Refrigerator car-unmanned aerial vehicle-rider collaborative path planning method considering fresh loss

Abstract

The invention discloses a refrigerator car-unmanned aerial vehicle-rider collaborative path planning method considering fresh loss, and relates to the technical field of path planning, and the method comprises the steps: defining path planning parameters of a refrigerator car, an unmanned aerial vehicle and a rider considering fresh loss; constructing a path planning model according to the path planning parameters; generating an initial solution of the path planning model based on a greedy insertion algorithm; performing local search improvement on the initial solution to obtain a target solution; and according to the target solution, the refrigerator car, the unmanned aerial vehicle and the rider are dispatched to cooperatively execute the fresh food delivery service. According to the method, the path planning model considering the fresh loss is constructed, the initial solution is generated, the target solution is searched, the delivery paths of the refrigerator car, the rider and the unmanned aerial vehicle can be synchronously decided, and the arrival freshness of fresh products can be improved and stabilized.

Description

Technical Field

[0001] This application relates to the field of route planning technology, and in particular to a collaborative route planning method for refrigerated trucks, drones, and riders that takes into account freshness loss. Background Technology

[0002] With the rapid development of e-commerce and new retail, the demand for instant delivery of fresh produce has increased dramatically. Fresh produce (such as vegetables, fruits, seafood, and meat) has the biological characteristics of being perishable, fragile, and having a short shelf life; its quality deteriorates over time during distribution. The fresh produce supply chain plays a crucial role in preserving the quality of perishable products.

[0003] However, current fresh food delivery services generally do not consider the freshness of the products, resulting in inconsistent freshness of the products received by users. Summary of the Invention

[0004] In view of this, embodiments of this application provide a refrigerated truck-drone-rider collaborative route planning method and related equipment that take into account freshness loss, so as to improve and stabilize the freshness of fresh products upon delivery.

[0005] One aspect of this application provides a collaborative path planning method for refrigerated truck-drone-rider considering freshness loss, the method comprising the following steps:

[0006] Define route planning parameters for refrigerated trucks, drones, and riders that take into account freshness loss;

[0007] Construct a path planning model based on the path planning parameters;

[0008] The initial solution of the path planning model is generated based on the greedy insertion algorithm;

[0009] The initial solution is improved by local search to obtain the target solution;

[0010] Based on the target solution, refrigerated trucks, drones, and riders are coordinated to carry out fresh food delivery services.

[0011] In some embodiments, the definition of route planning parameters for refrigerated trucks, drones, and riders that take into account freshness loss includes the following steps:

[0012] Constructing a directed graph And define the following parameters:

[0013] Fresh food delivery service providers have a central warehouse and its virtual endpoint A group of forward warehouses located near customer locations and its virtual endpoint The service provider has a set of orders, each with a customer's unloading location. Time window The time window includes the earliest arrival time. and latest arrival time Service hours and package weight ,in An index for customer collections; the central warehouse is equipped with a set of refrigerated trucks. and a group of drones Each front warehouse Equipped with a corresponding group of riders The rider team is Pre-positioned warehouse The location and quantity are predetermined;

[0014] They represent refrigerated trucks. rider and drones The maximum permissible payload; define the straight-line flight of the UAV, and calculate the Euclidean distance between any two nodes while the UAV is flying. ; Calculate the Manhattan distance as the refrigerated truck and rider travel. Assume the daily arrival time of fresh produce at the central warehouse is denoted as... Let α1 be the freshness loss rate of fresh products stored in central warehouses and forward warehouses, α2 be the freshness loss rate of fresh products stored in refrigerated trucks, and α3 be the freshness loss rate of fresh products stored in drones and riders' insulated boxes, where α1 < α2 < α3.

[0015] In some embodiments, constructing a path planning model based on the path planning parameters includes the following steps:

[0016] The objective function for constructing the path planning model includes:

[0017] ;

[0018] ;

[0019] ;

[0020] ;

[0021] ;

[0022] ;

[0023] Where min is the minimization function, and total cost Including refrigerated trucks rider and drones Transportation costs and freshness loss .

[0024] In some embodiments, generating an initial solution for the path planning model based on a greedy insertion algorithm includes the following steps:

[0025] Construct the first-gradient route, set the planned route using all refrigerated trucks according to the route planning model, and add the forward warehouse delivery task to each refrigerated truck in sequence;

[0026] Initialize the available drone and rider routes, insert each customer to be delivered into a location in the drone or rider route with the lowest insertion cost, thereby obtaining the initial solution; wherein, before calculating the current insertion position, load constraints, drone maximum flight distance constraints, and time window constraints are satisfied.

