Distribution method based on space-time clustering in remote rural co-distribution mode

Abstract

According to the method, the complexity of a remote rural distribution environment is considered, the high dispersion characteristic of order space-time distribution is considered, the differentiation characteristic of multi-vehicle-type transport capacity resources is also considered, and the order combination distribution method based on space-time clustering in a rural common distribution mode is provided. In the first stage, customer geographic positions and order time window constraints are comprehensively considered, reachability-guaranteed clustering is formed to realize order combination, and an initial path feasible solution is constructed based on a space-time clustering result; in the second stage, a simulated annealing framework is designed to be fused into adaptive large neighborhood search, a clustering result is dynamically adjusted according to multi-vehicle-type differential time-space characteristics, and an initial solution is improved; on-demand combination of orders, accurate path planning and multi-vehicle collaborative distribution are realized, and logistics enterprises are helped to cope with distribution challenges of scattered orders, fuzzy time windows and complex road networks in remote rural areas with lower cost on the premise of ensuring distribution timeliness and service quality.

Description

A spatiotemporal clustering-based delivery method for a shared delivery model in remote rural areas Technical Field

[0001] This invention relates to a spatiotemporal clustering-based delivery method under a shared delivery model in remote rural areas, belonging to the field of rural logistics and delivery technology. Background Technology

[0002] In 2023, the national postal delivery volume reached 162 billion pieces and the business revenue reached 1.5 trillion yuan. At the same time, more than 100 million express parcels are rapidly circulating in rural areas every day, and the demand for express delivery to every village is increasing. However, the efficiency of rural delivery is still lagging behind that of cities. This contradiction is particularly prominent in remote areas represented by the Tibet Autonomous Region, which also reveals the limitations of the traditional self-delivery model in special geographical and economic environments. However, joint delivery, through the resource integration method of multiple logistics companies forming an alliance to coordinate delivery, can reduce operating costs and improve delivery efficiency, providing a feasible path to solve the problem of "difficulty in getting industrial products to the countryside". At present, under the "1+N+N" type of joint delivery mechanism, remote rural areas represented by Tibet still face significant bottlenecks in route planning and operation: (1) Highly dispersed time and space demand: The sparse village layout and scattered herder settlements result in extremely dispersed locations for picking up and delivering parcels in space, making it difficult to accurately predict and aggregate in time, making it difficult to form an effective intensive delivery window, which greatly reduces vehicle loading rate and joint delivery efficiency. (2) Increased uncertainty of fuzzy time windows: Due to the combination of terrain barriers and the performance of vehicles with different capacities, delivery times become inaccurate. The "fuzzy time window" promised by the recipient (such as "morning" or "afternoon") becomes wider and more uncertain in actual operation. This not only increases the customer's waiting anxiety, but also puts the alliance under great pressure when dispatching vehicles, coordinating the handover of multi-brand parcels, and balancing multiple delivery methods. (3) Dynamic changes in road network conditions: Low-grade rural roads (mountain roads, dirt roads, etc.) that are easily affected by weather cause static path planning algorithms to fail, and travel time and path length increase sharply. It is necessary to adapt to sudden events such as landslides and snow accumulation in real time. Traditional algorithms based on Euclidean distance fail, path length, travel time, and safety risks increase sharply, and it is necessary to dynamically adapt to real-time changes in road conditions.

