Urban logistics unmanned aerial vehicle landing site selection planning method

Abstract

The application provides a city logistics unmanned aerial vehicle landing site location planning method, comprising the following steps: constructing a virtual space logistics demand dataset; extracting an unmanned aerial vehicle landing site candidate location set; dividing a region to be planned into a discrete space grid set, and constructing a multi-objective optimization location model of the unmanned aerial vehicle landing site; designing a double-layer iterative optimization algorithm to solve the multi-objective optimization location model, and obtaining a Pareto optimal location solution set of the unmanned aerial vehicle landing site that satisfies a constraint condition; and using the virtual space logistics demand dataset under different random disturbance intensities to verify and evaluate the multi-scenario robustness of the Pareto optimal location solution set, and determining a final location planning scheme. The application can effectively eliminate the multiple-level congestion pressure caused by excessive concentration of landing sites on the surrounding road network, and ensure that the final location planning scheme still has high coverage, low traffic impact and strong risk resistance stability when encountering a sudden demand peak or a spatial demand deviation.

Description

Technical Field

[0001] This invention belongs to the field of low-altitude logistics infrastructure planning technology, specifically relating to a method for planning the location of take-off and landing sites for urban logistics drones. Background Technology

[0002] As a crucial component of smart city logistics systems, low-altitude drone logistics demonstrates significant application value in last-mile delivery and instant logistics scenarios due to its rapid response and lack of terrain limitations. With the advancement of low-altitude airspace opening, constructing a scientifically sound network of drone landing sites (Vertiports / Landing Sites) has become a primary infrastructure guarantee for supporting the large-scale development of low-altitude logistics. However, existing landing site selection and planning technologies still have many significant shortcomings in practical applications:

[0003] First, in terms of logistics demand modeling and spatiotemporal characterization, traditional site selection methods rely heavily on real static historical order data or single mean estimation. This not only makes it difficult to obtain a complete dataset of urban low-altitude delivery, but also fails to reflect the random fluctuations and spatiotemporal volatility of logistics demand in real scenarios. Consequently, the foundation for planning model construction lacks generalization ability and spatiotemporal representativeness.

[0004] Second, regarding the coupling and impact assessment of ground traffic load, existing studies generally treat UAV take-off and landing nodes as isolated spatial physical sites, focusing only on airspace accessibility or the economic cost of building the sites themselves, while ignoring the traffic transmission mechanism between "UAVs and ground connections." When take-off and landing nodes handle goods in a concentrated manner, they inevitably induce a large number of last-mile delivery vehicles (such as express delivery tricycles and mini-trucks) to frequently enter and exit the roads around the sites. Existing site selection models fail to accurately convert the cargo throughput of take-off and landing points into ground connection traffic flow, and do not consider the cumulative effect of this connection traffic flow on traffic congestion in surrounding areas of different densities of Points of Interest (POIs), the continuous attenuation law with increasing spatial Euclidean distance, and the exponential congestion pressure on the road network when the radiation ranges of multiple take-off and landing points overlap.

[0005] Third, in terms of solution algorithm paradigm and decision robustness, existing multi-objective location selection algorithms mostly adopt traditional heuristic solutions in a flat manner, which makes it difficult to efficiently handle the double-layer nested relationship between "take-off and landing node selection" and "dynamic allocation of internal demand flow", and is very easy to get trapped in local optima. At the same time, most existing solutions lack robustness testing mechanisms for location solution sets under multiple scenarios and strong disturbances, resulting in poor adaptability of the final location selection scheme when facing sudden demand peaks or spatial demand shifts.

[0006] Therefore, how to organically integrate dynamic modeling of low-altitude logistics demand, coupled analysis of ground connection traffic pressure derived from take-off and landing point throughput, a two-layer iterative optimization mechanism, and multi-scenario robustness assessment into a single method for urban logistics drone take-off and landing site selection and planning that balances high demand coverage with low ground traffic congestion is a key technical challenge that urgently needs to be addressed in this field. Summary of the Invention

[0007] Purpose of the Invention: This invention aims to address the technical shortcomings of existing low-altitude drone take-off and landing facility planning, such as secondary traffic congestion on ground roads caused by neglecting last-mile connecting vehicles, difficulties in obtaining low-altitude logistics space demand data, and the lack of multi-scenario robustness in traditional site selection schemes. Specifically, this invention provides a method for planning take-off and landing sites for urban logistics drones. This method constructs a virtual space logistics demand dataset based on mixed probability distribution and Monte Carlo sampling; it combines geographical no-fly zones and safety isolation constraints to screen candidate site selection sets; it constructs a multi-objective optimization site selection model with the objectives of maximizing demand coverage and minimizing the ground traffic congestion pressure index fused with connecting traffic volume, POI density, distance attenuation, and overlap radiation penalty; it designs a two-layer iterative algorithm consisting of an outer-layer site selection master algorithm and an inner-layer flow allocation sub-algorithm to solve for the Pareto optimal solution set; and it uses datasets with different random disturbance intensities to conduct multi-scenario robustness evaluation and determine the final scheme. This invention solves the problems of missing low-altitude order data, secondary ground congestion caused by last-mile connecting vehicles, and weak anti-disturbance capability of site selection schemes, and is applicable to scenarios such as urban logistics planning and low-altitude economic infrastructure construction.

