Unmanned aerial vehicle transportation route distribution method for logistics multi-point transportation

Abstract

The invention relates to the field of intelligent logistics optimization, and discloses an unmanned aerial vehicle transportation route distribution method for logistics multi-point transportation, which comprises the following steps: carrying out coordinate-weight joint standardization on a distribution point set to generate a three-dimensional feature vector; constructing a topological complex based on the standardized data, and extracting a key ring structure through continuous coherence; encoding the ring structure into a topological constraint term of a quantum model, and constructing Hamiltonian containing distance, load and topological constraint; dividing quantum sub-blocks according to the topological ring, and executing block annealing solution through chain coupling constraint; and carrying out topology-guided privacy fine tuning under a federated learning framework by using gradient information of a quantum solution. According to the method, a complex path optimization problem is decomposed into sub-problems capable of being processed in parallel through a quantum topological coding and block annealing strategy, quantum bit grouping is guided through a topological ring structure, the calculation complexity is reduced, efficient solving of a large-scale logistics network is achieved, and the calculation speed is increased compared with a traditional optimization method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent logistics optimization, in particular to a UAV transportation route allocation method for logistics multi-point transportation. BACKGROUND

[0002] The existing UAV logistics path optimization method generally adopts a classical optimization algorithm or a quantum heuristic strategy, but it faces significant bottlenecks when dealing with multi-constrained dynamic scenarios. Although the traditional quantum encoding scheme can improve the calculation speed, it ignores the characteristics of the logistics network topology, resulting in low utilization of quantum bits and difficulty in effectively integrating multi-dimensional constraints such as load and distance. At the same time, the existing privacy protection mechanism uses homogenized noise injection in distributed optimization, which fails to identify the sensitivity differences of key path structures, causing an imbalance between privacy strength and optimization effectiveness.

[0003] More prominently, the lack of dynamic environment adaptability has become a key problem restricting technological development. Existing methods often rely on global recalculation mechanisms when dealing with new distribution points or sudden changes in road conditions, leading to waste of computing resources and delayed responses. In addition, traditional time-space conflict detection uses static safety thresholds, which cannot adapt to the real-time needs of multi-robot collaborative transportation, exacerbating the risk of path conflicts. These defects collectively make it difficult for existing technologies to meet the comprehensive requirements of modern logistics for efficiency, privacy, and robustness. SUMMARY

[0004] To address the shortcomings of existing technologies, the present application provides a UAV transportation route allocation method for logistics multi-point transportation, which solves the problems of existing UAV logistics path planning methods in multi-constrained collaborative optimization, privacy protection, and dynamic adaptability.

[0005] To achieve the above purpose, the present application realizes the following technical solutions: a UAV transportation route allocation method for logistics multi-point transportation, comprising the following steps:

[0006] S1, coordinate-weight joint standardization is performed on the distribution point set to generate a three-dimensional feature vector;

[0007] S2, a topological complex is constructed based on the standardized data, and a key ring structure is extracted through persistent homology;

[0008] S3, the ring structure is encoded as a topological constraint term of the quantum model, and a Hamiltonian containing distance, load, and topological constraints is constructed;

[0009] S4, quantum sub-blocks are divided according to the topological ring, and block annealing solving is performed through chain coupling constraints;

[0010] S5, gradient information of the quantum solution is used to perform topologically guided privacy fine-tuning under a federated learning framework;

[0011] S6, verifying triple constraint satisfaction, outputting authentication path set.

[0012] Preferably, the coordinate-weight joint normalization in step S1 comprises:

[0013] Maximizing the weight of goods for each distribution point;

[0014] Combining the normalized weight and the original coordinates into a three-dimensional feature vector.

[0015] Preferably, the step 2 of constructing a topological complex comprises the following steps:

[0016] Calculating the average distance between distribution points, the calculation rule is:

[0017]

[0018] Where N is the total number of distribution points; p i is the coordinate of the ith distribution point; ||p i -p j || is the Euclidean distance; j is the index variable of inner summation;

[0019] Constructing a Vietoris-Rips complex:

[0020]

[0021] Where, is the set of distribution points; σ is the vertex set of a simplex; ∈ is the average distance between distribution points; p u ,p v is any two vertices in the complex.

