Dynamic demand cold chain transportation path planning method based on improved ALNS algorithm

Abstract

The invention discloses a dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm, which is applied to the field of cold chain path optimization under dynamic demand, and comprises the following steps: converting a DVRPTW problem of cold chain transportation path planning into a C-DVRPTW problem, based on a mixed integer programming method, constructing a C-DVRPTW problem mathematical model which comprises fixed cost, transportation cost, refrigeration cost, goods damage cost and time window penalty cost and takes operation cost minimization as a target; a dynamic insertion strategy, a simulated annealing strategy and a multi-time-period rolling optimization strategy are fused, an improved ALNS algorithm with two novel operators of low-efficiency customer removal and cost reduction route removal is designed to solve a mathematical model, and an optimal cold chain transportation route is obtained. According to the method, the cold chain distribution path optimization under the complex scene of dynamic change of the cargo demand and the distribution time window is realized, and meanwhile, the timeliness and quality requirements of cold chain transportation are met.

Description

Technical Field

[0001] The present invention relates to the field of cold chain path optimization under dynamic demand, and particularly to a method for planning cold chain transportation paths based on an improved ALNS algorithm under dynamic demand. Background Art

[0002] With the continuous development of the economic society and the rapid improvement of the people's material level, the standards of "high quality" and "on-time delivery" are constantly being upgraded. Among them, cold chain logistics is crucial for the transportation of foods such as aquatic products, meat, poultry, eggs, and fresh vegetables, and is also an important guarantee for vaccine transportation. Therefore, strategically laying out the cold chain transportation industry is an important measure.

[0003] In recent years, many research scholars have studied the cold chain logistics VRP problem from different perspectives. Among them, in the research on the VRP problem considering dynamic demand, Hu Xiao et al. proposed a dynamic path planning method based on an improved Harris hawk optimization algorithm. By combining square grid diffusion and the dynamic window method, the real-time performance and global search ability of path planning were improved. However, this method mainly focuses on the dynamic nature of path planning and does not solve the complex constraint problems in cold chain logistics. To address this issue, Lu Fuqiang et al. proposed a distribution mode assisted by drones for riders. By designing a two-stage optimization model and improved AP clustering and tabu search algorithms, the path optimization efficiency and customer satisfaction were improved. However, the specific optimization of dynamic cold chain paths was not involved in their research. Wang et al. transformed the DVRP problem into a multi-period static VRP, and provided a simplified idea for dynamic demand optimization by establishing an exponential function model and using an integrated-evolution algorithm to solve it. In contrast, Zhang et al. directly established a mathematical model for the DVRP problem based on the Markov decision model and used the approximate dynamic programming algorithm to solve it, avoiding the complexity of static decomposition. However, the adaptability of this research to the specific requirements of cold chain still needs to be improved. For this reason, Wang et al. further improved the DVRP model through mixed integer programming and improved the solution efficiency through a hybrid heuristic algorithm, but the specific problems of dynamic demand in cold chain logistics have not been completely solved.

[0004] Therefore, how to provide a method for planning cold chain transportation paths based on an improved ALNS algorithm under dynamic demand, which can optimize cold chain distribution paths with the goal of minimizing the total operating cost in a complex scenario where the demand for goods and the delivery time window change dynamically, and at the same time meet the timeliness and quality requirements of cold chain transportation, is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention proposes a method for planning cold chain transportation paths based on an improved ALNS algorithm under dynamic demand.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm comprises: Step 1: Convert the DVRPTW problem of cold chain transportation path planning into a C-DVRPTW problem, and build a mathematical model of the C-DVRPTW problem based on mixed integer programming, which includes fixed cost, transportation cost, refrigeration cost, cargo damage cost and time window penalty cost, and aims to minimize operating cost; Step 2: By integrating dynamic insertion strategy, simulated annealing strategy and multi-time period rolling optimization strategy, an improved ALNS algorithm with two novel operators, inefficient customer removal and cost reduction route removal, is designed to solve the mathematical model and obtain the optimal cold chain transportation route.

[0007] Optionally, in step 1, the DVRPTW problem of cold chain transportation path planning is converted into a C-DVRPTW problem, specifically: The distribution time range [0, T] is divided into multiple uniform time periods, and the cold chain distribution demand in each time period is analyzed, and each time period is treated as a static problem; After completing the delivery in the previous time period, the current location of the vehicle is regarded as a temporary distribution center, and the delivery route for the next cycle is generated in combination with the newly added customer needs, thereby realizing the optimization of the multi-cycle static summation process.

[0008] Optionally, in step 1, the mathematical model of the C-DVRPTW problem is as follows: Among them, Z is the operating cost; P is the set of all time periods; is the fixed cost in the pth time period; is the transportation cost in the pth time period; is the cooling cost in the pth time period; is the cargo damage cost in the pth time period; is the time window penalty cost for the p-th time period; is the carbon tax cost in the pth time period; n is the number of cold chain delivery vehicles required in the pth time period; is the fixed cost of a unit cold chain distribution vehicle; M is the customer point set; is the driving distance from customer point i to customer point j; The transportation cost per unit distance of an empty cold chain distribution vehicle; is the loading rate of cold chain delivery vehicles in the pth time period; The transportation cost per unit distance of a fully loaded cold chain distribution vehicle; is a decision variable, which is 1 if the cold chain distribution vehicle travels from customer point i to customer point j within time period p, and 0 otherwise; is the transportation refrigeration cost per unit time when the cold chain distribution vehicle is loaded; is the load quantity from customer point i to customer point j; is the transportation time from customer point i to customer point j; is the refrigeration cost per unit time during the loading and unloading process of the cold chain distribution vehicle; is a decision variable, which is 1 if the cold chain distribution vehicle undergoes loading and unloading at customer point i within time period p, and 0 otherwise; is the unit price of the goods; is the freshness decay rate during the transportation process of the cold chain distribution vehicle; is the freshness decay rate during the unloading process of the cold chain distribution vehicle; N is the set of cold chain distribution vehicles; is the penalty coefficient for early delivery; is the expected service time of customer point i; is the time when cold chain distribution vehicle k arrives at customer point i within the p-th time period; is the penalty coefficient for late delivery; is the carbon emission tax rate per unit distance.