[0003] Existing research mainly focuses on four aspects: rural logistics route optimization methods, drone delivery technology applications, multi-dimensional analysis of rural logistics demand characteristics, and crowdsourced delivery solutions based on vehicle-cargo matching. The State Intellectual Property Office of the People's Republic of China patent number CN 118278848A, entitled "Rural Logistics Delivery Route Generation Method, Apparatus, and Equipment," designs a rural logistics route generation method, apparatus, computer equipment, storage medium, and computer program product. It enables drone delivery points to adopt a static vehicle-drone collaborative mode, vehicle delivery points to adopt a traditional vehicle delivery mode, and a dynamic vehicle-drone collaborative mode. It expands the drone flight mission through a bidirectional search algorithm, identifies flight intermediate points among flexible delivery points, and constructs collaborative delivery routes between drones and vehicles. However, high altitude and low air pressure environments severely affect the drone's payload capacity and endurance, limiting its applicability in remote rural areas. The State Intellectual Property Office of the People's Republic of China patent number CN 116797126A, entitled "A Rural Last-Mile Logistics Site Selection-Route Planning Method Based on Two-Level Planning," obtains rural POI and route information, and uses building distribution to represent the distribution of rural logistics demand. It constructs a logistics site selection model with rural residents and express delivery points as the upper-level decision-making entities, and establishes a two-level planning site selection-route planning method with logistics companies as the lower-level decision-making entities. However, in remote rural areas, there is a significant deviation between building distribution and actual logistics demand, and the low population density but long delivery distances limit the accuracy of this method. The State Intellectual Property Office of the People's Republic of China patent number CN 116258429A, entitled "A Vehicle-Cargo Matching Method for Rural Last-Mile Logistics," comprehensively considers distribution centers, carriers, and delivery points, establishing a vehicle-cargo matching model based on a crowdsourcing model, and solving it using the GA-C algorithm. The crowdsourcing model relies on sufficient carrier resources, but the number of carriers in remote areas is limited and unevenly distributed, requiring the construction of a multi-carrier collaborative delivery optimization mechanism adapted to resource constraints.

[0004] While existing research offers valuable insights, it lacks a systematic consideration of the timeliness constraints of orders in remote rural areas. The complex terrain of rural Tibet and the absence of clear rules for same-day and multi-day order delivery lead to a vicious cycle of delivery delays and order backlogs. Furthermore, current spatiotemporal clustering algorithms do not fully consider the differentiated spatiotemporal attributes of postal vehicles, passenger bus routes, and crowdsourced vehicles within the "county-township-village" delivery network. They also largely rely on Euclidean distance for clustering analysis, neglecting the crucial impact of actual road network distance on the spatiotemporal accessibility of rural order delivery. This makes it difficult to accurately characterize the spatiotemporal distribution characteristics of orders in the complex geographical environment of rural areas, thus limiting the effectiveness of optimizing shared delivery routes.

[0005] This invention considers the complexity of delivery environments in remote rural areas, taking into account both the highly dispersed spatiotemporal distribution of orders and the differentiated characteristics of multi-vehicle transport capacity resources. It proposes a spatiotemporal clustering-based order merging and delivery method under a rural collaborative delivery model. The first stage comprehensively considers customer geographic location and order time window constraints to form reachability-guaranteed clusters for order merging, and constructs an initial feasible path solution based on the spatiotemporal clustering results. The second stage designs a simulated annealing framework incorporating adaptive large neighborhood search to dynamically adjust the clustering results for the differentiated spatiotemporal characteristics of multi-vehicle types, thus improving the initial solution. This enables on-demand order merging, precise route planning, and collaborative delivery across multiple vehicle types, helping logistics companies address the challenges of dispersed orders, ambiguous time windows, and complex road networks in remote rural areas at a lower cost while ensuring delivery timeliness and service quality. Summary of the Invention

[0006] The purpose of this invention is to provide a spatiotemporal clustering-based delivery method for remote rural areas under a shared delivery model. It constructs a spatiotemporal clustering model based on actual road network data and uses a two-stage hybrid optimization algorithm to achieve order merging and route optimization. This addresses the technical challenges of order dispersion, uncertainties in fuzzy time windows, and multi-vehicle collaborative delivery in remote rural areas, thereby minimizing delivery costs and maximizing service timeliness.

[0007] Technical solution of the present invention

[0008] A spatiotemporal clustering-based delivery method for a shared delivery model in remote rural areas includes the following steps:

[0009] Step 1: Establishing a mixed-integer nonlinear programming model for the two-level shared distribution path planning problem. A "1+N+N" shared distribution model for express delivery in remote rural areas is proposed: Led by "1" enterprise (either the postal service or a leading local express delivery company), "N" express delivery companies form a shared distribution alliance to integrate express delivery services within the county. Then, through "N" delivery methods combining postal delivery, passenger delivery, and crowdsourced delivery, shared distribution tasks are achieved in the two-level distribution network of "county-township" and "township-village". This shared distribution model consists of one county-level distribution center, Township transit points The bus routes, which integrate passenger, freight, and mail services, are composed of these lines. Each village collection point is uniformly operated by the express delivery alliance. The county-level distribution center publishes cargo information through the platform, and crowdsourced vehicles publish their own route information. After obtaining information on these three types of vehicles, the county-level distribution center rationally selects the transportation mode and plans the route, pays service fees to the crowdsourced vehicles, and sets up a time penalty mechanism. Passenger-freight-mail integrated public transportation and self-operated freight trucks depart from and return to the county-level distribution center; crowdsourced vehicles have no fixed origin or destination and can detour to pick up goods without returning. The specific model design is as follows:

