A low-altitude traffic situation analysis method and system based on digital twin

Abstract

The present invention relates to the technical field of low-altitude traffic management, and discloses a low-altitude traffic situation analysis method and system based on digital twins, wherein a low-altitude traffic situation analysis method based on digital twins includes: constructing a non-uniform adaptive quadtree spatial index structure; generating a spatiotemporal risk probability field, and using multi-model fusion trajectory prediction to calculate the conflict risk probability; constructing a scene adaptive classification model, and dynamically adjusting the decision-making strategy according to the characteristics of the traffic scene; realizing collaborative decision optimization, and using a hierarchical decision structure and a right of way allocation algorithm based on an auction mechanism; establishing a digital twin ecological potential analysis platform, mapping the low-altitude traffic status in real time and providing decision support; the present invention realizes accurate perception, analysis and prediction of low-altitude traffic by organically combining technologies such as non-uniform adaptive quadtree spatial indexing, multi-model fusion trajectory prediction, spatiotemporal risk probability field calculation, hierarchical collaborative decision optimization and digital twin ecological potential analysis platform.

Description

Technical Field

[0001] The present invention relates to the technical field of low-altitude traffic management, and more specifically, to a low-altitude traffic situation analysis method and system based on digital twins. Background Art

[0002] With the rapid development of drone technology and the widespread commercial application, low-altitude traffic is experiencing explosive growth. According to statistics, the global civilian drone fleet is expected to exceed 42 million by 2025, a significant proportion of which will operate in urban low-altitude airspace. This trend poses a severe challenge to traditional low-altitude traffic management technologies.

[0003] The existing low-altitude traffic management technology has the following technical defects:

[0004] In terms of spatial indexing technology, existing systems generally use uniform grid division or fixed-depth quadtree structures for airspace management. This static division method is inefficient when dealing with dynamically changing traffic density. In high-density areas, coarse-grained division results in too many aircraft being included in a single grid, resulting in a quadratic increase in the time complexity of conflict detection. In low-density areas, fine-grained division results in a large number of empty grids, wasting storage space and computing resources. In terms of conflict prediction technology, existing methods mainly rely on simple trajectory extrapolation for conflict detection, ignoring the maneuverability of aircraft, changes in flight intentions, and the influence of environmental factors. This leads to insufficient prediction accuracy in complex flight scenarios and high false alarm and missed detection rates. In terms of obstacle avoidance decision-making technology, existing systems often use distributed obstacle avoidance strategies, which are prone to "decision oscillation" in densely populated environments. When multiple aircraft simultaneously detect a conflict and adjust their routes, the new routes may cause new conflicts with other aircraft, resulting in continuous route adjustments and system instability. In terms of situational awareness technology, existing low-altitude traffic management systems lack a unified digital twin platform, resulting in delayed information transmission and inconsistent situational understanding, affecting the timeliness and accuracy of overall decision-making.

[0005] Existing technologies also have shortcomings in scenario adaptability. Most systems use fixed safety parameters and decision-making rules, making it difficult to dynamically adjust management strategies based on different traffic scenarios, weather conditions, and mission requirements, making it difficult to achieve an optimal balance between safety and efficiency.

[0006] Therefore, there is an urgent need for a low-altitude traffic situation analysis technology that can adapt to changes in traffic density, improve the accuracy of conflict prediction, achieve collaborative decision-making optimization, and have scene adaptability to meet the growing needs of low-altitude traffic management. Summary of the Invention

[0007] The present invention provides a low-altitude traffic situation analysis method and system based on digital twins, which solve the technical problems in related technologies such as low spatial indexing efficiency, insufficient conflict prediction accuracy, frequent oscillation of obstacle avoidance decisions, lack of scene adaptability, and inconsistent situation perception.

[0008] The present invention provides a low-altitude traffic situation analysis method based on digital twins, comprising:

[0009] Construct a non-uniform adaptive quadtree spatial index structure and dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density;

[0010] Based on the spatial partitioning results output by the quadtree spatial index structure, a spatiotemporal risk probability field is generated, and the collision risk probability is calculated using trajectory prediction using multi-model fusion.

[0011] Based on the risk distribution characteristics of the spatiotemporal risk probability field, a scenario-adaptive classification model is constructed to dynamically adjust the decision-making strategy according to the characteristics of the traffic scenario;

[0012] Based on the scenario types and decision strategies output by the scenario adaptive classification model, collaborative decision optimization is achieved, using a hierarchical decision-making structure and a right-of-way allocation algorithm based on an auction mechanism;

[0013] Based on the results of collaborative decision-making optimization, the output results of the quadtree spatial index structure, the risk distribution characteristics and decision-making strategies of the spatiotemporal risk probability field, a digital twin potential analysis platform is established to map the low-altitude traffic status in real time and provide decision support.

[0014] Furthermore, in the step of constructing a non-uniform adaptive quadtree spatial index structure:

[0015] Setting splitting conditions based on the number of drones, traffic density, and speed variance within a node;

[0016] Set merging conditions based on the number of drones in a node, traffic density, and stabilization time;

[0017] Adopting an incremental update algorithm, only locally update the affected quadtree nodes;

[0018] Accelerate hot spot area queries through a multi-level caching mechanism.

[0019] Furthermore, the multi-model fusion trajectory prediction combines three models: track extrapolation, intention reasoning, and historical pattern matching, and dynamically adjusts the model weights according to different scenarios;

[0020] The trajectory extrapolation model is implemented using an improved Kalman filter, taking into account the UAV's dynamic characteristics and environmental constraints;

[0021] The intention inference model adopts a Bayesian network structure, which includes an observation layer, an intention hypothesis layer, and a trajectory prediction layer;

[0022] The historical pattern matching model is implemented based on similar trajectory retrieval and pattern recognition technology;

[0023] According to the risk probability value, risks are divided into four levels: critical, high, medium and low, and conflict warnings are generated.

[0024] Furthermore, in the step of constructing a scene-adaptive classification model:

[0025] Extract traffic density features, traffic flow features, spatial constraint features, meteorological condition features and urgency features from low-altitude traffic data;

[0026] Adopting a hierarchical classification architecture, coarse-grained scene classification is performed first, followed by fine-grained scene recognition;

[0027] Configure basic obstacle avoidance strategy parameter sets for each scenario type;

[0028] According to the characteristics of the scenario, the weight ratio of safety margin and path efficiency is dynamically adjusted.

[0029] Furthermore, the hierarchical classification architecture includes:

[0030] The coarse-grained classifier, based on the random forest algorithm, divides the scenes into three categories, including normal scenes, abnormal scenes, and transition scenes;

[0031] Fine-grained classifier, for each coarse-grained category, a specialized classifier is used to further subdivide it, using the gradient boosting decision tree algorithm;

[0032] The scene attribute analyzer uses a multi-layer perceptron to extract the key attributes of the scene, including the degree of congestion, risk level and meteorological impact intensity.