[0008] The method includes the following steps:

[0009] Step S1: Obtain the spatial geographic elements and logistics distribution characteristics of the urban area to be planned, and construct a virtual spatial logistics demand dataset based on the probability distribution model and Monte Carlo random sampling method.

[0010] Step S2: Based on the geographical no-fly zone and safety interval constraints, the region is spatially filtered, and the clustering characteristics of the virtual space logistics demand dataset are combined to extract a set of candidate locations for drone take-off and landing.

[0011] Step S3: Divide the area to be planned into a discrete spatial grid set. With the dual optimization objectives of maximizing demand coverage and minimizing the regional ground traffic congestion pressure index, construct a multi-objective optimization site selection model for UAV take-off and landing sites. The ground traffic congestion pressure index integrates the ground connection traffic flow per unit time derived from the cargo throughput of the take-off and landing points, the density of spatial grid points of interest (POI), the continuous decay function based on Euclidean distance, and the superposition amplification coefficient of the overlapping radiation pressure of multiple take-off and landing points.

[0012] Step S4: Design a two-layer iterative optimization algorithm to solve the multi-objective optimization location model and obtain the Pareto optimal location solution set of the UAV take-off and landing sites that meets the constraints.

[0013] Step S5: Using virtual space logistics demand datasets under different random disturbance intensities, the robustness of the Pareto optimal location solution set is verified and evaluated under multiple scenarios to determine the final location planning scheme.

[0014] Step S1 specifically includes:

[0015] A two-dimensional spatial coordinate system is established within the urban area. A hybrid probability distribution model is used to generate N virtual logistics demand points. The spatial coordinates of each virtual logistics demand point are randomly selected using a spatial density distribution function. The order demand volume of each virtual logistics demand point is generated according to a Poisson or normal distribution. A Monte Carlo algorithm is then used to simulate logistics demand fluctuations at different times, constructing a virtual spatial logistics demand matrix D.

[0016] ,

[0017] in Let i be the coordinates of the i-th virtual logistics demand point. Let represent the demand quantity corresponding to the i-th virtual logistics demand point.

[0018] Step S2 specifically includes the following steps:

[0019] Step S2.1: Perform cluster analysis on the virtual space logistics demand matrix D using kernel density estimation or density clustering algorithm, and extract the centroid coordinates of the clusters as initial candidate points.

[0020] Step S2.2: Import no-fly zones and high-risk safety buffer zones, and common logistics drone flight parameters from the geographic information system, and remove initial candidate points that fall within no-fly zones, high-risk safety buffer zones, and other unusable routes for drones;

[0021] Step S2.3: Calculate the Euclidean distance between the remaining initial candidate points. When the distance between any two candidate points is less than the preset safety isolation threshold, retain the candidate point with the larger required service coverage and remove the other candidate point, finally generating the candidate site selection set J for UAV take-off and landing.

[0022] ,

[0023] Where M represents the total number of final candidate take-off and landing sites.

[0024] In step S3, the optimization objective of the multi-objective location optimization model is established as follows:

[0025] Objective 1: Maximize demand coverage :

[0026] ,

[0027] Where I represents the set of virtual logistics demand points. As a decision variable, if the i-th virtual logistics demand point is located within the effective service radius of at least one selected take-off and landing location, then =1, otherwise =0;

[0028] Objective 2: Minimize the regional surface traffic congestion pressure index :

[0029] ,

[0030] Where K is the set of spatial grids for dividing the area to be planned. As a decision variable, if a landing site is built at candidate point j, then =1, otherwise =0; To cover the identifier variable, when the Euclidean distance between grid k and take-off / landing point j... When not greater than the preset radiation radius R =1 otherwise 0; Let J be the ground connection traffic flow generated at take-off and landing point j per unit time, and = , For the predicted total terminal cargo throughput allocated to takeoff and landing point j, For conversion factors of ground shuttle vehicles; The POI density index within grid K; For Euclidean distance A monotonically decreasing pressure decay function; The total number of selected take-off and landing points covering grid k; This is the superposition amplification factor for overlapping radiation pressure. >0.

[0031] In step S3, the pressure decay function Using an exponential decay model = , where e is the natural constant; This is the distance attenuation coefficient.

[0032] In step S3, the ground connection traffic flow The calculation formula is: = Based on the number of cargo throughput orders at the terminal point j and the preset average cargo load equivalent of the shuttle vehicles, the throughput order number is converted into the number of ground shuttle vehicles or the equivalent of standard small transport vehicles per unit time.