[0010] Self-operated truck transportation considers fixed vehicle costs; integrated passenger-freight-mail public transportation considers the unit product freight cost of outsourced transportation; crowdsourced vehicles adopt an open-loop transportation model and introduce outsourced transportation costs based on unit product and distance, as well as time-related penalties for vehicle defaults. The objective function is to minimize the total delivery cost, as follows:

[0011] .

[0012] In the formula, Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Passing through feasible arc distance, Indicates vehicle Cost per unit distance traveled Indicates vehicle Unit cargo transportation cost Represents village nodes The number of orders, Indicates vehicle Fixed costs. It is a 0-1 variable, representing when the vehicle From node Drive to the node The value is 1 if the time is right and 0 otherwise. To balance delivery efficiency and customer satisfaction, the model introduces a fuzzy time window constraint: requiring that the arrival time of vehicles at each node must be within the acceptable time range for that node. Within the system, for the order collection points in each village, the most urgent orders are selected and their on-time arrival times are recorded. This serves as the optimal service time window for that point. In the time penalty function, let... Penalty cost for arriving early per unit time and The objective function is the penalty cost per unit time for being late. The formula for the time-related penalty for vehicle violations is as follows:

[0013] .

[0014] The specific constraints are as follows:

[0015] Node constraints: Each vehicle must return to this point after departing from the county-level distribution center, and each vehicle can use this point at most once. ,

[0016] In the formula, the first-level transport vehicle set ( Intercity freight trains, A transportation network integrating passenger, freight, and mail services (public transport) from county-level distribution centers to township transfer points; a second-level collection of transport vehicles. ( Township trucks Integrated passenger, freight, and mail public transport Crowdsourced delivery vehicles serve the route from township transfer points to village collection points. Among them, the passenger, freight and mail integrated bus relies on urban and rural bus routes to connect county-level distribution centers, township transfer points and village collection points. Represents the entire set of vertices; These represent the county-level distribution center node set, the village collection node set, and the township transfer network node set, respectively. Indicates the first A collection of vehicles of all grades This represents the collection of all vehicles. . express A subset of serviceable nodes for vehicles of the same class; express A subset of village-level collection nodes; express Level path.

[0017] We guarantee that each customer will be served by only one vehicle:

[0018] .

[0019] The number of times the same type of vehicle can be served at each transfer point is limited to no more than once:

[0020] .

[0021] Vehicle-related constraints: Ensure that the vehicle load does not exceed the capacity limit. , Indicates when vehicle From node To the node The load, Indicates vehicle The maximum capacity it has.

[0022] Maximum driving time limit for each vehicle: In the formula, Vehicles are represented as continuous variables. Reaching the node Time, Indicates the maximum route duration.

[0023] Customer demand-related constraints: The load after a vehicle accesses a node is equal to the load before access minus the node's demand. , Representative node Order demand at the location.

[0024] Ensure the vehicle is empty when it completes the route: In the formula, This represents the total order demand from all customers.

[0025] The total volume of goods transported by a county-level distribution center must equal the sum of customer order demands. .

[0026] The departure load of a Tier 2 transport vehicle is limited to equal to the total order demand of its service customers: .

[0027] Second-level transport vehicles are required to arrive at township transfer points no earlier than first-level vehicles. .

[0028] The regulation stipulates that the departure time of first-level transport vehicles from the county-level distribution center is zero: .

[0029] Calculate the arrival time of vehicles at each node: , Indicates vehicle Passing through feasible arc Travel time, Represents a node Service hours.

[0030] Ensure the arrival time is within the fuzzy soft time window: .

