the Bayesian update mechanism and conduct stricter monitoring or flight restrictions in the area in future missions. This dynamic adjustment enables the system to better respond to emergencies and ensure the successful completion of each flight mission.

[0083] Through the above five steps, the system realizes comprehensive monitoring and management of UAV flight activities, significantly improving flight safety and efficiency. First, by integrating multi-source heterogeneous data, the system ensures the integrity and timeliness of the data; secondly, based on multi-agent system theory and distributed computing resources, the system can dynamically establish virtual isolation zones to effectively prevent potential conflicts between UAVs; then, using advanced machine learning and anomaly detection algorithms, the system improves the recognition accuracy of abnormal flight behaviors; thirdly, through geographic fencing technology, time window control strategy and multi-objective optimization path planning algorithm, the system formulates the optimal flight path and ensures the secure transmission of instructions through quantum encryption communication links; finally, through the Bayesian update mechanism, the system achieves self-optimization, further enhancing its adaptive ability and response speed. This whole set of processes provides solid technical support for the large-scale application of UAVs and ensures their safe operation in complex environments.

[0084] In order to further improve the prediction accuracy of the UAV flight intention and enhance the effectiveness of abnormal behavior detection, in some embodiments, the step 103 uses a deep recurrent neural network and a long short-term memory network to evaluate the flight intention of each UAV according to the virtual isolation zone and the real-time flight activity information, and uses an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of the abnormal flight behavior of the UAV, and obtain a warning signal, including:

[0085] The real-time flight activity information and the data of the virtual isolation zone are integrated and processed by using graph database technology to obtain a data set; historical UAV flight activity records from multiple sources are collected, and based on the data set and the historical UAV flight activity records, a flight intention prediction model of a deep recursive neural network and a long short-term memory network is constructed in combination with an attention mechanism, and an optimized prediction model is obtained by using transfer learning technology; the flight intention, flight path and initial behavior pattern of the UAV are predicted from the optimized prediction model, and the flight path and the initial behavior pattern are simulated and analyzed by a reinforcement learning algorithm to obtain a flight strategy, and the expected benefits under the flight strategy are simulated to evaluate the flight intention to obtain an optimal flight path and key behavior pattern; the optimal flight path and the key behavior pattern are compared and analyzed with the actual flight behavior of the UAV to obtain behavioral differences, and an isolation forest combined with anomaly detection algorithm is used to identify the target behavior pattern from the behavioral differences, and a local anomaly factor algorithm is introduced to mark the behavior of the target behavior pattern as abnormal flight behavior; a game theory model is used to predict a conflict scenario from the abnormal flight behavior, and a scenario simulation technology and a Monte Carlo method are combined to estimate the probability of the conflict scenario and generate an early warning signal.

[0086] In this embodiment, graph database technology is used to integrate real-time flight activity information and virtual isolation zone data to generate a comprehensive data set. This data set not only contains basic information such as the location and speed of the drone, but also includes its historical flight records and relevant information about the virtual isolation zone. By collecting historical drone flight activity records from multiple sources and combining the attention mechanism, a flight intention prediction model of deep recurrent neural network and long short-term memory network is constructed. This method can more accurately capture the complex flight patterns of drones, thereby improving prediction accuracy.

[0087] In an embodiment of the present application, first, the system uses graph database technology to process real-time flight activity information and data from virtual isolation zones to form a comprehensive data set. Then, based on this data set and multi-source historical flight records, combined with the attention mechanism, a flight intention prediction model is constructed. After optimizing the model through transfer learning technology, the system can predict the flight intention, path, and initial behavior pattern of the drone. Then, the reinforcement learning algorithm is used to simulate and analyze these paths and patterns to evaluate the optimal flight path and key behavior patterns. The optimal path is compared and analyzed with the actual flight behavior to identify behavioral differences, and an isolated forest combined with anomaly detection algorithm is used to mark abnormal behavior. Finally, a game theory model is used to predict conflict scenarios, and probability estimates are performed through scenario simulation technology and the Monte Carlo method to generate early warning signals.

[0088] The following is a specific embodiment:

[0089] In a smart city logistics and delivery project, drones are widely used to transport express parcels. To ensure that drones operate safely and efficiently in busy urban airspace, the system first integrates real-time flight activity information from ground stations, satellite links, and mobile networks, as well as dynamically established virtual isolation zone data, through graph database technology to form a comprehensive data set. This enables the system to fully understand the status of each drone and its surroundings.

[0090] Next, the system collected flight records of all drones in the past few months, combined with the current data set, and built a flight intention prediction model based on a deep recurrent neural network with an attention mechanism and a long short-term memory network. Through transfer learning technology, the model has been continuously optimized to improve its ability to understand complex flight patterns. For example, in a certain mission, when a drone was about to take off, the system used the optimized model to predict its flight intention and path, and simulated the expected benefits under different flight strategies through a reinforcement learning algorithm, and finally determined an optimal flight path.

