Logistics network global optimization method based on quantum annealing algorithm

Abstract

The invention belongs to the technical field of intelligent logistics, and discloses a logistics network global optimization method based on a quantum annealing algorithm, and the method comprises the steps: S1, data collection: collecting multi-source logistics data in real time, and accessing environment data; s2, constructing a quantum model: performing quantum mapping of the logistics network, and constructing an energy function; s3, optimizing the logistics network: performing decomposition optimization on the logistics network by adopting a mixed quantum-classical optimization strategy; decomposing the logistics network into a backbone path optimized by using a quantum annealing optimization algorithm and a tail end distribution path optimized by using a classical algorithm; s4, dynamic adjustment and post-processing are carried out; and S5, performing user interaction and scheme output. The method has the advantages that the optimization efficiency is improved, the dynamic robustness is enhanced, the multi-target optimization precision is high, and the large-scale problem processing capability is good.

Description

Technical Field

[0001] This invention belongs to the field of intelligent logistics technology, specifically relating to a global optimization method for logistics networks based on the quantum annealing algorithm. Background Technology

[0002] Logistics scheduling is a crucial technical area within logistics technology, involving the warehousing, outbound transportation, and handling of raw materials, semi-finished products, and finished products. In existing technologies, when applying artificial intelligence to optimize large-scale logistics networks (such as vehicle routing with over 500 nodes), limitations inherent in techniques like genetic algorithms and simulated annealing can lead to local optima, resulting in low global search efficiency. Furthermore, the increased complexity of modern logistics, involving multi-objective optimization, makes Pareto solution set generation inefficient and unable to cover the entire optimization space within a reasonable timeframe. For NP-hard problems in logistics optimization (such as the large-scale Traveling Salesman Problem (TSP), the time complexity of classic algorithms increases exponentially, failing to meet real-time optimization requirements. Moreover, existing technologies exhibit poor adaptability to dynamic scenarios, unable to respond in real-time to unexpected situations such as traffic congestion and abnormal weather, leading to insufficient robustness of routing solutions. Therefore, existing technologies require multifaceted improvements to logistics network optimization methods. Summary of the Invention

[0003] The purpose of this invention is to provide a global optimization method for logistics networks based on the quantum annealing algorithm, so as to solve the technical problems that existing technologies are prone to entering local optima, resulting in low global search efficiency and low computational efficiency in multi-objective optimization and complex NP problems.

[0004] The aforementioned global optimization method for logistics networks based on quantum annealing algorithm includes the following steps: S1. Data Acquisition: Real-time acquisition of multi-source logistics data and integration with environmental data; S2. Constructing a quantum model: Performing quantum mapping of the logistics network and constructing the energy function; S3. Optimize the logistics network: Use a hybrid quantum-classical optimization strategy to decompose and optimize the logistics network; decompose the logistics network into backbone paths optimized by quantum annealing and last-mile delivery paths optimized by classical algorithms; S4. Dynamic adjustment and post-processing; S5. Conduct user interaction and output solutions.

[0005] Preferably, step S2 includes: 1) Quantum mapping of logistics networks: Abstracting logistics networks into a quantum Ising model. i The node state of a node is expressed in qubits. s i ∈{-1,1} represents; 2) Energy function construction: Integrating multi-objective optimization objectives and constraints, as detailed below: Cost items: ,in, E cost For cost items, c ij For nodes i arrive j Transportation costs; Time item: in, E time For time terms, t ij For transportation time, l 1 represents the time weight; Constraints: These include the following constraint functions: Flow conservation constraint: ,in, E flow For total flow, d i For node demand, Q Total demand is used as the flow conservation threshold. m 1 represents traffic weight; The total energy function is: H t = E cost + E time + E flow .

[0006] Preferably, the energy function construction also includes: Carbon emissions: ,in, E carbon For carbon emissions, e ij For carbon emissions, l 2 represents the carbon emission weight; Time window constraints: ,in, E window For the total time window, [ a i , b i ] is a node i Time window, t arrive ( i ) is a node i Arrival time, m 2 represents the weight of the time window; The total energy function is: Ht = E cost + E time + E carbon + E flow + E window .