[0027] In some embodiments, the step of performing local search improvement on the initial solution to obtain the target solution includes the following steps:

[0028] Based on the path planning model, four removal operators and four repair operators are constructed; wherein, the removal operators include random removal operator, worst-cost removal operator, worst-freshness removal operator, and random front-end removal operator; the repair operators include random insertion operator, greedy insertion operator, regret value insertion operator, and freshness-based greedy insertion operator;

[0029] Based on the accumulated feature values, a logistic regression model is used to predict the selection probability of each operator, and the selection probability is used as the adaptive weight of the corresponding operator. The feature values ​​include solution state features, operator features, and stage features. The solution state features include the optimal solution improvement value, the current solution improvement value, route similarity, the number of times the optimal solution has been improved, and the number of times the current solution has been improved. The operator feature is the feasibility rate. The stage feature is the number of iterations.

[0030] The roulette wheel algorithm is used to select the removal and repair operators for each iteration based on the adaptive weights.

[0031] The local search is improved by using the selected removal and repair operators to obtain updated routes and feature values;

[0032] The above steps are iterated continuously until the preset termination condition is reached, and the target solution is obtained.

[0033] In some embodiments, the logistic regression model is trained through the following steps:

[0034] A static logistic regression model is trained in advance using training examples of different sizes and corresponding complete search trajectory data.

[0035] Alternatively, the initial parameters of the logistic regression model can be set to zero or a preset prior value, and the parameters of the logistic regression model can be updated periodically based on the accumulated feature value data increments during the process of solving the target solution.

[0036] Another aspect of this application embodiment provides a refrigerated truck-drone-rider collaborative route planning device that takes into account freshness loss, the device comprising:

[0037] The parameter definition unit is used to define the route planning parameters for refrigerated trucks, drones, and riders, taking into account freshness loss.

[0038] A model building unit is used to build a path planning model based on the path planning parameters;

[0039] An initial solution generation unit is used to generate an initial solution for the path planning model based on a greedy insertion algorithm.

[0040] The target solution search unit is used to perform local search improvement on the initial solution to obtain the target solution;

[0041] The delivery scheduling unit is used to coordinate refrigerated trucks, drones, and riders to perform fresh food delivery services according to the target.

[0042] Another aspect of this application embodiment provides an electronic device, including a processor and a memory;

[0043] The memory is used to store programs;

[0044] The processor executes the program to implement any of the methods described above.

[0045] Another aspect of this application provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described in any of the above embodiments.

[0046] This application includes at least the following beneficial effects:

[0047] This application defines route planning parameters for refrigerated trucks, drones, and delivery riders, taking into account freshness loss; constructs a route planning model based on these parameters; generates an initial solution for the route planning model using a greedy insertion algorithm; improves the initial solution through local search to obtain a target solution; and schedules refrigerated trucks, drones, and delivery riders to collaboratively execute fresh produce delivery services based on the target solution. By constructing a route planning model that considers freshness loss, generating an initial solution, and searching for a target solution, this application can simultaneously determine the delivery routes of refrigerated trucks, delivery riders, and drones, thereby improving and stabilizing the arrival freshness of fresh produce. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 A flowchart illustrating the refrigerated truck-drone-rider collaborative path planning method considering freshness loss, provided for an embodiment of this application;

[0050] Figure 2 A schematic diagram of a collaborative delivery model involving refrigerated trucks, drones, and delivery riders provided in an embodiment of this application;

[0051] Figure 3 This is a schematic diagram of the LR-ALNS process using an offline learning method provided in an embodiment of this application;

[0052] Figure 4 A schematic diagram of the LR-ALNS process using an online learning method is provided for an embodiment of this application;

[0053] Figure 5 This is a schematic diagram illustrating the changes in the freshness loss of a product as provided in an embodiment of this application.

[0054] Figure 6 Example diagram of operator working logic provided in the embodiments of this application;

[0055] Figure 7 A flowchart of the LR-ALNS algorithm based on an offline learning method is provided for embodiments of this application;

[0056] Figure 8 A flowchart of the LR-ALNS algorithm based on an online learning method is provided for embodiments of this application;

[0057] Figure 9 A diagram illustrating the offline and online learning mechanisms of the logistic regression algorithm provided in this application embodiment;

[0058] Figure 10 A structural block diagram of a refrigerated truck-drone-rider collaborative path planning device that takes into account freshness loss, provided in an embodiment of this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] Before providing a detailed description of the embodiments of this application, some related technologies involved in the embodiments of this application will be described first, as follows:

[0061] Terminology Explanation:

[0062] The low-altitude economy is a comprehensive economic form that drives the integrated development of related fields through various low-altitude flight activities of manned and unmanned aircraft. Its related products mainly include drones, eVTOL (electric vertical takeoff and landing aircraft), helicopters, and traditional fixed-wing aircraft, involving both residential consumption and industrial applications.

[0063] NP-hard: Non-deterministic Polynomial, a class of problems to which all NP problems can be reduced in polynomial time.