[0009] Optionally, in step 1, the constraint conditions of the mathematical model are as follows: where M is the set of customer points; is the demand quantity of customer point i; is a decision variable, which is 1 if customer point i is served by cold chain distribution vehicle k, and 0 otherwise; N is the set of cold chain distribution vehicles.

[0010] Optionally, in step 2, the improved ALNS algorithm is as follows: Use a heuristic algorithm to generate an initial solution, and perform operator deletion and operator insertion operations in sequence according to the preset deletion rules and preset insertion rules; among them, the preset deletion rules include: random removal, random route removal, worst removal, Shaw removal, inefficient customer removal, and cost reduction route removal; the preset insertion rules include: greedy insertion, Regret-3 insertion; Update the current solution and the optimal solution based on the temperature cooling mechanism of simulated annealing, and after completing the path optimization of each time period, use the newly added demand and the adjusted customer points as the initial path points for the next time period to cope with the dynamic demand of customers. When the maximum number of iterations is reached or the algorithm converges, output the current optimal solution as the final path planning scheme.

[0011] Optionally, use a heuristic algorithm to generate an initial solution, specifically as follows: Set the input variable to 0, sort according to the requirements of customer points from large to small, and divide customer points into a certain warehouse service point according to the distance size; According to the order arranged by distance, add the customer points within the initial period to the initial transportation route of a certain vehicle. The total weight of the nodes is less than or equal to the load capacity of the vehicle to obtain the initial transportation route of the vehicle, and obtain the decision variable , The value of; among them, Indicates 1 if the cold chain distribution vehicle performs loading and unloading at customer point i during time period p, otherwise 0; Indicates 1 if customer point i is served by cold chain distribution vehicle k, otherwise 0; According to the arrangement of the initial nodes, obtain the transportation route array , obtain the positions where new customer points can be inserted, and delete the insertable positions that do not meet the constraints according to the constraints of time windows and vehicle capacities. According to the comparison of the cost increments after adding the nodes at the insertable positions, take the minimum value to obtain the positions of the inserted nodes; The input of the heuristic algorithm is the set of customer points, and the output is the initial feasible solution of C-DVRPTW.

[0012] Optionally, update the current solution and the optimal solution based on the temperature cooling mechanism of simulated annealing. Specifically: Operator weight update: In the adaptive algorithm iteration, the update of the operator weight is dynamically applied to the periodic probability adjustment of the deletion and insertion operators. After the current period is completed and before the next period iteration starts, the operator weight Is updated to ; where r is the weight forgetting coefficient; Is the cumulative usage times of operator i in this period; Is the cumulative score of operator i in this period; the cumulative score depends on the quality of the new solution generated by operator i. There are three cases for the new solution: better than the global optimal solution, better than the current solution, worse than the current solution and will be updated; Temperature update: Use the temperature cooling mechanism of simulated annealing to update the temperature in each round.

[0013] As can be seen from the above technical solutions, compared with the prior art, the present invention proposes a dynamic-demand cold-chain transportation route planning method based on an improved ALNS algorithm. For the first time, the cold-chain path optimization problem under dynamic demand is reformulated as a "multi-period static summation model" (C-DVRPTW). Through dynamic time period division, soft time window mechanism and rolling optimization strategy, under multiple constraints such as comprehensive operating cost, time window violation penalty and carbon emissions, an accurate modeling of the dynamic cold-chain logistics path problem is achieved. An improved adaptive large neighborhood search (IALNS) algorithm is proposed, which integrates a dynamic insertion mechanism, a simulated annealing strategy and a multi-time period rolling optimization method, and designs two novel operators: inefficient customer removal and cost reduction route removal, significantly improving the global search ability and solution efficiency of the algorithm. Experimental results show that IALNS is superior to the traditional solver CPLEX in small-scale examples, and compared with ALNS in large-scale examples, the distribution cost is reduced by an average of 4.21% and the maximum reduction is up to 16.7%, and significant advantages are also shown in the objective function value, convergence speed and path optimization effect. The benefit analysis of the multi-period static summation model shows that the operating benefit is the best when the number of periods P = 5, achieving a balance between benefit and response speed. The present invention provides theoretical support and practical reference for the dynamic cold-chain path optimization problem, and has important academic value and application prospects for the green and sustainable development of cold-chain transportation enterprises. In the future, the applicability of the model in more complex scenarios can be further expanded by combining real-time traffic information and multi-agent collaborative optimization technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings according to the provided drawings without creative efforts.

[0015] Figure 1 It is a schematic flow chart of the method of the present invention.

[0016] Figure 2 It is a schematic diagram of an example of the C-DVRPTW problem of the present invention.

[0017] Figure 3 It is a schematic diagram of the integer coding method of the solution to the C-DVRPTW problem of the present invention.