[0007] The preferred quantum mapping algorithm for logistics networks is as follows: 1.1) Node-Bit Mapping: Each decision node corresponds to one qubit. Chimera graph embedding technology is used to adapt the quantum hardware topology. The mapping formula is as follows: ,in, v i Represents nodes in the logistics network i , ( x j , y j ) represents a node in the Chimera graph. j Two-dimensional grid coordinates; k Indicates sublattice index, k =0,1, indicating the node j Two child nodes within the same grid; 1.2) Edge weight mapping: Transportation costs, time, and carbon emissions are converted into coupling coefficients in the energy function, and reliability constraints are converted into transverse field parameters of the quantum system.

[0008] Preferably, step S3 includes: S3.1, Layered optimization: Decompose the logistics network into backbone paths optimized by quantum annealing and last-mile delivery paths optimized by classical algorithms; S3.2, Quantum optimization of backbone paths: The quantum annealing algorithm is used to solve the backbone paths; S3.3 Classic Optimization of Last-Mile Delivery Routes: An improved Clarke-Wright cost-saving algorithm is used to solve the last-mile delivery routes; In step S3.1, the mathematical model for dividing the backbone route and last-mile delivery is as follows: , Where, distance( i , j ) represents a node i arrive j Geographical distance; d i , d j They are nodes i and nodes jThe demand for goods; D thres Indicates the distance threshold; or Indicates the threshold for the proportion of demand; Q The total network demand is expressed by the following formula: ; The formula for calculating the last-mile delivery route is: That is, routes other than the backbone routes are classified as last-mile delivery routes.

[0009] Preferably, step S3 further includes: S3.4, based on historical data, this step can also apply incremental annealing optimization strategies to the backbone path; specifically including the following steps: S3.4.1. Perform historical solution caching. The formula for retaining the ground state of historical solutions used in this step is: ,in, Represents the historical best bit string. H old Represents the historical energy function; S3.4.2. Perform energy function update. The update model for the dynamically changing energy function applied in this step is as follows: The formula for the dynamically updated item is: , where Δ E Δ represents the set of edges affected. V Represents the set of nodes affected; Δ w ij Δ represents the change in edge weights along the path. b i Node bias; S3.4.3 Adjust the incremental annealing parameters. The formula for adjusting the incremental annealing parameters includes: 1) Annealing time adjustment: ,in, T full For global optimization time, |ΔE| and |ΔV| are the number of changed edges and nodes, respectively; 2) Lateral field strength restart strategy: , where Γ old Here, γ represents the final value of the historical horizontal field, and γ is the adjustment coefficient. This represents the average edge weight.

[0010] Preferably, in step S3.2, the quantum system is initialized first, then the annealing parameters are set, and then the specific flow of the quantum annealing optimization algorithm is as follows: S3.2.1 Input: Input the logistics network topology information into the quantum annealing optimization module; S3.2.2, Combining the total energy function, map it to the qubit system; S3.2.3 Set the annealing parameters T and Γ, and perform quantum adiabatic evolution, where T is the annealing time and Γ is the transverse field strength; S3.2.4 Measure the quantum state and obtain the optimal solution { }; S3.2.5 Perform path decoding and constraint verification; S3.2.6 Parameter Adjustment and Iteration; During training, repeat steps S3.2.3 to S3.2.6 until convergence.

[0011] Preferably, step S3.2.3 includes the following: Annealing time: ,in k For D-Wave hardware calibration coefficients, N The number of nodes; The formula for the transverse field intensity at time T is: , The initial transverse field strength is calculated as follows: α is the decay factor, which is optimized through reinforcement learning; Adiabatic evolution equation: Where s∈[0,1] represents the evolutionary progress. H is the initial Hamiltonian, and H is the target Hamiltonian.