[0033] Furthermore, in the steps of achieving collaborative decision optimization:

[0034] Construct a traffic flow topology model based on low-altitude traffic data, including node sets, edge sets and edge weights;

[0035] Build a hierarchical collaborative decision-making structure, with the macro level responsible for regional flow balance and the micro level responsible for individual route planning;

[0036] An auction-based right-of-way allocation algorithm is adopted, taking into account the time window utility value, drone priority and cost;

[0037] Achieve global optimization by minimizing overall latency and energy consumption.

[0038] Furthermore, the construction of the traffic flow topology model includes:

[0039] Extract key nodes from historical trajectory data using density clustering algorithm;

[0040] Determine the connection relationship between nodes based on trajectory frequency statistics;

[0041] Calculate the flow weight, capacity weight, delay weight and conflict weight for each edge;

[0042] A three-level update mechanism is adopted, including short-term updates, mid-term updates and long-term updates.

[0043] Furthermore, in the steps of establishing a digital twin potential analysis platform:

[0044] Build a digital twin model that includes a spatial model, a traffic entity model, a traffic rules model, and a meteorological environment model;

[0045] Conduct comprehensive situation analysis, including current situation assessment, situation evolution prediction, and decision-making support analysis;

[0046] Develop a multi-scale situation visualization interface that supports macro, meso, and micro views;

[0047] Establish a closed-loop system of situational awareness, analysis, decision-making and execution.

[0048] Furthermore, the digital twin model adopts a layered architecture, including:

[0049] The base layer contains geographic information system data, three-dimensional urban models, and terrain data;

[0050] The environmental layer, which contains meteorological data and environmental conditions;

[0051] Traffic layer, including drone entities, routes, and traffic rules;

[0052] The control layer includes management decisions, control instructions and response measures.

[0053] The present invention provides a low-altitude traffic situation analysis system based on digital twins, which is used to execute the above-mentioned low-altitude traffic situation analysis method based on digital twins, including:

[0054] A non-uniform adaptive quadtree spatial index module is used to dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density;

[0055] A spatiotemporal risk probability field generation module is used to calculate the conflict risk probability based on trajectory prediction using multi-model fusion;

[0056] A scenario-adaptive classification module, used to dynamically adjust decision-making strategies based on traffic scenario characteristics;

[0057] A collaborative decision-making optimization module for allocating right-of-way using a hierarchical decision-making structure and an auction-based mechanism;

[0058] Digital twin potential analysis platform for real-time mapping of low-altitude traffic status and providing decision support.

[0059] The beneficial effects of the present invention are: the conflict detection efficiency is improved and the memory usage is reduced through the non-uniform adaptive quadtree spatial index structure, which solves the problem of low efficiency of the traditional spatial index structure when processing dynamic traffic density;

[0060] Through the multi-model fusion trajectory prediction and spatiotemporal risk probability field calculation method, the conflict warning lead time is extended, the system response time is increased, and at the same time, the conflict prediction accuracy is improved, reducing false alarms and missed alarms.

[0061] Through scene classification and adaptive strategy adjustment technology based on multi-dimensional features, the present invention can dynamically adjust decision strategies according to different scene characteristics, optimize efficiency while ensuring safety, and improve adaptability. In particular, the present invention can still maintain normal processing capabilities under complex weather conditions.

[0062] Through a hierarchical collaborative decision-making structure and an auction-based right-of-way allocation algorithm, decision-making oscillation is effectively reduced, traffic efficiency in high-density intersections is improved, congestion duration is reduced, and a shift from local to global optimization is achieved, addressing the lack of coordination in traditional point-to-point obstacle avoidance.

[0063] By dynamically adjusting the spatial partitioning granularity and computing resource allocation, the present invention optimizes resource usage while ensuring performance, improves computing load balancing, and enables limited computing resources to support larger-scale low-altitude traffic management needs.

[0064] The digital twin platform integrates situational awareness, analysis, and decision support functions, eliminating information silos, improving consistency in situational understanding, and shortening the system's response time to emergencies, providing comprehensive situational awareness capabilities for low-altitude traffic management.

[0065] Through the spatiotemporal risk probability field and situation evolution prediction technology, the present invention can predict potential traffic congestion and risk hotspots in advance, thereby improving the foresight and initiative of management. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of a low-altitude traffic situation analysis method based on digital twins in the present invention;

[0067] Figure 2 This is a histogram comparing the non-uniform adaptive quadtree spatial index structure with the traditional indexing method on different performance indicators;

[0068] Figure 3 It is a line graph showing the performance trend under different architectures as the number of drones increases;

[0069] Figure 4 It is a scatter plot of the probability distribution of conflict risk under different traffic densities;

[0070] Figure 5 It is a radar chart showing the performance of the method of the present invention on five key indicators in different scenarios;

[0071] Figure 6 It is a pie chart showing the distribution of collaborative decision-making optimization effects. DETAILED DESCRIPTION

[0072] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0073] At least one embodiment of the present invention discloses a low-altitude traffic situation analysis method based on digital twins, such as Figure 1 Shown, including:

[0074] Step 1: Construct a non-uniform adaptive quadtree spatial index structure and dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density;

[0075] According to the embodiment of the present application, this step uses a non-uniform adaptive quadtree algorithm to dynamically partition the low-altitude traffic airspace, automatically adjust the granularity of the spatial division according to the traffic density, and generate a highly adaptable spatial index structure. Specifically, it includes:

[0076] Step 1.1, initialize the quadtree root node;

[0077] Initialize the monitored airspace area into a rectangular area as the root node of the quadtree, denoted as , which contains the boundary information of the entire monitoring area. It should be noted that the definition of the root node adopts the boundary coordinate method, which includes the coordinates of the lower left corner and upper right corner of the area.

[0078] Step 1.2, adaptive splitting based on traffic density;

[0079] According to one embodiment of the present application, quadtree node splitting is based on comprehensive criteria, taking into account factors such as the number of drones within the node, traffic density, and speed variance. It should be understood that the system will execute a node split when the number of drones within a node exceeds a preset threshold, or when both the traffic density and speed variance within the node exceed their respective thresholds.

[0080] The traffic density within a node is calculated as the ratio of the number of drones within the node to the node area. The speed variance reflects the degree of dispersion of the drone motion state within the node. A larger value indicates a more complex traffic flow.

