Multi-source data fusion unmanned aerial vehicle low-altitude air traffic flow distribution method and system

Abstract

This invention discloses a method and system for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion, relating to the field of UAV air traffic management technology. The method includes: identifying the type of a target non-cooperative UAV based on multi-source data and obtaining its performance parameters; assessing the avoidance risk coefficient by comprehensively considering the predicted communication latency and UAV density of the target area; delineating the potential risk range by combining the performance status of cooperative UAVs, and dynamically correcting it to a real-time risk range based on the real-time motion status of non-cooperative UAVs and the regional risk coefficient; and using a Q-learning reinforcement learning algorithm to integrate a 3D building map with the real-time risk range of each UAV, and collaboratively and dynamically allocating traffic flow for cooperative UAVs. This invention achieves accurate perception and global proactive avoidance of threats from non-cooperative UAVs, effectively improving the safety and operational efficiency of low-altitude airspace traffic management.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) air traffic management technology, specifically to a method and system for allocating low-altitude air traffic flow for UAVs based on multi-source data fusion. Background Technology

[0002] With the rapid expansion of drone applications, low-altitude airspace is becoming increasingly busy, placing higher demands on air traffic management capabilities. Non-cooperative drones, which cannot be identified or managed, pose a significant safety risk. The uncertainty of their behavior renders traditional traffic management methods, which rely on fixed routes and preset intervals, ineffective in dealing with such sudden intrusions, making it difficult to ensure the stability and safety of overall traffic flow.

[0003] For non-cooperative targets, existing methods mostly focus on using radar or optical equipment for detection and tracking, lacking an effective strategy to transform detection information into dynamic and coordinated adjustments to the entire traffic flow. Therefore, they can only make emergency avoidance to individual drones that directly conflict with non-cooperative drones. This localized and passive response is prone to triggering chain reactions, leading to larger-scale traffic chaos or new conflict risks, and cannot optimize the allocation of traffic flow from a global perspective. Summary of the Invention

[0004] This invention addresses the technical problems in existing technologies, such as the inability to effectively handle intrusions by non-cooperative drones, the lack of global dynamic traffic flow adjustment capabilities, and the potential for chain reactions caused by local avoidance strategies. It provides a method and system for allocating low-altitude air traffic flow for drones by fusing multi-source data.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides a method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion, including:

[0007] Based on multi-source data, the model of the target non-cooperative drone is obtained, and based on the model of the target non-cooperative drone, the performance parameters of the target non-cooperative drone are obtained.

[0008] Acquire drone prediction communication latency and drone density in the target area, and assess the avoidance risk coefficient of the target area;

[0009] Obtain the performance parameters and status information of all cooperative drones within the target area, and combine them with the performance parameters of the target non-cooperative drones to define the potential risk range of each cooperative drone;

[0010] The real-time motion status of the target non-cooperative drone is obtained. Combined with the performance parameters of the target non-cooperative drone and the avoidance risk coefficient of the target area, the potential risk range of the cooperative drone is dynamically corrected to obtain the real-time risk range.

[0011] Based on machine learning, the drone traffic flow of cooperative drones is dynamically corrected by combining a 3D building map of the target area and the real-time risk range of each cooperative drone, thus avoiding non-cooperative drones.

[0012] Secondly, this invention provides a multi-source data fusion-based UAV low-altitude air traffic flow distribution system, comprising:

[0013] The data acquisition and processing module is used to obtain the model of the target non-cooperative drone based on multi-source data, and to obtain the performance parameters of the target non-cooperative drone based on the model of the target non-cooperative drone.

[0014] The regional risk assessment module is used to obtain the predicted communication latency and drone density of the target area, and to assess the avoidance risk coefficient of the target area.

[0015] The static risk range delineation module is used to obtain the performance parameters and status information of all cooperative drones within the target area, and, in combination with the performance parameters of the target non-cooperative drones, delineate the potential risk range of each cooperative drone.

[0016] The dynamic risk range correction module is used to obtain the real-time motion status of the target non-cooperative drone, and combine the performance parameters of the target non-cooperative drone and the avoidance risk coefficient of the target area to dynamically correct the potential risk range of the cooperative drone to obtain the real-time risk range.

[0017] The traffic flow dynamic allocation module is used to dynamically correct the drone traffic flow of cooperative drones based on machine learning, combined with the 3D building map of the target area and the real-time risk range of each cooperative drone, so as to avoid non-cooperative drones.

[0018] The beneficial effects of this invention are:

[0019] Compared to existing technologies, this invention achieves accurate identification and performance analysis of non-cooperative drones through multi-source data fusion. First, it establishes a scientific avoidance risk coefficient by dynamically assessing regional risk based on comprehensive communication latency and drone density. Second, it constructs a dynamically updated individualized risk range model by combining the real-time motion status of non-cooperative drones. Finally, based on machine learning algorithms, it collaboratively optimizes and allocates traffic flow for cooperative drones in three-dimensional space, achieving a breakthrough from passive avoidance by individual drones to intelligent swarm scheduling, effectively improving the safety and operational efficiency of low-altitude traffic management. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the UAV low-altitude air traffic flow allocation method based on multi-source data fusion provided by this invention.

[0021] Figure 2 A schematic diagram of the structure of the UAV low-altitude air traffic flow distribution system with multi-source data fusion provided by the present invention.