[0022] Preferably, the step 2 of extracting key ring structure comprises:

[0023] Calculating the persistence interval set of one-dimensional persistent homology group, the calculation rule is:

[0024] PH1={(b m ,d m )∣m=1,...,K};

[0025] Where b m is the birth time of the mth ring structure; d m is the death time;

[0026] Screening topological rings that meet the persistence condition, the screening rule is:

[0027]

[0028] Where γ mThe mth persistent topological ring structure is composed of 1-dimensional holes in the simplicial complex;d m The death time of the mth ring structure, corresponding to the filter parameter value when the ring disappears;b m The birth time of the mth ring structure, corresponding to the filter parameter value when the ring appears;τ topo The persistent survival time threshold for screening significant topological features.

[0029] Preferably, the Hamiltonian in step 3 includes a distance constraint term, a load constraint term, and a topological constraint term.

[0030] Preferably, the block annealing solution in step 4 includes:

[0031] Convert the Hamiltonian to QUBO form:

[0032]

[0033] Where q i is a quantum bit; Q is the total number of quantum bits; h i is the bias coefficient of the ith quantum bit, converted from the linear term of the Hamiltonian; J ij is the coupling coefficient of the quantum bit pair, converted from the quadratic term of the Hamiltonian;

[0034] Divide the quantum sub-blocks according to the topological ring:

[0035]

[0036] Where γ m is the extracted topological ring structure; q k is the set of quantum bits associated with the distribution point p k ; The mth quantum sub-block corresponds to the topological ring γ m ;

[0037] Apply chain coupling constraints:

[0038]

[0039] Where A is the chain strength coefficient; is the total number of key topological rings; (q i ,q j ) is a pair of quantum bits within the same sub-block.

[0040] Preferably, the federated learning framework in step 5 includes regional model parameter encryption, global model aggregation, and topological constraint synchronization.

[0041] Preferably, the topologically guided privacy fine-tuning in step 5 includes:

[0042] Computing topological sensitivity:

[0043]

[0044] where |γ m | is the number of distribution points contained in the mth topological ring; d max is the maximum distribution point spacing within the ring; d min is the minimum distribution point spacing within the ring;

[0045] Adaptive noise injection:

[0046]

[0047] where θ is the original model parameter vector; ∈ is a preset privacy budget parameter; is a standard Gaussian distributed noise; is the total number of extracted key topological rings;

[0048] Privacy constraint verification:

[0049]

[0050] where Δf is the maximum parameter change; is the integrated global topological ring set.

[0051] Preferably, the step 6 of verifying the satisfaction of the triple constraints includes load constraint verification, restricted area constraint verification, and topological ring constraint verification.

[0052] A UAV transportation route allocation system for logistics multi-point transportation, comprising:

[0053] An input module for receiving distribution points, UAVs, and environmental parameters;

[0054] A topological analysis module for extracting the topological ring structure of the distribution network through persistent homology computation;

[0055] A quantum modeling module for encoding the logistics path planning problem into a quantum solvable form;

[0056] A solving module for performing block quantum annealing optimization;

[0057] A federated learning module for realizing multi-region collaborative optimization and privacy protection;

[0058] A verification output module for checking constraints and generating path certificates.

[0059] The present application provides a UAV transportation route allocation method for logistics multi-point transportation. It has the following advantages:

[0060] 1、The application decomposes the complex path optimization problem into sub-problems that can be processed in parallel by quantum topological encoding and block annealing strategy, and then uses the topological ring structure to guide the grouping of quantum bits, thereby reducing the computational complexity, achieving efficient solution of large-scale logistics network, and improving the operation speed compared with traditional optimization methods.

[0061] 2、The composite Hamiltonian designed in the application integrates distance, load and topological constraints, ensures that key constraints are satisfied in priority through a weight layering mechanism, and simultaneously optimizes multi-dimensional objectives in the quantum annealing process, thereby balancing transportation cost and capacity limit and solving the multi-objective conflict problem.

[0062] 3、The application adopts a federated learning framework and homomorphic encryption technology to realize secure parameter exchange between distributed nodes, and a topologically guided noise injection mechanism that protects sensitive information while maintaining the optimization effect of key path features, thereby meeting the strict requirements of modern logistics for data privacy.

[0063] 4、The application adopts an incremental re-planning system combined with time-space conflict detection to achieve rapid response under environmental changes, and through a topological inheritance strategy and local quantum parameter update, path adjustment is completed under the premise of maintaining the stability of the core network structure, thereby enhancing the robustness of the system in responding to emergencies.