[0064] The relevant technical details are as follows:

[0065] As a newly developed product in recent years, drones have moved from concept to reality in the fresh produce supply chain, enabling lower economic and environmental costs. Drones offer the advantage of direct delivery, improving efficiency compared to ground transportation by providing faster, more direct routes and reducing reliance on road infrastructure. This is particularly ideal for the logistics and distribution of perishable foods such as fresh produce. Drones also provide significant flexibility in leveraging the opportunities presented by collaborative transportation. However, due to technological bottlenecks such as limited flight distance and insufficient carrying capacity, the service coverage of drone delivery is limited. Multimodal transport, on the other hand, demonstrates significant advantages throughout the transportation chain. By introducing different modes of transport such as trucks, riders, and robots, it can utilize the strengths of different modes and compensate for their weaknesses, thereby improving the efficiency, reliability, and sustainability of the supply chain. Therefore, utilizing drones in conjunction with traditional delivery methods is a key research direction in modern logistics networks.

[0066] Drone-vehicle collaborative scheduling technology is a key technological means to realize collaborative delivery models and solve delivery problems. Commonly used technologies include exact algorithms, heuristic algorithms, metaheuristic algorithms, and machine learning algorithms. Currently, heuristic and metaheuristic algorithms are widely used. However, research on machine learning algorithms, such as deep learning, reinforcement learning, and meta-learning, in drone-based collaborative delivery is relatively limited. Nevertheless, they have significant advantages in handling complex problems, dynamic environments, and large-scale datasets, thus becoming an important direction for future research.

[0067] Currently, most existing technologies in the fresh food delivery industry only consider the collaborative delivery of refrigerated trucks and riders. After consumers place orders through digital platforms, refrigerated trucks are responsible for transporting goods from a central cold storage facility to pre-positioned warehouses located within a 2-3 kilometer radius of the consumer, with riders ultimately completing the last-mile delivery. However, this system fails to effectively address the issues of idle capacity and congestion caused by order demand in areas not covered by pre-positioned warehouses and the uneven distribution of orders. Therefore, the introduction of drone technology can help cope with the spatially dispersed and temporally fluctuating nature of fresh food orders.

[0068] Existing technologies for drone delivery include both independent delivery and drone-vehicle collaborative delivery. Current technology indicates that independent drone delivery is somewhat idealistic, limited by payload capacity and delivery range, making its independent deployment in large-scale delivery systems impractical. Therefore, drone-vehicle collaborative delivery is a current technological direction for improving delivery efficiency. Considering the parallel delivery of drones with traditional transportation methods such as refrigerated trucks and riders can help improve delivery efficiency and reduce freshness loss.

[0069] Heuristic and metaheuristic algorithms can efficiently solve NP-hard problems. While they may not always find the optimal solution, they are superior in terms of efficiency. However, metaheuristic algorithms lack sufficient exploitation of explicit information during the search process, thus limiting their search efficiency. Machine learning, on the other hand, relies on algorithms learning from experience to gradually improve their performance on specific tasks.

[0070] Based on this, the key technical problems that this application needs to solve are as follows: how to simultaneously decide the delivery routes of refrigerated trucks, riders and drones for fresh food delivery with high freshness requirements, and to use learning-based heuristic algorithms to explore the solution efficiency.

[0071] Reference Figure 1 This application provides a refrigerated truck-drone-rider collaborative path planning method that takes into account freshness loss, specifically including the following steps S100~S140:

[0072] S100: Defines the route planning parameters for refrigerated trucks, drones, and riders, taking into account freshness loss;

[0073] S110: Construct a path planning model based on the path planning parameters;

[0074] S120: Generate the initial solution of the path planning model based on the greedy insertion algorithm;

[0075] S130: Perform a local search improvement on the initial solution to obtain the target solution;

[0076] S140: Based on the target solution, refrigerated trucks, drones, and riders are coordinated to perform fresh food delivery services.

[0077] The following section will provide a detailed introduction and explanation of the solutions in the embodiments of this application, using specific application examples.

[0078] I. Overview of the Plan.

[0079] First, consider a fresh food logistics network equipped with a central warehouse and multiple forward warehouses. The central warehouse has a fleet of refrigerated trucks and a fleet of drones, while each forward warehouse has a corresponding team of riders. The problem mainly involves two delivery modes: The first mode is a two-tiered logistics fleet, where refrigerated trucks deliver goods from the central warehouse to a group of forward warehouses and then return to the central warehouse. Riders then depart from the forward warehouses for delivery and return to their original forward warehouses, or drones pick up goods from the forward warehouses for delivery. The second mode is independent drone delivery, where drones depart from the central warehouse for delivery and ultimately return to the central warehouse (see details). Figure 2 ).

[0080] Therefore, we designed a collaborative delivery model for refrigerated trucks, drones, and riders based on the vehicle routing problem. Key considerations included capacity limitations for refrigerated trucks, drones, and riders; the maximum flight distance of drones; and time alignment constraints for collaboration among refrigerated trucks, drones, and riders. Furthermore, we considered the changes in freshness loss at each delivery stage to obtain the task and routing decisions for this routing problem.

[0081] Finally, a learning-based adaptive large neighborhood search algorithm was designed. This algorithm obtains initial path plans through a greedy insertion algorithm. Then, it learns from historical search data using logistic regression to dynamically predict the performance of different operators at specific stages. A roulette wheel algorithm is then used to select the remove and repair operators for each iteration, thereby accelerating the search for better collaborative path planning schemes for refrigerated trucks, drones, and riders. We experimented with both offline and online learning settings; the technology roadmaps are shown below. Figure 3 and Figure 4 .