[0033] In step S3, the constraints of the multi-objective optimization location selection model include:

[0034] Limit on the number of takeoff and landing sites: The total number of selected takeoff and landing sites shall not exceed the preset upper limit. And not lower than the preset lower limit (At least ensure complete coverage of the selected area);

[0035] Terminal throughput constraint: The total predicted terminal cargo throughput at any takeoff and landing point j. Maximum daily processing capacity not exceeding take-off and landing point j ;

[0036] Drone single-trip delivery radius constraint: The straight-line distance from take-off and landing point j to the i-th virtual logistics demand point served by take-off and landing point j shall not exceed the set delivery radius. .

[0037] In step S4, the two-layer iterative optimization algorithm includes an outer-layer location decision main algorithm and an inner-layer flow allocation sub-algorithm:

[0038] The outer-layer location decision master algorithm is used to process decision variables. Perform global heuristic search and population evolution update;

[0039] The inner-layer flow allocation sub-algorithm is used to match and map each virtual logistics demand point to its corresponding take-off and landing point based on the principles of proximity and no capacity overload, given the known outer-layer location combination, and to calculate the real-time end-point throughput of each take-off and landing point. And update the coverage status of the demand points. The calculated demand coverage rate and regional ground traffic congestion pressure index Feedback is provided to the outer layer location decision master algorithm.

[0040] In step S4, the two-layer iterative optimization algorithm specifically includes the following steps:

[0041] Step S4.1: The outer-layer addressing decision master algorithm randomly initializes the population and generates individuals containing binary addressing codes;

[0042] Step S4.2: Input each individual in the outer population into the inner flow allocation sub-algorithm. The inner flow allocation sub-algorithm performs greedy matching based on the nearest distance principle to determine the service assignment of each virtual logistics demand point and update the decision variables. and predicted terminal throughput at takeoff and landing points ;

[0043] Step S4.3: Combine the POI density data around the candidate sites to calculate the demand coverage rate corresponding to each site selection scheme. Regional ground traffic congestion pressure index Construct the bi-objective fitness vector corresponding to each location scheme. ;

[0044] Step S4.4, the outer-layer location decision master algorithm determines the bi-objective fitness based on the calculated fitness ( Perform non-dominated sorting and crowding calculation, and generate offspring populations through selection, crossover and mutation operators;

[0045] Step S4.5: Repeat steps S4.2 to S4.4 until the preset maximum number of iterations or Pareto preface convergence index is met, and output the Pareto optimal address solution set.

[0046] Step S5 includes:

[0047] By adjusting the mixed probability distribution parameters and random seeds, M sets of test virtual datasets with different spatial distribution patterns and peak demand intensities are generated in batches. The Pareto optimal site selection solution set output in step S4 is then substituted into the M sets of test virtual datasets for simulation. The variance of demand coverage and the maximum value of ground traffic congestion pressure of each site selection scheme under multiple scenario fluctuations are statistically analyzed. The site selection solution with the smallest comprehensive standard deviation and the best mean is selected as the final UAV take-off and landing site selection planning scheme.

[0048] Beneficial effects: This invention introduces an air-to-ground traffic transmission mechanism, which derives the throughput of take-off and landing points into ground connecting traffic flow. It also comprehensively considers the spatial grid POI density, Euclidean distance attenuation, and overlap radiation penalty, effectively eliminating the multi-fold congestion pressure on the surrounding road network caused by the excessive concentration of take-off and landing sites.

[0049] Through multi-scenario random disturbance testing and a dimensionless comprehensive robustness evaluation function, the final site selection scheme is ensured to maintain high coverage, low traffic impact, and strong risk resistance stability even when encountering sudden demand peaks or spatial demand shifts. Attached Figure Description

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0051] Figure 1 This is a schematic diagram of the process of the present invention.

[0052] Figure 2 This is a schematic diagram of the take-off and landing site selection results of the present invention.

[0053] Figure 3 This is a schematic diagram illustrating the convergence of the two-layer algorithm of this invention. Detailed Implementation

[0054] This invention provides a method for site selection and planning of take-off and landing sites for urban logistics drones, comprising the following steps:

[0055] Step S1, to address the issue of missing or non-public high-precision low-altitude logistics order data in real-world urban areas, this step employs a method based on a mixed probability distribution model and Monte Carlo random sampling to generate a virtual logistics demand dataset D with spatiotemporal distribution characteristics. The specific process is as follows:

[0056] Step S1.1, Spatial Coordinate Generation: Establish a two-dimensional Cartesian coordinate system in the area to be planned, and use a two-dimensional Gaussian mixture probability density function constructed based on spatial geographic elements (such as the density of residential areas and commercial areas). The spatial coordinates of N virtual logistics demand points are generated through random sampling. , where i = 1, 2, ..., N.