[0022] In the attached diagram, the components represented by each number are as follows:

[0023] The module includes: data acquisition and processing module 11, regional risk assessment module 12, static risk range delineation module 13, dynamic risk range correction module 14, and traffic flow dynamic allocation module 15. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0027] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion, including:

[0028] S10: Based on multi-source data, obtain the model of the target non-cooperative drone, and based on the model of the target non-cooperative drone, obtain the performance parameters of the target non-cooperative drone;

[0029] Based on multi-source data, the model of the target non-cooperative drone is obtained, and based on the model of the target non-cooperative drone, the performance parameters of the target non-cooperative drone are obtained, including:

[0030] Based on the detection drone or cooperative drones within the target area, collect the shape and signal characteristics of the target non-cooperative drones;

[0031] Retrieve a pre-trained UAV model identifier, input the aforementioned shape features and signal features, and output the model information of the target non-cooperative UAV;

[0032] Based on the model information, a search is performed in a preset model-parameter mapping table to obtain the performance parameters of the target non-cooperative UAV. The performance parameters of the target non-cooperative UAV include at least the maximum flight speed and the maximum climb speed.

[0033] First, multi-source data collection is conducted on non-cooperative drones targeting the area by deploying dedicated reconnaissance drones or utilizing cooperative drones already on missions within the target area. The target area refers to a specific low-altitude airspace where air traffic flow monitoring and management are required; this area is typically dynamically defined based on the location of the non-cooperative drones and their potential impact zone. Reconnaissance drones are drone platforms specifically equipped with sensors and deployed for reconnaissance and surveillance missions. Cooperative drones are controllable drones that are integrated into the air traffic management system and capable of receiving commands and sharing data.

[0034] Specifically, the collected multi-source data includes visual shape features acquired by optical sensors and radio frequency signal features captured by radio monitoring equipment. The aim is to comprehensively acquire target identification information from both physical appearance and electromagnetic spectrum dimensions, laying a data foundation for subsequent model identification.

[0035] Secondly, a pre-trained drone model identifier is invoked. This drone model identifier is a classification model trained using machine learning algorithms based on sample feature data from a large number of known drone models. Specifically, the training process of this drone model identifier first requires building a sample database covering multiple drone models. Each sample in this database contains its accurate model label and corresponding standardized shape feature vector and signal feature vector. Subsequently, an appropriate classification algorithm, such as a deep convolutional neural network or support vector machine, is selected. These labeled feature vectors are used as the training set for model training. The internal parameters of the model are continuously adjusted through iterative optimization until the model can accurately map the corresponding drone model based on the input features, thus completing the construction of the identifier.

[0036] The collected shape and signal features are then used as input vectors to the UAV type identifier. This identifier performs feature extraction and pattern matching analysis, ultimately outputting the identification result of the specific type of the non-cooperative UAV. Non-cooperative UAVs refer to UAV targets that have not established an effective communication connection with the current air traffic management system and are therefore not directly dispatched or commanded by the system.

[0037] Furthermore, after obtaining the specific model information of the target non-cooperative UAV, a pre-defined model-parameter mapping table is consulted. This model-parameter mapping table systematically stores the correspondence between various known UAV models and their key performance parameters. By searching, the set of performance parameters matching the model can be accurately obtained. Among these performance parameters, at least the maximum flight speed and maximum climb speed of the target non-cooperative UAV are included. Specifically, the maximum flight speed refers to the highest instantaneous speed that the UAV model can achieve in horizontal flight, which directly determines the horizontal distance it can cover per unit time. The maximum climb speed refers to the maximum vertical speed that the UAV model can achieve when performing a vertical climb maneuver, which reflects its vertical maneuverability. Obtaining these dynamic performance indicators provides a core basis for accurately predicting the potential activity range of non-cooperative UAVs, assessing the level of conflict risk, and calculating safe and efficient avoidance strategies for cooperative UAVs.

[0038] S20: Obtain the predicted communication latency and drone density of the target area, and assess the avoidance risk coefficient of the target area;

[0039] Obtain predicted communication latency and drone density for the target area, and assess the avoidance risk coefficient of the target area, including:

[0040] The range of the target area is determined based on the performance parameters of the target non-cooperative drone;

[0041] Retrieve existing air traffic flow allocation information, predict the maximum number of drones in the target area within a preset time window, and calculate the drone density in the target area. The preset time window is inversely proportional to the maximum flight speed of the target non-cooperative drones.

[0042] Based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted and used as the predicted communication delay of the UAV.

[0043] The predicted communication latency and drone density of the target area are weighted and fused to obtain the avoidance risk coefficient of the target area within a preset time window.

[0044] First, based on the obtained performance parameters of the target non-cooperative UAV, especially its maximum flight speed, the spatial range of the target area requiring risk assessment is determined. This spatial range must be delineated to ensure it covers the airspace the non-cooperative UAV might reach within a specific reaction time, thus making the risk assessment forward-looking and targeted. Specifically, the spatial range of the target area is obtained through the following calculation process:

[0045] Using the target non-cooperative UAV's current real-time position as the geometric center and its maximum flight speed multiplied by a preset safe reaction time as the radius, a dynamically updated spherical airspace model is constructed. This safe reaction time is a key parameter, its length taking into account data processing time, command issuance delays, and the shortest time required for cooperative UAVs to perform maneuvers. The spatial range defined in this way ensures that all possible airspace reached by the non-cooperative UAV during the system's entire perception, decision-making, and response cycle is fully encompassed, providing the necessary geographical boundary conditions for effective early warning and avoidance.