[0081] Optionally, in some implementations, the splitting criteria can further consider the acceleration distribution characteristics of drones within a node, enabling earlier identification of potential traffic flow changes. Furthermore, in environments with complex meteorological conditions, such as areas with strong winds, the system can incorporate wind field influencing factors to adjust the splitting threshold, enhancing adaptability to adverse weather conditions.

[0082] Step 1.3, dynamic merging mechanism;

[0083] To avoid wasting resources due to overly detailed airspace partitioning, this application provides a dynamic merging mechanism. When the four child nodes of a quadtree node simultaneously meet the following conditions, they are merged into a parent node: the number of drones in the node is lower than the merge threshold, the traffic density in the node is lower than the density merge threshold, and the time since the last split exceeds the stability time threshold. This mechanism effectively prevents frequent split-merge oscillations.

[0084] The node merging judgment function is expressed as:

[0085] ;

[0086] in, Represents the node merging judgment function; Indicates the quadtree node to be judged; Indicates the execution of a merge operation; Indicates that no merge operation is performed; Indicates the number of drones in the node; The number of drones representing the merging threshold; represents the traffic density within the node; represents the density merging threshold; Indicates the current time; Indicates the last split time; represents the stabilization time threshold; Represents the logical AND operator; Indicates other situations, that is, the default situation when the above conditions are not met.

[0087] In some implementations, an adaptive threshold adjustment method based on historical statistics can be used to dynamically adjust the merge threshold based on the historical merge-split frequency. Furthermore, to further improve system stability, a predictive merge strategy can be introduced to predict future node loads based on traffic flow trends and execute merge operations in advance.

[0088] Step 1.4, incremental update algorithm;

[0089] According to one embodiment of the present application, when a drone's location is updated, an incremental update algorithm is used to update only the affected quadtree nodes, avoiding a global rebuild. The incremental update process includes: first locating the leaf node where the drone is located; then updating the drone list and related statistics within the node; then checking and performing split or merge operations; and finally updating the statistics of the parent node along the path upward. This algorithm significantly improves the update efficiency of the spatial index structure, enabling the system to adapt to high-frequency location updates.

[0090] Step 1.5, multi-level cache mechanism;

[0091] This application provides a multi-level caching mechanism to accelerate queries in frequently accessed hotspots. The mechanism includes a first-level cache that stores recently accessed leaf nodes; a second-level cache that stores leaf nodes in high-density areas; and a third-level cache that stores frequently split / merged nodes. It should be noted that the system adaptively adjusts the cache size at each level based on the cache hit rate to further optimize query performance.

[0092] Step 2: Based on the spatial partitioning results output by the quadtree spatial index structure, a spatiotemporal risk probability field is generated, and the collision risk probability is calculated using trajectory prediction using multi-model fusion.

[0093] The second step proposed in this application is to build a trajectory prediction model based on the current state and historical trajectory data of the drone, calculate the conflict risk probability of each point in the airspace, and form a spatiotemporal risk probability field for conflict warning. Specifically, it includes:

[0094] Step 2.1, trajectory prediction model construction;

[0095] According to one embodiment of the present application, the system uses a multi-model fusion trajectory prediction method that combines three models: track extrapolation, intent reasoning, and historical pattern matching. It should be noted that the fusion method uses a weighted summation method, where the weight coefficients are dynamically adjusted based on the prediction accuracy of each model in different scenarios.

[0096] The specific implementation of the trajectory prediction model includes:

[0097] Track Extrapolation Model: This model utilizes a modified Kalman filter, taking into account the UAV's dynamic characteristics and environmental constraints. This model takes the UAV's current position, velocity, acceleration, and heading as input and outputs a sequence of possible future positions. It should be noted that this model incorporates a wind compensation term to improve prediction accuracy in complex weather conditions.

[0098] The Intention Inference Model utilizes a Bayesian network architecture consisting of three layers: observation, intention hypothesis, and trajectory prediction. This model uses historical drone behavior, mission type, and communication information to infer potential flight intent and generates trajectory predictions based on that intent. The model trains parameters using the EM algorithm (Expectation Maximization) and employs variational inference to achieve real-time inference.

[0099] Historical Pattern Matching Model: This model is based on similar trajectory retrieval and pattern recognition technology. This model maintains a historical trajectory database. When a prediction is needed, it quickly retrieves similar historical trajectories using Locality Sensitive Hashing (LSH). It then calculates trajectory similarity using the Dynamic Time Warping (DTW) algorithm, ultimately fusing multiple similar trajectories to generate a prediction result.

[0100] The weight coefficients of the three sub-models are adaptively updated based on the current scenario characteristics and historical prediction accuracy. Furthermore, the trajectory prediction model exhibits different characteristics in different application scenarios: in regular flight areas, the track extrapolation model has a higher weight; in high-density intersection areas, the intention inference model has an increased weight; and in special mission areas, the historical pattern matching model has an increased weight.

[0101] Step 2.2, calculation of four-dimensional risk probability field;

[0102] Based on the core technology of this application, the system calculates the collision risk probability for each point in four-dimensional space (three-dimensional space + time dimension) based on the predicted trajectory. It should be understood that the overall collision risk probability is derived by combining the collision probabilities between all possible drone pairs. The collision probability of a single drone pair is determined by the probability of their occurrence at a specific point in time and space and the probability of their overlap at a safe distance.

[0103] The calculation formula for the overall conflict risk probability is:

[0104] ;

[0105] in, Indicates a point in time and space The overall conflict risk probability at the location; Represent three-dimensional space coordinates respectively; Represents the time dimension; Represents the multiplication operator; Indicates the total number of drones in the monitoring area; Indicates the The first drone and the drone at a point in time and space The probability of conflict at Indicates the The first drone and the The probability that the two drones will not collide at this point in time and space; represents the probability that at least one pair of drones will collide.

[0106] In some implementations, to improve the accuracy of risk calculations, the drone's motion state information can be considered, introducing a risk adjustment factor based on relative speed. Furthermore, for different types of drones, risk weightings based on performance characteristics can be adjusted. For example, a collision between a manned aircraft and an autonomous drone could be assigned a higher risk weight; for emergency drones, the risk weight could be dynamically adjusted based on the mission's urgency.

[0107] Step 2.3, risk level classification;

[0108] This application divides the calculated risk probability values ​​into multiple levels, including critical, high, medium, and low risk levels. It should be noted that the risk level division is based on preset probability thresholds, which can be adjusted according to application scenarios and management needs.

[0109] In some embodiments, the risk level threshold can be dynamically adjusted based on historical statistical data to adapt to the risk tolerance of different regions and time periods. For example, a lower risk threshold can be set over densely populated areas, while a more relaxed risk threshold standard can be used over open areas.