[0057] Step S1.2, Demand Assignment: For each generated virtual i-th virtual logistics demand point, a Poisson distribution is used to simulate its initial order demand within a set time period. Taking the Poisson distribution as an example, the demand... probability mass function Represented as:

[0058] ,

[0059] in, Let X be the benchmark mean of spatiotemporal demand intensity in the region where the i-th virtual logistics demand point is located. X represents a random variable with a mean of 1 / 2. The Poisson distribution.

[0060] Step S1.3 involves setting up a Monte Carlo random sampling cycle to further characterize the dynamic fluctuations in demand across different time periods. In the t-th sampling experiment, a spatiotemporal fluctuation coefficient is introduced. The implementation demand of each virtual logistics demand point. Updated to .

[0061] Finally, a virtual space logistics demand dataset matrix is ​​constructed:

[0062] ,

[0063] Step S2 involves spatially filtering the region based on geographical no-fly zones and safety interval constraints, and extracting a set of candidate locations for UAV take-off and landing sites by combining the clustering features of the virtual space logistics demand dataset; specifically including:

[0064] Step S2.1: Use density clustering algorithm to perform spatial clustering on the virtual demand matrix D generated in step S1, identify high-frequency demand clustering areas, and extract the centroid coordinates of each cluster as initial candidate sites.

[0065] Step S2.2 involves combining the initial candidate sites with the actual spatial constraints of the urban area to eliminate initial candidate sites that fall within no-fly zones or high-risk safety buffer zones. Simultaneously, based on preset common logistics drone cruise performance parameters, invalid sites that exceed the effective route connection range are eliminated.

[0066] Step S3: Divide the area to be planned into a set of discrete spatial grids. With the dual optimization objectives of maximizing demand coverage and minimizing the regional ground traffic congestion pressure index, construct a multi-objective optimization site selection model for UAV take-off and landing sites. The ground traffic congestion pressure index integrates the ground connection traffic flow per unit time derived from the cargo throughput of the take-off and landing points, the spatial grid POI density, the continuous decay function based on Euclidean distance, and the superposition amplification coefficient of the overlapping radiation pressure of multiple take-off and landing points.

[0067] The area to be planned is divided into a set of K spatial square grids with equal side lengths, K = {1, 2, ..., K}. An optimization model is constructed with the dual objectives of maximizing demand coverage and minimizing the surface traffic congestion pressure index.

[0068] Objective 1: Maximize demand coverage :

[0069] ,

[0070] Where I is the set of all virtual logistics demand points. ; Let i be the order demand quantity for the i-th virtual logistics demand point; Let the demand cover state decision variable be denoted as . If the i-th virtual logistics demand point is within the effective radius R of at least one selected take-off and landing point, then ;on the contrary =0.

[0071] Objective 2: Minimize the regional surface traffic congestion pressure index :

[0072] ,

[0073] in For the location decision variable, if the station is built at candidate point j, then... ,otherwise . For spatial grid correlation variables, the Euclidean distance between the center of grid k and candidate point j is... hour ,otherwise . Let the ground connection traffic flow derived from take-off and landing point j per unit time be defined as... = . The predicted terminal cargo throughput is allocated to the take-off and landing point j for the inner layer; Ground connection conversion factor (representing the equivalent of a standard small truck generated per unit order). is the POI density index within grid k, used to characterize the inherent sensitivity and baseline congestion level of the current road traffic system on the ground within this network. The continuous negative exponential pressure decay function based on Euclidean distance is expressed as: = λ > 0 is the spatial decay rate constant. To determine the total number of selected takeoff and landing sites covering grid k, the calculation formula is as follows: . This represents the superposition amplification factor of the radiation pressure from multiple overlapping take-off and landing points. When grid k falls within the radiation range of multiple take-off and landing points simultaneously, this term generates a dynamic penalty to prevent excessively dense deployment of take-off and landing fields in the same area.

[0074] Step S4: Design a two-layer iterative optimization algorithm to solve the multi-objective optimization location model and obtain the Pareto optimal location solution set of the UAV take-off and landing site that meets the constraints.

[0075] Given the location decision variables With demand flow allocation There is a strong two-layer coupling between them. This step adopts a solution mechanism that uses the outer-layer main addressing algorithm and the inner-layer flow allocation sub-algorithm to evolve together:

[0076] Step S4.1, Outer Main Algorithm Encoding and Initialization: The outer location selection decision main algorithm uses binary chromosome encoding, with the gene bit length equal to the total number of candidate points M. A value of 1 in the binary string indicates that the candidate point is selected, and 0 indicates that it is not selected. The random generation scale is... The initial site selection population.

