CN-121766569-B - Route optimization and flow management method and system based on AI

CN121766569BCN 121766569 BCN121766569 BCN 121766569BCN-121766569-B

Abstract

The embodiment of the invention relates to the technical field of low-altitude unmanned aerial vehicle management, and particularly discloses an AI-based route optimization and flow management method and system. According to the embodiment of the invention, a plurality of regular three-dimensional grid units are divided, a double adjacency matrix is constructed and used as input of a space-time diagram neural network, traffic situation prediction in a future time window is carried out through a cloud server in a macroscopic layer, a multi-target particle swarm optimization algorithm is utilized by an edge computing node in a microscopic layer to generate a 4D track route of each unmanned aerial vehicle, the edge computing node carries out intention feedback, and when an airborne sensor detects an abnormal state, re-planning processing is triggered. The method can accurately pre-judge the future flow density of the air-domain grid, and realize the dynamic adjustment of the air-domain capacity based on predictability, thereby actively controlling the flow before the occurrence of congestion, guaranteeing the safety margin of the air-domain operation, and providing core technical support for realizing a green, efficient and sustainable low-altitude three-dimensional traffic network.

Inventors

XIE ZHIGUO
LIANG JINGHUA
Chu Jiahao
CHE KUN
NIU GUANCHONG
ZENG MINSI
SUN RAN

Assignees

广州安粤信息科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260303

Claims (6)