[0110] Optionally, risk levels can be further subdivided, adding "very high risk" and "low risk" levels to provide a more nuanced description of risk. For risks associated with manned aircraft, a stricter risk classification standard can be used; while for small drones performing non-critical tasks such as data collection, a more relaxed risk classification standard can be used.

[0111] Step 2.4: Conflict warning based on risk threshold;

[0112] According to an embodiment of the present application, the system generates conflict warnings by combining risk probability and time factors. When the maximum collision risk probability for a pair of drones within a predicted timeframe exceeds the warning threshold, the system triggers a conflict warning. For conflicts that trigger a warning, the system calculates the warning priority and records relevant information, including the location and time of the conflict, the identities of the drones involved, and the risk level.

[0113] In addition, the conflict warning mechanism provided by this application also takes the time dimension into account. It not only calculates the probability of conflict risk, but also evaluates the time urgency of the conflict, so that the system can give priority to high-risk conflicts with urgent time.

[0114] Step 3: Based on the risk distribution characteristics of the spatiotemporal risk probability field, a scenario-adaptive classification model is constructed to dynamically adjust the decision-making strategy according to the characteristics of the traffic scenario;

[0115] According to one embodiment of the present application, this step uses multi-dimensional features to build a low-altitude traffic scene classification model, automatically identifies the current scene type, and establishes a scene-strategy mapping relationship to achieve scene adaptation of the obstacle avoidance strategy. Specifically, it includes:

[0116] Step 3.1, scene feature extraction;

[0117] The method provided in this application extracts multidimensional scene feature vectors from low-altitude traffic data, including traffic density features, traffic flow features, spatial constraint features, meteorological condition features, and urgency features. It should be noted that traffic density features include average density, density variance, and density gradient; traffic flow features include flow rate, flow consistency, and speed distribution; spatial constraint features describe airspace geometry and obstacle distribution; meteorological condition features include wind speed, visibility, and humidity; and urgency features indicate the presence and severity of an emergency situation.

[0118] These multi-dimensional features together form the basis of scene description, enabling the system to fully grasp the characteristics of the current low-altitude traffic environment.

[0119] Step 3.2, scene classification model training;

[0120] The core technology of this application uses a multi-feature fusion-based scene classification model to classify low-altitude traffic scenarios into several typical categories, including high-density intersection areas, low-density open areas, narrow corridors, areas around temporary no-fly zones, areas affected by severe weather, and emergency response scenarios. The classification model uses a hierarchical classification architecture, first performing coarse-grained scene classification and then fine-grained scene recognition to improve classification accuracy and efficiency.

[0121] The scene classification model is implemented as an integrated model of multi-layer perceptron and gradient boosting decision tree. Its structure and training method are as follows:

[0122] Feature preprocessing: The input scene feature vector is first normalized and denoised. Missing values ​​are supplemented using context-aware interpolation to ensure feature data quality. Continuous time series features (such as traffic flow changes and density trends) are statistically extracted using a sliding window and converted to a fixed-dimensional representation.

[0123] Hierarchical classification architecture:

[0124] The first layer: a coarse-grained classifier, based on the random forest algorithm, divides the scenes into three categories: normal scenes, abnormal scenes, and transition scenes;

[0125] The second layer: fine-grained classifier, for each coarse-grained category, uses a specialized classifier to further subdivide, using the Gradient Boosting Decision Tree (GBDT) algorithm;

[0126] The third layer: scene attribute analyzer uses a multi-layer perceptron (MLP) to extract key attributes of the scene, such as congestion level, risk level, and meteorological impact intensity.

[0127] Classification result fusion: A weighted voting mechanism is used to fuse the classification results of each layer. The weights are automatically adjusted based on the performance of each classifier on the validation set. The final output is the scene type and attribute label.

[0128] The model, trained through supervised learning, demonstrated over 90% scene classification accuracy in tests across various cities and weather conditions, demonstrating its generalization and adaptability. Furthermore, through transfer learning and few-shot learning techniques, the system can quickly adapt to the scene characteristics of new cities, reducing deployment and adaptation time from months with traditional methods to weeks.

[0129] Step 3.3: Establish scenario-strategy mapping relationship;

[0130] According to the embodiments of the present application, the system configures an optimized set of basic obstacle avoidance strategy parameters for each scenario type, including safety margin parameters, maneuver constraint parameters, priority rules, and collaborative mode parameters. The strategy mapping relationship is constructed based on prior knowledge and scenario simulation optimization results and stored in a strategy library.

[0131] This mapping relationship enables the system to automatically select the most appropriate decision-making strategy for different scenario characteristics, such as reducing the obstacle avoidance maneuver amplitude in narrow channel scenarios to maintain traffic efficiency, and increasing the safety margin in areas affected by severe weather.

[0132] Step 3.4, dynamic adjustment of security-efficiency trade-off;

[0133] The method provided in this application dynamically adjusts the weight ratio of safety margin to path efficiency based on scenario characteristics. It should be noted that the adjustment process considers factors such as the current scenario type, traffic density, weather conditions, and urgency level, and calculates the optimal weight ratio through an adaptive function.

[0134] This adaptive function ensures an optimal balance between safety and efficiency in different scenarios, for example, appropriately reducing the safety margin to ensure efficient traffic flow in high-density areas, while increasing the safety margin in adverse weather conditions. This dynamic adjustment mechanism allows the system to flexibly respond to real-time scenario changes, avoiding the limitations of fixed-parameter strategies.

[0135] Step 4: Based on the scenario types and decision strategies output by the scenario adaptive classification model, collaborative decision optimization is achieved, using a hierarchical decision-making structure and an auction-based right-of-way allocation algorithm.

[0136] According to the embodiment of the present application, this step constructs a traffic flow topology model in the intersection area, establishes a hierarchical collaborative decision-making structure, and uses a right-of-way allocation algorithm based on an auction mechanism to optimize the overall traffic flow. Specifically, it includes:

[0137] Step 4.1, traffic flow topology model construction;

[0138] The method provided in this application constructs a traffic flow topology model for intersection areas based on low-altitude traffic data. This model is represented as a graph structure, consisting of a set of nodes, a set of edges, and edge weights. It should be noted that nodes represent intersections, entry points, and exit points, while edges represent corridors connecting nodes. Edge weights include attributes such as flow rate, capacity, and delay. By analyzing the topological structure, the system can identify key conflict points and bottleneck areas.

[0139] The specific implementation of the traffic flow topology model consists of three main components: dynamic topology map construction, topology property analysis, and a dynamic update mechanism. The dynamic topology map construction process first extracts key nodes from historical trajectory data using a density clustering algorithm. Then, the connectivity between nodes is determined based on trajectory frequency statistics. Finally, a multidimensional weight is calculated for each edge. Topology property analysis includes node centrality calculation, community detection, and bottleneck analysis, which identify key nodes and potential bottlenecks in the network.