[0077] Step S4.2: Transfer the addressing combination scheme passed from the outer layer. Access the inner layer traffic allocation algorithm. The inner layer traffic allocation algorithm prioritizes closest proximity and allows for capacity not exceeding a certain limit. The dual principle is used to iterate and allocate all virtual i-th virtual logistics demand points:

[0078] Find the virtual logistics demand point that is closest to the i-th virtual logistics demand point and has not reached its capacity limit. Selected take-off and landing points ;

[0079] Match the i-th virtual logistics demand point to the destination and update the cumulative throughput of that origin and destination point. ;

[0080] Update the coverage state variable of the i-th virtual logistics demand point. If there are no take-off and landing points that meet the conditions, then .

[0081] Step S4.3, Fitness Assessment and Target Calculation: Utilize the throughput of each takeoff and landing point obtained from the inner flow allocation algorithm. Exported connection traffic = Combined with the POI density of each grid Substitute the values ​​into the formula in step S3 to calculate the demand coverage rate for each site selection. Ground traffic congestion pressure index The bi-objective function values ​​are combined to form the bi-objective fitness vector F=( This serves as the basis for subsequent evolutionary evaluation and selection.

[0082] Step S4.4: The outer main algorithm uses the fast domination sorting algorithm to perform Pareto hierarchical partitioning of the current population based on the bi-objective fitness vector F of each individual; the uniformity of solution distribution is evaluated by crowding distance. The next generation of offspring population is generated using tournament selection operator, single / multi-point crossover operator, and basic position operator.

[0083] Step S4.5, Algorithm Convergence and Output: Merge the parent and child populations, and repeat steps S4.2-S4.4 until the algorithm reaches the maximum number of iterations. Or the Pareto optimal frontier index in continuous Once no further significant changes occur, the algorithm terminates and outputs the optimal set of UAV take-off and landing sites that satisfies multiple constraints.

[0084] Step S5: Using virtual space logistics demand datasets under different random disturbance intensities, perform multi-scenario robustness verification and evaluation on the Pareto optimal site selection solution set to determine the final site selection planning scheme.

[0085] In multi-objective programming, the Pareto optimal location solution set obtained in step S4 contains multiple candidate solutions that compromise between demand coverage and traffic congestion. Since actual urban logistics is subject to strong dynamic disturbances due to holidays, extreme weather, or unforeseen events, to prevent the selected solution from experiencing a sharp drop in coverage or excessive local pressure when encountering demand fluctuations, this step uses a spatial logistics demand dataset under different random disturbance intensities to conduct a multi-scenario robustness evaluation and final decision on the Pareto solution set. The specific implementation process is as follows:

[0086] Keeping the two-dimensional spatial coordinate system and basic geographic constraints unchanged in step S1, batch generation is achieved by adjusting the statistical distribution parameters and random number seed of the demand generation model. The group contains virtual test datasets with varying peak intensities and spatiotemporal fluctuations. (where m=1,2,..., And set To ensure the sample size requirement for statistical robustness under the law of large numbers.

[0087] For the m-th test dataset Introducing a peak intensity amplification factor for global demand. ( (Used to simulate fluctuations in total order volume), updating the order volume at each virtual logistics demand point to... , expressed as:

[0088] ,

[0089] Let the Pareto optimal address solution set output in step S4 be represented as follows: , where P is the total number of solutions in the Pareto solution set. Let p represent the set of specific take-off and landing sites, where p = 1, 2, ..., P.

[0090] Each site selection scheme Substitute into Group test dataset In the next step, the inner flow molecular algorithm from step S4 is run again to re-match the demand and calculate the real-time throughput of each take-off and landing point. Then, the scheme is calculated. Demand coverage in the m-th demand scenario Ground traffic congestion pressure index .

[0091] Construct a robustness evaluation index system:

[0092] For each site selection option ,statistics exist Based on the statistical characteristics of demand fluctuations, the following indicators were constructed:

[0093] Demand coverage expected value ;

[0094] in Indicates the site selection scheme Requirements coverage in m test scenarios.

[0095] Fluctuation Standard Deviation ;

[0096] in Reflection Plan Long-term average service guarantee capability; The standard deviation of the demand coverage rate is denoted as σ. The smaller the value, the stronger the service stability of the location scheme.

[0097] Extreme traffic pressure (1≤m≤ );

[0098] average pressure ;

[0099] in This reflects the maximum traffic impact that the plan will have on surrounding roads during peak demand periods; This represents the traffic load under normal operating conditions.

[0100] Construct a comprehensive robustness evaluation function Each site selection scheme was scored using a dimensionless method.

[0101] ,

[0102] in For the plan The overall robustness score indicates that the higher the score, the better the overall performance of the solution.

[0103] Each denominator term is used to normalize evaluation indicators of different dimensions to... interval;

[0104] , , , The preset decision preference weight coefficients are given, and the sum of the four terms equals 1.

[0105] Iterate through and compute all solutions in the Pareto solution set. The site selection scheme S with the highest comprehensive robustness score (i.e., large average coverage, small service fluctuation, and low average and peak traffic pressure) was selected as the final site selection plan for urban logistics drone take-off and landing sites.