[0046] Secondly, existing air traffic flow allocation plan information is retrieved. Based on this information, the maximum number of drones expected to enter or exist in the target area within a preset future time window is predicted. Then, using the ratio of this predicted number to the airspace volume of the target area, the average drone density within that time window can be calculated. Average drone density = maximum number of drones expected to exist in the target area within the preset time window / total three-dimensional airspace volume of the area. This quantifies the congestion level of the airspace; higher density means smaller safe intervals between drones, a higher potential probability of conflict, and more strained available airspace resources for avoidance operations. It is worth noting that the length of this preset time window is not a fixed value, but rather inversely proportional to the maximum flight speed of the target non-cooperative drones. Since the faster the non-cooperative drones, the faster the potential conflicts they may cause may occur, it is necessary to shorten the prediction window to focus on immediate and urgent risks and ensure the timeliness of risk assessment.

[0047] Furthermore, a pre-built and trained communication delay predictor is invoked. This predictor comprehensively analyzes historical and real-time communication environment data to predict the maximum possible communication delay between all UAVs and the control system in the target area within the same preset time window. This maximum communication delay is defined as the predicted communication delay of the UAVs, reflecting the degree of information lag that may be encountered when executing avoidance commands in that area.

[0048] Specifically, based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted, including:

[0049] Historical environmental parameters, historical base station parameters, and historical drone distribution parameters on the target drone traffic line are obtained as sample environmental data, and the communication delay of each drone on the target drone traffic line under the corresponding historical parameter conditions is obtained as supervisory data.

[0050] The communication delay predictor is built based on machine learning, and the sample environment data and supervision data are used to supervise the training of the communication delay predictor until convergence.

[0051] Input the current environmental parameters, base station parameters, and UAV distribution parameters into the communication delay predictor, predict and output the communication delay of each UAV in the target area within the preset time window, and take the maximum value as the maximum communication delay of the target area within the preset time window.

[0052] First, data preparation is performed. Historical environmental parameters, historical base station parameters, and historical drone distribution parameters along the target drone traffic line are acquired as sample environmental data. This includes: historical environmental parameters, such as weather conditions and spatial electromagnetic interference intensity; historical base station parameters, such as base station workload and channel allocation status; and historical drone distribution parameters, such as the number and spatial location of drones at different times. These together constitute the sample environmental data used for training the communication delay predictor model. Simultaneously, the actual communication delay data recorded by each drone along the target drone traffic line when the aforementioned historical environmental parameters actually occur are acquired as supervisory data, i.e., the target value for the communication delay predictor model to learn.

[0053] Secondly, appropriate machine learning algorithms, such as gradient boosting decision trees or long short-term memory networks, are selected to build a communication delay prediction model.

[0054] For example, a Long Short-Term Memory (LSTM) network can be selected as the core prediction model. LSTM is a special type of recurrent neural network that can effectively process time-series data and capture long-term dependencies, making it suitable for time-series prediction problems where communication delays are affected by historical states.

[0055] The core of this Long Short-Term Memory (LSTM) network lies in its unique memory cell structure and gating mechanism. Memory cells, as the primary information transmission channels, are responsible for maintaining and transmitting long-term pattern information from historical communication states. Three gating units work together to precisely regulate the information flow: the forget gate determines which historical communication state information needs to be discarded based on the current input and the previous time step; the input gate controls which new network state features are stored in the memory cell; and the output gate, based on the updated cell state, determines which information needs to be output to the next time step.

[0056] In constructing the communication delay predictor, a time series consisting of historical environmental parameters, historical base station parameters, and historical UAV distribution parameters is received as input. The design incorporates a two-layer Long Short-Term Memory (LSTM) network structure, with 128 neurons in each layer. Its gating mechanism delves into the complex temporal dependencies and long-term dynamic features within the multi-source parameter sequences. The final output is a high-dimensional feature vector that comprehensively represents the cumulative impact of historical and current network conditions on communication delay. In the output section, the temporal feature vector extracted by the LSM network is passed to a fully connected layer for final mapping. This fully connected layer uses a linear activation function to convert the high-dimensional features into specific communication delay prediction values.

[0057] In the specific training process, the initial learning rate was set to 0.001, the maximum number of training epochs was 200, and the batch size was set to 128. All input historical parameter sequences were standardized and preprocessed, and all data were randomly divided into training, validation, and test sets in a ratio of 7:2:1. Mean squared error was used as the loss function during training, and backpropagation was performed using the Adam optimization algorithm to update the network weights. After each training cycle, the model performance was evaluated using an independent test set. Training was terminated when the prediction error stabilized and the prediction reached a preset accuracy threshold, such as 90%. The trained communication delay predictor can accurately learn the nonlinear mapping relationship between multi-source parameters and communication delay, achieving reliable prediction of communication delay in the target area within a future time window, providing crucial data support for the calculation of the risk avoidance coefficient.

[0058] Furthermore, the collected environmental parameters, base station parameters, and UAV distribution parameters for the current moment are input into the communication delay predictor. Based on the input parameters, the communication delay predictor outputs a predicted communication delay for each individual UAV expected to exist in the target area within the current time window. The maximum value is selected from all predicted values ​​and formally determined as the maximum communication delay for the target area within the preset time window. This maximum communication delay represents the most unfavorable communication conditions that may be faced during this period, providing a crucial and conservative communication condition indicator for subsequent risk avoidance assessments.

[0059] Finally, the calculated drone density and the predicted maximum communication delay are weighted and fused. Using preset weighting coefficients, the two are unified in scale and merged into a comprehensive evaluation index, thus obtaining the avoidance risk coefficient for the target area within the preset time window. The avoidance risk coefficient equals the normalized drone density multiplied by its weight in the fusion model, plus the normalized maximum communication delay multiplied by its corresponding weight. The weight of drone density is typically set based on the core bottlenecks of current airspace traffic management; if airspace capacity is the primary constraint, it is given a higher weight. The weight of maximum communication delay is determined based on control link reliability requirements; if the real-time performance of command transmission is extremely sensitive, it is given a higher weight. The higher the value of this avoidance risk coefficient, the greater the overall avoidance risk in the area due to both poor communication and traffic congestion.