[0018] Figure 4 It is a schematic flow chart of the improved ALNS algorithm of the present invention.

[0019] Figure 5 It is a schematic diagram of the insertable positions of the initial nodes of the present invention.

[0020] Figure 6 Schematic diagram for comparing the iterative processes of IALNS and ALNS of the present invention.

[0021] Figure 7 Schematic diagram for comparing large-scale examples of five algorithms of the present invention.

[0022] Figure 8 Schematic diagram for comparing small-scale examples of five algorithms of the present invention.

[0023] Figure 9 Schematic diagram of the trajectory of large-scale problems of the HHO algorithm of the present invention.

[0024] Figure 10 Schematic diagram of the trajectory of large-scale problems of the HOA algorithm of the present invention.

[0025] Figure 11 Schematic diagram of the trajectory of large-scale problems of the GA algorithm of the present invention.

[0026] Figure 12 Schematic diagram of the trajectory of large-scale problems of the SA algorithm of the present invention.

[0027] Figure 13 Schematic diagram of the trajectory of large-scale problems of the IALNS algorithm of the present invention. Detailed implementation manners

[0028] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] Embodiment 1: Embodiment 1 of the present invention discloses a dynamic demand cold chain transportation route planning method based on an improved ALNS algorithm, as Figure 1 shown, including: Step 1: Convert the DVRPTW problem of cold chain transportation route planning into a C-DVRPTW problem, and construct a mathematical model of the C-DVRPTW problem including fixed costs, transportation costs, refrigeration costs, cargo damage costs, and time window penalty costs, with the goal of minimizing operating costs, based on the mixed integer programming method.

[0030] Converting the DVRPTW problem of cold chain transportation route planning into a C-DVRPTW problem simplifies the complexity of dynamic demand. Specifically: Divide the delivery time range [0, T] into multiple equal time periods, analyze the cold chain delivery demand within each time period, and treat each time period as a static problem; After completing the delivery in the previous time period, the current location of the vehicle is regarded as a temporary distribution center, and the delivery route for the next cycle is generated by combining the newly added customer demands, so as to optimize the multi-cycle static summation process.

[0031] An instance of the C-DVRPTW problem, as Figure 2 shown. If it is known that there are three delivery stations {1, 2, 3}, and the time range is divided into two cycles. In the first cycle, the known customer points are {4, 5, 6, 7, 8}, and in the second cycle, the newly added customer points are {9, 10, 11}, then the decisions S0 and S1 can be obtained respectively. Therefore, the decision S0 can be expressed as 1→4→5→1 and 2→6→7→8→2. The decision S1 after T1 is 5→9→1, 7→10→8→2, and 3→11→3. Among them, the customer points {5, 7} are used as the temporary stations in the second cycle, and transportation starts from this point after adding the customer points in the next cycle. This method effectively addresses the complexity of dynamic demands by introducing dynamic time division and per-cycle optimization, providing a solution that combines theory and practice.

[0032] Considering the characteristics of the C-DVRPTW problem, the following assumptions are made: The total demand of customer points on each delivery route cannot exceed the maximum load capacity of the delivery vehicle; Assume that changes in traffic conditions, environmental problems such as weather are predictable and follow a certain pattern or rule, although there may be unexpected situations in reality; The demand of each customer point is satisfied and can only be transported by one vehicle; Each customer has a service time window, and a penalty is imposed for deliveries outside the service window time; Customer points can only be delivered by the nearest station; Only consider the vehicle loading process with load, and do not calculate the vehicle empty loading process; All vehicles are homogeneous; The vehicles departing from each vehicle section must return to the original vehicle section.

[0033] The mathematical model of the C-DVRPTW problem is as follows: Fixed cost , including driver's salary, vehicle depreciation cost, equipment wear cost, etc. The fixed cost is only related to the type and quantity of cold chain delivery vehicles, as follows: Transportation cost , mainly refers to the fuel cost consumed during transportation, which is closely related to the transportation distance, transportation speed, vehicle engine emissions, etc., including the dynamic cost of empty and full loads, as follows: Cooling costs Since cold chain products are perishable and easy to deteriorate, the temperature requirements during the distribution process are high, so a certain amount of refrigeration costs will be incurred. The refrigeration costs are divided into two parts: one is the refrigeration cost during transportation; the other is the refrigeration cost generated by the heat exchange of the compartment door opening and closing during the loading and unloading process of the vehicle, as follows: Cargo damage cost The perishability of fresh products during cold chain transportation will be affected by factors such as temperature, humidity, and oxygen concentration in the storage environment, and will cause certain losses over time, as follows: Time window penalty cost Cold chain logistics distribution has high requirements for timeliness, and customers also attach great importance to the arrival time of delivery. Therefore, it is necessary to consider the penalty cost of time window breach, as follows: Carbon tax costs Compared with ordinary vehicles, refrigerated vehicles have higher carbon dioxide emissions. In addition, each cold chain logistics company will obtain a certain carbon emission quota every year. If the actual emissions exceed the quota, the company needs to purchase the excess emission rights from the carbon trading market, otherwise it will face high fines. The present invention calculates the carbon tax cost based on carbon emissions as follows: Among them, Z is the operating cost; P is the set of all time periods; is the fixed cost in the pth time period; is the transportation cost in the pth time period; is the cooling cost in the pth time period; is the cargo damage cost in the pth time period; is the time window penalty cost for the p-th time period; is the carbon tax cost in the pth time period; n is the number of cold chain delivery vehicles required in the pth time period; is the fixed cost of a unit cold chain distribution vehicle; M is the customer point set; is the driving distance from customer point i to customer point j; The transportation cost per unit distance of an empty cold chain distribution vehicle; is the loading rate of cold chain delivery vehicles in the pth time period; is the transportation cost per unit distance for a cold chain distribution vehicle at full load; is a decision variable, which is 1 if the cold chain distribution vehicle travels from customer point i to customer point j within time period p, and 0 otherwise; is the transportation refrigeration cost per unit time when the cold chain distribution vehicle is loaded; is the load from customer point i to customer point j; is the transportation time from customer point i to customer point j; is the refrigeration cost per unit time during the loading and unloading process of the cold chain distribution vehicle; is a decision variable, which is 1 if the cold chain distribution vehicle undergoes loading and unloading at customer point i within time period p, and 0 otherwise; is the unit price of the goods; is the freshness decay rate during the transportation process of the cold chain distribution vehicle; is the freshness decay rate during the unloading process of the cold chain distribution vehicle; N is the set of cold chain distribution vehicles; is the penalty coefficient for early delivery; is the expected service time of customer point i; is the time when the cold chain distribution vehicle k arrives at customer point i within the p-th time period; is the penalty coefficient for late delivery; is the carbon emission tax rate per unit distance.