[0012] Preferably, in step S3.2.6, the parameters are adjusted and iterated in the following manner: Annealing time: Where δ is the error coefficient. T old and T new To adjust the annealing time before and after; Traffic penalty coefficient: ,in, and To adjust the traffic penalty coefficient before and after; Time window penalty coefficient: Where maxDelay is the maximum time window delay. and This is to adjust the time window penalty coefficient before and after.

[0013] The technical advantages of this invention are as follows: 1. Improved efficiency: For vehicle routing problems with more than 1,000 nodes, the method provided by this invention improves the solution speed by 40%-60% compared to the classic simulated annealing algorithm, and increases the probability of obtaining the global optimal solution by 30%.

[0014] 2. Enhanced dynamic robustness: In case of sudden situations (such as 20% road congestion), this invention is based on the quantum annealing algorithm for calculation, and the corresponding replanning time is shortened to 1 / 3 of that of the classical algorithm, and the path reliability is improved by 25%.

[0015] 3. Multi-objective optimization accuracy: Pareto solution set coverage is improved by 50%, and when carbon emission optimization is included in the calculation, this scheme reduces emissions by 15%-20% compared to traditional methods.

[0016] 4. Large-scale problem handling capability: It can effectively handle logistics network optimization problems with more than 5,000 nodes, with time complexity controlled within O(N^2). 3 (N is the number of nodes), which is far lower than the exponential complexity of classic algorithms. Attached Figure Description

[0017] Figure 1 This is a basic flowchart of a global optimization method for logistics networks based on quantum annealing algorithm according to the present invention.

[0018] Figure 2 This is a diagram of the hybrid quantum-classical optimized architecture in this invention.

[0019] Figure 3 The flowchart of the quantum annealing optimization algorithm in this invention. Detailed Implementation

[0020] The following detailed description of the embodiments, with reference to the accompanying drawings, will further illustrate the specific implementation of the present invention, in order to help those skilled in the art to have a more complete, accurate, and in-depth understanding of the inventive concept and technical solution of the present invention.

[0021] like Figure 1-Figure 3 As shown, this invention provides a global optimization method for logistics networks based on the quantum annealing algorithm, comprising the following steps.

[0022] S1. Data Acquisition: Real-time acquisition of multi-source logistics data and integration with environmental data. This includes the following:

[0023] 1) Real-time acquisition of multi-source logistics data: including transportation node (warehouse, hub) capacity, real-time status of transportation network (congestion index, flight delay), and cargo demand data (weight, volume, time window).

[0024] 2) Access to environmental data: such as meteorological data, policies and regulations (such as traffic restriction policies), and energy prices (fuel and electricity costs).

[0025] S2. Constructing a quantum model: Performing quantum mapping of the logistics network and constructing the energy function. This includes the following:

[0026] 1) Quantum mapping of logistics networks: Abstracting logistics networks into a quantum Ising model. i The node state of a node is expressed in qubits. s i ∈{-1,1} means (1 is selected, -1 is not selected).

[0027] The specific quantum mapping algorithm for logistics networks is as follows: 1.1) Node-Bit Mapping: Each decision node (e.g., whether to choose a certain path) corresponds to a qubit. Large-scale networks use graph embedding techniques (such as Chimera graph embedding) to adapt the quantum hardware topology. Specifically, for large-scale logistics networks (N>1000 nodes), Chimera graph embedding technology is used to adapt the quantum hardware topology, and the mapping formula is: ,in, v i Represents nodes in the logistics network i ;( x j , y j ) represents a node in the Chimera graph. j Two-dimensional grid coordinates; k Indicates sublattice index, k =0,1, indicating the node j Two child nodes within the same grid.

[0028] 1.2) Edge weight mapping: Transportation costs, time, and carbon emissions are converted into coupling coefficients in the energy function, and reliability constraints are converted into transverse field parameters of the quantum system.