[0060] S30: Obtain the performance parameters and status information of all cooperative drones within the target area, and in conjunction with the performance parameters of the target non-cooperative drones, delineate the potential risk range of each cooperative drone;

[0061] Specifically, the performance parameters and status information of all cooperative drones within the target area are obtained, and combined with the performance parameters of the target non-cooperative drones, the potential risk range of each cooperative drone is defined, including:

[0062] Based on the performance parameters of the target non-cooperative drone, an initial risk range is set for the cooperative drone;

[0063] Retrieve the models of all cooperative drones within the target area, and obtain the performance parameters of the cooperative drones based on their models. The performance parameters of the cooperative drones include at least the minimum turning radius and the maximum climb speed.

[0064] Retrieve the mission information of all cooperative drones within the target area and obtain the status information of the cooperative drones, which includes at least real-time speed and real-time payload.

[0065] Based on the preset calculation rules and the performance parameters and status information of the cooperative drones in the target area, multiple risk range adjustment coefficients corresponding to all cooperative drones in the target area are calculated.

[0066] The initial risk range is corrected using the risk range adjustment coefficient to obtain the potential risk range for each cooperative drone within the target area.

[0067] First, based on the key performance parameters of the target non-cooperative UAV, namely maximum flight speed and maximum climb speed, a unified initial risk range is established for the cooperative UAV. This initial risk range represents a basic safety buffer zone, and specifically, it is obtained through the following calculation process:

[0068] Using the target non-cooperative UAV's current real-time position as the geometric center and its maximum flight speed multiplied by a preset safe reaction time as the radius, a dynamically updated spherical airspace model is constructed as the initial risk range. This safe reaction time is a key parameter, its length taking into account data processing time, command issuance delays, and the shortest time required for cooperative UAVs to perform maneuvers. The initial risk range defined in this way ensures coverage of the airspace that the non-cooperative UAV may reach within the system's basic reaction cycle.

[0069] Secondly, the model information of all cooperative drones within the target area is retrieved. Based on the model information, a pre-stored drone performance database is consulted to accurately obtain the performance parameters of each cooperative drone. These performance parameters include at least the minimum turning radius and maximum climb rate. The minimum turning radius refers to the minimum circular path radius required for a drone to perform a stable turning maneuver at a specific flight speed; this parameter directly determines the drone's flexibility and required space for changing course horizontally. The maximum climb rate refers to the maximum stable rate of ascent that a drone can achieve vertically; this parameter reflects its ability to rapidly change altitude. The minimum turning radius and maximum climb rate together determine the limits of the cooperative drone's maneuverability in the horizontal and vertical directions when performing emergency evasive maneuvers, and are key performance indicators for assessing its ability to escape the risk area in a timely and effective manner.

[0070] Next, the mission information of all cooperating drones within the target area is retrieved to obtain their status information. This status information includes at least the drone's real-time flight speed and real-time payload weight. Flight speed affects its kinetic energy and braking distance, while payload weight directly relates to its maneuverability and acceleration performance.

[0071] Furthermore, based on pre-defined calculation rules and the obtained performance parameters and status information of the cooperative drones, a unique risk range adjustment coefficient is calculated for each cooperative drone. This calculation rule comprehensively considers the impact of the cooperative drone's maneuverability and its current flight status on avoidance efficiency. Risk range adjustment coefficient = basic maneuverability coefficient × speed influence factor × load influence factor.

[0072] The basic maneuverability coefficient is calculated as follows: (Maximum climb rate of the cooperative UAV / Maximum climb rate of the standard UAV) × (Minimum turning radius of the standard UAV / Minimum turning radius of the cooperative UAV). A higher maximum climb rate and a smaller minimum turning radius for the cooperative UAV indicate better basic maneuverability and a larger basic maneuverability coefficient, suggesting a narrower inherent risk range. The speed influence factor is calculated as: (1 + Absolute value of the cooperative UAV's real-time speed / Maximum level flight speed). Higher real-time speeds require greater braking distance or turning radius to change motion, increasing the difficulty of evasion and thus increasing the speed influence factor. The load influence factor is calculated as: (1 + Real-time load weight of the cooperative UAV / Maximum permissible load weight). Heavier loads result in greater inertia and lower maneuverability, thus increasing the load influence factor. The standard UAV refers to a pre-selected benchmark model, whose maximum climb rate and minimum turning radius are used as the benchmark values ​​for normalized calculations.

[0073] For example, consider a heavily loaded, high-speed cooperative drone. Because both the speed and load factors are greater than one, the calculated risk range adjustment coefficient will be significantly larger than the base maneuverability coefficient. This means that a larger safety buffer range must be set for it to compensate for the delay in evasive reaction caused by its high inertia and limited maneuverability.

[0074] Finally, the initial risk range is corrected using the calculated risk range adjustment coefficient to obtain the potential risk range for each cooperative UAV within the target area. Potential risk range = Risk range adjustment coefficient × Initial risk range. This results in a tailored potential risk range for each cooperative UAV. This potential risk range is a spherical airspace centered on the real-time location of the cooperative UAV and with a radius equal to the potential risk range value, accurately reflecting the minimum safe distance required to safely avoid non-cooperative UAVs in a specific scenario.