[0034] The constraint conditions of the mathematical model are as follows: It means that the total demand of any delivery route does not exceed the full load weight of the vehicle; It means that all customer points are only served once; It means that the distribution center is the end point of all vehicles; where M is the set of customer points; is the demand of customer point i; is a decision variable, which is 1 if customer point i is served by cold chain distribution vehicle k, and 0 otherwise; N is the set of cold chain distribution vehicles.

[0035] Step 2: Use the improved ALNS algorithm that combines the dynamic insertion strategy, simulated annealing strategy, and multi-time period rolling optimization strategy, and designs two novel operators: inefficient customer removal and cost reduction route removal, to solve the mathematical model and obtain the optimal cold chain transportation route.

[0036] As an extension of the vehicle routing problem (VRP), the cold chain dynamic path optimization problem not only inherits the essence of its NP-hard problem, but also significantly enhances the complexity and solution difficulty of the problem due to the comprehensive impact of dynamic customer demands, time window changes, and route dynamic adjustments on transportation costs. The greatest innovation of this invention lies in systematically integrating these dynamic factors into the cold chain path optimization model for the first time, and proposing a new solution that breaks through the limitations of existing algorithms from both theoretical and practical levels. Existing exact algorithms are difficult to obtain high-quality solutions within a reasonable time due to high computational complexity in medium and large-scale transportation scenarios, and the research of this invention aims to completely change this dilemma. On this basis, this invention starts with the adaptive large neighborhood search (ALNS) algorithm and transforms its core mechanism. Since the ALNS algorithm was proposed by Ropke and Pisinger, it has been widely applied to complex optimization problems, but there are still significant deficiencies in its adaptability to dynamic demands and computing power efficiency. Therefore, this invention proposes two innovative deletion operations, namely inefficient customer deletion operator and cost-reducing route removal, for the first time to more precisely address the dynamic characteristics in cold chain path optimization. Compared with traditional operators, these new operators not only optimize the path planning, but also significantly reduce the interference of inefficient computational operations on global search. In addition, to solve the problem of computing power waste caused by repeated insertions in the ALNS algorithm, this invention innovatively introduces the simulated annealing (SA) mechanism, which is deeply integrated with the dynamic insertion strategy. By significantly improving the search efficiency and diversity, it significantly enhances the ability of the algorithm to jump out of local optima, thereby achieving an efficient exploration of the global optimal solution. At the same time, this invention designs a multi-time period rolling optimization strategy, which effectively responds to the complex changes brought by dynamic demands by dynamically updating the optimization process in real time. This strategy not only has breakthroughs in methodology, but also provides theoretical support and practical guidance for the efficient solution of the dynamic cold chain path optimization problem.

[0037] For the solution representation of the C-DVRPTW problem in this invention, an intuitive and efficient integer coding method is adopted, such as Figure 3 shown, the transportation path of the vehicle is represented in the form of an array with variable length, and each array corresponds to the delivery route of a vehicle. This coding method can clearly describe the path planning of the vehicle, and at the same time has good flexibility and scalability, which is convenient for operation and adjustment in the subsequent optimization process.

[0038] Improve the ALNS algorithm, as Figure 4 shown, specifically: Generate an initial solution using a heuristic algorithm, and successively perform operator deletion and operator insertion operations according to preset deletion rules and preset insertion rules; among them, the preset deletion rules include: random removal, random route removal, worst removal, Shaw removal, inefficient customer removal, and cost-reducing route removal; the preset insertion rules include: greedy insertion, Regret-3 insertion.

[0039] Generate an initial solution using a heuristic algorithm, specifically: Set the input variable to 0, sort the customer points in descending order according to their demands, and divide the customer points into a certain warehouse service point according to the distance. According to the order of the distances, add the customer points within the initial period to the initial transportation route of a certain vehicle. The total weight of the nodes is less than or equal to the load capacity of the vehicle to obtain the initial transportation route of the vehicle, and obtain the decision variable , The value of; among them, Indicates 1 if the cold chain distribution vehicle performs loading and unloading at customer point i during time period p, otherwise 0; Indicates 1 if customer point i is served by cold chain distribution vehicle k, otherwise 0; According to the arrangement of the initial nodes, obtain the transportation route array , obtain the positions where new customer points can be inserted, as shown in Figure 5 , and delete the insertable positions that do not meet the constraints according to the constraints of the time window and vehicle capacity. Compare the cost increments after adding the nodes at the insertable positions and take the minimum value to obtain the positions of the inserted nodes; The input of the heuristic algorithm is the set of customer points, and the output is the initial feasible solution of C-DVRPTW.