[0091] However, during the execution, the system found that the drone deviated from the planned path and its behavior did not match the previous prediction. By comparing and analyzing the optimal path with the actual flight behavior, the system identified the behavioral differences and used the isolation forest combined with the anomaly detection algorithm to mark the drone's behavior as abnormal flight behavior. Subsequently, the system used game theory models to predict potential conflict scenarios and estimated the probability of conflict through scenario simulation technology and Monte Carlo methods. Based on these analysis results, the system generated early warning signals and promptly notified the operator to take measures, such as adjusting the flight plans of other drones or dispatching ground personnel to intervene, to ensure the smooth operation of the entire logistics distribution system.

[0092] In order to further improve the prediction accuracy of the UAV flight intention and enhance the effectiveness of the flight strategy evaluation, in some embodiments, the step 103 predicts the UAV flight intention, flight path and initial behavior pattern from the optimized prediction model, uses a reinforcement learning algorithm to simulate and analyze the flight path and the initial behavior pattern to obtain a flight strategy, simulates the expected benefits under the flight strategy, evaluates the flight intention, and obtains the optimal flight path and key behavior pattern, and also includes:

[0093] Based on the optimized prediction model combined with the attention mechanism and the introduction of the graph convolutional network, the flight intention, flight path and initial behavior pattern of each drone are predicted from the data set and the historical drone flight activity records; the flight path and the initial behavior pattern are simulated and analyzed by the reinforcement learning algorithm to obtain the flight strategy, and a multi-objective optimization framework is introduced to simulate the flight strategy to obtain a flight strategy with an expected return higher than a preset threshold; the initial behavior performance is identified from the initial behavior pattern, and the initial behavior performance and expected return of the drone in the flight strategy are simulated in combination with the virtual isolation zone and the preset airspace usage rules, and a flight strategy evaluation report is generated through scenario simulation technology and Monte Carlo method; based on the flight strategy evaluation report, the flight intention is evaluated to obtain an initial evaluation result, and a multimodal fusion technology combined with a hierarchical clustering algorithm is introduced to classify the initial evaluation result to generate a comprehensive evaluation result; based on the comprehensive evaluation result, a collision detection algorithm combined with a genetic algorithm is applied to determine the optimal flight path and key behavior pattern. Optionally, the method comprises: using a reinforcement learning algorithm to simulate and analyze the flight path and the initial behavior pattern to obtain a flight strategy, introducing a multi-objective optimization framework to simulate the flight strategy, and obtaining a flight strategy with an expected return higher than a preset threshold, including: using a reinforcement learning algorithm to simulate and analyze the flight path and the initial behavior pattern to obtain an initial flight strategy; using a multi-objective optimization framework to simulate the initial flight strategy to generate a target flight strategy, and based on the target flight strategy, applying a Monte Carlo method in combination with a Bayesian optimization algorithm to generate a flight strategy evaluation report; and according to a pre-set condition that the expected return is higher than a preset threshold, based on the condition in combination with a differential evolution algorithm, selecting the optimal flight strategy from the flight strategy evaluation report.

[0094] In this embodiment, based on the optimized prediction model combined with the attention mechanism, and the introduction of the graph convolutional network, the flight intention, flight path and initial behavior pattern of each drone are predicted from the data set and historical drone flight activity records. The graph convolutional network can effectively process complex graph structure data, such as the relative position relationship between drones and their relationship with the virtual isolation zone. In addition, through the multi-objective optimization framework, the system can evaluate the expected benefits of different flight strategies and ensure that they are higher than the preset threshold. The flight strategy evaluation report combines the results of scenario simulation technology and Monte Carlo method to analyze the feasibility and risks of flight strategies in detail.

[0095] In an embodiment of the present application, first, based on the optimized prediction model combined with the attention mechanism and the graph convolutional network, the system predicts the flight intention, path and initial behavior pattern of each drone from the data set and historical flight records. Then, the reinforcement learning algorithm is used to simulate and analyze these paths and behavior patterns to generate an initial flight strategy. Then, the initial flight strategy is simulated through a multi-objective optimization framework to generate a target flight strategy, and the Monte Carlo method is combined with the Bayesian optimization algorithm to generate a flight strategy evaluation report. According to the condition that the pre-set expected return is higher than the preset threshold, combined with the differential evolution algorithm, the optimal flight strategy is screened out from the flight strategy evaluation report. Finally, based on the comprehensive evaluation results, the collision detection algorithm is combined with the genetic algorithm to determine the optimal flight path and key behavior pattern.

[0096] The following is a specific embodiment:

[0097] In a smart agricultural monitoring project, drones are widely used for tasks such as farmland monitoring and pesticide spraying. To ensure the safe and efficient operation of drones in a busy and complex agricultural environment, the system first uses an optimized prediction model combined with an attention mechanism and a graph convolutional network to integrate real-time flight activity information from ground stations, satellite links, and mobile networks, as well as historical flight records, to predict the flight intention, path, and initial behavior pattern of each drone.