[0082] II. Model Preparation.

[0083] 1. Model description.

[0084] We assume a fresh produce distribution company delivers fresh products to customers and there are no stockouts. We are using a directed graph... The above defines the problem: a fresh food delivery service provider has a central warehouse. and its virtual endpoint A group of forward warehouses located near customer locations and its virtual endpoint The service provider has a set of orders, each with a customer's unloading location. Time window (earliest arrival time) and latest arrival time , Index for customer collection), service time ( (Index for the customer set) and package weight ( (Index for customer collections). The central warehouse has a fleet of refrigerated trucks. and a group of drones Each front warehouse Equipped with a corresponding group of riders Therefore, the rider team is The location and number of these facilities are predetermined. To ensure a viable solution is obtained, we assume that there are a sufficient number of refrigerated vehicles in each instance. They represent refrigerated trucks. rider and drones The maximum permissible payload. We assume the UAV flies in a straight line, and calculate the Euclidean distance between any two nodes while the UAV is in motion. (E is an abbreviation for Euclidean distance); Calculate the Manhattan distance as the refrigerated truck and rider travel. (M is an abbreviation for Manhattan distance). Let the daily arrival time of fresh produce at the central warehouse be denoted as... They believe that fresh produce is completely fresh at this point (e.g., Figure 5 (As shown). We determine the penalty cost by calculating the freshness loss at different stages to ensure the freshness and quality of the product. Freshness loss comes from the following three stages:

[0085] (1) For the storage of fresh products, large-scale refrigeration equipment will be used in the warehouse to ensure the freshness of the products. Therefore, the central warehouse and the forward warehouse are equipped with the same preservation facilities, and the freshness loss is minimized. The freshness loss rate is set as α1.

[0086] (2) Due to space and cost limitations, the refrigeration equipment of the refrigerated trucks used is less effective than the preservation facilities of warehouses. The freshness loss rate is set as α2.

[0087] (3) The preservation facilities for drones and riders are usually insulated boxes, and their preservation measures are far less effective than those of cold storage warehouses and refrigerated trucks. Therefore, the freshness loss is the highest, and the freshness loss rate is set as α3.

[0088] 2. Model building.

[0089] Table 1 Model Parameter Table

[0090]

[0091]

[0092] Specifically, the nonlinear mixed-integer programming model considering freshness loss is as follows:

[0093] Objective function:

[0094] (1)

[0095] (2)

[0096] (3)

[0097] (4)

[0098] (5)

[0099] (6)

[0100] Where min is the minimization function, and total cost Refrigerated trucks were considered. rider and drones Transportation costs and freshness loss As shown in equation (1). Transportation costs This is expressed as the sum of the unit distance cost multiplied by the travel distance, as shown in equation (2). In equation (3), freshness loss... The freshness loss rate multiplied by the delivery time at each stage. The summation is expressed by equations (4) to (5).

[0101] (7)

[0102] (8)

[0103] (9)

[0104] (10)

[0105] (11)

[0106] (12)

[0107] (13)

[0108] (14)

[0109] (15)

[0110] Among them, constraint (7) stipulates that any customer must be served exactly once by either a drone or a refrigerated truck-rider fleet; constraints (8) to (10) define that the refrigerated truck, drone, and rider complete a round trip between their respective origin and destination; constraint (11) ensures that riders only depart from their home warehouse; constraints (12) to (14) are flow balancing constraints to ensure that the route is within the node. The front warehouse is continuous; constraint (15) ensures that each front warehouse can be accessed by at most one vehicle.

[0111] (16)

[0112] (17)

[0113] (18)

[0114] (19)

[0115] (20)

[0116] (twenty one)

[0117] (twenty two)

[0118] (twenty three)

[0119] (twenty four)

[0120] (25)

[0121] Among them, constraints (16) and (17) require that if a customer's package from the central warehouse is transported to the forward warehouse by a refrigerated truck, the truck must visit both the central warehouse and the forward warehouse; constraint (18) connects the two-level fleet routes and stipulates that a customer's package delivered to the forward warehouse can only be received by one rider or one drone to complete the second-tier delivery; constraints (19) and (20) require that if a customer's package from the forward warehouse is received by a rider, the rider must visit both the forward warehouse and the customer node; constraints (21) and (22) require that if a customer's package from the central warehouse is received by a drone, the drone must visit both the central warehouse and the customer node; constraints (23) to (25) require that if a customer's package from the forward warehouse is received by a drone, the drone must visit the central warehouse, the forward warehouse, and the customer node in sequence.