[0029] 2) Energy function construction: Integrating multi-objective optimization objectives and constraints, as detailed below: 2.1) Cost Items: ,in, E cost For cost items, c ij For nodes i arrive j Transportation costs.

[0030] 2.2) Time Item: in, E time For time terms, t ij For transportation time, l 1 represents the time weight.

[0031] 2.3) Carbon emissions: ,in, E carbon For carbon emissions, eij For carbon emissions, l 2 represents the carbon emission weight.

[0032] 2.4) Constraints: These include the following constraint functions.

[0033] 2.4.1) Flow conservation constraint: ,in, E flow For total flow, d i For node demand, Q Total demand is used as the flow conservation threshold. m 1 represents traffic weight.

[0034] 2.4.2) Time window constraint: ,in, E window For the total time window, [ a i , b i ] is a node i Time window, t arrive ( i ) is a node i Arrival time, m 2 represents the weight of the time window.

[0035] 2.5) The total energy function is: H t = E cost + E time + E carbon + E flow + E window That is, the sum of cost, time, carbon emissions, flow conservation constraints, and time window constraints.

[0036] S3. Optimize the logistics network: A hybrid quantum-classical optimization strategy is used to decompose and optimize the logistics network. This step includes the following sub-steps: S3.1, Layered Optimization: Decompose the logistics network into backbone paths optimized using the quantum annealing optimization algorithm and last-mile delivery paths optimized using the classical algorithm.

[0037] The mathematical model for dividing backbone routes and last-mile delivery is as follows: , Where, distance( i , j ) represents a nodei arrive j Geographical distance (km); d i , d j They are nodes i and nodes j The demand for goods (tons); D thres This indicates the distance threshold, which is 500km by default. It can be adjusted to 1000km for cross-border logistics. or This represents the threshold for the proportion of demand. In this example, the default value is 0.3, which means that the sum of the demands of the two nodes exceeds 30% of the total demand. Q The total demand of the network (in tons) is expressed by the following formula: .

[0038] The formula for calculating the last-mile delivery route is: That is, routes other than the backbone routes are classified as last-mile delivery routes.

[0039] S3.2, Quantum optimization of backbone paths: The quantum annealing algorithm is used to solve the backbone paths.

[0040] Before applying the quantum annealing optimization algorithm, the quantum system is initialized: an initial Hamiltonian H0 is constructed, and the system transitions to the target Hamiltonian H through adiabatic evolution. s =H, where H is the target Hamiltonian.

[0041] Next, set the annealing parameters. The annealing parameters include: Annealing time T: has a polynomial relationship with problem size. T ∝ N 2 ( N (Number of nodes).

[0042] Transverse field strength Γ: controls quantum tunneling efficiency and is dynamically adjusted through an adaptive algorithm.

[0043] Quantum state measurement: Obtain the ground state quantum state and map it to a logistics network optimization scheme.

[0044] The specific process of the quantum annealing optimization algorithm is as follows: S3.2.1 Input: Input the logistics network topology information into the quantum annealing optimization module, such as the number of nodes N, edge weight matrix [ c ij , t ij , e ij Multi-objective weights l 1, l 2. Constraints. Constraints include the flow conservation threshold. Q,node i Time window a i , b i Data such as [etc.].

[0045] S3.2.2. Combining the total energy function, map it to the qubit system. The total energy function is: H t = E cost + E time + E carbon + E flow + E window .

[0046] When calculating the total energy function of the backbone pathway, carbon emissions and time window constraints can be ignored because they are relatively stable over a certain period of time and lack variation.

[0047] S3.2.3. Set the annealing parameters T and Γ, and perform quantum adiabatic evolution. This includes the following:

[0048] Annealing time: ,in k =0.012, which is the D-Wave hardware calibration coefficient.

[0049] The formula for the transverse field intensity at time T is: , The initial transverse field strength is calculated as follows: α is the decay factor, which is 1.2 in this embodiment and is optimized through reinforcement learning.