[0098] For example, in a farmland monitoring mission, the system uses the optimized model to predict that a drone will perform a monitoring mission in a specific area and plans a preliminary flight path. Next, the system uses a reinforcement learning algorithm to simulate and analyze this path and the initial behavior pattern to generate an initial flight strategy. Through the multi-objective optimization framework, the system simulates the initial flight strategy, generates multiple target flight strategies, and applies the Monte Carlo method combined with the Bayesian optimization algorithm to generate a detailed flight strategy evaluation report. The report analyzes the expected benefits and potential risks of each strategy in detail. Based on the condition that the expected benefits are higher than the preset threshold, the system combines the differential evolution algorithm to select the optimal flight strategy from the flight strategy evaluation report. Subsequently, the system combines the virtual isolation zone and the preset airspace use rules to simulate the initial behavior performance and expected benefits in the flight strategy in detail and generate a flight strategy evaluation report. Based on this report, the system uses multimodal fusion technology combined with a hierarchical clustering algorithm to classify the initial evaluation results and generate a comprehensive evaluation result.

[0099] Finally, based on the comprehensive evaluation results, the system applied collision detection algorithms combined with genetic algorithms to determine the optimal flight path and key behavior patterns. This not only ensures that the drone can operate safely in complex environments, but also maximizes mission efficiency and benefits. For example, in actual operation, when the system finds that a drone deviates from the predetermined path, it can quickly adjust the flight strategy, replan the path, and ensure that the drone can successfully complete the mission while avoiding conflicts with other drones. This series of steps significantly improves the flexibility and reliability of the drone system and provides strong support for smart agricultural monitoring.

[0100] In order to further improve the accuracy and real-time adjustment capability of flight path planning, in some embodiments, the step 104 identifies the flight path from the flight intention based on the warning signal, obtains the flight path planning through geo-fencing technology, time window control strategy and multi-objective optimization-based path planning algorithm, and uses a quantum encryption communication link to send the warning signal and the flight path planning to the corresponding drone to obtain real-time control feedback information, including:

[0101] Based on the flight intention, the warning signal is analyzed to identify the flight path related to the warning signal and obtain a preliminary flight path plan; the geo-fence technology is applied to set the restricted area of ​​the virtual isolation zone, and the information of the restricted area is integrated into the preliminary flight path plan to generate a flight path, and based on the flight path combined with a time window control strategy, a path planning algorithm based on multi-objective optimization is used to generate a flight path plan; using a quantum encryption communication link, the warning signal and the flight path plan are sent to the corresponding UAV to obtain a real-time adjusted flight status, and based on the flight status, real-time management and control feedback information is obtained. Optionally, the application of geo-fence technology is used to set the restricted area of ​​the virtual isolation zone, and the information of the restricted area is integrated into the preliminary flight path plan to generate a flight path. Based on the flight path and a time window control strategy, a path planning algorithm based on multi-objective optimization is used to generate a flight path plan, including: applying geo-fence technology to set the restricted area of ​​the virtual isolation zone; based on the restricted area, integrating the preliminary flight path plan, and inputting the information of the restricted area into the preliminary flight path plan to generate a flight path; according to the flight path, in combination with predicted airspace usage rules, a time window control strategy is introduced to obtain an optimized time window arrangement; based on the optimized time window arrangement, a path planning algorithm based on multi-objective optimization is used to optimize the flight path and generate a flight path plan.

[0102] In this embodiment, the warning signal is analyzed based on the flight intention to identify the flight path related to the warning signal and generate a preliminary flight path plan. Geofencing technology is used to set the restricted areas of the virtual isolation zone, and the information of these restricted areas is integrated into the preliminary flight path plan to generate a detailed flight path. Combined with the time window control strategy, the path planning algorithm based on multi-objective optimization is used to generate the final flight path plan. The quantum encrypted communication link ensures the secure transmission of the warning signal and the flight path plan, so as to obtain the real-time adjusted flight status and control feedback information.

[0103] In an embodiment of the present application, first, the system analyzes the warning signal based on the flight intention, identifies the flight path related to it, and generates a preliminary flight path plan. Next, the geo-fence technology is used to set the restricted areas of the virtual isolation zone, and the information of these restricted areas is integrated into the preliminary flight path plan to generate a detailed flight path. Then, according to the flight path combined with the predicted airspace usage rules, a time window control strategy is introduced to optimize the time window arrangement. Based on the optimized time window arrangement, the path planning algorithm based on multi-objective optimization is used to further optimize the flight path and generate the final flight path plan. Finally, the quantum encrypted communication link is used to securely send the warning signal and flight path planning to the corresponding drone, and obtain the real-time adjusted flight status and control feedback information.

[0104] The following is a specific embodiment:

[0105] In a smart city logistics and delivery project, drones are widely used to transport express parcels. To ensure that drones operate safely and efficiently in busy urban airspace, the system first analyzes warning signals based on flight intent, identifies the flight paths associated with them, and generates a preliminary flight path plan. For example, when the system detects that a drone may enter a no-fly zone, it will issue a warning signal and generate a preliminary flight path that avoids the no-fly zone.