[0122] (26)

[0123] (27)

[0124] (28)

[0125] (29)

[0126] (30)

[0127] (31)

[0128] Among them, constraint (26) is the capacity constraint of the refrigerated truck, which stipulates that the total amount of goods transported shall not exceed its maximum capacity; constraint (27) is also the capacity constraint of the rider; constraints (28) to (31) together constitute the capacity constraint of the drone, where constraint (28) limits the initial load of the drone when it departs from the central warehouse, and constraint (29) stipulates that if the drone passes through the front warehouse node Receiving a customer's package increases the load capacity, and the increase is equal to the customer's package demand. Constraint (30) stipulates that if the drone passes through a customer node... When delivery is completed, the load is reduced by the amount of the customer's package demand. Constraint (31) ensures that the real-time load of the UAV in any origin-destination pair does not exceed its maximum capacity during the current scheduling cycle.

[0129] (32)

[0130] Among them, constraint (32) considers the maximum straight-line flight distance of the drone. We assume that any drone has zero power consumption when leaving the drone dispatch center, that is, the drone must leave with a full charge.

[0131] (33)

[0132] (34)

[0133] (35)

[0134] (36)

[0135] (37)

[0136] (38)

[0137] (39)

[0138] (40)

[0139] (41)

[0140] (42)

[0141] (43)

[0142] Among them, constraint (33) ensures that if a refrigerated truck departs from the central warehouse, its departure time must not be earlier than the time when the fresh produce arrives at the central warehouse, otherwise it is 0; constraint (34) stipulates that if the arc On the refrigerated truck route, the vehicle arrives at the node. The time is no less than its arrival node Timing and travel arc The sum of required times; constraint (35) limits the time if the vehicle does not pass through the node. If the corresponding variable is 0, then constraint (36) ensures that the time when the rider picks up the customer's package at the pre-positioning warehouse is no earlier than the time when the refrigerated truck arrives at the pre-positioning warehouse; constraint (37) stipulates that if the arc If the rider is located on the rider's path, then the rider has arrived at the node. The time is no less than its arrival node Time, travel arc Time and service hours The sum; constraint (38) forces riders to reach customer nodes. The time falls within the delivery time window specified by the node; constraints (39) and (33) correspond to the same earliest departure time for the UAV; constraints (40)-(43) correspond to constraints (36)-(38) respectively, applying node access sequence, first-to-last delivery and time window constraints to the UAV path.

[0143] 3. Solution algorithm.

[0144] This problem is NP-hard; exact algorithms, dynamic programming, or integer programming can find the optimal solution, but the complexity increases exponentially with the problem size, making it infeasible to obtain the global optimum within a reasonable time. Therefore, efficient solutions to this problem rely on heuristic or metaheuristic methods. The algorithm designed in this embodiment is an adaptive large neighborhood search algorithm (LR-ALNS) that incorporates a learning mechanism. By learning from historical search trajectories, it adaptively adjusts operator selection and search strategies, thereby dynamically narrowing the solution space, guiding the search direction, significantly improving solution efficiency, and obtaining a better solution.

[0145] First, this embodiment generates an initial feasible solution based on a greedy insertion algorithm. The specific steps are as follows:

[0146] (1) Construct the first gradient route. Considering the characteristics of the model, the vehicle route requires the use of all refrigerated trucks. Add the front warehouse delivery task to each truck in sequence.

[0147] (2) Initialize the available drone and rider routes. The algorithm attempts to insert each customer into a location on the drone or rider's current route with the lowest insertion cost. Before calculating the current insertion location, load limits, drone maximum flight distance limits, and time window limits must be met.

[0148] Next, after obtaining the initial solution, a local search is performed to improve the current initial solution. Based on the specific characteristics of the model, this embodiment develops four removal operators and four repair operators related to the nature of the problem. A logistic regression model is used to predict the selection probability of each operator and uses this as an adaptive weight for each operator. Then, a roulette wheel algorithm is used to select the removal and repair operators used in each iteration. This process is iterated until a preset termination criterion is met.

[0149] For the logistic regression model, we experimented with both offline and online learning frameworks (see details). Figure 6 ):

[0150] (1) Offline learning pre-trains a static logistic regression model using training examples of varying sizes and their corresponding complete search trajectory data. This model completes parameter learning before the algorithm runs, saving training time each time a problem is solved. This means that the data collection and training phases can be completed in advance. The flowchart of the LR-ALNS algorithm based on offline learning is shown below. Figure 7 As shown.

[0151] (2) Unlike traditional offline learning that relies on static predefined datasets, online learning mechanisms generate training data in real time during the algorithm's solution process. The model is initially empty or contains only weak prior information. During the search process, the model parameters are periodically updated incrementally based on accumulated experience data, allowing the model to continuously improve its understanding of "which operators are most effective in a specific search state." The flowchart of the LR-ALNS algorithm based on online learning is shown below. Figure 8 As shown.

[0152] Referring to Table 2, this embodiment prioritizes general characteristics independent of problem type to enhance the adaptability and transferability of the algorithm across multiple optimization problems.

[0153] Table 2 Features used in the LR-ALNS algorithm

[0154]

[0155] The working logic of the operator designed in this embodiment is as follows (see example diagram for details). Figure 9 ):

[0156] (1) Remove operator.

[0157] Random Removal Operator: This process randomly removes n customers from the current solution. The removal rules of the random removal operator are relatively simple, which can diversify the neighborhood search and make it easier to escape local optima, but it has greater uncertainty and may produce worse solutions.