[0040] Combined with the dynamic demand characteristics of the cold chain path, the key point of the deletion operator is to balance the reduction of various cost elements. The preset deletion rules are specifically: Random removal: This operator randomly selects customers from the given solution.

[0041] Random route removal: This operator randomly selects routes from the given solution, extracts the customers related to these routes and transfers them to the request list L. The operator repeats the selection until the number of removed customers reaches or exceeds .

[0042] Worst removal: Calculate the difference in the objective function before and after each customer node is removed in the current solution. The larger the difference, the greater the degree of deterioration of the objective function after the node is inserted. Prioritize removing customer nodes with larger differences until the number of removed customers reaches or exceeds . The method of determining the number of removals is the same as that of random removal.

[0043] Shaw Removal: Randomly remove a customer node from the current solution, calculate the correlation between the remaining nodes and this node, and then remove nodes with high correlation multiple times until a certain number of removals is reached. The method for determining the number of removals is the same as the method for random removal. Until the number of removed customers reaches or exceeds , Method for calculating the correlation between nodes: .

[0044] Inefficient Customer Removal: Calculate the service efficiency of each customer, calculated as demand / transportation time, sort the service efficiency from low to high, and remove the least efficient customers until the number of removed customers reaches or exceeds .

[0045] Cost Reduction Route Removal: During transportation, the operating cost increases with the increase in the number of dispatched vehicles. Therefore, in this invention, it is specifically proposed to preferentially remove the route with the minimum cost. If the current solution contains one route, no removal is performed; if the current solution contains two routes, the shortest one is removed; if the current solution contains three or more routes, the two shortest routes are removed. Until the number of removed customers reaches or exceeds .

[0046] Preset insertion rules, specifically: Greedy Insertion: Calculate the cost increment after inserting each insertable node, and take the minimum value as the inserted node. , where k represents the inserted node.

[0047] Regret-3 Insertion: Similar to greedy insertion, for each unserved customer node i, calculate its cost of insertion into all possible positions and select the three positions with the lowest costs. Then, calculate the difference between the costs of these positions and the cost of the sub-optimal position as the regret value, and the customer node with the highest regret value will be preferentially inserted. The Regret-3 insertion method can usually find high-quality solutions within a reasonable time and is applicable to large-scale problems. The Regret value is defined as the difference in the objective function after the node is inserted into these two positions. Compare the optimal insertion positions of all customer nodes and insert the node with the largest Regret value into the current solution. Repeat the above process until the remaining nodes are inserted.

[0048] Update the current solution and the optimal solution based on the temperature cooling mechanism of simulated annealing, and after completing the path optimization for each time period, use the new demand and the adjusted customer points as the initial path points for the next time period (multi-time period rolling optimization) to cope with the dynamic demand of customers. When the maximum number of iterations is reached or the algorithm converges, output the current optimal solution as the final path planning scheme.

[0049] Update the current solution and the optimal solution based on the temperature cooling mechanism of simulated annealing, specifically as follows: Update of operator weights: In the iteration of the adaptive algorithm, the update of operator weights is dynamically applied to the periodic probability adjustment of the deletion and insertion operators. After the completion of the previous cycle and before the start of the next cycle iteration, the operator weights are updated to ; where r is the weight forgetting coefficient; is the cumulative usage times of the operator in the i-th current cycle; is the cumulative score of the operator i in the current cycle; the cumulative score depends on the quality of the new solution generated by the operator i. There are three cases for the new solution: better than the global optimal solution, better than the current solution, worse than the current solution and will be updated; Temperature update: Utilize the temperature cooling mechanism of simulated annealing to update the temperature in each round.

[0050] Example 2: Example 2 of the present invention discloses a performance analysis of a dynamic demand cold chain transportation route planning method based on an improved ALNS algorithm, as follows: The experimental research of the present invention is carried out in the Matlab 2024b environment, where all computing tasks are based on the AMDRyzen™ 5 5500U processor (Radeon™ Graphics, main frequency 2.10GHz) and completed on the Windows10 operating system. The total number of iterations designed for the experiment is set to 200 times to ensure the stability and reliability of the results. To systematically verify the effectiveness of the C-DVRPTW mathematical model proposed by the present invention and the solution efficiency of the improved algorithm, the experimental design is divided into the following three parts: First, the analysis of the operation results is carried out. The solution results are respectively compared with the CPLEX solver and the IALNS algorithm on small and large-scale examples, and the accuracy of the model and the solution efficiency of the IALNS in small-scale problems are analyzed. Through the performance comparison between the IALNS and the traditional ALNS algorithm on large-scale examples, the solution efficiency and optimization ability of the improved algorithm in complex scenarios are evaluated in detail; Second, the solution performance of the IALNS algorithm in large and small-scale problems is compared and analyzed with the newly proposed HOA, HHO algorithms and the classical GA, SA algorithms, and its advantages in terms of convergence speed and solution quality are explored. Finally, the benefit analysis of the multi-cycle static summation model involved in the present invention is evaluated. By analyzing different cycle number configurations of the model, the balance effect on operation benefits and response speed is evaluated, so as to further verify the practical application value of the proposed model.