[0140] Furthermore, this application employs a multi-level update mechanism, including short-term updates (minutes), medium-term updates (hours), and long-term updates (days), to adapt to traffic pattern changes at different time scales. This multi-level update strategy enables the system to adapt to long-term traffic pattern evolution while maintaining real-time performance.

[0141] Step 4.2: Establishment of hierarchical collaborative decision-making structure;

[0142] According to one embodiment of the present application, the system employs a two-layer collaborative decision-making structure: a macro-layer responsible for regional traffic flow balance and entry rhythm control, and a micro-layer responsible for refined individual route planning. The macro-layer monitors the overall traffic flow situation, calculates the optimal flow control parameters for each entrance, and adjusts the spatiotemporal distribution of traffic flow to avoid local congestion. The micro-layer generates obstacle avoidance trajectories for individual drones based on the current situation, coordinates conflict resolution between adjacent drones, and performs local optimization to meet global constraints.

[0143] The two-layer decision-making achieves collaboration through information sharing and constraint transmission: the macro layer transmits flow control constraints and priority information to the micro layer, and the micro layer feeds back execution status and local situation to the macro layer.

[0144] Alternatively, in areas with particularly high traffic complexity, an intermediate-level decision-making structure can be introduced, forming a three-tiered decision-making architecture: a macro-level (responsible for inter-regional traffic balancing and global resource allocation), a meso-level (responsible for intra-regional traffic flow organization and sub-regional coordination), and a micro-level (responsible for individual route planning and local conflict resolution). The introduction of a meso-level allows for finer granularity in decision-making, making it particularly suitable for managing large, intersecting areas with complex internal structures.

[0145] In some implementations, the hierarchical decision-making structure can employ an adaptive adjustment mechanism to dynamically adjust the weights and intervention levels of each layer based on traffic complexity. It should be noted that when traffic flows are low and regular, the system can reduce macro-level intervention and grant more autonomy to the micro-level. During peak traffic periods or unusual circumstances, the system can strengthen macro-level control to ensure overall system stability.

[0146] Step 4.3, allocation of right of way based on auction mechanism;

[0147] Based on the core technology of this application, the system uses an auction-based algorithm to allocate access rights in high-conflict areas. This algorithm designs a bidding function, where drones bid for time windows based on a weighted combination of the window's utility value, the drone's priority, and the cost of selecting that window. The auction execution process consists of four phases: initialization, bidding rounds, price updates, and allocation confirmation. Ultimately, each drone is assigned a time window and route based on the auction results.

[0148] The formula for calculating competitive value is:

[0149] ;

[0150] in, Indicates the drones for the time window competitive value; The weight coefficient representing the utility value of the time window; Indicates the drones for the time window Utility value of The weight coefficient indicating the priority of the drone; Indicates the Priority of drones; The weight coefficient representing the cost; Indicates the Drone selection time window cost.

[0151] In some implementations, the bidding function can incorporate collaborative factors to encourage formation flight among drones with similar objectives or belonging to the same mission group. Furthermore, for emergency missions or special circumstances, a hybrid allocation mechanism can be employed, combining priority rules and an auction mechanism. For example, in a natural disaster relief scenario, search and rescue drones could be granted priority, while drones performing routine monitoring missions would have their airtime allocated through an auction mechanism.

[0152] Another option is a dynamic bidding coefficient adjustment mechanism that dynamically adjusts bidding weights based on the overall system state. This dynamic adjustment mechanism can automatically increase the weight of efficiency-related parameters when the system is congested, optimizing overall traffic efficiency.

[0153] Step 4.4, global optimization goal is achieved;

[0154] The method provided in this application achieves global optimization by minimizing overall delay time and energy consumption. Under the premise of meeting safety interval constraints, capacity constraints, and time window constraints, the system solves the optimal passage plan through a distributed optimization algorithm, maximizing global traffic flow efficiency while ensuring safety.

[0155] It's important to note that global optimization involves a multi-objective trade-off. The system dynamically adjusts the weights of different objectives based on actual application needs. For example, it balances safety and efficiency in normal situations, prioritizes efficiency in high-density scenarios, and prioritizes safety in inclement weather. This dynamic trade-off allows the system to adapt to the optimization needs of different scenarios.

[0156] Step 5: Based on the results of collaborative decision-making optimization, the output of the quadtree spatial index structure, the risk distribution characteristics of the spatiotemporal risk probability field, and the decision-making strategy, a digital twin potential analysis platform is established to map the low-altitude traffic status in real time and provide decision support.

[0157] According to the embodiment of this application, this step integrates the above-mentioned technology into the digital twin platform to map the low-altitude traffic status in real time, analyze the current situation and predict future evolution, and support management decision-making. Specifically, it includes:

[0158] Step 5.1, digital twin model construction;

[0159] The method proposed in this application constructs a digital twin model of low-altitude traffic, which includes core components such as a spatial model, a traffic entity model, a traffic rules model, and a meteorological environment model. The digital twin model is updated in real time through multi-source data fusion technology to maintain synchronization with the physical world.

[0160] The specific implementation of the digital twin model adopts a layered architecture, including the foundation layer, environment layer, traffic layer, and control layer. It should be understood that the foundation layer contains geographic information system data, three-dimensional urban models, and terrain data, and uses multi-resolution modeling technology to dynamically adjust the model accuracy based on the area of ​​interest; the environment layer contains meteorological data and environmental conditions, using a simplified computational fluid dynamics model to simulate urban microclimate and wind field distribution; the traffic layer contains dynamic elements such as drone entities, routes, and traffic rules, and uses an agent-based modeling approach to represent various traffic participants; the control layer contains management decisions, control instructions, and response measures, and uses an event-driven state machine model to represent the control logic.

[0161] In addition, this application implements an efficient data integration and synchronization mechanism, ensuring the consistency of the digital twin model with the physical world through multi-source data fusion, time synchronization, spatial alignment, and data quality assessment. At the same time, state update and prediction technology is adopted to support real-time state updates, short-term state predictions, and long-term trend predictions, thereby improving the predictability of the model. It should be noted that the system also introduces a virtual-reality mapping accuracy assurance mechanism, which continuously optimizes the accuracy of the digital twin model through difference detection, parameter self-calibration, and model verification.

[0162] Step 5.2, situation comprehensive analysis algorithm;

[0163] According to one embodiment of the present application, the system implements comprehensive situation analysis based on a digital twin model, encompassing three aspects: current situation assessment, situation evolution prediction, and decision-making support analysis. Current situation assessment primarily involves traffic flow analysis, risk hotspot identification, and abnormal behavior detection; situation evolution prediction includes short-term (0 to 5 minutes), medium-term (5 to 15 minutes), and long-term (15 to 30 minutes) forecasts; and decision-making support analysis provides intervention effect simulation, emergency plan evaluation, and capacity optimization recommendations.