[0106] Next, the system applies geo-fencing technology to set restricted areas of the virtual isolation zone and integrates the information of these restricted areas into the preliminary flight path plan. Assuming that there is a temporary no-fly zone in a certain area, the system will avoid this area in the preliminary flight path and generate a detailed flight path. Based on this flight path, combined with the predicted airspace usage rules (such as flight restrictions during a specific time period), the system introduces a time window control strategy to optimize the time window arrangement. For example, during peak hours, the system may arrange drones to perform tasks during off-peak hours to reduce conflicts with other aircraft. Based on the optimized time window arrangement, the system uses a path planning algorithm based on multi-objective optimization to further optimize the flight path and generate the final flight path plan. This not only ensures that the drone can avoid the no-fly zone, but also maximizes the efficiency of the mission. For example, the system may choose a path that is slightly longer but safer and meets the time window requirements.

[0107] Finally, using quantum encrypted communication links, the system securely sends warning signals and the final flight path planning to the corresponding drones. After receiving the instructions, the drones perform the mission according to the new path planning and provide real-time feedback on their current flight status. Based on this real-time feedback information, the system can further adjust the flight path or take other necessary measures to ensure the smooth operation of the entire logistics distribution system.

[0108] In order to further improve the processing efficiency of flight activity information and dynamically adjust the virtual isolation zone to ensure the safe operation of the UAV in a complex environment, in some embodiments, the flight activity information is processed in parallel based on multi-agent system theory and distributed computing resources in step 102 to establish a virtual isolation zone, including:

[0109] By using the multi-agent system theory, the flight activity information is distributed to multiple agents in the multi-agent system theory to obtain flight activity information fragments; the flight activity information fragments are preprocessed to generate preprocessed flight activity information; the area that needs to be dynamically adjusted is identified from the preprocessed flight activity information to obtain an evaluation result; according to the evaluation result, in combination with a distributed algorithm, a virtual isolation zone is constructed using space partitioning technology and geographic information system tools.

[0110] In this embodiment, the flight activity information is distributed to multiple agents using the theory of multi-agent systems, and each agent is responsible for processing a specific part of the information fragments. These information fragments contain key data such as the position, speed, and altitude of the drone, which are used to evaluate the status of the flight activity. By preprocessing these information fragments, preprocessed flight activity information is generated. Next, the system identifies the areas that need to be dynamically adjusted from the preprocessed flight activity information, and combines distributed algorithms to construct virtual isolation areas using spatial partitioning technology and geographic information system (GIS) tools. This method not only improves processing efficiency, but also enhances the system's adaptive ability.

[0111] In an embodiment of the present application, first, the system uses multi-agent system theory to distribute flight activity information to multiple agents, and each agent is responsible for processing a portion of the information. Next, these information fragments are preprocessed to generate standardized flight activity information. Then, the system identifies areas that need dynamic adjustment from the preprocessed information, such as drone-dense areas or potential conflict areas, and generates evaluation results. Based on these evaluation results, combined with distributed algorithms (such as MapReduce or Spark), spatial partitioning technology and GIS tools are used to construct a virtual isolation zone. This step ensures that the drone can fly safely in a dynamically changing environment and avoid collisions with other drones.

[0112] The following is a specific embodiment:

[0113] In a smart agricultural monitoring project, drones are widely used for tasks such as farmland monitoring and pesticide spraying. To ensure that drones operate safely and efficiently in large areas of farmland, the system first uses multi-agent system theory to distribute flight activity information to multiple agents. Each agent is responsible for processing drone flight data in a specific area, such as location, speed, and altitude. For example, an agent may be responsible for processing data from all drones in a specific farmland area. Next, the system preprocesses these information fragments to generate standardized flight activity information. This step includes operations such as data cleaning, format conversion, and outlier detection to ensure the accuracy of subsequent analysis. For example, during a mission, the system detected that the speed of a drone increased abnormally, which may be caused by an increase in wind speed. The system will mark this outlier and take it into account in subsequent analysis.

[0114] The system then identifies areas that need to be dynamically adjusted from the pre-processed flight activity information. Assuming that a large number of drones are operating simultaneously in a certain farmland area, the system will identify this area as a high-risk area and generate an assessment result. Based on these assessment results, the system combines distributed algorithms (such as Apache Spark) with spatial partitioning technology and GIS tools to build virtual isolation zones. For example, the system may divide the high-risk area into several small isolation zones, and each drone is assigned to a different isolation zone to avoid mutual interference.

[0115] In this way, the system can not only monitor the flight status of drones in real time, but also dynamically adjust the virtual isolation zone according to the actual situation to ensure the safe operation of drones in complex farmland environments. For example, during a pesticide spraying mission, when the system finds that the number of drones in a certain farmland area suddenly increases, it will immediately adjust the virtual isolation zone and reallocate the drones' working areas to ensure that each drone can complete the mission efficiently and safely.