[0064] 5、The application reduces hardware resource occupation through block quantum computing, improves annealing process convergence efficiency through dynamic temperature scheduling, and avoids UAV conflicts through space-time cube constraints to reduce airspace resource competition loss, thereby improving overall logistics network resource utilization. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 is a method step diagram of the application;

[0066] Figure 2 is a system module diagram of the application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the specification of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0068] As shown in Figure 1 , a UAV transportation route allocation method for multi-point transportation of logistics can include the following steps:

[0069] S1, coordinate-weight joint standardization is performed on the distribution point set to generate a three-dimensional feature vector;

[0070] S2, constructing a topological complex based on the standardized data, and extracting a key ring structure through persistent homology;

[0071] S3, encoding the ring structure as a topological constraint term of a quantum model, and constructing a Hamiltonian containing distance, load, and topological constraints;

[0072] S4, dividing quantum sub-blocks according to the topological ring, and performing block annealing solving through chain coupling constraints;

[0073] S5, using gradient information of quantum solutions to perform topologically guided privacy fine-tuning under a federated learning framework;

[0074] S6, verifying triple constraint satisfaction, and outputting an authentication path set

[0075] The following is a detailed description of each step in the method of the present application, which fully describes the specific implementation principles, technical details and processes of each step.

[0076] For step S1, in this embodiment, the coordinate-weight joint standardization step converts heterogeneous logistics data into a unified representation form through multi-dimensional feature fusion technology. Specifically, first, the original distribution point data set is obtained, each distribution point containing two types of heterogeneous parameters: geographic coordinates and cargo weight. The geographic coordinates are represented in the latitude-longitude coordinate system, and the cargo weight is the actual measured value, with significant differences in dimension and numerical range.

[0077] For the preprocessing of the cargo weight parameter, the maximum value normalization method is used to eliminate the dimensional effect. Specifically, the maximum value of all the cargo weight values of the distribution points is identified as the normalization reference. The normalized weight of each distribution point is calculated as follows: the original weight value is divided by the maximum weight value, so that the normalized weight parameter is compressed to the range [0, 1]. This process can be represented by the following mathematical formula:

[0078]

[0079] where w i is the original cargo weight of the i-th distribution point; is the maximum weight value among all distribution points; is the normalized weight parameter after normalization.

[0080] After completing the weight normalization, the processed weight parameter is combined with the original geographic coordinates. The three-dimensional feature vector of each distribution point is constructed as follows: the original latitude and longitude coordinates are retained as the first two dimensions, and the normalized weight is used as the third dimension. The three-dimensional feature vector generated in this way can be represented as:

[0081]

[0082] wherein x i , y i are the longitude and latitude coordinates of the i-th delivery point, respectively; is the corresponding normalized weight value.

[0083] This feature construction method realizes the organic integration of spatial position information and cargo attribute information, providing a unified data basis for subsequent topological analysis.

[0084] In the specific implementation process, the geographic coordinate system preferably adopts the WGS84 standard coordinate system to ensure the global universality of spatial position data. The normalization processing of the weight parameter is completed in the data preprocessing stage, and the calculation result is stored in the form of floating-point numbers. The generated three-dimensional feature vector is arranged in the order of delivery point number to form a feature matrix:

[0085]

[0086] wherein N is the total number of delivery points, and each row of the matrix corresponds to the standardized feature of a delivery point.

[0087] Preferably, when there is a special cargo type, a weight correction coefficient can be introduced to adjust the normalized weight. The specific formula adjustment is:

[0088]

[0089] wherein η is the correction coefficient; is the normalized weight before adjustment; is the normalized weight after adjustment.

[0090] This correction coefficient can be dynamically set according to the cargo danger level or transportation priority, for example, η = 0.8 for fragile goods and η = 1.0 for ordinary goods. This extended design enhances the adaptability of the method to complex logistics scenarios.

[0091] Through the coordinate-weight joint normalization processing of this step, the following technical effects are achieved: First, the scale difference between different dimension parameters is eliminated, avoiding the deviation caused by different parameter magnitudes in the subsequent calculation process; second, the three-dimensional feature vector retains the spatial position relationship and cargo attribute characteristics at the same time, providing multi-dimensional data support for topological structure analysis; finally, the standardization process has linear computational complexity, ensuring the processing efficiency of large-scale logistics networks.

[0092] For step S2, in this embodiment, the topology complex construction and key ring extraction step extracts persistent topological features from the standardized distribution point data by a computational topology method. In actual implementation, first, based on the three-dimensional feature vector generated in step S1, the spatial relationship between the distribution points is calculated, and then the topology complex structure reflecting the connection characteristics of the logistics network is constructed. In the average distance calculation stage, the spatial relationship between the distribution points is measured by using the Euclidean distance. The specific calculation formula is:

[0093]

[0094] wherein N is the total number of distribution points; p i is the coordinates of the i-th distribution point; ||p i -p j || is the Euclidean distance; j is the index variable of the inner summation.