[0164] It should be noted that the situation analysis results are output in the form of structured data, including situation indicators, risk warnings and decision-making recommendations, providing managers with comprehensive situation awareness and decision-making support.

[0165] Step 5.3, multi-scale situation visualization;

[0166] The method proposed in this application develops a multi-scale situation visualization interface that supports flexible switching from a global overview to local details. This visualization interface includes a macro view (displaying overall traffic flow distribution, regional density heat maps, and the status of major corridors), a meso view (focusing on specific areas, displaying local traffic situation and conflict warnings), and a micro view (presenting detailed information on individual drones and specific conflict details).

[0167] Furthermore, the visualization interface supports a variety of interactive operations, including perspective switching (plan view, 3D view, cross-section view, etc.), time control (historical playback, current situation, forecast simulation, etc.), and information filtering (by risk level, drone type, region, etc.). These rich interactive features enable managers to comprehensively and intuitively grasp the low-altitude traffic situation.

[0168] Step 5.4, real-time response and decision support;

[0169] According to the embodiments of this application, the system establishes a closed-loop system of situational awareness, analysis, decision-making, and execution, including five steps: situation monitoring threshold setting, automatic alarm triggering, decision solution generation, solution evaluation and selection, and instruction issuance and execution. The decision support system adopts a hierarchical authorization model, automatically determining whether manual intervention is required based on the complexity of the situation.

[0170] It's important to note that the real-time response system automatically adjusts its response strategy based on the urgency of the situation. For routine changes, the system provides recommendations, leaving manual decision-making to the user. For sudden changes, the system automatically executes emergency responses within authorized scope. In extreme cases, the system triggers the highest level of alert and activates the emergency plan. This hierarchical response mechanism ensures the system's level of automation while preserving necessary channels for human intervention.

[0171] A low-altitude traffic situation analysis system based on digital twins is used to execute the above-mentioned low-altitude traffic situation analysis method based on digital twins, comprising:

[0172] A non-uniform adaptive quadtree spatial index module is used to dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density;

[0173] A spatiotemporal risk probability field generation module is used to calculate the conflict risk probability based on trajectory prediction using multi-model fusion;

[0174] A scenario-adaptive classification module, used to dynamically adjust decision-making strategies based on traffic scenario characteristics;

[0175] A collaborative decision-making optimization module for allocating right-of-way using a hierarchical decision-making structure and an auction-based mechanism;

[0176] Digital twin potential analysis platform for real-time mapping of low-altitude traffic status and providing decision support.

[0177] Here, the present invention provides an implementation example:

[0178] The digital twin-based low-altitude traffic situation analysis method proposed in this application has been applied to air traffic management in a large urban center. This area, covering approximately 40 square kilometers, includes commercial centers, residential areas, and several open spaces. Daily drone traffic averages 200 to 300 per hour, with peaks reaching 600 to 800.

[0179] The area has the following typical characteristics: 8 major low-altitude traffic corridors intersect, forming 4 high-density intersection areas; multiple high-rise building complexes create complex terrain restrictions and wind field environments; large-scale events are held regularly, requiring temporary adjustments to airspace management strategies; and there are significant seasonal meteorological changes, including heavy rainfall, strong winds and changes in visibility.

[0180] The application deployment environment includes: an edge computing server cluster with a 32-core CPU, 128GB of memory, and four GPU units; a monitoring equipment network with 12 low-altitude surveillance radars, 24 ADS-B receiving stations, and 18 optoelectronic sensor arrays; a communication network with a 5G private network and a backup communication system; and a visual control center equipped with a multi-level situation display screen and interactive operation terminals.

[0181] The system deployment adopts a center-edge collaborative architecture, which includes:

[0182] Perception layer deployment:

[0183] Monitoring equipment is deployed in a distributed manner, giving priority to covering intersections and traffic-intensive areas;

[0184] The average coverage radius of a single monitoring point is 2.5 kilometers, and the monitoring overlap rate is no less than 40%;

[0185] Data collection frequency: radar 3Hz, ADS-B 1Hz, photoelectric sensor 10Hz.

[0186] Computing layer deployment:

[0187] Lightweight processing units are deployed at edge nodes and are responsible for data preprocessing and emergency local decision-making;

[0188] The core algorithm runs on the central server, providing global situation analysis and collaborative decision-making;

[0189] Adopt elastic computing resource allocation and dynamically adjust computing power distribution according to traffic load.

[0190] Storage layer deployment:

[0191] Real-time data uses a distributed memory database to support high-frequency reading and writing;

[0192] Historical data uses a time series database, optimized by time and space indexes;

[0193] The knowledge base uses a graph database to store airspace rules and decision strategies.

[0194] The actual implementation process of the method of this application in the above application scenario is as follows:

[0195] When the system is initialized, the entire 40 square kilometers area is divided into a root node, and the initial depth is set to 3, forming an 8×8 grid. As the system runs over time, the system dynamically adjusts the quadtree structure according to traffic density:

[0196] Over busy commercial areas, the minimum cell size reaches 75 meters by 75 meters (5th layer depth), ensuring high-precision conflict detection;

[0197] In relatively open areas, the cells are maintained at 300m × 300m (depth of the third layer) to save computing resources;

[0198] The system dynamically monitors and records cell loads. During peak periods, approximately 18% of cells are at depth 5, 45% are at depth 4, and the rest remain at depth 3.

[0199] The system employs an incremental update strategy: when a drone's position is updated, only the nodes within its affected range are recalculated. In tests, a single drone's position update triggered the recalculation of only two to three leaf nodes on average, saving 96% of the computational effort compared to a global update.

[0200] The multi-level cache configuration is as follows:

[0201] Level 1 cache: the 200 most recently accessed leaf nodes;

[0202] Second-level cache: leaf nodes ranked in the top 500 by traffic density;

[0203] Level 3 cache: 100 nodes that are frequently split / merged.

[0204] In actual operation, the cache hit rate reached 82%, significantly reducing query latency.

[0205] The system implements trajectory prediction based on multi-model fusion based on the traffic characteristics of the city:

[0206] The track extrapolation model uses an improved Kalman filter and adds a wind field compensation term for complex wind fields in urban areas;

[0207] The intention inference model is based on a Bayesian network and identifies seven typical flight intention patterns in the area by analyzing historical data;

[0208] The historical pattern matching model maintains more than 100,000 historical trajectories and uses the LSH algorithm to accelerate the retrieval of similar trajectories.