[0158] Worst-Cost Removal Operator: This operator selects the customer with the highest cost. First, the removal cost for each customer is calculated as the difference between the target value for routes with and without the customer. Then, n customers are removed in descending order of their removal costs. Note that the removal cost may change after some customers are removed. Therefore, when a customer is removed, the removal cost for the remaining customers is recalculated.

[0159] The worst freshness removal operator: calculates the freshness loss of each customer on the current route and sorts them, then selects the n customers with the largest freshness loss to remove.

[0160] Random Front-End Warehouse Removal Operator: Given the unique architecture of this model, although the fleet routes in the first tier do not directly deliver to customer nodes, they still affect customer deliveries. We designed a random front-end warehouse removal operator to randomly select n front-end warehouses on the refrigerated truck routes for removal, allowing redundant front-end warehouses to be removed through this operator. It is important to note that when a front-end warehouse node on a refrigerated truck route is removed, it will affect customer goods transported by riders and drones at that front-end warehouse node. Therefore, the route planning for riders and drones also needs to be adjusted synchronously to remove customer transport nodes affected by the front-end warehouse nodes.

[0161] (2) Repair operator.

[0162] Random Insertion Operator: This operator includes insertions for removed customers and inactive front-end warehouses. First, n inactive front-end warehouse nodes are randomly selected, prioritizing insertion into refrigerated truck routes without delivery tasks. Then, the set of removed customers is randomly selected sequentially for insertion into rider or drone routes. Furthermore, drone delivery is flexible: on one hand, it can pick up goods directly from the central warehouse; on the other hand, depending on actual needs, it can also retrieve transshipment goods from front-end warehouses before delivery. Therefore, front-end warehouse nodes that meet specific insertion rules are allowed to insert drone routes with a certain probability to explore potential transshipment and delivery needs for drones.

[0163] Greedy insertion operator: Calculate the insertion cost of all possible insertion positions for each removed customer, insert the customer with the lowest insertion cost into the corresponding node, then recalculate the insertion cost of the remaining customers, and repeat the above process until all removed customers are inserted into the route.

[0164] Regret-based insertion operator: Calculates the position with the lowest insertion cost and the second lowest insertion cost for each customer; the difference is defined as the customer's regret value. The operator then finds the customer with the highest regret value and inserts them into the position with the lowest cost. This process is repeated until all customer insertion routes have been removed.

[0165] Freshness-based greedy insertion operator: For each uninserted customer, calculate the freshness loss for insertion into all possible positions, and select the customer with the least penalty for insertion. Repeat the above process until all removed customers are inserted into the insertion route.

[0166] In summary, this embodiment includes the following key technical solutions:

[0167] 1. Model Structure. Based on the background of fresh food logistics where product quality loss is severe, this embodiment designs a nonlinear mixed integer programming model that considers the loss of freshness for refrigerated trucks, drones, and riders. This model can effectively address the high-quality requirements of fresh products and explore the path planning for collaborative delivery by refrigerated trucks, drones, and riders.

[0168] 2. Design of a learning-based adaptive large neighborhood search algorithm. The algorithm designed in this embodiment can find the global optimum in a runtime far faster than that of commercial solvers. It is applicable not only to the model proposed in this embodiment but also to all VRP problems and their variants.

[0169] It should be noted that, provided the model constraints are met, different machine learning algorithms and operators can be used to implement the algorithm iteration in this embodiment.

[0170] This embodiment addresses the common freshness loss characteristics in the fresh food delivery industry, considering the collaborative delivery route optimization problem between traditional delivery methods (refrigerated trucks and delivery riders) and emerging delivery technologies (drones), and establishes a model for research, including the following beneficial effects:

[0171] (1) The model in this embodiment is based on the current situation of the fresh food delivery industry. The model is a hierarchical cold chain delivery system centered on three core components: central cold storage, forward warehouse and last delivery unit. It innovatively considers the collaborative delivery of refrigerated trucks, riders and drones. This delivery mode can effectively improve delivery efficiency and product quality.

[0172] (2) This embodiment takes into account the pain points of the real fresh food e-commerce industry delivery. By measuring the freshness loss rate of different delivery links, it explores the interaction between preservation measures and transportation scheduling, and provides a solution for enterprises to balance cost and product quality.

[0173] (3) For the model in this embodiment, the learning-based heuristic algorithm provides more accurate and adaptive guidance for the search strategy by continuously utilizing new data. This algorithm can find the global optimum in a runtime that is much faster than that of the commercial solver CPLEX.