[0095] The average distance parameter ∈ represents the overall connection density of the logistics network, which serves as the scale reference for subsequent complex construction.

[0096] Based on the calculated average distance, the Vietoris-Rips complex is constructed to capture the topological association between the distribution points. The complex construction rule is defined as:

[0097]

[0098] wherein is the set of distribution points; σ is the vertex set of the simplex; ∈ is the average distance between the distribution points; p u ,p v are any two vertices in the complex.

[0099] The subset of distribution points constitutes a simplex in the complex only if the distance between any two points in the subset does not exceed ∈. This construction method ensures that the topological structure reflects the reachability relationship in the actual logistics network.

[0100] In the persistent homology calculation stage, the persistence of topological features is dynamically analyzed by filtering parameter sequences. In actual implementation, an increasing filtering parameter sequence is generated to gradually expand the complex structure. For each filtering parameter, the corresponding one-dimensional persistent homology group is calculated, and the birth time and death time of the ring structure are recorded to form a persistent interval set:

[0101] PH1={(b m ,d m )∣m=1,...,K};

[0102] wherein b m is the birth time of the m-th ring structure; d m is the death time.

[0103] The screening of the key ring structure is based on the survival time threshold of the ring, and the specific screening condition is:

[0104]

[0105] wherein γ m is the mth persistent topological ring structure, which is composed of 1-dimensional holes in the simplicial complex; d m is the death time of the mth ring structure, which corresponds to the filter parameter value when the ring disappears; b m is the birth time of the mth ring structure, which corresponds to the filter parameter value when the ring appears; τ topo is a survival time threshold, which is used to screen significant topological features.

[0106] The threshold parameter τ topo is preferably calculated as follows:

[0107]

[0108] wherein N is the total number of distribution points, which is a positive integer; is the three-dimensional feature vector of the ith distribution point; is the three-dimensional feature vector of the jth distribution point; is the Euclidean distance between the distribution points i and j.

[0109] The threshold setting principle ensures that only topological rings with a survival time significantly longer than the average connection distance of the network are retained, filtering out transient noise structures. The key ring set obtained by screening is stored as a nested list structure of distribution point indices, and each ring γ m is represented as an ordered set of a group of distribution point numbers.

[0110] Preferably, a weight correction factor is introduced during the construction of the complex, and the distance determination condition is adjusted to:

[0111] ||p u -p v ||2≤β·∈;

[0112] wherein ||p u -p v ||2 is the original Euclidean distance (not standardized) between the distribution points u and v; β is the weight correction factor; and ∈ is the average distance parameter.

[0113] This factor can be dynamically adjusted according to the actual complexity of the road network, with a smaller value in complex areas to increase the sparsity of the complex, and a larger value in simple areas to enhance connectivity. This extended design improves the adaptability of the method to heterogeneous logistics environments.

[0114] Through the topology analysis of this step, the following technical effects are achieved: first, the Vietoris-Rips complex construction maps the discrete distribution points to a continuous topological space, effectively capturing the connected mode of potential transportation paths; second, the persistent homology calculation identifies ring structures with significant persistence through multi-scale analysis, avoiding local noise interference; finally, the key ring set provides a topological feature basis for subsequent quantum constraint coding, ensuring that the path planning conforms to the overall structure characteristics of the network.

[0115] For step S3, in this embodiment, the quantum Hamiltonian construction step maps the logistics path optimization problem into a quantum solvable model through multi-constraint fusion coding technology. In specific implementation, based on the key topological ring structure extracted in step S2, a composite Hamiltonian is constructed by integrating distance, load and topological constraints to provide an optimization objective function for quantum annealing.

[0116] In the distance constraint term construction stage, the set of connectable edges is defined as:

[0117]

[0118] Where ∈ is the average distance parameter calculated in step S2; is the three-dimensional normalized feature vector of the ith distribution point; is the three-dimensional normalized feature vector of the jth distribution point; ||.||2 is the Euclidean norm operator, which calculates the geometric distance in three-dimensional space.

[0119] The mathematical expression of the distance constraint term is:

[0120]

[0121] Where s ij is the edge selection quantum bit; is the three-dimensional normalized feature vector of the ith distribution point; is the three-dimensional normalized feature vector of the jth distribution point; ||.||2 is the Euclidean norm operator.