[0116] This application considers that in the process of drone flight management and monitoring, accurately predicting the flight intention of drones and timely identifying abnormal behaviors are the key to ensuring flight safety. Traditional flight management methods often rely on static settings and predefined rules, which are difficult to cope with complex and changing practical application needs. In order to improve the accuracy of predicting drone flight intentions and enhance the effectiveness of abnormal behavior detection, this application proposes a new optional solution, which includes:

[0117] The method of evaluating the flight intention of each UAV using a deep recurrent neural network and a long short-term memory network based on the virtual isolation zone and the real-time flight activity information, and using an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV and obtain a warning signal includes:

[0118] Using graph database technology, the real-time flight activity information and the data of the virtual isolation zone are integrated and processed to obtain a data set;

[0119] Historical drone flight activity records are collected from multiple sources. Based on the data set and the historical drone flight activity records, a flight intention prediction model of a deep recurrent neural network and a long short-term memory network is constructed in combination with an attention mechanism. The optimized prediction model is obtained through transfer learning technology. In this process, the parameters of the optimized prediction model are calculated using the following formula:

[0120] ;

[0121] in, is the model weight matrix or parameter vector, which contains all the parameters that need to be trained. represents the optimized model parameters, and Respectively represent The input features and labels of samples, is the loss function, is the number of training samples, is the regularization coefficient, It is a broad norm, where and are different values, such as and , to accommodate different types of weight decay, is the weight of transfer learning, is the Kullback-Leibler divergence, is the weight of the additional constraint, is a regularization term based on the attention mechanism, is the attention parameter;

[0122] The following is a detailed explanation of each parameter:

[0123] Represents the model weight matrix or parameter vector, which contains all the parameters that need to be trained. It is learned from the data through optimization algorithms (such as gradient descent).

[0124] Represents the number of training samples, which is directly determined by the size of the dataset.

[0125] Represents the loss function, which measures the difference between the model prediction value and the true label. Usually, cross entropy loss or mean square error are selected. The design of the loss function is based on the specific task requirements. For example, the classification task uses cross entropy loss, and the regression task uses mean square error.

[0126] Represents the regularization coefficient, which is used to control the degree of weight decay to prevent overfitting. The optimal value is usually determined by cross-validation.

[0127] Indicates a general norm, used for regularization. Different and The values ​​are adapted to different types of data and model complexity. For example, and Effectively control the complexity of the model.

[0128] Represents the Kullback-Leibler divergence, which measures the difference between two distributions. In transfer learning, it is used to evaluate the similarity between the current model parameters and the pre-trained model parameters. Pre-trained model parameters Usually comes from an already trained model.

[0129] Represents the weight of transfer learning, controlling the importance of the KL divergence term. Adjusted through experiments or set based on experience.

[0130] Represents the regularization term based on the attention mechanism, which enhances the model's attention to important features. Attention parameter Learned through the training process.

[0131] Represents the weight of the additional constraint, controlling the influence of the regularization term. Usually adjusted through cross-validation or experiments.

[0132] The following is an introduction to the design reasons of each sub-item:

[0133] Loss function term The reason for the design of is to measure the difference between the model's predicted value and the actual label. By minimizing this term, the prediction accuracy of the model can be improved.

[0134] Regularization term The reason for designing is that the regularization term prevents the model from overfitting, especially when the model is too complex. norm, which can flexibly control different types of weight decay.

[0135] KL divergence term The reason for the design is that in transfer learning, the KL divergence term ensures that the current model parameters are close to the pre-trained model parameters, thereby utilizing the knowledge of the pre-trained model to accelerate convergence and improve generalization ability.

[0136] Attention Regularization Term The reason for the design of is that the attention mechanism helps the model better focus on important features and reduce noise interference. By introducing regularization terms, the robustness and interpretability of the model can be enhanced.

[0137] The purpose of adding up the sub-items is to directly optimize the prediction performance of the model through the loss function. The regularization term prevents the model from overfitting and improves the generalization ability. The KL divergence term maintains the stability of the model parameters in transfer learning and utilizes existing knowledge. The attention regularization term enhances the model's focus on key features and improves the robustness and interpretability of the model. By adding up these sub-items, multiple optimization objectives can be considered simultaneously in a single objective function to achieve a comprehensive optimization effect.

[0138] Predicting the flight intention, flight path and initial behavior pattern of the UAV from the optimized prediction model, simulating and analyzing the flight path and the initial behavior pattern using a reinforcement learning algorithm, obtaining a flight strategy, simulating the expected benefits under the flight strategy, evaluating the flight intention, and obtaining an optimal flight path and key behavior pattern;

[0139] The optimal flight path and the key behavior pattern are compared and analyzed with the actual flight behavior of the UAV to obtain the behavior difference. The target behavior pattern is identified from the behavior difference by using the isolation forest combined with the anomaly detection algorithm. The local anomaly factor algorithm is introduced to mark the behavior of the target behavior pattern as abnormal flight behavior. In this process, the local anomaly factor is used to quantify the abnormal degree of the behavior of the target behavior pattern:

[0140] ;

[0141] in, Yes The local anomaly factor, Yes The local reachable density of Yes of Neighbor set, yes The size of the neighbor set, Yes The average distance to its nearest neighbors, is the standard deviation of the distance, Yes and its neighbors The distance between is the standard deviation;

[0142] The following is a detailed explanation of each parameter:

[0143] Indicate point The local anomaly factor is used to measure the point The degree of deviation from the density of its neighbors. The larger the value, the more abnormal the point is.

[0144] Indicate point of Neighbor set, including distance points Recent It is obtained by calculating the Euclidean distance or other distance measurement methods between points.

[0145] Indicate point The local reachability density of the point The density of surrounding points. It is determined by the average reachable distance between the points and their nearest neighbors.