[0075] S40: Obtain the real-time motion status of the target non-cooperative drone, and dynamically correct the potential risk range of the cooperative drone by combining the performance parameters of the target non-cooperative drone and the avoidance risk coefficient of the target area, so as to obtain the real-time risk range.

[0076] Specifically, the real-time motion state of the target non-cooperative drone is acquired, and combined with the performance parameters of the target non-cooperative drone and the avoidance risk coefficient of the target area, the potential risk range of the cooperative drone is dynamically corrected to obtain the real-time risk range, including:

[0077] The real-time flight speed and direction of the target non-cooperative drone are obtained as the real-time motion state;

[0078] Starting from the real-time motion state of the target non-cooperative drone, and based on the performance parameters of the target non-cooperative drone, the maximum activity range of the target non-cooperative drone within a preset time window is calculated as the theoretical risk range.

[0079] The maximum risk range is the union of the theoretical risk range and the potential risk range.

[0080] The maximum risk range is corrected using the avoidance risk coefficient of the target area to obtain the real-time risk range.

[0081] First, the real-time flight speed and direction of the target non-cooperative drone are continuously acquired through detection equipment, serving as its real-time motion state. Second, using the real-time motion state as initial conditions, and combining it with the known performance parameters of the target non-cooperative drone, the set of all possible spatial locations it can reach within the next preset time window is calculated, thus forming a theoretical maximum range of activity, i.e., the theoretical risk range.

[0082] Specifically, the theoretical risk range is obtained by predicting its reachable dynamic envelope, which is an irregular three-dimensional airspace whose specific shape is highly dependent on its real-time motion state. The calculation starts with the real-time position of the target non-cooperative UAV, uses its real-time flight direction as the main axis, and comprehensively considers the maneuverability allowed by its performance parameters. The boundary of this irregular body is determined by the following factors: along the current flight direction, the maximum forward distance is the product of the maximum flight speed and a preset time window; simultaneously, considering the UAV's maximum possible turning rate in the horizontal plane and its maximum possible climb or descent rate in the vertical plane, the maximum lateral and vertical distances that can deviate from the main axis within the preset time window are calculated. Connecting the outer boundaries of the spatial position point clouds generated by all possible maneuver paths, the minimum convex hull or boundary volume formed is the theoretical risk range representing its maximum activity range. This dynamic irregular body accurately characterizes the threat airspace that a non-cooperative UAV may constitute in the near future based on its current state and capabilities.

[0083] Secondly, the calculated theoretical risk range is geometrically joined with the potential risk range defined for cooperative drones in the previous steps. This ensures the integrity of the risk range, encompassing both newly added risk areas based on future dynamic predictions of non-cooperative drones and preserving the original safety buffer zone required based on the maneuverability characteristics of the cooperative drones themselves. This joined area is defined as the maximum risk range.

[0084] Finally, the previously assessed target area avoidance risk coefficient is introduced to make a final correction to the maximum risk range. The avoidance risk coefficient comprehensively reflects the communication reliability and traffic congestion level of the area. Specifically, the correction method involves multiplying the avoidance risk coefficient as a scaling factor by the radius or volume of the maximum risk range. When the regional risk coefficient is high, it indicates a poor overall avoidance environment, requiring an expanded safety margin. Therefore, the maximum risk range is enlarged to obtain the final implemented real-time risk range. The final real-time risk range is a dynamically updated safety boundary that comprehensively considers target behavior, its own capabilities, and environmental factors.

[0085] S50: Based on machine learning, it dynamically corrects the drone traffic flow of cooperative drones by combining a 3D building map of the target area and the real-time risk range of each cooperative drone, thus avoiding non-cooperative drones.

[0086] Specifically, based on machine learning, and combining a 3D building map of the target area with the real-time risk range of each cooperative drone, the drone traffic flow of cooperative drones is dynamically corrected to avoid non-cooperative drones, including:

[0087] Acquire a 3D building map of the target area, a real-time updated risk range, and the status information of cooperative drones within the target area, as a state set S, wherein the status information includes real-time coordinates and flight speed;

[0088] Obtain the performance parameters of the cooperative drones within the target area, define basic actions based on the performance parameters, and form a basic action space A;

[0089] Based on the real-time risk range, set reward and penalty functions for the flight routes of cooperative drones;

[0090] Create a Q(S,A) table, where rows correspond to state information and columns correspond to basic actions, and record the expected cumulative reward for performing a specific action in each state as the Q value;

[0091] Randomly initialize the state information, select the basic action according to the ε-greedy strategy, update the Q-table with the Q-value update formula, and repeat the iteration until the Q-table converges.

[0092] The cooperating drone selects the action with the largest Q value from the Q table to execute based on real-time status information;

[0093] The same operation is performed synchronously on all cooperating drones within the target area to dynamically adjust traffic flow and avoid non-cooperating drones.

[0094] First, a 3D building map of the target area, a real-time updated risk range, and the status information of cooperative drones within the target area are acquired, forming a state set S. This state information includes real-time coordinates and flight speed. State set S contains all environmental information affecting decision-making, ensuring that the drones can comprehensively perceive static obstacles, dynamic threats, and the status of themselves and neighboring drones. Second, the performance parameters of the cooperative drones within the target area are acquired. Based on these performance parameters, basic actions are defined, forming a basic action space A. Basic action space A includes limited forward / backward, left / right, and up / down displacements, small-amplitude speed adjustments, and hovering commands. All actions are ensured to be within the drones' physical maneuverability.