[0122] When the edge (i, j) is selected, it takes the value of 1, otherwise 0. This constraint term is weighted by the squared distance to encourage the quantum annealing process to prefer short-range paths.

[0123] The load constraint term construction considers the unmanned aerial vehicle carrying capacity limitation, and defines the distribution variable x ij to represent whether the ith distribution point is served by the jth unmanned aerial vehicle. The load constraint term expression is:

[0124]

[0125] Where, is the normalized weight parameter; c jis the rated carrying capacity of the jth UAV; λ1is the constraint weight coefficient; M is the total number of UAVs; and N is the total number of distribution points.

[0126] The square term design ensures that a penalty energy growth occurs when the actual carrying capacity exceeds the limit.

[0127] The topological constraint term encodes the key ring structure extracted in step S2, with each ring corresponding to a constraint term:

[0128]

[0129] wherein, is the Pauli X operator acting on the kth qubit; is the total number of key rings; λ2is the topological constraint weight; is 1 when all qubits in the ring flip simultaneously; γ m is the mth key topological ring, stored as a set of distribution point indexes.

[0130] The design ensures that the path planning does not incompletely contain any topological ring structure through the qubit product term.

[0131] Preferably, the weight coefficient setting follows the principle of λ1> λ2> 1, ensuring that the carrying capacity constraint is satisfied before the topological constraint. In specific implementation, λ1may be taken as 10 3 orders of magnitude, and λ2is taken as 10 2 orders of magnitude, with the best value determined through experimental tuning.

[0132] The final form of the composite Hamiltonian is a linear combination of three constraint terms:

[0133]

[0134] wherein, is the distance constraint term; is the carrying capacity constraint term; is the topological constraint term.

[0135] The expression converts the logistics path optimization problem into a quantum system problem of finding the ground state energy, laying a foundation for subsequent quantum annealing solution.

[0136] Through the constraint encoding of this step, the following technical effects are achieved: first, the distance constraint term guides the algorithm to preferentially select an economic path; second, the carrying capacity constraint term ensures the transport capacity limit through a quadratic penalty function; and finally, the topological constraint term encodes the key ring structure as quantum interaction, avoiding the path from falling into a local optimum. The synergistic effect of the three ensures that the final solution simultaneously satisfies the multi-dimensional constraint conditions.

[0137] For step S4, in this embodiment, the block quantum annealing solving step decomposes the complex path optimization problem into subproblems that can be processed in parallel through a topology-guided quantum bit partitioning strategy. In actual implementation, first, the composite Hamiltonian constructed in step S3 is converted into a QUBO form suitable for execution by a quantum processor, and then quantum bits are grouped based on the key topological ring structure extracted in step S2, and finally efficient solving is achieved through chain coupling constraints.

[0138] In the QUBO conversion stage, the Hamiltonian is reconstructed in the form of a quadratic unconstrained binary optimization. The converted expression is:

[0139]

[0140] where q i is a quantum bit; Q is the total number of quantum bits; h i is the bias coefficient of the i-th quantum bit, converted from the linear term of the Hamiltonian; J ij is the coupling coefficient of the quantum bit pair, converted from the quadratic term of the Hamiltonian.

[0141] This conversion process realizes the expansion of the constraint term into a quadratic polynomial, ensuring that the quantum annealing machine can directly process the optimization problem.

[0142] The total number of quantum bits is determined by the edge selection variables and the distribution variables, and the specific calculation formula is:

[0143] Q = |ε| + M x N;

[0144] where |ε| is the total number of connectable edges; M x N is the total number of UAV-distribution point distribution variables.

[0145] This design ensures that all decision variables are encoded as quantum bits.

[0146] The topology ring-based quantum bit grouping strategy divides the quantum system into multiple subblocks, each corresponding to a key ring structure. The subblock division rule is defined as:

[0147]

[0148] where γ m is the extracted topological ring structure; q k is the set of quantum bits associated with the distribution point p k ; is the m-th quantum subblock, corresponding to the topological ring γ m .

[0149] This division method maintains the local coupling of topologically associated quantum bits and reduces the computational overhead caused by cross-subblock interaction.

[0150] The introduction of chain coupling constraints ensures the inter-subblock collaborative optimization, and the specific expression is:

[0151]

[0152] wherein A is a chain strength coefficient; is the total number of key topological rings;(q i ,q j is a pair of qubits in the same subblock.

[0153] This setting principle ensures that the strength of the chain constraint matches the order of magnitude of the original coupling coefficient, avoiding the decline in solution quality caused by too strong or too weak constraints.