[0031] Step 2: Propose an adaptive large neighborhood search hybrid simulated annealing algorithm (HSAALNS_TC) that considers multi-vehicle collaboration and spatiotemporal distance. The algorithm consists of two stages: The first stage uses spatiotemporal clustering with capacity constraints to construct an initial solution and achieves preliminary vehicle matching through a greedy strategy; the second stage uses a hybrid optimization algorithm combining SA and ALNS, embedding a dynamic vehicle update mechanism during the iteration process to systematically iteratively optimize the initial solution to approximate the global optimum. The specific steps are as follows:

[0032] Step 2.1: Data Definition and Preprocessing. Acquire logistics network data, as well as sets of multiple vehicle types and their attributes. Specifically, for crowdsourced vehicles, set a relatively small mileage limit. It also boasts a high level of timeliness and responsiveness. Order data from each village's collection point is acquired, and a three-dimensional spatiotemporal coordinate system is constructed with latitude and longitude as the spatial axis and service time windows as the time axis.

[0033] Step 2.2: Construct the spatiotemporal distance feature matrix. Using a query strategy bound to the service time window, request actual road network travel time from the Baidu Maps API, and calculate the spatiotemporal distance by combining this with the distance difference within the order time window. Specifically, the calculation of spatiotemporal distance involves: And normalize it using the formula:

[0034]

[0035] In the formula, the weighting coefficients α and β are both 0.5. The time distance is defined as the relative offset between the start and end points of the time window, and is normalized to construct the spatiotemporal feature matrix of the village collection nodes.

[0036] Step 2.3: Data Feature Identification. Principal component analysis is performed on the normalized spatiotemporal distance matrix constructed in Step 2.2 to identify the spatial distribution characteristics, time window differences, and spatiotemporal interaction characteristics in the spatiotemporal matrix among village collection nodes.

[0037] Step 2.4: Spatiotemporal clustering based on genetic algorithm. Data analysis in Step 2.3 revealed that the logistics network in remote rural areas is primarily affected by uneven geographical distribution. Traditional distance metrics cannot reflect the complexity of terrain and transportation conditions. Therefore, a clustering algorithm considering spatiotemporal clustering was designed. Traditional K-means and DBSCAN algorithms encounter numerous unidentifiable noise points when handling low-density distributions, and the clustering results have low silhouette coefficients, leading to clustering results that do not reflect delivery conditions and significantly increase delivery costs. Genetic algorithms offer great flexibility, allowing for the design and optimization of complex objective functions. Therefore, a genetic algorithm was used for spatiotemporal clustering, with the basic objective of minimizing intra-cluster distance and maximizing inter-cluster distance.

[0038] The specific implementation process of spatiotemporal clustering is as follows:

[0039] Step 2.4.1: Determine the number of clusters This is assumed to be true, with the upper limit not exceeding the total number of candidate township transit points and the lower limit within a reasonable clustering range, and the iteration... Choose an appropriate clustering unit.

[0040] Step 2.4.2: Calculate the spatiotemporal distance matrix according to the formula.

[0041] Step 2.4.3: Population Initialization. Create a population containing... The initial population of chromosomes, each chromosome consisting of... A candidate clustering scheme is composed of randomly selected representative customer order IDs.

[0042] Step 2.4.4: Fitness Evaluation and Capacity Constraint Check. Assign the remaining orders to the most recent representative order, and sum the subproblem costs as the chromosome fitness value. The fitness function is designed as follows: ,in For the spatiotemporal distance within a cluster, The distance between clusters, To avoid division by zero, a validity check is performed before calculating fitness: for each cluster... Total demand .like If the clustering requirement exceeds the load limit of the largest vehicle in the fleet, then that individual is deemed infeasible, and the process is iteratively executed until the maximum number of generations is reached. .

[0043] Step 2.4.5: Perform selection, crossover, and mutation operations. Using roulette wheel selection, single-point crossover, and constrained mutation operations, each cluster contains a transfer point and its assigned set of village collection points. Superior individuals are selected for the next generation based on fitness values; probabilistic... Select parent generation pairs, apply the crossover operator to generate offspring, and replace parent generation with offspring; probability. The chromosomes in the population are mutated to maintain the diversity of the population; the newly generated population enters the next round of iteration, returns to step 2.4.4 to evaluate fitness, and terminates the iteration when the maximum number of generations is reached.