[0050] Adiabatic evolution equation: Where s∈[0,1] represents the evolutionary progress. H is the initial Hamiltonian, and H is the target Hamiltonian.

[0051] S3.2.4 Measure the quantum state and obtain the optimized scheme (bit string) { }

[0052] S3.2.5 Perform path decoding and constraint verification.

[0053] Path decoding: The node sequence with a value of 1 is mapped to the logistics path, and the actual arrival time is calculated. t arrive ( i ) and total transportation cost E cost Then, constraint verification is performed.

[0054] The constraint verification formulas include: Flow conservation: ϵ1 is the error threshold, ϵ1=0.01Q.

[0055] Time window: t arrive ( i )∈[ai,bi]±ϵ2, ϵ2 is the time tolerance, ϵ2=10 minutes.

[0056] S3.2.6 Parameter Adjustment and Iteration: Perform parameter adjustment and iteration in the following manner.

[0057] If the constraints are not met or the objective function does not reach its target (e.g., energy fluctuation ΔH > 0.05H), min Adjust the parameters using the following formula: Annealing time: Where δ is the error coefficient, and δ = 0.2 when the constraint is not met. T old and T new To adjust the annealing time before and after.

[0058] Traffic penalty coefficient: ,in, and This refers to the traffic penalty coefficient before and after adjustment.

[0059] Time window penalty coefficient: Where maxDelay is the maximum time window delay. and This is to adjust the time window penalty coefficient before and after.

[0060] During training, repeat steps S3.2.3 to S3.2.6 until convergence.

[0061] S3.3 Classic Optimization of Last-Mile Delivery Routes: An improved Clarke-Wright cost-saving algorithm is used to solve the last-mile delivery routes.

[0062] Savings i , j The formula for calculating ) is: ,in, C 0i , C 0j , C ij From distribution center to node i Cost, from center to node j Costs and nodes i To the node jBased on the cost, routes are merged in descending order of savings, and the last-mile delivery routes must simultaneously meet the following requirements: 1) Vehicle capacity constraints: Where Cap represents the maximum vehicle capacity of the node; 2) Time window constraint: t arrive ( i )∈[ a i , b i ], [ a i , b i The symbol ] represents the corresponding time window.

[0063] S3.4. Based on historical data, this step can also apply incremental annealing optimization strategies to the backbone path. Specifically, it includes the following steps.

[0064] S3.4.1 Perform historical decaching.

[0065] The formula for preserving the ground state of the historical solution used in this step is: ,in, Represents the historical best bit string. H old This represents the historical energy function.

[0066] S3.4.2, Update the energy function.

[0067] The update model for the dynamically changing energy function applied in this step is as follows: The formula for the dynamically updated item is: , where Δ E Δ represents the set of edges affected (e.g., congested road sections). V Represents the set of nodes affected (e.g., temporarily closing a warehouse); Δ w ij This represents the change in edge weights of a path (e.g., the increase in transportation costs Δ during congestion). w ij =0.2 c ij ); Δ b i Node bias (e.g., Δbi forced when the warehouse is closed) s i =-1); S3.4.3. Adjust the incremental annealing parameters. The formula for adjusting the incremental annealing parameters includes: 1) Annealing time adjustment: ,in, T fullFor global optimization of time, |ΔE| and |ΔV| represent the number of changing edges and nodes, respectively. For example, when 20% of the road segment is congested... .

[0068] 2) Lateral field strength restart strategy: , where Γ old The final value of the historical horizontal field is given, and γ is the adjustment coefficient (γ=0.5). To achieve average edge weights, under unforeseen circumstances, Γ inc It can improve efficiency by 30%-50% to accelerate tunneling.

[0069] An example of an incremental annealing process is as follows: 1) Detect dynamic events: For example, if congestion on a certain road segment increases transportation time by 40%, the corresponding edge ( i , j )∈Δ E Δ t ij =0.4 t ij .

[0070] 2) Update the energy function: .