[0154] In the annealing execution phase, a staged annealing strategy is adopted: first, local annealing is independently performed on each subblock to obtain a preliminary solution; then, global annealing optimization is performed to adjust the coupling relationship between subblocks. Preferably, the annealing time parameter adopts an adaptive adjustment mechanism, and a longer annealing time (such as 200 μs) is set in the initial stage to ensure sufficient exploration of the solution space, and the time is gradually shortened in the later stage to improve the convergence speed.

[0155] Through the block-based solution of this step, the following technical effects are achieved: first, the topological-guided qubit division reduces the problem dimension and alleviates the qubit resource limitation; second, the chain coupling maintains the relevance between subblocks, avoiding the loss of solution quality caused by segmentation; finally, the staged annealing strategy balances global exploration and local development, improving the solving efficiency.

[0156] For step S5, in this embodiment, the federated learning privacy fine-tuning step realizes the global optimization of the path scheme under the premise of ensuring data privacy through a distributed collaborative optimization mechanism. In specific implementation, based on the quantum solution obtained in step S4, a federated learning framework is established between multiple regional nodes, and model fine-tuning is completed through encrypted parameter transmission and topologically guided noise injection for privacy protection.

[0157] In the regional model updating phase, each participating node calculates gradient information based on local data and performs mask processing on the gradient. The gradient mask process adopts a random projection technique to map high-dimensional gradient vectors to a low-dimensional space:

[0158]

[0159] wherein Φ k is the random projection matrix of region k; is the loss function gradient; θ (t) is the t-th round of model parameters.

[0160] This projection operation realizes data anonymization while reducing communication overhead.

[0161] The encryption transmission stage protects the masked gradient using a homomorphic encryption algorithm:

[0162]

[0163] where HE.Enc(.) is a homomorphic encryption function; is a Gaussian noise term.

[0164] After decrypting the received regional gradient, the global aggregation server performs a weighted average:

[0165]

[0166] where K is the total number of participating nodes; is the aggregated gradient, reflecting the global optimization direction while eliminating region-specific bias; is the gradient encrypted and transmitted by the kth node.

[0167] In the topology-guided noise injection stage, the sensitivity weight is calculated based on the key ring structure extracted in step S2. For each topology ring, its sensitivity weight is defined as:

[0168]

[0169] where diam(γ m ) is the diameter of the ring; |γ m | is the number of distribution points contained in the ring.

[0170] This weight reflects the sensitivity of the ring structure to gradient changes, and small-diameter dense rings have higher sensitivity.

[0171] The final noise-injected gradient update formula is:

[0172]

[0173] where α is the learning rate; ξ m is an independent noise vector; ∈ is the global privacy parameter; θ (t) is the model parameter vector in the tth round; is the global aggregated gradient; ω m is the sensitivity weight of the mth topology ring; is the total number of key topology rings.

[0174] This design ensures that weaker noise is applied on sensitive ring structure-related parameters, while stronger noise is applied in non-sensitive areas, achieving a balance between privacy and utility.

[0175] Preferably, a dynamic privacy budget allocation strategy is adopted, and the privacy parameter is exponentially decayed according to the training round t:

[0176] ∈ t = ∈max ·e -λt ;

[0177] where λ is the decay coefficient; ∈ max is the initial maximum privacy budget; t is the current training round, starting from 0 and increasing; e is the base of the natural logarithm.

[0178] This strategy allows larger gradient update amplitude at the beginning of training, and gradually tightens noise injection to improve privacy protection at the later stage.

[0179] Through the implementation of this step, the following technical effects are achieved: first, random projection and homomorphic encryption double protection ensure regional data privacy; second, noise injection guided by topological sensitivity weight maintains the optimization effect of the key path structure; finally, the dynamic privacy budget mechanism balances the optimization needs and privacy requirements at different training stages.

[0180] For step S6, in this embodiment, the dynamic path re-planning step realizes continuous optimization of the logistics network through a real-time feedback mechanism and incremental quantum optimization. In specific implementation, based on the path scheme fine-tuned in step S5, an environmental change monitoring system is established, and when a change in the distribution point or road conditions is detected, an incremental quantum annealing process is triggered to make local path adjustments while maintaining the stability of the core topological structure.

[0181] In the environmental change detection stage, define the dynamic event trigger condition:

[0182]

[0183] where and are the new and old sets of distribution points, respectively; is the symmetric difference set operator; τ change is the change threshold.