[0044] Step 2.5: Construct the initial solution using a two-stage decimal encoding scheme. Each individual contains a first-level and a second-level cluster set and path set: the cluster set length is equal to the node set, representing service relationships; the path set represents the access order. Cluster sets allow for quick extraction of nodes from the same service center, facilitating demand calculation and neighborhood operations. Based on the above encoding scheme, the algorithm employs a combined backward and forward initial solution construction strategy: first, construct the second-level and first-level allocation sets, and solve for the first-level path set; then calculate vehicle arrival and departure times, and solve for the second-level path set. The specific steps for constructing the backward-forward initial solution using spatiotemporal clustering are as follows:

[0045] Step 2.5.1: Initialization and Path Segmentation. Based on the clustering scheme determined in Step 2.4 (i.e., the set of village collection points served by each township transfer station), each transfer station within the cluster and all customers within its service area are regarded as independent paths, and an initial set of secondary paths is constructed.

[0046] Step 2.5.2: Calculate the spatiotemporal distance savings. Based on the spatiotemporal distance calculation formula, calculate the spatiotemporal savings between nodes, generate a list of savings values ​​that integrates normalized spatial distance savings and spatiotemporal distance savings, and sort them in descending order.

[0047] Step 2.5.3: Path merging based on vehicle model adaptation logic. The savings value list is traversed sequentially using a greedy strategy, attempting to merge paths. During each merging attempt, the vehicle model adaptation logic is executed, and the state is calculated: the merged path is calculated. Cumulative load Total driving distance and the estimated completion time Feasibility screening: Traverse the set of vehicle models. Filter out those that simultaneously meet the capacity constraint ( ) and mileage constraints ( Fable vehicle subset Optimal vehicle model matching: If For non-empty data, calculate the comprehensive cost of each vehicle type in the subset and select the minimum value: For self-operated trucks, the cost is calculated as follows: Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Passing through feasible arc Travel time, Indicates vehicle Cost per unit time travel distance Indicates vehicle Cost per unit distance traveled; Indicates vehicle Fixed costs; Indicates vehicle Passing through feasible arc The distance. The cost of the freight passenger vehicle is Crowdsourced car service is The fare for passenger, freight, and mail integrated public transportation is calculated based on the unit volume of goods transported. All of them have time penalties added. If a match is found, the vehicle model is selected and the merge is executed; otherwise, the merge is abandoned and the list remains unchanged until the list has been traversed.

[0048] Step 2.5.4: Primary Path Construction. Based on the secondary path set, determine the set of activated transit stations and use them as demand points. Use the nearest neighbor heuristic algorithm to construct primary paths, and apply the vehicle model adaptation logic from Step 2.5.3 to match the optimal vehicle model to the primary paths. Finally, integrate the primary and secondary path sets to output an initial solution containing complete vehicle model and path information.

[0049] Step 2.6: Neighborhood Structure Design. Through the clustering neighborhood search operator in the second stage of the algorithm, one of them is randomly selected to generate a neighborhood solution after each iteration of spatiotemporal clustering to construct the initial solution. Specifically, it includes: (1) Transfer point exchange operator: randomly select two activated township transfer points and exchange the subset of customers they serve; (2) Transfer point reset operator: randomly close one activated transfer point and activate one from the remaining transfer points, and redistribute the customers served by the original transfer point; (3) Customer point exchange operator: randomly select two customers belonging to different transfer points and exchange their affiliation.

[0050] Step 2.7: ALNS Operator Design. In this process, all operators involve spatiotemporal distance as the distance metric, and the cost calculations are all based on the optimal vehicle comprehensive cost obtained by solving the vehicle adaptation logic described in Step 2.5.3. Specifically, it includes the following four types of destructive operators and six types of repair operators: (1) Random removal operator, from path Random selection (2) Importance removal operator, based on the importance index of customer points. The selection is made based on this formula: , Indicates the quantity demanded. Indicates the width of the time window. Indicates the distance to the transfer station. (3) Worst-case removal operator, removing items via random factors. [0.8, 1.2] The node with the worst cost contribution after the disturbance; (4) The correlation removal operator removes the nodes with the highest correlation to the selected nodes in turn, and prioritizes nodes that are not on the same path. Specifically, the correlation removal process is as follows: randomly select a customer from the customer set and add it to the set to be removed; then, according to the formula Calculate the remaining customers and selected customers in the customer set. The correlation, On behalf of clients With customers The distance between, Indicates customer The maximum distance between the remaining customers and the customer cluster. This is a binary variable used to determine if two clients are on the same path. If the client... With customers If the path is the same, select 1; otherwise, select 0. Finally, if the set to be removed is not full... For each customer, select the customer with the highest relevance and add them to the set, then treat them as the new current customer and repeat the previous step; otherwise, terminate.