[0051] Example Introduction: Since the C-DVRPTW belongs to a new type of problem and there is no ready-made standard example available, the present invention selects the Solomon dataset for preprocessing and extension to construct experimental data applicable to different-scale examples. The preprocessed dataset is divided into six categories: C1 / C2, R1 / R2, RC1 / RC2. Among them, the distribution of customer points in the C category shows the characteristic of block aggregation, the distribution of customer points in the R category is random, and the RC category combines the characteristics of block aggregation and random distribution. This classification method not only ensures the diversity of experimental data but also facilitates the comprehensive evaluation of the algorithm performance under different characteristic scenarios.

[0052] Analysis of Running Results: Analysis of Running Results for Small-Scale Examples: Three instances are randomly selected from the six categories of examples, totaling 18 small-scale examples, and the solver CPLEX and the IALNS algorithm are used for solving respectively. In the experiment, the optimal solutions obtained by the CPLEX solver within the limited time and the running results of the IALNS algorithm are shown in Table 1. Among them, RN represents the number of vehicle operation routes for this example, TC represents the total cost, , and a negative Gap indicates that the solving effect of the IALNS algorithm is better than that of CPLEX.

[0053] Table 1 Comparison Results of Small-Scale Problems The experimental results show that in small-scale examples, the IALNS algorithm shows excellent solving effects. Among the 18 groups of experiments, the results of 14 groups of examples are consistent with CPLEX, indicating that IALNS can approach the optimal solution. In addition, the IALNS results of 4 groups of examples are better than CPLEX, and the maximum cost reduction is up to 11.67%. Notably, the RN value of IALNS in 4 examples is less than that of CPLEX, which means that IALNS can complete the distribution task with fewer vehicles, thus further reducing the operating cost. This result verifies the efficiency and resource optimization ability of IALNS in solving the C-DVRPTW problem.

[0054] Solving Analysis for Large-Scale Examples: To evaluate the solving efficiency of the IALNS algorithm in complex scenarios, a comparative experiment with the ALNS algorithm is designed. Five instances are randomly selected from each of the six categories of examples, totaling 30 large-scale examples for testing. The experimental results are shown in Table 2. Among them, RN represents the number of vehicle operation routes for this example, TC represents the total cost of the optimal solution, RT represents the solving time, and a negative Gap indicates that the effect of the IALNS algorithm is better than that of the ALNS algorithm.

[0055] Table 2 Comparison Results of Large-Scale Problems As can be seen from Table 2, in most cases, IALNS is superior to ALNS in terms of distribution cost. Among them, the average distribution cost of IALNS is reduced by 4.21%, and the maximum reduction is 16.7%. IALNS also shows advantages in the number of vehicles used. For example, in case C103, IALNS used 8 vehicles, while ALNS used 10 vehicles, reducing by 2 vehicles. This not only reduces the fixed cost but also the additional operating costs brought by vehicle operation, fully demonstrating its advantages in optimizing vehicle allocation and reducing distribution cost. Although the average solution time of IALNS is slightly higher than that of ALNS, this cost is acceptable because the final solution of IALNS is significantly better than that of ALNS, showing stronger global optimization ability and path allocation efficiency. This performance advantage makes IALNS more practical and valuable for promotion in dealing with complex dynamic demand problems in practical applications.

[0056] Comparison of the iterative processes of IALNS and ALNS, as Figure 6 shown, Figure 6 clearly demonstrates the significant advantages of IALNS: In the initial optimization stage, the objective function value of IALNS drops rapidly, showing high search efficiency and global optimization ability. In contrast, the convergence speed of ALNS is slower, and the objective function value shows a gradually decreasing trend; the objective function curve of IALNS quickly stabilizes after the decline, indicating that it can quickly find the global optimal or approximate optimal solution, demonstrating the high stability and reliability of the algorithm; during the entire iterative process, the objective function value of IALNS is always lower than that of ALNS, and the final solution is significantly better than the latter, further verifying the optimization ability of IALNS.

[0057] Algorithm comparison experiment: To further verify the advantages of the improved algorithm (IALNS) of the present invention in solving the C-DVRPTW problem, the newly proposed HHO and HOA algorithms, as well as the classical GA and SA algorithms, are used in the experiment to solve and compare large-scale and small-scale cases respectively. The comparison of the algorithm test results for large-scale and small-scale cases is shown respectively in Figure 7 and 8As shown in the figure. In small-scale examples, all algorithms exhibit a relatively fast convergence rate. However, the IALNS algorithm stands out particularly in the initial stage, being able to quickly approach the optimal solution and stabilize at a relatively low objective function value, demonstrating excellent global search capabilities and efficient convergence performance. In contrast, although other algorithms have certain competitiveness in some stages, the overall solution quality is slightly inferior to that of IALNS. For large-scale examples, the convergence rate of all algorithms slows down, but the IALNS algorithm still maintains a significant performance advantage. In most iterative processes, IALNS is always superior to other algorithms, and its final objective value is significantly lower than that of other algorithms. This indicates that IALNS can more effectively find high-quality solutions when dealing with complex problems, and has strong adaptability and stability. For small-scale examples, the IALNS algorithm maintains the lowest objective function value throughout the process, indicating that it can effectively find better solutions in small-scale problems. The specific comparative analysis is as follows: In small-scale examples, IALNS maintains the lowest objective function value in all iterative stages, proving that it can efficiently find better solutions. The HHO and GA algorithms show certain competitiveness in some stages, but there is still a gap in the solution quality and stability compared with IALNS.