[0184] When the number of added or removed distribution points exceeds the threshold, it is determined that re-planning needs to be performed.

[0185] In the incremental Hamiltonian update stage, the states of the quantum bits in the original path scheme that are not affected are retained, and only the quantum bits related to the changed area are re-encoded. The updated Hamiltonian expression is:

[0186]

[0187] where is the stable term not affected by the dynamic change; is the set of affected quantum bit indices; is the incremental update amount of the corresponding term.

[0188] This design reduces the computational complexity through local adjustment.

[0189] In terms of topology inheritance, the stability of the key ring set extracted in the maintenance step S2 is maintained. When a new delivery point falls into the neighborhood of an existing topology ring, a ring expansion strategy is adopted:

[0190] When the delivery point falls into the neighborhood of an existing topology ring, a ring expansion strategy is adopted:

[0191]

[0192] where γ' is the mth topology ring after expansion; γ is the original topology ring; p is the coordinate of the new delivery point; and d is the minimum distance between the new point and all points in the ring. m m new

[0193] This strategy ensures that the new point is integrated into the existing topology structure, avoiding the computational overhead caused by complete recalculation.

[0194] Preferably, a quantum annealing temperature scheduling strategy is adopted, and the initial temperature is set to:

[0195]

[0196] where I is the set of quantum bit indices affected by dynamic changes; N is the total number of affected quantum bits; and ΔHk is the Hamiltonian increment corresponding to the kth quantum bit.

[0197] This temperature parameter is positively related to the local Hamiltonian change amplitude, ensuring that the annealing process fully explores possible new path combinations.

[0198] The path verification stage introduces a spatiotemporal conflict detection mechanism, defining spatiotemporal cube constraints:

[0199]

[0200] where t i , t j are the timestamps of the arrival of unmanned aerial vehicles i and j at their respective target points; v max is the maximum flight speed; δ safe is the safety time interval; d i , d j are different unmanned aerial vehicle identifiers; U is the set of unmanned aerial vehicles; p i , p j are the target point coordinates of unmanned aerial vehicles i and j; ||.|| is the absolute value operator; and ||.||2 is the Euclidean distance operator.

[0201] ​​​​​​​​The constraint ensures that the paths of different UAVs are conflict-free in the time-space dimension.

[0202] Through the implementation of the step, the following technical effects are achieved: first, the incremental optimization greatly reduces the consumption of quantum computing resources; second, the topology ring expansion strategy maintains the continuity of the network structure; and finally, the space-time conflict detection improves the safety of multi-UAV collaborative transportation. The dynamic re-planning mechanism enables the system to respond in real time to environmental changes, ensuring the continuous and efficient operation of the logistics network.

[0203] In summary, the present application achieves efficient path planning through multi-dimensional data fusion and quantum topology optimization. The specific implementation includes the following innovative systems: a three-dimensional feature tensor is constructed to integrate space-time logistics data, and standardized processing is used to eliminate dimensional differences; key ring structures are extracted through topological persistent homology analysis to establish a logistics network connectivity model; a composite quantum Hamiltonian is designed to encode distance, load, and topology constraints, converting path optimization into a quantum solvable problem; a block quantum annealing strategy is implemented to divide quantum bits based on topology rings and introduce chain coupling constraints; a federated learning framework is used to achieve distributed optimization with privacy protection, and dynamic noise injection is used to balance privacy and utility; an incremental re-planning mechanism is established to respond to dynamic changes using space-time conflict detection and topology inheritance strategies; the technical effects of the present application include transportation cost optimization, multi-constraint collaborative satisfaction, and privacy security enhancement, providing an intelligent solution for complex logistics scenarios.

[0204] The logistics multi-point transportation-oriented UAV transportation route allocation system described below can be correspondingly referred to the logistics multi-point transportation-oriented UAV transportation route allocation method described above.

[0205] Please refer to the attached Figure 2 The present application also provides a logistics multi-point transportation-oriented UAV transportation route allocation system, comprising:

[0206] An input module for receiving distribution points, UAVs, and environmental parameters;

[0207] A topology analysis module for extracting the topology ring structure of the distribution network through persistent homology calculation;

[0208] A quantum modeling module for encoding the logistics path planning problem into a quantum solvable form;

[0209] A solving module for performing block quantum annealing optimization;

[0210] A federated learning module for achieving multi-region collaborative optimization and privacy protection;

[0211] A verification output module for checking constraints and generating path certificates.

[0212] The system of the embodiment can be used to execute the method embodiments described above, which have similar principles and technical effects, and will not be described here again.