[0071] 3) Calculate incremental annealing time: If |ΔE|=50, N=1000, then Tinc=83 seconds·(50 / 1000) 2 ≈4.15 milliseconds.

[0072] 4) Perform annealing: As the initial state, adjust T inc The subsequent fast solution reduces the replanning time by 95% compared to full optimization.

[0073] S4. Dynamic Adjustment and Post-processing. This includes the following:

[0074] 1) Feasibility verification: Check whether the solution meets all constraints (such as vehicle capacity and time window). If not, adjust it using the classic repair algorithm.

[0075] 2) Local optimization: For the quantum annealing results, classical algorithms such as 2-opt and 3-opt are used to fine-tune the path and improve the quality of the solution.

[0076] 3) Real-time monitoring and replanning: 3.1) Deploy sensors to monitor the status of the logistics network in real time. When changes exceed a threshold (such as a 30% increase in the congestion index), trigger dynamic replanning.

[0077] 3.2) Incremental quantum annealing is used to update only the energy function of the affected part, and new schemes are generated quickly.

[0078] S5. Conduct user interaction and solution output, specifically including the following:

[0079] 1) Visual interface: Displays indicators such as optimization path, cost distribution, and carbon emissions, and supports comparison of multiple options (such as the fastest path and the lowest carbon path).

[0080] 2) Preference Input and Feedback: Users can customize weights (e.g., λ1=0.6 indicates time priority), and the system records feedback and updates the preference model.

[0081] Compared with existing quantum annealing algorithms, this scheme has the following technical innovations: 1. Global search mechanism: Compared with the existing fixed parameter annealing, which is prone to falling into the pseudo ground state, the adaptive tunneling control + topology optimization embedding of this invention improves the probability of ground state acquisition.

[0082] 2. Multi-objective quantum energy function design: For the current single-objective or static weighted solutions, the Pareto solution coverage is less than 30%. This invention improves the coverage by 50%+ through constraint-objective quantum encoding + dynamic weights.

[0083] 3. Dynamic scene response: Incremental annealing + layered optimization, with a processing time of less than 30ms.

[0084] 4. Strong parameter adaptability: Compared with existing experience-based parameter setting, which takes a long time to optimize, the machine learning-based dynamic generation of parameters in this invention reduces optimization time.

[0085] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A logistics network global optimization method based on a quantum annealing algorithm, characterized in that: The method comprises the following steps: S1, data collection: real-time collection of multi-source logistics data and access to environmental data; S2, quantum model construction: quantum mapping of the logistics network and construction of an energy function; S3, optimization of the logistics network: decomposition and optimization of the logistics network using a hybrid quantum-classical optimization strategy; the logistics network is decomposed into a backbone path optimized using a quantum annealing optimization algorithm and an end distribution path optimized using a classical algorithm; S4, dynamic adjustment and post-processing; S5, user interaction and scheme output. 2.The method of claim 1, wherein: Step S2 comprises: 1) Quantum mapping of logistics network: abstract logistics network as quantum Ising model, i The node state of a node is represented by a quantum bit σ i ∈{-1,1} represents; 2) Energy function construction: fusion of multi-objective optimization objectives and constraint conditions, as follows: Cost item: wherein, E cost is a cost item, c ij is a node i to j the transportation cost; Time term: wherein, E time is a time term, t ij is a transport time, λ 1 is a time weight; Constraint terms include the following constraint functions: Flow conservation constraint term: where, E flow is the total flow, d i is the node demand, Q is the total demand, used as a flow conservation threshold, μ 1 is the flow weight; The total energy function is: H t = E cost + E time + E flow .

3. The method of claim 2, wherein: Energy function construction also includes: Carbon emission item: wherein, E carbon is a carbon emission item, e ij is a carbon emission amount, λ 2 is a carbon emission weight; Time window constraint term: wherein, E window is the total time window, a i , b i is the time window of node i , t arrive is the arrival time of node i , i is the arrival time of node μ 2 is the time window weight; The total energy function is: H t = E cost + E time + E carbon + E flow + E window .