[0058] In large-scale examples, IALNS demonstrates excellent performance. The objective value drops rapidly in the early stage of iteration, and the final solution quality is far better than that of other algorithms. The HHO and GA algorithms have poor convergence in large-scale problems, with a slow decline in the objective value and the final solution deviating from the optimal solution, showing insufficient adaptability in complex scenarios. The performance of the HOA and SA algorithms is relatively stable, but it is difficult to surpass IALNS in terms of solution quality.

[0059] In addition, the trajectory diagrams of large-scale problems for the HHO, HOA, GA, SA, and IALNS algorithms are shown in Figure 9 , 10 , 11, 12, and 13 respectively. It can be seen that the number of optimal solution paths (RN values) of the five algorithms is 15 (HHO), 19 (GA and SA), 12 (HOA), and 9 (IALNS) in sequence. IALNS uses the fewest number of paths, significantly optimizing the vehicle allocation plan and further reducing the operating cost during the distribution process. This result highlights the unique advantages of IALNS in path optimization and resource utilization.

[0060] The experimental results show that the IALNS algorithm exhibits the characteristics of fast convergence rate, high solution quality, adaptability to complex constraints, and strong path optimization ability in solving the C-DVRPTW problem. Compared with other algorithms, IALNS achieves a good balance in terms of convergence efficiency and solution quality, and is an efficient tool for solving the vehicle routing optimization problem with dynamic demands, especially suitable for practical application scenarios with complex constraints.

[0061] Multi - period Static Summation Model Evaluation: This invention focuses on the optimization problem of cold - chain routes considering dynamic demands. Aiming at the addition of dynamic customer points and dividing the distribution operation time into time periods, a multi - period static summation model is proposed. Theoretical analysis shows that through the decomposition of time periods, the more the number of cycles (P value), the closer the model solution result is to real - time decision - making; at the same time, the shorter the time period within a cycle, the faster the response speed, but it may bring additional operating costs. Therefore, in order to further verify the performance of this model under multi - period conditions, the following experiment is designed: Taking the cycle P = 10 as a reference benchmark, 18 large - scale instances in 6 types of cases are selected, and the objective function values when P = 1, 2, 5, 10, 15, 20 are solved respectively to analyze the influence of different P - value changes on the objective function value. The experimental results are shown in Table 3.

[0062] Table 3 Table of Function Value Changes with Cycle Changes It can be seen from Table 3 that when the cycle P = 5, the objective function value reaches the minimum, indicating the best operating efficiency at this time; however, when the P value exceeds 15, the objective function value gradually increases, indicating that the efficiency begins to decline; at the same time, when the P value is less than 5 and gradually decreases, the objective function value also shows an increasing trend, indicating that too short a cycle leads to over - response, thus increasing the operating cost. This shows that for the C - DVRPTW (Dynamic Vehicle Routing Problem with Time Windows) model, there is not a simple monotonic relationship between the objective function value and the number of cycles, but there is a balance point, and a trade - off is needed between efficiency and response speed. Although accelerating the response speed can improve the service level, too frequent adjustments may increase the operating cost and thus reduce the overall efficiency. Therefore, in practical applications, the cycle P value needs to be reasonably selected according to the operating scenario and demand dynamics to achieve the optimal balance between efficiency and response speed.

[0063] An embodiment of the present invention proposes a dynamic demand cold chain transportation route planning method based on an improved ALNS algorithm. For the first time, the cold chain path optimization problem under dynamic demand is reformulated as a "multi-period static summation model" (C-DVRPTW). Through dynamic time period division, soft time window mechanism and rolling optimization strategy, an accurate modeling of the dynamic cold chain logistics path problem is achieved under multiple constraints such as comprehensive operating costs, time window violation penalties, and carbon emissions. An improved adaptive large neighborhood search (IALNS) algorithm is proposed, which integrates a dynamic insertion mechanism, a simulated annealing strategy, and a multi-time period rolling optimization method. Two novel operators, namely inefficient customer removal and cost reduction route removal, are designed, significantly improving the global search ability and solution efficiency of the algorithm. Experimental results show that IALNS is superior to the traditional solver CPLEX in small-scale examples. Compared with ALNS in large-scale examples, the distribution cost is reduced by an average of 4.21%, and the maximum reduction is 16.7%. It also shows significant advantages in terms of objective function value, convergence speed, and path optimization effect. The benefit analysis of the multi-period static summation model shows that the best operating benefit is achieved when the number of periods P = 5, achieving a balance between benefit and response speed. The present invention provides theoretical support and practical reference for the dynamic cold chain path optimization problem, and has important academic value and application prospects for the green and sustainable development of cold chain transportation enterprises. In the future, it can be further combined with real-time traffic information and multi-agent collaborative optimization technology to expand the applicability of the model in more complex scenarios.

[0064] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method part.

[0065] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined in the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown in the present invention, but will conform to the widest scope consistent with the principles and novel features disclosed in the present invention.

Claims

1. A dynamic demand cold chain transportation path planning method based on improved ALNS algorithm, characterized in that: include: Step 1: Convert the DVRPTW problem of cold chain transportation path planning into a C-DVRPTW problem, and construct a mathematical model of the C-DVRPTW problem based on mixed integer programming, which includes fixed cost, transportation cost, refrigeration cost, cargo damage cost and time window penalty cost, and aims to minimize operating cost; Step 2: Utilize the fusion of dynamic insertion strategy, simulated annealing strategy and multi-time period rolling optimization strategy, and design an improved ALNS algorithm with two novel operators: inefficient customer removal and cost reduction route removal to solve the mathematical model and obtain the optimal cold chain transportation route.