[0095] Furthermore, based on the real-time risk range, a reward and penalty function is set for the cooperative drone flight path to quantitatively evaluate the quality of each action, thereby guiding the learning direction. Specifically, the reward and penalty function includes: basic reward: +1 point for flying along the original mission path; risk penalty: -50 points for entering the real-time risk range or colliding with buildings; obstacle avoidance reward: +10 points for successfully bypassing the risk area; efficiency reward: +5 points for shortening the mission completion time.

[0096] The core of the learning process lies in the iterative update of the Q-table. The Q-table is a lookup table with all possible states S as row indexes and all optional basic actions A as column indexes. Each cell stores the Q(S, A) value, which quantifies the expected long-term cumulative reward that can be obtained by choosing action A in a specific state S.

[0097] During the algorithm initialization phase, all values ​​in the Q-table are set to a uniform initial value, such as zero. Then, a simulation training loop begins. In each iteration, the drone agent selects an action based on the current environmental state S using an ε-greedy strategy: it randomly selects an action for exploration with a small probability ε, and selects the action with the highest Q-value corresponding to state S in the current Q-table with a high probability 1-ε. After executing action A, the environment transitions to a new state S', and an immediate reward R calculated according to the reward-penalty function is immediately obtained. Then, the Q-learning update formula is used to correct the Q-value: Q(S, A) ← Q(S, A) + α[R + γmaxQ(S', a) - Q(S, A)]. In this formula, the learning rate α controls the rate at which new information overwrites old knowledge, and is typically set to 0.1 to ensure learning stability; the discount factor γ balances the importance of immediate rewards versus future rewards, and is typically set to 0.9 to emphasize medium- to long-term gains. Through numerous simulation iterations, when the update frequency of the Q-table falls below a preset threshold (e.g., the average change in Q-values ​​falls below a relative threshold, such as 0.001), the Q-table is considered to have converged. At this point, for any state S, the action 'a' with the highest Q-value is considered the optimal strategy. This strategy maximizes the long-term benefits of the UAV in completing its mission while ensuring safe obstacle avoidance.

[0098] During the real-time control phase, each cooperating drone only needs to perceive the current real-time state S, then query the Q table to find which action A has the largest Q value under that state, and execute that action. By synchronously executing this "query-execution" process for all cooperating drones in the area, distributed and intelligent dynamic correction of the entire traffic flow can be achieved, jointly avoiding non-cooperating drones and static obstacles, thereby completing the flight mission safely and efficiently.

[0099] In summary, the embodiments of this application have at least the following technical effects:

[0100] Compared to existing technologies, this invention firstly accurately identifies non-cooperative drones and obtains their performance parameters through multi-source data fusion, providing a reliable data foundation for risk assessment. Secondly, it innovatively introduces the concept of a regional avoidance risk coefficient, comprehensively assessing the overall airspace risk level by integrating communication latency and drone density, thus making risk perception more comprehensive. Thirdly, by combining the real-time motion status of non-cooperative drones with the performance of cooperative drones, it achieves precise correction from a static potential risk range to a dynamic real-time risk range, making risk modeling more closely aligned with actual flight situations. Finally, based on machine learning algorithms, it integrates three-dimensional geographical constraints and individualized real-time risk ranges to collaboratively optimize and dynamically allocate the entire low-altitude traffic flow, achieving a fundamental shift from local passive avoidance to global proactive guidance, effectively preventing chain reactions and significantly improving the overall safety and operational efficiency of low-altitude airspace.

[0101] Example 2, as Figure 2 As shown, based on the same inventive concept as the UAV low-altitude air traffic flow allocation method using multi-source data fusion provided in Embodiment 1, this embodiment of the invention also provides a UAV low-altitude air traffic flow allocation system using multi-source data fusion, comprising:

[0102] The data acquisition and processing module 11 is used to obtain the model of the target non-cooperative drone based on multi-source data, and to obtain the performance parameters of the target non-cooperative drone based on the model of the target non-cooperative drone.

[0103] Regional risk assessment module 12 is used to obtain the predicted communication delay and drone density of the target area and to assess the avoidance risk coefficient of the target area.

[0104] The static risk range delineation module 13 is used to obtain the performance parameters and status information of all cooperative drones in the target area, and, in combination with the performance parameters of the target non-cooperative drones, delineate the potential risk range of each cooperative drone.

[0105] The dynamic risk range correction module 14 is used to obtain the real-time motion status of the target non-cooperative UAV, and combine the performance parameters of the target non-cooperative UAV and the avoidance risk coefficient of the target area to dynamically correct the potential risk range of the cooperative UAV and obtain the real-time risk range.

[0106] The traffic flow dynamic allocation module 15 is used to dynamically correct the drone traffic flow of cooperative drones based on machine learning, combined with the 3D building map of the target area and the real-time risk range of each cooperative drone, to avoid non-cooperative drones.

[0107] Specifically, the data acquisition and processing module 11 is used for:

[0108] Based on multi-source data, the model of the target non-cooperative drone is obtained, and based on the model of the target non-cooperative drone, the performance parameters of the target non-cooperative drone are obtained, including:

[0109] Based on the detection drone or cooperative drones within the target area, collect the shape and signal characteristics of the target non-cooperative drones;

[0110] Retrieve a pre-trained UAV model identifier, input the aforementioned shape features and signal features, and output the model information of the target non-cooperative UAV;

[0111] Based on the model information, a search is performed in a preset model-parameter mapping table to obtain the performance parameters of the target non-cooperative UAV. The performance parameters of the target non-cooperative UAV include at least the maximum flight speed and the maximum climb speed.