4. The method of claim 3, wherein: The logistics network quantum mapping algorithm is as follows: 1.1) Node-bit mapping: Each decision node corresponds to a qubit, and the Chimera graph embedding technique is used to adapt to the quantum hardware topology, and the mapping formula is: wherein, v i represents the logistics network node i , x j , y j represents the two-dimensional grid coordinates of the node j in the Chimera graph; k represents the sub-lattice index, k =0,1, identifies two sub-nodes of the node j within the same grid; 1.2) Edge weight mapping: transportation cost, time, and carbon emissions are converted into coupling coefficients in the energy function, and reliability constraints are converted into transverse field parameters of the quantum system.

5. The method of claim 3, wherein: Step S3 comprises: S3.1, hierarchical optimization: decomposition of the logistics network into a backbone path optimized using a quantum annealing optimization algorithm and an end distribution path optimized using a classical algorithm; S3.2, quantum optimization of the backbone path: solving the backbone path using a quantum annealing algorithm; S3.3, classical optimization of the end distribution path: solving the end distribution path using an improved Clarke-Wright saving algorithm; In step S3.1, the mathematical division model of the backbone path and the end distribution is as follows: ， wherein distance (i, j) represents the geographic distance from node i to node j; i , j i j d i , d j i j D thres η Q ;​​​​​​​​​ The formula for the end delivery route is: That is, the route other than the backbone route is classified into the end delivery route.

6. The method of claim 5, wherein: Step S3 also comprises S3.4, based on historical data, this step can also apply an incremental annealing optimization strategy to the backbone path; specifically comprising the following steps: S3.4.1, perform history uncache, the history uncache base state reserved formula used in this step is: wherein, represents the history optimal bit string, H old represents the history energy function; S3.4.2, performing energy function update, the update model of the dynamic change energy function applied in this step is: wherein the calculation formula of the dynamic update term is: wherein Δ E represents the affected edge set, Δ V represents the affected node set; Δ w ij represents the edge weight change amount of the path; Δ b i : node bias amount; S3.4.3, incremental annealing parameter adjustment, the incremental annealing parameter adjustment formula includes: 1) Annealing time adjustment: where, T full is the global optimization time, |ΔE| and |ΔV| are the number of changed edges and nodes, respectively; 2) Transverse field intensity restart strategy: where Γ old is the historical transverse field final value, γ is the adjustment coefficient, is the average edge weight.

7. The method of claim 5, wherein: In step S3.2, first perform quantum system initialization, then set the annealing parameters, and then the flow of the quantum annealing optimization algorithm is as follows: S3.2.1, input: input the logistics network topology information into the quantum annealing optimization module; S3.2.2, combine the total energy function and map to the quantum bit system; S3.2.3, set the annealing parameters T and Γ, perform quantum adiabatic evolution, T is the annealing time, and Γ is the transverse field strength; S3.2.4, measuring quantum states, obtaining optimization scheme }; S3.2.5, path decoding and constraint verification; S3.2.6, parameter adjustment and iteration; During training, steps S3.2.3 to S3.2.6 are repeated until convergence.

8. The method of claim 7, wherein: Step S3.2.3 comprises the following: Annealing time: where k is a D-Wave hardware calibration coefficient, N is the number of nodes; The transverse field intensity formula at time T is: , is the initial transverse field intensity, and the formula is: , and a is a decay factor, and the value is optimized by reinforcement learning; Adiabatic evolution equation: where s e [0, 1] is the evolution progress, is the initial Hamiltonian, H is the target Hamiltonian. 9.The method of claim 5, wherein: In step S3.2.6, parameter adjustment and iteration are performed as follows: Annealing time: where δ is an error coefficient, T old and T new is the annealing time before and after adjustment; flow penalty coefficient: wherein, and is the flow penalty coefficient before adjustment; and time window penalty coefficient: where maxDelay is the maximum time window delay, and is the time window penalty coefficient before and after adjustment.

Images