1. The AI-based route optimization and flow management method is characterized by comprising the following steps: Three-dimensional voxelization modeling is carried out, an urban low-altitude area is divided into a plurality of regular three-dimensional grid units, and a double adjacent matrix is constructed to serve as input of a space-time diagram neural network, wherein the double adjacent matrix comprises a physical adjacent matrix and a functional adjacent matrix; at a macroscopic layer, a space-time diagram neural network is operated through a cloud server, global space-time data are acquired and processed, traffic situation prediction in a future time window is carried out, and a space grid admission capacity threshold map is generated and issued; At a microscopic layer, an edge computing node receives the airspace grid admission capacity threshold map, and a multi-target particle swarm optimization algorithm is utilized to generate a 4D track route of each unmanned aerial vehicle; The edge computing node feeds back intention, a plurality of 4D track airlines are transmitted back to a space-time database of the cloud server in real time, and when an airborne sensor detects an abnormal state, re-planning processing is triggered; The functional adjacency matrix is constructed based on historical OD flow direction statistics, and if two three-dimensional grid units are far in physical distance but have direct route connection, connection is established; The method for obtaining and processing the global space-time data, predicting traffic situation in a future time window, and generating and issuing a space grid admission capacity threshold map specifically comprises the following steps: In the space dimension, historical traffic flow data are acquired, space feature extraction is carried out, a four-dimensional tensor is constructed, the four-dimensional tensor is input into the space-time diagram neural network, and a high-dimensional space embedded vector is output; introducing a multi-head time sequence attention mechanism and a serial structure of a gating circulation unit in a time dimension, and performing time evolution modeling and prediction output to obtain congestion probabilities of a plurality of three-dimensional grid units; according to the high-dimensional space embedded vector and the congestion probabilities, active flow control is executed, a nonlinear mapping relation between an admission capacity threshold of the airspace grid and a risk factor is established, and an admission capacity threshold map of the airspace grid is generated and issued; the edge computing node receives the airspace grid admission capacity threshold map, and generates a 4D track route of each unmanned aerial vehicle by utilizing a multi-target particle swarm optimization algorithm, wherein the method specifically comprises the following steps of: An edge computing node receives the airspace grid admission capacity threshold map; substituting the energy consumption model of the dynamic wind field, and automatically searching an energy efficiency optimal path; Substituting a population weighted noise propagation model, and planning a vertical lifting or horizontal winding strategy; Introducing an inflection point selection mechanism, and automatically selecting a balance point as a final execution route; generating a 4D track route of each unmanned aerial vehicle; the expression of the dynamic wind field energy consumption model is as follows: ; Wherein, the Is the total energy consumption in the unmanned plane flight process, For the total duration of the flight mission, The base power required to maintain a quasi-stationary hover state for the drone, As a characteristic power factor for a rotor propulsion system, For the rotational linear velocity of the rotor wing tip, For the speed of flight of the drone relative to the ground, Is the angle between the ground speed vector and the wind direction vector, For the real-time wind speed in the current environment, Is a time variable; the expression of the population weighted noise propagation model is: ; Wherein, the For the weighted noise pollution evaluation index generated by the route to the ground, For route discretized waypoint index, For the total number of waypoints contained in a single route, As a weight of the population density of the ground, Is a path point At the horizontal geographic coordinates of the ground projection, Is a predetermined smoothing factor, which is a predetermined smoothing factor, Is the reference sound pressure level of the unmanned aerial vehicle airborne sound source, In order to be able to take the altitude of the flight, Is the absorption coefficient of air to sound waves, Is a path point Linear propagation distance from the affected ground area.
2. The AI-based route optimization and traffic management method of claim 1, wherein the performing three-dimensional voxelized modeling divides the urban low-void area into a plurality of regular three-dimensional grid cells and constructs a dual adjacency matrix as an input to the space-time diagram neural network, comprising a physical adjacency matrix and a functional adjacency matrix specifically comprises the steps of: Determining a low-altitude area of the city, dividing the low-altitude area of the city into a plurality of regular three-dimensional grid units, and maintaining a multidimensional feature tensor inside each three-dimensional grid unit; A dual adjacency matrix is constructed as an input to the space-time diagram neural network, the dual adjacency matrix comprising a physical adjacency matrix and a functional adjacency matrix.
3. The AI-based route optimization and flow management method of claim 2, wherein the multidimensional feature tensor comprises a plurality of key physical quantities including flow density, an average speed field and environmental impedance, wherein the flow density is the number of aircrafts located in the three-dimensional grid unit at the current moment, the average speed field is the average value of all the aircraft speed vectors in the three-dimensional grid unit and represents microscopic kinetic energy of traffic flow, and the environmental impedance is a scalar value integrating wind speed resistance, visibility limitation and no-fly zone identification.
4. The AI-based route optimization and traffic management method of claim 1, wherein the expression of spatial feature extraction is: ; Wherein, the Is the first The feature matrix output by the layer graph neural network, Is the first The feature matrix of the layer input is used, To activate the function, for introducing a nonlinear expression capability, For the scaled laplace matrix, Is that The order chebyshev polynomials, Is the first Layer of the first layer A learnable convolution kernel weight parameter corresponding to the order chebyshev term; The expression of the nonlinear mapping relation between the airspace grid admission capacity threshold and the risk factor is as follows: ; Wherein, the For the index number of the stereoscopic grid cell, Is that Time of day A dynamic admission capacity threshold for a stereoscopic grid cell, For the physical limit capacity of the device, Is that Time of day The weather severity index of the solid grid cell, As a coefficient of sensitivity, a reference number, As a weighting factor for the congestion risk factor, Is the weight coefficient of the weather risk factor, Is that Time of day Predicted congestion probability for a stereoscopic grid cell.
5. The AI-based route optimization and traffic management method of claim 1, wherein the edge computing node performs intent feedback, transmits a plurality of the 4D trajectory routes back to a spatio-temporal database of a cloud server in real time, and triggers a re-planning process when an on-board sensor detects an abnormal state, comprising the steps of: The edge computing node performs intention feedback, and a plurality of 4D track airlines are transmitted back to a space-time database of a cloud server in real time and are used as occupied states in the future to be overlapped in a flow chart; When the airborne sensor detects sudden wind shear or abnormal state of the unreported obstacle, a perception fusion algorithm based on a dynamic diagram converter is triggered to carry out re-planning processing, and the reinforced learning strategy network is utilized to guide sampling distribution, so that an avoidance path is quickly generated.
6. AI-based route optimization and traffic management method and system for performing the AI-based route optimization and traffic management method of any of claims 1-5, wherein the system comprises a cloud server, a plurality of edge computing nodes, and a plurality of on-board sensors, wherein: The cloud server is used for carrying out three-dimensional voxel modeling, dividing an urban low-altitude area into a plurality of regular three-dimensional grid units, constructing a double adjacency matrix as input of a space-time diagram neural network, wherein the double adjacency matrix comprises a physical adjacency matrix and a functional adjacency matrix; the plurality of airborne sensors are used for detecting abnormal states; The system comprises a cloud server, a plurality of boundary computing nodes, a multi-target particle swarm optimization algorithm, a cloud server and an airborne sensor, wherein the cloud server is used for acquiring a space grid admission capacity threshold map of an unmanned aerial vehicle, the boundary computing nodes are used for receiving the space grid admission capacity threshold map at a microscopic layer, generating a 4D track route of each unmanned aerial vehicle by utilizing the multi-target particle swarm optimization algorithm, carrying out intention feedback, transmitting the 4D track routes back to the space-time database of the cloud server in real time, and triggering re-planning when the airborne sensor detects an abnormal state.