[0051] Repair operators: (1) Random repair operator: Randomly select one of the feasible paths that meet the capacity constraints and insert the customer to be assigned into the random position. (2) Greedy repair operator: Traverse all feasible insertion positions and calculate the cost increment caused by the insertion operation. The cost calculation requires re-execution of the vehicle model adaptation process to determine the optimal vehicle model required after the load increase and its corresponding fixed and variable costs. The customer point insertion position is selected based on the minimum cost increase after insertion. (3) Random greedy repair operator: Add random perturbation when calculating transportation costs to ensure the diversity of solutions obtained in the repair stage. (4) Importance repair operator: After sorting according to customer importance, prioritize inserting customers with high importance into the path according to the greedy principle. (5) Maximum regret repair operator: Calculate the cost difference between inserting the customer into the optimal position and the second-best position, and prioritize inserting the customer with the largest cost difference. The cost calculation of the optimal and second-best positions strictly follows the dynamic matching principle of multiple vehicle models to avoid forced vehicle model upgrades or high penalties due to delayed insertion. (6) First-level path repair operator: Use a greedy strategy to insert the transfer station into the optimal position of the first-level path.

[0052] Step 2.8: The algorithm introduces an adaptive weight adjustment mechanism. In the initial search phase, all destruction and repair operators are assigned the same initial weights. During iteration, the algorithm tracks the performance of each operator and scores it according to a three-level scoring rule: finding the globally optimal solution... By dividing the neighborhood, a better solution can be found. The solution that is not improved is accepted. Points. Algorithm iterations. Then, it will be based on the scores of each operator during this period. and number of times used The weights are dynamically adjusted using a weight update formula:

[0053] ,

[0054] In the formula As a response factor, it is used to adjust the influence of historical experience and current performance on weight updates, thereby achieving adaptive evolution of the search strategy.

[0055] Step 2.9: Iterative Convergence Determination and Global Optimal Solution Output. Simulated annealing and cooling are performed, along with termination condition checks, to control the algorithm's convergence process and output the final delivery plan. Step 2.9.1: Acceptance Judgment. The cost difference is calculated. Define the probability of acceptance Generate random numbers ,like Therefore, according to the Metropolis criterion, this inferior solution is also accepted. If so, then update the current solution; Update the global optimum. Regardless of acceptance, update the operator's weight in ALNS based on its performance in this step. Step 2.9.2: Cooling termination. Complete the current temperature... After the second internal circulation, cooling and temperature reduction are performed: Then return to step 2.9.1 to proceed to the next temperature iteration. If If the condition is met, the algorithm terminates and outputs the final globally optimal solution.

[0056] Beneficial effects of the present invention

[0057] To address the challenges of logistics and delivery in remote rural areas, this invention constructs a county-township-village collaborative delivery system based on a "1+N+N" model. The core innovation lies in using real road network data instead of the traditional Euclidean distance model, and employing a spatiotemporal clustering algorithm to accurately align customer needs, completely resolving the problem of route planning failure in complex mountainous environments. The solution introduces a multi-vehicle collaborative delivery mechanism, achieving optimal allocation of transportation resources and significantly shortening order delivery times to villages. This research is particularly applicable to high-altitude mountainous areas such as Tibet, significantly reducing time-window default rates and providing a highly practical technical approach and decision-making reference for solving the "last mile" problem in rural logistics. Attached Figure Description

[0058] Figure 1 is a basic flowchart of the present invention.

[0059] Figure 2 shows the "1+N+N" model of the shared distribution network design for remote rural areas according to an embodiment of the present invention.

[0060] Figure 3 shows a two-stage hybrid algorithm for the problem of shared delivery route planning in remote rural areas according to an embodiment of the present invention.

[0061] Figure 4 is a distribution map of examples of the present invention for remote rural areas.