[0112] Specifically, the regional risk assessment module 12 is used for:

[0113] Obtain predicted communication latency and drone density for the target area, and assess the avoidance risk coefficient of the target area, including:

[0114] The range of the target area is determined based on the performance parameters of the target non-cooperative drone;

[0115] Retrieve existing air traffic flow allocation information, predict the maximum number of drones in the target area within a preset time window, and calculate the drone density in the target area. The preset time window is inversely proportional to the maximum flight speed of the target non-cooperative drones.

[0116] Based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted and used as the predicted communication delay of the UAV.

[0117] The predicted communication latency and drone density of the target area are weighted and fused to obtain the avoidance risk coefficient of the target area within a preset time window.

[0118] Specifically, based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted, including:

[0119] Historical environmental parameters, historical base station parameters, and historical drone distribution parameters on the target drone traffic line are obtained as sample environmental data, and the communication delay of each drone on the target drone traffic line under the corresponding historical parameter conditions is obtained as supervisory data.

[0120] The communication delay predictor is built based on machine learning, and the sample environment data and supervision data are used to supervise the training of the communication delay predictor until convergence.

[0121] Input the current environmental parameters, base station parameters, and UAV distribution parameters into the communication delay predictor, predict and output the communication delay of each UAV in the target area within the preset time window, and take the maximum value as the maximum communication delay of the target area within the preset time window.

[0122] The static risk range delineation module 13 is specifically used for:

[0123] Obtain the performance parameters and status information of all cooperative drones within the target area, and combine this information with the performance parameters of the target non-cooperative drones to define the potential risk range for each cooperative drone, including:

[0124] Based on the performance parameters of the target non-cooperative drone, an initial risk range is set for the cooperative drone;

[0125] Retrieve the models of all cooperative drones within the target area, and obtain the performance parameters of the cooperative drones based on their models. The performance parameters of the cooperative drones include at least the minimum turning radius and the maximum climb speed.

[0126] Retrieve the mission information of all cooperative drones within the target area and obtain the status information of the cooperative drones, which includes at least real-time speed and real-time payload.

[0127] Based on the preset calculation rules and the performance parameters and status information of the cooperative drones in the target area, multiple risk range adjustment coefficients corresponding to all cooperative drones in the target area are calculated.

[0128] The initial risk range is corrected using the risk range adjustment coefficient to obtain the potential risk range for each cooperative drone within the target area.

[0129] The dynamic risk range correction module 14 is specifically used for:

[0130] The real-time motion state of the target non-cooperative UAV is obtained. Combined with the performance parameters of the target non-cooperative UAV and the avoidance risk coefficient of the target area, the potential risk range of the cooperative UAV is dynamically corrected to obtain the real-time risk range, including:

[0131] The real-time flight speed and direction of the target non-cooperative drone are obtained as the real-time motion state;

[0132] Starting from the real-time motion state of the target non-cooperative drone, and based on the performance parameters of the target non-cooperative drone, the maximum activity range of the target non-cooperative drone within a preset time window is calculated as the theoretical risk range.

[0133] The maximum risk range is the union of the theoretical risk range and the potential risk range.

[0134] The maximum risk range is corrected using the avoidance risk coefficient of the target area to obtain the real-time risk range.

[0135] The traffic flow dynamic allocation module 15 is specifically used for:

[0136] Based on machine learning, and combining a 3D building map of the target area with the real-time risk range of each cooperative drone, the drone traffic flow of cooperative drones is dynamically corrected to avoid non-cooperative drones, including:

[0137] Acquire a 3D building map of the target area, a real-time updated risk range, and real-time coordinates and flight speeds of cooperative drones within the target area, as a state set S;

[0138] Obtain the performance parameters of the cooperative drones within the target area, define basic actions based on the performance parameters, and form a basic action space A;

[0139] Based on the real-time risk range, set reward and penalty functions for the flight routes of cooperative drones;

[0140] Create a Q(S,A) table, where rows correspond to state information and columns correspond to basic actions, and record the expected cumulative reward for performing a specific action in each state as the Q value;

[0141] Randomly initialize the state information, select the basic action according to the ε-greedy strategy, update the Q-table with the Q-value update formula, and repeat the iteration until the Q-table converges.

[0142] The cooperating drone selects the action with the largest Q value from the Q table to execute based on real-time status information;

[0143] The same operation is performed synchronously on all cooperating drones within the target area to dynamically adjust traffic flow and avoid non-cooperating drones.

[0144] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0145] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0146] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion, characterized in that, include: Based on multi-source data, the model of the target non-cooperative drone is obtained, and based on the model of the target non-cooperative drone, the performance parameters of the target non-cooperative drone are obtained. Acquire drone prediction communication latency and drone density in the target area, and assess the avoidance risk coefficient of the target area; Obtain the performance parameters and status information of all cooperative drones within the target area, and combine them with the performance parameters of the target non-cooperative drones to define the potential risk range of each cooperative drone; The real-time motion state of the target non-cooperative UAV is obtained. Combined with the performance parameters of the target non-cooperative UAV and the avoidance risk coefficient of the target area, the potential risk range of the cooperative UAV is dynamically corrected to obtain the real-time risk range. This includes: obtaining the real-time flight speed and direction of the target non-cooperative UAV as the real-time motion state; starting from the real-time motion state of the target non-cooperative UAV, and based on the performance parameters of the target non-cooperative UAV, calculating the maximum activity range of the target non-cooperative UAV within a preset time window as the theoretical risk range; taking the union of the theoretical risk range and the potential risk range as the maximum risk range; and correcting the maximum risk range using the avoidance risk coefficient of the target area to obtain the real-time risk range. Based on machine learning, the drone traffic flow of cooperative drones is dynamically corrected by combining a 3D building map of the target area and the real-time risk range of each cooperative drone, thus avoiding non-cooperative drones.