[0174] One possible specific embodiment is as follows:

[0175] The dataset used in this implementation consists of a set of randomly generated test instances based on the nature of the problem. Each warehouse and customer is located within a square of 10 km × 10 km / 12 km × 12 km / 20 km × 20 km. Refrigerated trucks in urban areas have a speed of 30 km / h, a capacity of 650 kg, and a unit distance cost of 5.67 yuan / km. Drones have a capacity of 2.5 kg and a maximum flight distance of 10 km. The average speed of drones is 40 km / h, and the unit distance cost is 0.8 yuan / km. Riders have an average speed of 20 km / h, a capacity of 15 kg, and a unit distance cost of 1.45 yuan / km. We set the freshness loss rate α2 for fresh products during refrigerated truck transportation to 0.1, the freshness loss rate α1 in warehouses to 0.05, and the freshness loss rate α3 for rider and drone delivery to 0.5. The algorithm parameters mainly consist of parameters from the adaptive large neighborhood search algorithm and the logistic regression algorithm, as shown in Table 3.

[0176] Table 3 Parameter values ​​used in the algorithm

[0177]

[0178] First, to verify the effectiveness of the LR-ALNS algorithm, the results were compared with those of the Cplex solver and the standard Adaptive Large Neighborhood Search (ALNS) algorithm. L-On S L-Off S C and S A T is the objective value solved by LR-ALNS (online and offline learning), Cplex, and ALNS. L-On T L-Off T C and T A These represent the solution times (in seconds) for LR-ALNS (online and offline learning), Cplex, and ALNS, respectively. Gap OO =(S L-On -S L-Off ) / S L-Off Gap LC =(S L-On -S C ) / S C and Gap LA =(S L-On -S A ) / S A The value represents the optimality gap, where a positive value indicates that the LR-ALNS algorithm found a better solution. We use abbreviations to represent instances, such as 2-2-2-6-8, which represents 2 transshipment warehouses, 2 refrigerated trucks, 2 drones, 6 riders, and 8 customers.

[0179] Table 4 Results for small-scale instances of LR-ALNS, ALNS, and the solver

[0180]

[0181] For small-scale problems, Table 4 reports the results. Regarding solution quality, both Cplex and LR-ALNS learning settings achieved the best known solutions across all small-scale problem instances. The ALNS algorithm performed worse than LR-ALNS in terms of solution quality across four examples, with an average difference of 3.11%. In terms of computation time, LR-ALNS used the least computation time for most instances. It was also observed that the solution time for both Cplex and ALNS algorithms varied significantly with the number of clients. As the scale increased, online learning took longer to solve than offline learning because offline learning did not have time for data collection and training phases.

[0182] Table 5 Results of LR-ALNS, Cplex, and ALNS in large-scale instances.

[0183]

[0184] For instances with 10 and 12 customers, CPLEX failed to find a feasible solution within the 7200-second time limit (in one case), and none of the solutions achieved optimality. As the problem size increased, CPLEX could not find a feasible solution for instances with 20 to 50 customers. In contrast, both LR-ALNS and ALNS algorithms found feasible solutions for large-scale problem instances. Observations revealed that the online-learning LR-ALNS algorithm achieved the best optimality: its average optimality difference compared to the offline-learning LR-ALNS and CPLEX was 0.65% and 3.94%, respectively, and its average difference compared to ALNS was 9.21%. This indicates that the learning algorithm provides effective guidance for the exploration process, and online learning performs better on large-scale instances.

[0185] Learning models are highly sensitive to input features; therefore, identifying and extracting the key information that has the greatest impact on the solution space is crucial for improving model prediction performance. We evaluated the LR-ALNS method with different feature subsets to determine the contribution of each feature to the algorithm's performance. Each column represents the feature set used, with the first option showing the result when all seven features are considered. The subsequent options show the result after removing a certain feature, and the average optimality gap (avgGap) between each column and the optimal solution was calculated, as shown in Table 6.

[0186] Table 6 compares the impact of different feature sets on learning.

[0187]

[0188] Considering all features outperforms all other subsets. Features related to the optimal solution perform best, with average differences of 5.42% and 4.06% between the improvement value and the number of improvements, respectively, predicting more promising neighborhood subsets in global search. Features related to the current solution show average differences of 4.06% and 4.38% between the improvement value and the number of improvements, respectively, finding a better solution at the current stage. The iteration count feature performs worst, with an average difference of 2.77%. Route similarity and feasibility also show some improvement, with average differences of 3.47% and 3.38%, respectively. We also evaluated some combinations that simultaneously remove route similarity, feasibility, and iteration count from the column, with average differences of 2.31%, 3.47%, and 2.58%, respectively. Generally, we observed more significant differences in large instances.

[0189] Reference Figure 10 This application provides a refrigerated truck-drone-rider collaborative path planning device that takes into account freshness loss, including:

[0190] The parameter definition unit is used to define the route planning parameters for refrigerated trucks, drones, and riders, taking into account freshness loss.

[0191] A model building unit is used to build a path planning model based on the path planning parameters;

[0192] An initial solution generation unit is used to generate an initial solution for the path planning model based on a greedy insertion algorithm.

[0193] The target solution search unit is used to perform local search improvement on the initial solution to obtain the target solution;

[0194] The delivery scheduling unit is used to coordinate refrigerated trucks, drones, and riders to perform fresh food delivery services according to the target.