[0146] Indicate point The local reachable density of Similar, but for different points .

[0147] Indicate point to The average distance of neighboring points. The distance between it and all its neighboring points is averaged.

[0148] Indicates the standard deviation of the distance, used for normalization , so that it is not affected by the data scale. It is usually obtained by calculating the standard deviation of the distance between all points.

[0149] Indicate point and its neighbors The distance between them is calculated using Euclidean distance or other suitable distance metric.

[0150] Represents the standard deviation, used to standardize points The sum of the distances to its nearest neighbors, making it unaffected by the scale of the data. Usually obtained by calculating the standard deviation of the distances between all points.

[0151] The following is an introduction to the design reasons of each sub-item:

[0152] Locally accessible density ratio The reason for the design is that the local reachable density ratio is used to compare the points and its neighbors If The density of is significantly lower than that of its neighbors, then More likely an outlier.

[0153] Exponential decay term The reason for the design of is that the exponential decay term is used to adjust the influence of the local reachable density ratio. When the average distance to its neighboring points is large, this term will reduce the impact of the local reachable density ratio, thereby reducing the impact of noise on anomaly detection.

[0154] Average local reachability density ratio The design reason is that by averaging the local reachability density ratios of all neighboring points, the point All neighboring points around to avoid excessive influence of a single neighboring point.

[0155] Exponential decay term of the sum of distances The reason for the design is that the exponential decay term is used to adjust the point The total distance to its neighbors. A larger total distance means that the point The farther away from other neighboring points, the more likely they are outliers.

[0156] The reason for multiplying the sub-items together is that they each provide different information. The local reachability density ratio reflects the The density difference relative to its neighboring points is the core basis for judging anomalies. The exponential decay term adjusts the influence of the local reachable density ratio, so that the influence of points with larger distances on other points is weakened, reducing noise interference. The exponential decay term of the distance sum further emphasizes the point The sum of the distances to its nearest neighbors enhances the ability to identify isolated points.

[0157] Here is a specific example:

[0158] Assume that in a smart agricultural monitoring project, the system needs to monitor and manage multiple drones to perform farmland monitoring tasks. The following is the specific numerical substitution and calculation process:

[0159] Dataset construction: Assume that the system receives 100 drone flight samples , each sample contains features such as position, speed, etc. ( ) and its corresponding label .

[0160] Model optimization: Use transfer learning techniques to optimize model parameters, assuming regularization coefficients , broad sense Norm and , transfer learning weights , additional constraint weight Assuming the Kullback-Leibler divergence , regularization term based on attention mechanism . Substitute into the formula to calculate:

[0161] ;

[0162] Flight strategy evaluation: Assume that the system predicts that a drone’s flight intention is to fly from point A to point B. After simulation analysis, the optimal flight path and key behavior patterns are determined.

[0163] Abnormal behavior detection: If a drone deviates from the planned path during a flight, the system will compare its behavior with the normal pattern and find the difference in behavior. The degree of abnormality is calculated using the local anomaly factor algorithm:

[0164] ;

[0165] Assumptions , calculated:

[0166] ;

[0167] Assumptions , calculated , indicating that the drone's behavior was significantly abnormal.

[0168] Conflict scenario prediction: Assuming that the system predicts that the drone may collide with other drones, the probability of conflict is estimated through scenario simulation technology and Monte Carlo method. The system generates an early warning signal to alert the operator to take action.

[0169] From the calculation results, it can be seen that the system not only improves the prediction accuracy of the drone's flight intention, but also enhances the ability to identify abnormal behavior. Specifically, when a drone is detected to deviate from the predetermined path, the system calculates its local anomaly factor (LOF) value as , indicating that the drone's behavior is significantly abnormal, and then a warning signal is issued, and the flight plans of other drones are adjusted according to the prediction results to avoid potential conflicts. This series of steps significantly improves the flexibility and reliability of the drone system, ensuring that drones can operate safely and efficiently in complex environments, while effectively preventing possible safety hazards.

[0170] The game theory model is used to predict the conflict scenario generated from the abnormal flight behavior. The scenario simulation technology and Monte Carlo method are combined to estimate the probability of the conflict scenario and generate an early warning signal.

[0171] Figure 2 A schematic diagram of the structure of a networked drone identification and control system is provided for an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0172] A receiving module 21, for receiving real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly obtained from drones;

[0173] Establishing module 22, for parallel processing of the flight activity information based on multi-agent system theory and distributed computing resources to establish a virtual isolation zone;

[0174] An evaluation module 23 is used to evaluate the flight intention of each drone using a deep recurrent neural network and a long short-term memory network according to the virtual isolation zone and the real-time flight activity information, and to use an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the drone and obtain a warning signal. The flight intention is to input the real-time flight activity information and the multi-source historical drone flight activity records of the drone into a machine learning model to generate a prediction of the expected behavior of the drone;

[0175] The identification module 24 is used to identify the flight path from the flight intention based on the warning signal, obtain the flight path planning through the geo-fencing technology, the time window control strategy and the path planning algorithm based on multi-objective optimization, and send the warning signal and the flight path planning to the corresponding UAV by using the quantum encryption communication link to obtain real-time control feedback information;

[0176] The generation module 25 is used to generate a networked drone identification and control process based on the real-time control feedback information combined with the early warning signal through a Bayesian update mechanism.