2. The method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion according to claim 1, characterized in that, Based on multi-source data, the model of the target non-cooperative drone is obtained, and based on the model of the target non-cooperative drone, the performance parameters of the target non-cooperative drone are obtained, including: Based on the detection drone or cooperative drones within the target area, collect the shape and signal characteristics of the target non-cooperative drones; Retrieve a pre-trained UAV model identifier, input the aforementioned shape features and signal features, and output the model information of the target non-cooperative UAV; Based on the model information, a search is performed in a preset model-parameter mapping table to obtain the performance parameters of the target non-cooperative UAV. The performance parameters of the target non-cooperative UAV include at least the maximum flight speed and the maximum climb speed.

3. The method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion according to claim 1, characterized in that, Obtain predicted communication latency and drone density for the target area, and assess the avoidance risk coefficient of the target area, including: The range of the target area is determined based on the performance parameters of the target non-cooperative drone; Retrieve existing air traffic flow allocation information, predict the maximum number of drones in the target area within a preset time window, and calculate the drone density in the target area. The preset time window is inversely proportional to the maximum flight speed of the target non-cooperative drones. Based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted and used as the predicted communication delay of the UAV. The predicted communication latency of drones in the target area and the drone density are weighted and fused to obtain the avoidance risk coefficient of the target area within a preset time window.

4. The method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion according to claim 3, characterized in that, Based on a preset communication delay predictor, the maximum communication delay of the target area within a preset time window is predicted, including: Historical environmental parameters, historical base station parameters, and historical drone distribution parameters on the target drone traffic line are obtained as sample environmental data, and the communication delay of each drone on the target drone traffic line under the corresponding historical parameter conditions is obtained as supervisory data. The communication delay predictor is built based on machine learning, and the sample environment data and supervision data are used to supervise the training of the communication delay predictor until convergence. Input the current environmental parameters, base station parameters, and UAV distribution parameters into the communication delay predictor, predict and output the communication delay of each UAV in the target area within the preset time window, and take the maximum value as the maximum communication delay of the target area within the preset time window.

5. The method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion according to claim 1, characterized in that, Obtain the performance parameters and status information of all cooperative drones within the target area, and combine this information with the performance parameters of the target non-cooperative drones to define the potential risk range for each cooperative drone, including: Based on the performance parameters of the target non-cooperative drone, an initial risk range is set for the cooperative drone; Retrieve the models of all cooperative drones within the target area, and obtain the performance parameters of the cooperative drones based on their models. The performance parameters of the cooperative drones include at least the minimum turning radius and the maximum climb speed. Retrieve the mission information of all cooperative drones within the target area and obtain the status information of the cooperative drones, which includes at least real-time speed and real-time payload. Based on the preset calculation rules and the performance parameters and status information of the cooperative drones in the target area, multiple risk range adjustment coefficients corresponding to all cooperative drones in the target area are calculated. The initial risk range is corrected using the risk range adjustment coefficient to obtain the potential risk range for each cooperative drone within the target area.

6. The method for allocating low-altitude air traffic flow for unmanned aerial vehicles (UAVs) based on multi-source data fusion according to claim 1, characterized in that, Based on machine learning, and combining a 3D building map of the target area with the real-time risk range of each cooperative drone, the drone traffic flow of cooperative drones is dynamically corrected to avoid non-cooperative drones, including: Acquire a 3D building map of the target area, a real-time updated risk range, and the status information of cooperative drones within the target area, as a state set S, wherein the status information includes real-time coordinates and flight speed; Obtain the performance parameters of the cooperative drones within the target area, define basic actions based on the performance parameters, and form a basic action space A; Based on the real-time risk range, set reward and penalty functions for the flight routes of cooperative drones; Create a Q(S,A) table, where rows correspond to state information and columns correspond to basic actions, and record the expected cumulative reward for performing a specific action in each state as the Q value; Randomly initialize the state information, select the basic action according to the ε-greedy strategy, update the Q-table with the Q-value update formula, and repeat the iteration until the Q-table converges. The cooperating drone selects the action with the largest Q value from the Q table to execute based on real-time status information; The same operation is performed synchronously on all cooperating drones within the target area to dynamically adjust traffic flow and avoid non-cooperating drones.

7. A multi-source data fusion-based UAV low-altitude air traffic flow distribution system, characterized in that, The UAV low-altitude air traffic flow allocation method for performing multi-source data fusion as described in any one of claims 1-6 includes: The data acquisition and processing module is used to obtain the model of the target non-cooperative drone based on multi-source data, and to obtain the performance parameters of the target non-cooperative drone based on the model of the target non-cooperative drone. The regional risk assessment module is used to obtain the predicted communication latency and drone density of the target area, and to assess the avoidance risk coefficient of the target area. The static risk range delineation module is used to obtain the performance parameters and status information of all cooperative drones within the target area, and, in combination with the performance parameters of the target non-cooperative drones, delineate the potential risk range of each cooperative drone. The dynamic risk range correction module is used to obtain the real-time motion status of the target non-cooperative drone, and combine the performance parameters of the target non-cooperative drone with the avoidance risk coefficient of the target area to dynamically correct the potential risk range of the cooperative drone to obtain the real-time risk range; the traffic flow dynamic allocation module is used to dynamically correct the drone traffic flow of the cooperative drone based on machine learning, combined with the 3D building map of the target area and the real-time risk range of each cooperative drone, to avoid non-cooperative drones.

Images