Description

Route optimization and flow management method and system based on AI Technical Field The invention belongs to the technical field of low-altitude unmanned aerial vehicle management, and particularly relates to an AI-based route optimization and flow management method and system. Background In low-altitude unmanned aerial vehicle management, space-time flow prediction and capacity management are core bases of a macroscopic scheduling layer, and as low-altitude flight activities are increasingly frequent, a traditional static airspace management mode based on rules is difficult to cope with sudden large-scale traffic flows. The multi-objective route planning technology aims to solve the problem of path optimization under complex constraint, and particularly relates to a balance strategy when multiple conflict targets such as energy consumption, noise control and dynamic weather are involved. Traditional route planning often only pursues that the distance is shortest or the time is minimum, neglects the influence of unmanned aerial vehicle flight to ground sound environment and the nonlinear loss of bad weather to battery duration. The comprehensive cost function comprising the noise propagation model, the wind resistance energy consumption model and the meteorological safety factor is constructed, the intelligent optimization algorithm is utilized to find the pareto optimal solution (ParetoOptimality) among a plurality of targets, and the unmanned aerial vehicle is ensured to give consideration to running efficiency, safety and environmental friendliness when executing tasks. At present, three typical technical paths are mainly adopted for a route planning and flow management system for low-altitude unmanned aerial vehicle operation, wherein the three typical technical paths are static management based on a fixed airspace structure, single-target path planning based on geometric rules and reactive conflict solution strategies. Although these solutions can maintain basic flight order under low-density, simple weather conditions, they still expose a number of technical bottlenecks when faced with dynamic scenarios of high frequency, complex weather and multiple constraint limits, in particular: static management scheme based on fixed airspace structure relies on pre-defined isolation airspace or 'air corridor' to perform task allocation, and is generally used for logistics trunks with wide terrain and simple environment, and the method has definite management rules and lower deployment cost, but lacks dynamic expansion capability on the airspace capacity, when a certain area suddenly meets large-scale logistics demand or encounters local bad weather, the stiff airspace structure cannot flexibly adjust the airway resources, and local sector congestion, a large number of flight delays and serious uneven airspace resource utilization rate are extremely easy to cause; geometric rule-based single-target path planning focuses on finding the shortest geometric path or the fastest time path between starting points, and is widely applied to A Classical algorithms such as Dijkstra and the like have higher calculation efficiency in a static environment, but often neglect the critical multi-physical-field constraint in low-altitude flight, and the existing algorithms mostly do not bring additional energy consumption caused by a dynamic wind field, noise pollution generated by a flying sensitive area and sudden weather conditions into a cost function, so that the generated route has the shortest theoretical distance, but in actual implementation, electric quantity is possibly consumed due to upwind, or complaints are caused due to noise disturbance, and comprehensive balance of a multidimensional target is lacking; The reactive conflict solution strategy mainly depends on intervention of a central dispatching system or an onboard anti-collision system when a flight conflict is about to occur, a 'first-come' or simple horizontal spiral avoidance rule is generally adopted, the 'passive response' mechanism lacks prediction capability for future traffic situation, when the traffic flow density is higher, frequent temporary avoidance not only greatly increases the energy consumption of the unmanned aerial vehicle, but also possibly causes a interlocked congestion effect, and the traffic efficiency of the whole low-altitude network is obviously reduced. Disclosure of Invention The embodiment of the invention aims to provide an AI-based route optimization and flow management method and system, and aims to solve the problems in the background technology. In order to achieve the above object, the embodiment of the present invention provides the following technical solutions: the AI-based route optimization and flow management method specifically comprises the following steps: Three-dimensional voxelization modeling is carried out, an urban low-altitude area is divided into a plurality of regular three-dimensional grid units, and a d