[0177] Figure 2 The networked drone identification and control system can perform Figure 1 The implementation principle and technical effect of the networked drone identification and control method described in the embodiment are not described in detail. The specific way in which each module and unit performs operations in the networked drone identification and control system in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0178] In one possible design, Figure 2 A networked drone identification and control system of the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0179] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0180] The processing component 32 is used for the above Figure 1 A method for identifying and controlling a networked drone in the embodiment.

[0181] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0182] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0183] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0184] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0185] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0186] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0187] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for identifying and controlling a networked drone according to the illustrated embodiment.

[0188] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0189] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0190] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying and controlling networked drones, characterized in that: include: receiving real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly from the drone; Based on multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone; According to the virtual isolation zone and the real-time flight activity information, the flight intention of each drone is evaluated using a deep recurrent neural network and a long short-term memory network, and an anomaly detection algorithm is used to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the drone to obtain a warning signal, wherein the flight intention is to input the real-time flight activity information and the multi-source historical drone flight activity records of the drone into a machine learning model to generate a prediction of the expected behavior of the drone; Based on the warning signal, the flight path is identified from the flight intention, and the flight path planning is obtained through the geo-fencing technology, the time window control strategy and the path planning algorithm based on multi-objective optimization. The warning signal and the flight path planning are sent to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information; Based on the real-time control feedback information combined with the early warning signal, a networked drone identification and control process is generated through a Bayesian update mechanism; The method of evaluating the flight intention of each UAV using a deep recurrent neural network and a long short-term memory network based on the virtual isolation zone and the real-time flight activity information, and using an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV and obtain a warning signal includes: Using graph database technology, the real-time flight activity information and the data of the virtual isolation zone are integrated and processed to obtain a data set; Collect historical drone flight activity records from multiple sources, and build a flight intention prediction model of deep recursive neural network and long short-term memory network based on the data set and the historical drone flight activity records, and obtain an optimized prediction model through transfer learning technology; Predicting the flight intention, flight path and initial behavior pattern of the UAV from the optimized prediction model, simulating and analyzing the flight path and the initial behavior pattern using a reinforcement learning algorithm, obtaining a flight strategy, simulating the expected benefits under the flight strategy, evaluating the flight intention, and obtaining an optimal flight path and key behavior pattern; Compare and analyze the optimal flight path and the key behavior pattern with the actual flight behavior of the UAV to obtain behavior differences, use an isolation forest combined with anomaly detection algorithm to identify the target behavior pattern from the behavior differences, introduce a local anomaly factor algorithm, and mark the behavior of the target behavior pattern as abnormal flight behavior; The game theory model is used to predict conflict scenarios from the abnormal flight behavior, and the probability of the conflict scenarios is estimated by combining scenario simulation technology and Monte Carlo method to generate early warning signals.

2. The method according to claim 1, characterized in that The optimized prediction model is used to predict the flight intention, flight path and initial behavior pattern of the UAV, and the flight path and initial behavior pattern are simulated and analyzed by the reinforcement learning algorithm to obtain the flight strategy, and the expected benefits under the flight strategy are simulated to evaluate the flight intention to obtain the optimal flight path and key behavior pattern, including: Based on the optimized prediction model combined with the attention mechanism and the introduction of a graph convolutional network, the flight intention, flight path and initial behavior pattern of each drone are predicted from the data set and the historical drone flight activity records; Using a reinforcement learning algorithm, simulating and analyzing the flight path and the initial behavior pattern to obtain a flight strategy, introducing a multi-objective optimization framework to simulate the flight strategy, and obtaining a flight strategy with an expected benefit higher than a preset threshold; Identify initial behavior performance from the initial behavior pattern, simulate the initial behavior performance and expected benefits of the drone in the flight strategy in combination with the virtual isolation zone and the preset airspace use rules, and generate a flight strategy evaluation report through scenario simulation technology and Monte Carlo method; Based on the flight strategy evaluation report, the flight intention is evaluated to obtain an initial evaluation result, and a multimodal fusion technology is introduced in combination with a hierarchical clustering algorithm to classify the initial evaluation result to generate a comprehensive evaluation result; Based on the comprehensive evaluation results, a collision detection algorithm is applied in combination with a genetic algorithm to determine the optimal flight path and key behavior patterns.

3. The method according to claim 2, characterized in that The method utilizes a reinforcement learning algorithm to simulate and analyze the flight path and the initial behavior pattern to obtain a flight strategy, introduces a multi-objective optimization framework, simulates the flight strategy, and obtains a flight strategy with an expected benefit higher than a preset threshold, including: Using a reinforcement learning algorithm, simulating and analyzing the flight path and the initial behavior pattern to obtain an initial flight strategy; Using a multi-objective optimization framework, the initial flight strategy is simulated to generate a target flight strategy, and based on the target flight strategy, a Monte Carlo method combined with a Bayesian optimization algorithm is applied to generate a flight strategy evaluation report; According to a pre-set condition that the expected benefit is higher than a preset threshold, based on the condition and in combination with a differential evolution algorithm, an optimal flight strategy is screened out from the flight strategy evaluation report.