[0106] The technical approach adopted in this patent is as follows: Figure 1 As shown.

[0107] This embodiment takes a typical urban core area as the planning object, and the specific planning and site selection implementation steps are as follows:

[0108] Step 1: Construct a virtual space logistics demand dataset;

[0109] Based on the 300 virtual logistics demand points generated above, spatial clustering analysis was first performed using a density clustering algorithm to accurately identify high-frequency logistics demand clusters. Density clustering parameters were set (neighborhood radius Eps = 1.2 km, minimum sample size MinPts = 5), and the centroid coordinates of these clusters were extracted as initial candidate stations for takeoff and landing (a total of 25 high-demand centroids were extracted). Then, the spatial filtering stage began:

[0110] Import the city GIS layer, in this example, the core area is 10km × 10km ( Within this framework, two geographically defined no-fly zones, each 1km x 1km, are explicitly established, and initial candidate points falling within the restricted airspace are forcibly eliminated; in this example, no-fly zone A is randomly generated. B After spatial matching, four initial candidate points that fell within the no-fly zone were eliminated, leaving 21. Furthermore, the Euclidean spatial distances between the remaining candidate sites were calculated. In this embodiment, a preset safety isolation threshold was set to 2km. When the distance between any two candidate points is less than 2km, the system will trigger a filtering mechanism, retaining advantageous candidate points that can cover a larger volume of logistics demand and eliminating disadvantageous points (a total of 6 nodes were eliminated). Finally, a set of candidate locations for UAV take-off and landing sites was compiled and output.

[0111] Step 2: Extraction of candidate locations for UAV take-off and landing;

[0112] Based on the 300 virtual logistics demand points generated above, a density clustering algorithm is first used to perform spatial clustering analysis to accurately identify areas of high-frequency logistics demand. The centroid coordinates of these clusters are then extracted as initial candidate stations for take-off and landing. The spatial filtering stage then follows.

[0113] Import the city GIS layer, in this example, the core area is 10km × 10km ( Within the system, two geographically defined no-fly zones, each 1km x 1km, are explicitly established to forcibly eliminate initial candidate sites falling within these restricted airspaces. Furthermore, the Euclidean spatial distances between the remaining candidate sites are calculated. In this embodiment, a preset safety isolation threshold is set to 2km. When the distance between any two candidate sites is less than 2km, the system triggers a filtering mechanism, retaining advantageous candidate sites that can cover a larger volume of logistics demand and eliminating disadvantageous sites, ultimately refining and outputting a set of candidate locations for UAV take-off and landing.

[0114] Step 3: Construction of a multi-objective optimization location selection model;

[0115] The 20km × 20km area to be planned is discretized into a grid, with each grid cell having a side length of 1km. The total plan is divided into K = 400 square spatial grids of 1km × 1km each. Within this spatial framework, a dual optimization objective model is constructed to "maximize demand coverage" and "minimize the regional ground traffic congestion pressure index."

[0116] Regarding the constraint of demand coverage targets, the economic operating one-way delivery radius of current mainstream multi-rotor logistics drones is set as follows: =5km. The model aims to include as many order demands as possible within the 5km radius of existing stations. Meanwhile, while ensuring effective coverage of the study area, the total number of take-off and landing sites is strictly controlled between a lower limit K=5 and an upper limit K=15.

[0117] Regarding minimizing the regional ground traffic congestion pressure index: this indicator incorporates the air-to-ground traffic transmission mechanism. In the example, the maximum daily processing capacity of the take-off and landing points... Set at 2000 orders / day. The system calculates the throughput at the take-off and landing points based on the ground connection conversion factor. =0.0125 (i.e., 80 last-mile delivery packages are equivalent to one shuttle bus trip in a standard minivan) to calculate the resulting ground shuttle traffic flow. The congestion pressure caused by this shuttle flow to the surrounding grid decreases exponentially with increasing straight-line distance, and the spatial decay rate constant λ is set to 0.5. To reflect the differences in road foundation bearing capacity, the POI density index of each grid is... After normalization, values ​​are taken between [0.1, 1.0] (1.0 represents the core business district, and 0.1 represents a low-density building area). Specifically, when a grid is overlapped by two or more takeoff and landing fields, its traffic pressure will be... A superimposed penalty is applied using an amplification factor of 0.3 to prevent excessively dense deployment of take-off and landing fields in the same area.

[0118] Step 4: Solve using a two-level iterative optimization algorithm;

[0119] A two-level nested mechanism is used for collaborative optimization of the constructed location model.