[0209] In actual deployment, the model weights are adjusted dynamically:

[0210] Under normal weather conditions: the weight of the track extrapolation model is 0.5, the weight of the intention inference model is 0.3, and the weight of the historical pattern matching model is 0.2;

[0211] In strong wind conditions: the weight of the track extrapolation model is 0.3, the weight of the intention inference model is 0.4, and the weight of the historical pattern matching model is 0.3;

[0212] During large-scale events: the track extrapolation model has a weight of 0.4, the intent inference model has a weight of 0.3, and the historical pattern matching model has a weight of 0.3.

[0213] The risk level thresholds set by the system are:

[0214] Critical risk: probability value ≥ 0.75;

[0215] High risk: probability value between 0.5 and 0.75;

[0216] Medium risk: probability value between 0.25 and 0.5;

[0217] Low risk: probability value <0.25.

[0218] Implementation example of scene adaptive classification:

[0219] The system identified and classified a variety of typical scenarios in the area: high-density intersection scenarios in commercial areas; high-density scenarios above residential areas; low-density scenarios in open spaces; temporary activity area scenarios; and scenarios affected by severe weather.

[0220] The scene classification model uses a hierarchical structure, with a random forest classifier in the first layer, the GBDT algorithm in the second layer, and a five-layer fully connected neural network in the third layer. The model was trained using 12 months of historical data from the region, containing over 5 million labeled samples.

[0221] In different scenarios, the system automatically adjusts the weight ratio of safety margin and efficiency: in high-density commercial area scenarios, the safety / efficiency ratio is 1.2:1; in residential area scenarios, the safety / efficiency ratio is 1.5:1; in open space scenarios, the safety / efficiency ratio is 1:1; in temporary activity scenarios, the safety / efficiency ratio is 2:1; in severe weather scenarios, the safety / efficiency ratio is 3:1.

[0222] The system constructed a traffic flow topology model covering the entire urban area, consisting of 35 nodes and 52 edges. Through topological analysis, it identified four key conflict points (with a centrality index exceeding 0.75 and a historical conflict frequency exceeding 10 per hour) and six potential bottleneck areas (with a traffic flow / capacity ratio exceeding 0.8).

[0223] In the hierarchical collaborative decision-making structure: the macro layer performs a global traffic balance analysis every 30 seconds and calculates the optimal traffic control parameters for each entrance; the meso layer (enabled in high-density areas) performs regional traffic organization optimization every 10 seconds; the micro layer updates the route planning for all drones every 3 seconds.

[0224] The right-of-way allocation algorithm based on the auction mechanism is applied in high-conflict areas, and the bidding function parameters are as follows: utility weight α = 0.4 (normal situation) to 0.6 (congested situation); priority weight β = 0.3 (normal situation) to 0.5 (emergency situation); cost weight γ = 0.3 (normal situation) to 0.1 (congested situation).

[0225] In actual application in a high-density intersection area, the system processes about 1,200 auction processes per hour, with an average of 12 drones involved in each auction. The average time for an auction to be completed is 0.8 seconds, and the success rate of right-of-way allocation reaches 98.5%.

[0226] The system builds a digital twin model consisting of the following components:

[0227] Spatial model: Integrates a 3D geographic model of the urban area with a resolution of 1 meter;

[0228] Traffic Entity Model: supports simultaneous simulation and tracking of up to 1,000 drones;

[0229] Traffic rules model: includes 72 airspace management rules and more than 200 restrictions;

[0230] Meteorological environment model: Integrates meteorological monitoring data and simulates wind field distribution in real time with a grid resolution of 100 meters.

[0231] The comprehensive situation analysis algorithm implements the following functions:

[0232] Current situation assessment: Global situation indicators are updated every second, including regional density heat maps, risk distribution maps, and abnormal behavior markers;

[0233] Situation evolution prediction: Short-term prediction (0 to 5 minutes) accuracy reaches over 90%, and medium-term prediction (5 to 15 minutes) accuracy is about 80%;

[0234] Decision-making support analysis: The system can simulate the effects of 10 different control measures to assist managers in making decisions.

[0235] The multi-scale situation visualization interface provides three levels of views:

[0236] Macro view: covers the entire 40 square kilometer area and displays traffic density and risk distribution in the form of a heat map;

[0237] Mesoscopic view: can focus on any area of ​​1 to 5 square kilometers, showing drone tracks and conflict warnings;

[0238] Micro View: Provides accurate visibility down to the individual drone level, displaying detailed status information and predicted trajectory.

[0239] The proposed method has been used in the aforementioned practical application for over six months and has achieved significant technical results. The following is detailed verification data for the two core technical results:

[0240] Through adaptive spatial index structure and multi-level caching mechanism, the system significantly improves computing efficiency while maintaining high precision:

[0241] Conflict detection efficiency: The system can process 12,500 UAV collision detections per second, an 83.8% improvement over the traditional fixed grid method (6,800 per second).

[0242] Memory usage: At the same detection accuracy, the memory usage of the adaptive quadtree structure is 52% of that of the traditional method;

[0243] Query response time: Average query response time was reduced from 18.5 milliseconds with traditional methods to 6.2 milliseconds, a 66.5% reduction.

[0244] Incremental update efficiency: Position update processing speed increased by 96%, from 3,200 times per second to 62,800 times per second;

[0245] Resource utilization: Average CPU utilization was reduced from 85% with traditional methods to 48%, and peak utilization was reduced from 98% to 76%.

[0246] In tests during rush hour traffic, the system was able to maintain stable performance, with query response time increasing by only 15% when the number of drones increased from 300 to 800, compared to a 125% increase for traditional methods.

[0247] Through a hierarchical collaborative decision-making structure and an auction-based right-of-way allocation algorithm, the system significantly improves traffic flow efficiency and stability:

[0248] Traffic efficiency: The average time to pass through high-density intersections was reduced from 68 seconds to 43 seconds, an improvement of 36.8%;

[0249] Congestion reduction: The duration of congestion decreased by 42.5%, from an average of 85 minutes per day to 49 minutes;

[0250] Decision oscillation: The number of repeated revisions to obstacle avoidance decisions was reduced by 78%, from an average of 224 to 49 per hour.

[0251] Energy efficiency: Average drone energy consumption was reduced by 22%, primarily due to smoother trajectory planning.

[0252] Conflict Warning: The system can warn of potential conflicts 28 seconds in advance, 15 seconds earlier than traditional methods;

[0253] Safety: Near misses (close encounters without violating safety distances) were reduced by 62%, with no safety incidents occurring in six months.