4. The method according to claim 1, characterized in that: Based on the warning signal, the flight path is identified from the flight intention, and the flight path planning is obtained through the geo-fencing technology, the time window control strategy and the path planning algorithm based on multi-objective optimization. The warning signal and the flight path planning are sent to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information, including: Based on the flight intention, analyzing the warning signal to identify a flight path related to the warning signal and obtain a preliminary flight path plan; Applying geo-fencing technology to set a restricted area of ​​the virtual isolation zone, integrating information of the restricted area into the preliminary flight path plan to generate a flight path, and generating a flight path plan based on the flight path combined with a time window control strategy and a path planning algorithm based on multi-objective optimization; By using a quantum encrypted communication link, the warning signal and the flight path planning are sent to the corresponding UAV to obtain a real-time adjusted flight status, and based on the flight status, real-time control feedback information is obtained.

5. The method according to claim 4, characterized in that The application of geo-fencing technology, setting the restricted area of ​​the virtual isolation zone, integrating the information of the restricted area into the preliminary flight path plan, generating a flight path, and generating a flight path plan based on the flight path combined with a time window control strategy and a path planning algorithm based on multi-objective optimization, including: Applying geo-fencing technology to set a restricted area of ​​the virtual isolation zone; Based on the restricted area, the preliminary flight path plan is integrated and processed, and information of the restricted area is input into the preliminary flight path plan to generate a flight path; According to the flight path, combined with the predicted airspace use rules, a time window control strategy is introduced to obtain an optimized time window arrangement; Based on the optimized time window arrangement, the flight path is optimized using a path planning algorithm based on multi-objective optimization to generate a flight path plan.

6. The method according to claim 1, characterized in that Based on the multi-agent system theory and distributed computing resources, the flight activity information is processed in parallel to establish a virtual isolation zone, including: Using multi-agent system theory, the flight activity information is distributed to multiple agents in the multi-agent system theory to obtain flight activity information fragments; Preprocessing the flight activity information fragments to generate preprocessed flight activity information; Identifying the area requiring dynamic adjustment from the preprocessed flight activity information to obtain an evaluation result; According to the evaluation results, a virtual isolation zone is constructed by combining distributed algorithms, utilizing space partitioning technology and geographic information system tools.

7. A networked drone identification and control system, characterized in that: include: A receiving module for receiving real-time flight activity information from multiple heterogeneous data sources, including ground stations, satellite links, mobile networks, and data directly obtained from drones; Establishing a module for parallel processing of the flight activity information based on multi-agent system theory and distributed computing resources to establish a virtual isolation zone; An evaluation module, for evaluating the flight intention of each UAV using a deep recurrent neural network and a long short-term memory network according to the virtual isolation zone and the real-time flight activity information, and using an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV and obtain a warning signal, wherein the flight intention is to input the real-time flight activity information and the multi-source historical UAV flight activity records of the UAV into a machine learning model to generate a prediction of the expected behavior of the UAV; An identification module is used to identify a flight path from the flight intention based on the warning signal, obtain a flight path plan through geo-fencing technology, a time window control strategy, and a path planning algorithm based on multi-objective optimization, and send the warning signal and the flight path plan to the corresponding UAV using a quantum encryption communication link to obtain real-time control feedback information; A generation module, for generating a networked drone identification and control process based on the real-time control feedback information combined with the warning signal through a Bayesian update mechanism; The method of evaluating the flight intention of each UAV using a deep recurrent neural network and a long short-term memory network based on the virtual isolation zone and the real-time flight activity information, and using an anomaly detection algorithm to compare the flight intention with the actual flight behavior to identify and warn of abnormal flight behavior of the UAV and obtain a warning signal includes: Using graph database technology, the real-time flight activity information and the data of the virtual isolation zone are integrated and processed to obtain a data set; Collect historical drone flight activity records from multiple sources, and build a flight intention prediction model of deep recursive neural network and long short-term memory network based on the data set and the historical drone flight activity records, and obtain an optimized prediction model through transfer learning technology; Predicting the flight intention, flight path and initial behavior pattern of the UAV from the optimized prediction model, simulating and analyzing the flight path and the initial behavior pattern using a reinforcement learning algorithm, obtaining a flight strategy, simulating the expected benefits under the flight strategy, evaluating the flight intention, and obtaining an optimal flight path and key behavior pattern; Compare and analyze the optimal flight path and the key behavior pattern with the actual flight behavior of the UAV to obtain behavior differences, use an isolation forest combined with anomaly detection algorithm to identify the target behavior pattern from the behavior differences, introduce a local anomaly factor algorithm, and mark the behavior of the target behavior pattern as abnormal flight behavior; The game theory model is used to predict conflict scenarios from the abnormal flight behavior, and the probability of the conflict scenarios is estimated by combining scenario simulation technology and Monte Carlo method to generate early warning signals.

8. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a networked drone identification and control method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, a networked drone identification and control method as described in any one of claims 1 to 6 is implemented.

Images