2. According to claim 1, a dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm is characterized in that: In step 1, the DVRPTW problem of cold chain transportation path planning is converted into a C-DVRPTW problem, specifically: The distribution time range [0, T] is divided into multiple uniform time periods, and the cold chain distribution demand in each time period is analyzed, and each time period is treated as a static problem; After completing the delivery in the previous time period, the current location of the vehicle is regarded as a temporary distribution center, and the delivery route for the next cycle is generated in combination with the newly added customer needs, thereby realizing the optimization of the multi-cycle static summation process.

3. According to claim 1, a dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm is characterized in that: In step 1, the mathematical model of the C-DVRPTW problem is as follows: Among them, Z is the operating cost; P is the set of all time periods; is the fixed cost in the pth time period; is the transportation cost in the pth time period; is the cooling cost in the pth time period; is the cargo damage cost in the pth time period; is the time window penalty cost for the p-th time period; is the carbon tax cost in the pth time period; n is the number of cold chain delivery vehicles required in the pth time period; is the fixed cost of a unit cold chain distribution vehicle; M is the customer point set; is the driving distance from customer point i to customer point j; The transportation cost per unit distance of an empty cold chain distribution vehicle; is the loading rate of cold chain delivery vehicles in the pth time period; The transportation cost per unit distance of a fully loaded cold chain distribution vehicle; is a decision variable, which means that if the cold chain delivery vehicle travels from customer point i to customer point j within time period p, it is 1, otherwise it is 0; The transport refrigeration cost per unit time when the cold chain distribution vehicle is loaded; is the load from customer point i to customer point j; is the transportation time from customer point i to customer point j; The refrigeration cost per unit time during the loading and unloading process of cold chain distribution vehicles; is a decision variable, which means that if the cold chain delivery vehicle loads or unloads at customer point i within time period p, it is 1, otherwise it is 0; is the unit price of the goods; It is the freshness decay rate during the transportation process of cold chain distribution vehicles; is the freshness decay rate during the unloading process of the cold chain delivery vehicle; N is the set of cold chain delivery vehicles; is the penalty factor for early delivery; The expected service time for customer point i; is the time when the cold chain delivery vehicle k arrives at the customer point i in the pth time period; is the penalty factor for delayed delivery; is the carbon emission tax rate per unit distance.

4. According to claim 1, a dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm is characterized in that: In step 1, the constraints of the mathematical model are as follows: Among them, M is the customer point set; is the demand of customer point i; is a decision variable, which means that if customer point i is served by cold chain delivery vehicle k, it is 1, otherwise it is 0; N is the set of cold chain delivery vehicles.

5. According to claim 1, a dynamic demand cold chain transportation path planning method based on improved ALNS algorithm is characterized in that: In step 2, the improved ALNS algorithm is specifically: Generate an initial solution using a heuristic algorithm, and perform operator deletion and operator insertion operations in sequence according to preset deletion rules and preset insertion rules; wherein the preset deletion rules include: random removal, random route removal, worst removal, Shaw removal, inefficient customer removal, and cost reduction route removal; the preset insertion rules include: greedy insertion and Regret-3 insertion; The temperature cooling mechanism based on simulated annealing updates the current solution and the optimal solution. After completing the path optimization in each time period, the newly added demand and the adjusted customer points are used as the initial path points for the next time period to meet the dynamic needs of customers. When the maximum number of iterations is reached or the algorithm converges, the current optimal solution is output as the final path planning solution.

6. According to claim 5, a dynamic demand cold chain transportation path planning method based on an improved ALNS algorithm is characterized in that: The initial solution is generated using a heuristic algorithm, specifically: Set the input variables to 0, sort the customer points from large to small according to their needs, and divide the customer points into a certain warehouse service point according to the distance; According to the order of distance arrangement, the customer points in the initial period are added to the initial transportation route of a vehicle. The sum of the weights of the nodes is less than or equal to the load capacity of the vehicle. The initial transportation route of the vehicle is obtained, and the decision variables are obtained. , The value of ; where It means that if the cold chain delivery vehicle loads or unloads at customer point i within time period p, it is 1, otherwise it is 0; Indicates that if customer point i is served by cold chain delivery vehicle k, it is 1, otherwise it is 0; According to the arrangement of the initial nodes, the transportation route array is obtained , get the position where the new customer point can be inserted, and delete the insertable position that does not meet the constraints according to the time window and vehicle capacity constraints, and compare the cost increment after the node in the insertable position is added, and take the minimum value to get the position of the inserted node; The input of the heuristic algorithm is a set of customer points, and the output is an initial feasible solution of C-DVRPTW.

7. According to claim 5, a dynamic demand cold chain transportation path planning method based on improved ALNS algorithm is characterized in that: The temperature cooling mechanism based on simulated annealing updates the current solution and the optimal solution, specifically: Operator weight update: In the adaptive algorithm iteration, the operator weight update is dynamically applied to the periodic probability adjustment of the deletion and insertion operators. After the current cycle is completed and before the next cycle iteration begins, the operator weight Updated to ; Among them, r is the weight forgetting coefficient; is the cumulative number of times the operator is used in this cycle i; is the cumulative score of operator i in this cycle; the cumulative score depends on the quality of the new solution generated by operator i. The new solution has three conditions: better than the global optimal solution, better than the current solution, and worse than the current solution and will be updated; Temperature update: The temperature is updated in each round using the temperature cooling mechanism of simulated annealing.

Images