[0120] The outer-layer site selection decision algorithm uses binary encoding to globally evolve the construction status of all candidate sites. In this embodiment, the initial population size is set to... The maximum number of iterations is set to The inner traffic allocation sub-algorithm is responsible for evaluating each location chromosome combination passed from the outer layer and performing real-time traffic allocation. Following the dual principles of proximity and not exceeding the site's maximum capacity (2000 orders / day), the inner layer maps 300 virtual logistics demand points to corresponding take-off and landing points, calculates the real-time throughput of each take-off and landing point, and dynamically updates the coverage status. The bi-objective evaluation function value calculated by the inner layer is fed back to the outer main algorithm. Based on fitness feedback, the outer main algorithm performs non-dominated sorting and congestion distance evaluation, evolving for 500 generations and outputting a set of all Pareto optimal location solutions that satisfy both safety and capacity constraints.

[0121] The algorithm of this patent converges as follows: Figure 2 As shown.

[0122] Step 5: Multi-scenario robustness verification and solution decision-making;

[0123] To ensure the final solution is resilient to sudden surges in demand, this step involves multi-scenario stress testing of the obtained Pareto solution set. The system is configured... =30 sets of virtual datasets with different fluctuation intensities, re-run traffic allocation, and statistically analyze the expected coverage, variance of coverage, maximum traffic pressure and mean of each site selection scheme under dynamic disturbance.

[0124] Finally, a dimensionless score is given to each scheme using a comprehensive robustness evaluation function. The decision preference weights set in the example are:

[0125] Daily average demand coverage =0.4,

[0126] Demand coverage stability =0.2,

[0127] Off-peak traffic pressure =0.2,

[0128] Resisting extreme peak pressure =0.2.

[0129] Table 1 shows a comparison of the robustness of some solution schemes.

[0130] Table 1

[0131]

[0132] After 500 generations of evolution and 30 sets of scenario robustness tests, the system finally selected the optimal location scheme with the highest comprehensive robustness score (0.915). This scheme has a total of 8 UAV take-off and landing points, and the final accurate virtual two-dimensional coordinates and spatial verification results are shown in Table 2.

[0133] Table 2

[0134]

[0135] The final takeoff and landing site selection results are as follows: Figure 3 As shown.

[0136] In summary, this invention balances the needs of urban drone logistics with the demands of traffic congestion, providing a solution to the problem of selecting take-off and landing sites for urban logistics drones.

[0137] This invention provides a method for site selection and planning of take-off and landing sites for urban logistics drones. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for site selection and planning of take-off and landing sites for urban logistics drones, characterized in that, Includes the following steps: Step S1: Obtain the spatial geographic elements and logistics distribution characteristics of the urban area to be planned, and construct a virtual spatial logistics demand dataset based on the probability distribution model and Monte Carlo random sampling method. Step S2: Based on the geographical no-fly zone and safety interval constraints, the region is spatially filtered, and the clustering characteristics of the virtual space logistics demand dataset are combined to extract a set of candidate locations for drone take-off and landing. Step S3: Divide the area to be planned into a set of discrete spatial grids. With the dual optimization objectives of maximizing demand coverage and minimizing the regional ground traffic congestion pressure index, construct a multi-objective optimization site selection model for UAV take-off and landing sites. The ground traffic congestion pressure index integrates the ground connection traffic flow per unit time derived from the cargo throughput of the take-off and landing points, the density of points of interest (POIs) in the spatial grid, the continuous decay function based on Euclidean distance, and the superposition amplification coefficient of the overlapping radiation pressure of multiple take-off and landing points. Step S4: Design a two-layer iterative optimization algorithm to solve the multi-objective optimization location model and obtain the Pareto optimal location solution set of the UAV take-off and landing sites that meets the constraints. Step S5: Using virtual space logistics demand datasets under different random disturbance intensities, the robustness of the Pareto optimal location solution set is verified and evaluated under multiple scenarios to determine the final location planning scheme.

2. The method according to claim 1, characterized in that, Step S1 specifically includes: A two-dimensional spatial coordinate system is established within the urban area. A hybrid probability distribution model is used to generate N virtual logistics demand points. The spatial coordinates of each virtual logistics demand point are randomly selected using a spatial density distribution function. The order demand volume of each virtual logistics demand point is generated according to a Poisson or normal distribution. A Monte Carlo algorithm is then used to simulate logistics demand fluctuations at different times, constructing a virtual spatial logistics demand matrix D. ， in Let i be the coordinates of the i-th virtual logistics demand point. Let represent the demand quantity corresponding to the i-th virtual logistics demand point.

3. The method according to claim 2, characterized in that, Step S2 specifically includes the following steps: Step S2.1: Perform cluster analysis on the virtual space logistics demand matrix D using kernel density estimation or density clustering algorithm, and extract the centroid coordinates of the clusters as initial candidate points. Step S2.2: Import no-fly zones and high-risk safety buffer zones, and common logistics drone flight parameters from the geographic information system, and remove initial candidate points that fall within no-fly zones, high-risk safety buffer zones, and unusable drone routes; Step S2.3: Calculate the Euclidean distance between the remaining initial candidate points. When the distance between any two candidate points is less than the preset safety isolation threshold, retain the candidate point with the larger required service coverage and remove the other candidate point, finally generating the candidate site selection set J for UAV take-off and landing. ， Where M represents the total number of final candidate take-off and landing sites.