[0195] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0196] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0197] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0198] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0199] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0200] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0201] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0202] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0203] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0204] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A collaborative path planning method for refrigerated trucks, drones, and riders considering freshness loss, characterized in that, The method includes the following steps: Define route planning parameters for refrigerated trucks, drones, and riders that take into account freshness loss; Construct a path planning model based on the path planning parameters; The initial solution of the path planning model is generated based on the greedy insertion algorithm; The initial solution is improved by local search to obtain the target solution; Based on the target solution, refrigerated trucks, drones, and riders are coordinated to carry out fresh food delivery services.

2. The refrigerated truck-drone-rider collaborative path planning method considering freshness loss according to claim 1, characterized in that, The definition takes into account the route planning parameters of refrigerated trucks, drones, and riders regarding freshness loss, and includes the following steps: Constructing a directed graph And define the following parameters: Fresh food delivery service providers have a central warehouse and its virtual endpoint A group of forward warehouses located near customer locations and its virtual endpoint The service provider has a set of orders, each with a customer's unloading location. Time window The time window includes the earliest arrival time. and latest arrival time Service hours and package weight ,in An index for customer collections; the central warehouse is equipped with a set of refrigerated trucks. and a group of drones Each front warehouse Equipped with a corresponding group of riders The rider team is Pre-positioned warehouse The location and quantity are predetermined; They represent refrigerated trucks. rider and drones The maximum permissible payload; define the UAV's straight-line flight path, and calculate the Euclidean distance between any two nodes while the UAV is in flight. ; Calculate the Manhattan distance as the refrigerated truck and rider travel. Assume the daily arrival time of fresh produce at the central warehouse is denoted as... Let α1 be the freshness loss rate of fresh products stored in central warehouses and forward warehouses, α2 be the freshness loss rate of fresh products stored in refrigerated trucks, and α3 be the freshness loss rate of fresh products stored in drones and riders' insulated boxes, where α1 < α2 < α3.

3. The refrigerated truck-drone-rider collaborative path planning method considering freshness loss according to claim 2, characterized in that, The process of constructing a path planning model based on the path planning parameters includes the following steps: The objective function for constructing the path planning model includes: ； ； ； ； ； ； Where min is the minimization function, and total cost Including refrigerated trucks rider and drones Transportation costs and freshness loss .

4. The refrigerated truck-drone-rider collaborative path planning method considering freshness loss according to claim 1, characterized in that, The process of generating the initial solution for the path planning model based on the greedy insertion algorithm includes the following steps: Construct the first-gradient route, set the planned route using all refrigerated trucks according to the route planning model, and add the forward warehouse delivery task to each refrigerated truck in sequence; Initialize the available drone and rider routes, insert each customer to be delivered into a location in the drone or rider route with the lowest insertion cost, thereby obtaining the initial solution; wherein, before calculating the current insertion position, load constraints, drone maximum flight distance constraints, and time window constraints are satisfied.

5. The refrigerated truck-drone-rider collaborative path planning method considering freshness loss according to claim 1, characterized in that, The process of improving the initial solution through local search to obtain the target solution includes the following steps: Based on the path planning model, four removal operators and four repair operators are constructed; wherein, the removal operators include random removal operator, worst-cost removal operator, worst-freshness removal operator, and random front-end removal operator; the repair operators include random insertion operator, greedy insertion operator, regret value insertion operator, and freshness-based greedy insertion operator; Based on the accumulated feature values, a logistic regression model is used to predict the selection probability of each operator, and the selection probability is used as the adaptive weight of the corresponding operator. The feature values ​​include solution state features, operator features, and stage features. The solution state features include the optimal solution improvement value, the current solution improvement value, route similarity, the number of times the optimal solution has been improved, and the number of times the current solution has been improved. The operator feature is the feasibility rate. The stage feature is the number of iterations. The roulette wheel algorithm is used to select the removal and repair operators for each iteration based on the adaptive weights. The local search is improved by using the selected removal and repair operators to obtain updated routes and feature values; The above steps are iterated continuously until the preset termination condition is reached, and the target solution is obtained.

6. The refrigerated truck-drone-rider collaborative path planning method considering freshness loss according to claim 5, characterized in that, The logistic regression model is trained through the following steps: A static logistic regression model is trained in advance using training examples of different sizes and corresponding complete search trajectory data. Alternatively, the initial parameters of the logistic regression model can be set to zero or a preset prior value, and the parameters of the logistic regression model can be updated periodically based on the accumulated feature value data increments during the process of solving the target solution.

7. A refrigerated truck-drone-rider collaborative path planning device considering freshness loss, characterized in that, The device includes: The parameter definition unit is used to define the route planning parameters for refrigerated trucks, drones, and riders, taking into account freshness loss. A model building unit is used to build a path planning model based on the path planning parameters; An initial solution generation unit is used to generate an initial solution for the path planning model based on a greedy insertion algorithm. The target solution search unit is used to perform local search improvement on the initial solution to obtain the target solution; The delivery scheduling unit is used to coordinate refrigerated trucks, drones, and riders to perform fresh food delivery services according to the target.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 6.

Images