[0213] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for allocating unmanned aerial vehicle (UAV) transport routes for multi-point logistics transportation, characterized in that, Includes the following steps: S1. Perform coordinate-weight joint standardization on the delivery point set to generate a three-dimensional feature vector; S2. Construct a topological complex based on standardized data and extract the key loop structure through continuous homology; S3. Encode the ring structure into a topological constraint term of a quantum model, and construct a Hamiltonian containing distance, load and topological constraints; S4. Divide the quantum blocks according to the topological ring and perform block annealing to solve the problem through chain coupling constraints; S5. Utilize the gradient information of quantum solutions to perform privacy-preserving fine-tuning guided by topology within a federated learning framework; S6. Verify the satisfaction of the triple constraint and output the authentication path set.

2. The method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The coordinate-weight joint standardization in step S1 includes: Normalize the maximum value of the goods weight at each delivery point; The normalized weight is combined with the original coordinates to form a three-dimensional feature vector.

3. The method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, Step 2, which involves constructing the topological complex, includes the following steps: The average distance between delivery points is calculated according to the following rules: Where N is the total number of delivery points; p i Let p be the coordinates of the i-th delivery point; i -p j || represents the Euclidean distance; j is the index variable for the inner summation; Constructing the Vietoris-Rips complex: in, Let be the set of delivery points; σ be the set of vertices of the simplex; ∈ be the average distance between delivery points; p u ,p v Let be any two vertices in the complex.

4. The method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The extraction of the critical loop structure in step 2 includes: The set of persistent intervals of a one-dimensional persistent homology group is calculated using the following rules: PH1={(b m ,d m )∣m=1,...,K}; Among them, b m d represents the birth time of the m-th ring structure. m Time of death; To filter topological cycles that satisfy the persistence condition, the filtering rules are as follows: Where, γ m The m-th persistent topological ring structure is composed of 1-dimensional holes in a simple complex; d m b represents the death time of the m-th ring structure, corresponding to the filter parameter value when the ring disappears. m τ represents the birth time of the m-th ring structure, corresponding to the filter parameter value when the ring appears. topo This is a duration threshold used to filter significant topological features.

5. The method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The Hamiltonian in step 3 includes distance constraints, load constraints, and topology constraints.

6. A method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The block annealing solution in step 4 includes: Convert the Hamiltonian to QUBO form: Where, q i For qubits; Q is the total number of qubits; h i J is the bias coefficient of the i-th qubit, obtained by transforming the Hamiltonian linear term; ij The coupling coefficient of the qubit pair is obtained by transformation from the quadratic term of the Hamiltonian; Quantum blocks are divided according to topological rings: Where, γ m The extracted topological ring structure; q k To cooperate with the delivery point p k A set of associated qubits; For the m-th quantum block, the corresponding topological ring γ m ; Apply chain coupling constraints: Where A is the chain strength coefficient; The total number of critical topological rings; (q) i ,q j () represents a pair of qubits within the same sub-block.

7. A method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The federated learning framework in step 5 includes regional model parameter encryption, global model aggregation, and topology constraint synchronization.

8. A method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, The privacy fine-tuning guided by the topology in step 5 includes: Calculate topology sensitivity: Where, |γ m | represents the number of delivery points contained in the m-th topological ring; d max d represents the maximum distance between delivery points within the ring road. min This represents the minimum distance between delivery points within the ring road. Adaptive noise injection: Where θ is the original model parameter vector; ∈ is the preset privacy budget parameter; Standard Gaussian noise; The total number of key topological rings extracted; Privacy constraint verification: Where Δf is the maximum change in the parameter; For the integrated global topological ring set.

9. A method for allocating unmanned aerial vehicle (UAV) transportation routes for multi-point logistics transportation according to claim 1, characterized in that, Step 6 verifies the satisfaction of the triple constraints, including load constraint verification, no-fly zone constraint verification, and topology loop constraint verification.

10. A drone transportation route allocation system for multi-point logistics transportation, used to execute a drone transportation route allocation method for multi-point logistics transportation as described in any one of claims 1-9, characterized in that, include: The input module is used to receive parameters from delivery points, drones, and the environment. The topology analysis module extracts the topological ring structure of the delivery network through continuous homology calculation; The quantum modeling module is used to encode the logistics route planning problem into a quantum-solvable form; The solver module is used to perform block quantum annealing optimization; The federated learning module is used to achieve multi-regional collaborative optimization and privacy protection. The verification output module is used to check constraints and generate path certificates.

Images