4. The method according to claim 3, characterized in that, In step S3, the optimization objective of the multi-objective location optimization model is established as follows: Objective 1: Maximize demand coverage : ， Where I represents the set of virtual logistics demand points. As a decision variable, if the i-th virtual logistics demand point is located within the effective service radius of at least one selected take-off and landing location, then =1, otherwise =0; Objective 2: Minimize the regional surface traffic congestion pressure index : ， Where K is the set of spatial grids for dividing the area to be planned. As a decision variable, if a landing site is built at candidate point j, then =1, otherwise =0; To cover the identifier variable, when the Euclidean distance between grid k and take-off / landing point j... When not greater than the preset radiation radius R =1 otherwise 0; Let J be the ground connection traffic flow generated at take-off and landing point j per unit time, and = , For the predicted total terminal cargo throughput allocated to takeoff and landing point j, For conversion factors of ground shuttle vehicles; The POI density index within grid K; For Euclidean distance A monotonically decreasing pressure decay function; The total number of selected take-off and landing points covering grid k; This is the superposition amplification factor for overlapping radiation pressure. >

0.

5. The method according to claim 4, characterized in that, In step S3, the pressure decay function Using an exponential decay model = , where e is the natural constant; This is the distance attenuation coefficient.

6. The method according to claim 5, characterized in that, In step S3, the ground connection traffic flow The calculation formula is: = Based on the number of cargo throughput orders at the terminal point j and the preset average cargo load equivalent of the shuttle vehicles, the throughput order number is converted into the number of ground shuttle vehicles or the equivalent of standard small transport vehicles per unit time.

7. The method according to claim 6, characterized in that, In step S3, the constraints of the multi-objective optimization location selection model include: Limit on the number of takeoff and landing sites: The total number of selected takeoff and landing sites shall not exceed the preset upper limit. And not lower than the preset lower limit ; Terminal throughput constraint: The total predicted terminal cargo throughput at any takeoff and landing point j. Maximum daily processing capacity not exceeding the take-off and landing point j ; Drone single-trip delivery radius constraint: The straight-line distance from take-off and landing point j to the i-th virtual logistics demand point served by take-off and landing point j shall not exceed the set delivery radius. .

8. The method according to claim 7, characterized in that, In step S4, the two-layer iterative optimization algorithm includes an outer-layer location decision main algorithm and an inner-layer flow allocation sub-algorithm: The outer-layer location decision master algorithm is used to process decision variables. Perform global heuristic search and population evolution update; The inner-layer flow allocation sub-algorithm is used to match and map each virtual logistics demand point to its corresponding take-off and landing point based on the principles of proximity and no capacity overload, given the known outer-layer location combination, and to calculate the real-time end-point throughput of each take-off and landing point. And update the coverage status of the demand points. The calculated demand coverage rate and regional ground traffic congestion pressure index Feedback is provided to the outer layer location decision master algorithm.

9. The method according to claim 8, characterized in that, In step S4, the two-layer iterative optimization algorithm specifically includes the following steps: Step S4.1: The outer-layer addressing decision master algorithm randomly initializes the population and generates individuals containing binary addressing codes; Step S4.2: Input each individual in the outer population into the inner flow allocation sub-algorithm. The inner flow allocation sub-algorithm performs greedy matching based on the nearest distance principle to determine the service assignment of each virtual logistics demand point and update the decision variables. and predicted terminal throughput at take-off and landing points ; Step S4.3: Combine the POI density data around the candidate sites to calculate the demand coverage corresponding to each site selection scheme. Regional ground traffic congestion pressure index Construct the bi-objective fitness vector corresponding to each location scheme. ; Step S4.4, the outer-layer location decision master algorithm determines the bi-objective fitness based on the calculated fitness ( Perform non-dominated sorting and crowding calculation, and generate offspring populations through selection, crossover and mutation operators; Step S4.5: Repeat steps S4.2 to S4.4 until the preset maximum number of iterations or Pareto preface convergence index is met, and output the Pareto optimal address solution set.

10. The method according to claim 9, characterized in that, Step S5 includes: By adjusting the mixed probability distribution parameters and random seeds, M sets of test virtual datasets with different spatial distribution patterns and peak demand intensities are generated in batches. The Pareto optimal site selection solution set output in step S4 is then substituted into the M sets of test virtual datasets for simulation. The variance of demand coverage and the maximum value of ground traffic congestion pressure of each site selection scheme under multiple scenario fluctuations are statistically analyzed. The site selection solution with the smallest comprehensive standard deviation and the best mean is selected as the final UAV take-off and landing site selection planning scheme.

Images