[0254] In emergency scenario tests, when a certain area is suddenly banned from flying, the system can complete global traffic flow reorganization within 27 seconds, ensuring that all drones can safely adjust their routes, while traditional methods require more than 75 seconds to stabilize.

[0255] It can be seen from the above verification data that the low-altitude traffic situation analysis method based on digital twins proposed in this application has achieved good technical results in practical applications, especially in terms of computational efficiency and collaborative decision-making optimization, which proves the practicality and advancement of this method.

[0256] like Figures 2 to 6 The following are histograms comparing the non-uniform adaptive quadtree spatial index structure and the traditional indexing method in terms of different performance indicators; a line graph showing the performance change trend under different architectures as the number of drones increases (the processing time benchmark is 4ms); a scatter plot showing the distribution of conflict risk probability under different traffic densities; a radar chart showing the performance of the method of the present invention on five key indicators in different scenarios; and a pie chart showing the distribution of collaborative decision-making optimization effects.

[0257] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A low-altitude traffic situation analysis method based on digital twins, characterized in that: include: Construct a non-uniform adaptive quadtree spatial index structure and dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density; Based on the spatial partitioning results output by the quadtree spatial index structure, a spatiotemporal risk probability field is generated, and the collision risk probability is calculated using trajectory prediction using multi-model fusion. Based on the risk distribution characteristics of the spatiotemporal risk probability field, a scenario-adaptive classification model is constructed to dynamically adjust the decision-making strategy according to the characteristics of the traffic scenario; Based on the scenario types and decision strategies output by the scenario adaptive classification model, collaborative decision optimization is achieved, using a hierarchical decision-making structure and a right-of-way allocation algorithm based on an auction mechanism; Based on the results of collaborative decision-making optimization, the output results of the quadtree spatial index structure, the risk distribution characteristics and decision-making strategies of the spatiotemporal risk probability field, a digital twin potential analysis platform is established to map the low-altitude traffic status in real time and provide decision support.

2. A low-altitude traffic situation analysis method based on digital twin according to claim 1, characterized in that: In the step of constructing the non-uniform adaptive quadtree spatial index structure: Setting splitting conditions based on the number of drones, traffic density, and speed variance within a node; Set merging conditions based on the number of drones in a node, traffic density, and stabilization time; Adopting an incremental update algorithm, only locally update the affected quadtree nodes; Accelerate hot spot area queries through a multi-level caching mechanism.

3. The low-altitude traffic situation analysis method based on digital twin according to claim 1 is characterized in that: The multi-model fusion trajectory prediction combines three models: track extrapolation, intention reasoning, and historical pattern matching, and dynamically adjusts the model weights according to different scenarios; The trajectory extrapolation model is implemented using an improved Kalman filter, taking into account the UAV's dynamic characteristics and environmental constraints; The intention inference model adopts a Bayesian network structure, which includes an observation layer, an intention hypothesis layer, and a trajectory prediction layer; The historical pattern matching model is implemented based on similar trajectory retrieval and pattern recognition technology; According to the risk probability value, risks are divided into four levels: critical, high, medium and low, and conflict warnings are generated.

4. The low-altitude traffic situation analysis method based on digital twin according to claim 1 is characterized in that: In the step of constructing the scene adaptive classification model: Extract traffic density features, traffic flow features, spatial constraint features, meteorological condition features and urgency features from low-altitude traffic data; Adopting a hierarchical classification architecture, coarse-grained scene classification is performed first, followed by fine-grained scene recognition; Configure basic obstacle avoidance strategy parameter sets for each scenario type; According to the characteristics of the scenario, the weight ratio of safety margin and path efficiency is dynamically adjusted.

5. The low-altitude traffic situation analysis method based on digital twin according to claim 4 is characterized in that: The hierarchical classification architecture includes: The coarse-grained classifier, based on the random forest algorithm, divides the scenes into three categories, including normal scenes, abnormal scenes, and transition scenes; Fine-grained classifier, for each coarse-grained category, a specialized classifier is used to further subdivide it, using the gradient boosting decision tree algorithm; The scene attribute analyzer uses a multi-layer perceptron to extract the key attributes of the scene, including the degree of congestion, risk level and meteorological impact intensity.

6. The low-altitude traffic situation analysis method based on digital twin according to claim 1 is characterized in that: In the steps of achieving collaborative decision optimization: Construct a traffic flow topology model based on low-altitude traffic data, including node sets, edge sets and edge weights; Build a hierarchical collaborative decision-making structure, with the macro level responsible for regional flow balance and the micro level responsible for individual route planning; An auction-based right-of-way allocation algorithm is adopted, taking into account the time window utility value, drone priority and cost; Achieve global optimization by minimizing overall latency and energy consumption.

7. The low-altitude traffic situation analysis method based on digital twin according to claim 6 is characterized in that: The construction of the traffic flow topology model includes: Extract key nodes from historical trajectory data using density clustering algorithm; Determine the connection relationship between nodes based on trajectory frequency statistics; Calculate the flow weight, capacity weight, delay weight and conflict weight for each edge; A three-level update mechanism is adopted, including short-term updates, mid-term updates and long-term updates.

8. The low-altitude traffic situation analysis method based on digital twin according to claim 1 is characterized in that: In the steps of establishing a digital twin potential analysis platform: Build a digital twin model that includes a spatial model, a traffic entity model, a traffic rules model, and a meteorological environment model; Conduct comprehensive situation analysis, including current situation assessment, situation evolution prediction, and decision-making support analysis; Develop a multi-scale situation visualization interface that supports macro, meso, and micro views; Establish a closed-loop system of situational awareness, analysis, decision-making and execution.

9. The low-altitude traffic situation analysis method based on digital twin according to claim 8 is characterized in that: The digital twin model adopts a layered architecture, including: The base layer contains geographic information system data, three-dimensional urban models, and terrain data; The environmental layer, which contains meteorological data and environmental conditions; Traffic layer, including drone entities, routes, and traffic rules; The control layer includes management decisions, control instructions and response measures.

10. A low-altitude traffic situation analysis system based on digital twins, characterized in that: A method for analyzing low-altitude traffic situation based on digital twins, used to perform any one of claims 1-9, comprising: A non-uniform adaptive quadtree spatial index module is used to dynamically adjust the spatial partitioning granularity according to the low-altitude traffic density; A spatiotemporal risk probability field generation module is used to calculate the conflict risk probability based on trajectory prediction using multi-model fusion; A scenario-adaptive classification module, used to dynamically adjust decision-making strategies based on traffic scenario characteristics; A collaborative decision-making optimization module for allocating right-of-way using a hierarchical decision-making structure and an auction-based mechanism; Digital twin potential analysis platform for real-time mapping of low-altitude traffic status and providing decision support.

Images