[0062] Figure 5 is a three-dimensional cluster distribution map of service points for the problem of shared distribution route planning in remote rural areas according to an embodiment of the present invention. Detailed Implementation

[0063] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0064] As shown in Figure 1, a spatiotemporal clustering-based delivery method for a shared delivery model in remote rural areas comprises the following steps:

[0065] Step 1: Establish a mixed-integer nonlinear programming model for the two-level shared distribution path planning problem. Propose the "1+N+N" shared distribution model for remote rural express delivery shown in Figure 2: Led by "1" enterprise (either the postal service or a leading local express delivery company), a shared distribution alliance is formed with "N" other express delivery companies to integrate express delivery services within the county. Then, shared distribution is achieved through "N" delivery methods combining postal delivery, passenger delivery, and crowdsourced delivery. The shared distribution network consists of one county-level distribution center, Township transit points The bus routes, which integrate passenger, freight, and mail services, are composed of these lines. Each administrative village is managed by a unified express delivery alliance. The county-level distribution center publishes cargo information through the platform, while crowdsourced vehicles publish their own route information. After obtaining information on these three types of vehicles, the county-level distribution center rationally selects transportation methods and plans routes, pays service fees to the crowdsourced vehicles, and sets up a time penalty mechanism. Passenger-freight-postal integrated public transportation and self-operated freight trucks depart from and return to the county-level distribution center; crowdsourced vehicles have no fixed origin or destination and can detour to pick up goods without returning.

[0066] Self-operated truck transportation considers fixed vehicle costs; integrated passenger-freight-mail public transportation considers the unit product freight cost of outsourced transportation; crowdsourced delivery adopts an open-loop transportation model and introduces outsourced transportation costs based on unit product and distance, as well as time-related penalties for vehicle-related delays. The objective function is to minimize the total delivery cost, as follows:

[0067] .

[0068] In the formula, Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Passing through feasible arc distance, Indicates vehicle Cost per unit distance traveled Indicates vehicle Unit cargo transportation cost Indicates village collection point The number of orders, Indicates vehicle Fixed costs. It is a 0-1 variable, representing when the vehicle From node Drive to the node The value is 1 if the time is right and 0 otherwise. To balance delivery efficiency and customer satisfaction, the model introduces a fuzzy time window constraint: requiring that the arrival time of vehicles at each node must be within the acceptable time range for that node. Within the system, for the order collection points in each village, the most urgent orders are selected and their on-time arrival times are recorded. This serves as the optimal service time window for that point. In the time penalty function, let... Penalty cost for arriving early per unit time and The objective function is the penalty cost per unit time for being late. The formula for the time-related penalty for vehicle violations is as follows:

[0069] .

[0070] The specific constraints are as follows:

[0071] Node constraints: Each vehicle must return to this point after departing from the county-level distribution center, and each vehicle can use it at most once. In the formula, the first-level transport vehicle set ( Intercity freight trains, A transportation network integrating passenger, freight, and mail services (public transport) from county-level distribution centers to township transfer points; a second-level collection of transport vehicles. ( Township trucks Integrated passenger, freight, and mail public transport Crowdsourced delivery vehicles serve the route from township transfer points to village collection points. Among them, the passenger, freight and mail integrated bus relies on urban and rural bus routes to connect county-level distribution centers, township transfer points and village collection points. Represents the entire set of vertices; These represent the county-level distribution center node set, the village collection node set, and the township transfer network node set, respectively. Indicates the first A collection of vehicles of all grades This represents the collection of all vehicles. . express A subset of serviceable nodes for vehicles of the same class; express A subset of village-level collection nodes; express Level path.

[0072] We guarantee that each customer will be served by only one vehicle:

[0073]

[0074] The number of times the same type of vehicle can be served at each transfer point is limited to no more than once:

[0075]

[0076] Vehicle-related constraints: Ensure that the vehicle load does not exceed the capacity limit. .

[0077] Indicates when vehicle From node To the node The load, Indicates vehicle Maximum load capacity.

[0078] Maximum driving time limit for each vehicle: In the formula, Vehicles are represented as continuous variables. Reaching the node Time, Indicates the maximum route duration.

[0079] Customer demand-related constraints: The load after a vehicle accesses a node is equal to the load before access minus the node's demand. .

[0080] The initial load of a second-level transport vehicle is set equal to the sum of the demand from all service village collection nodes along its route, and its load is reduced to zero after the vehicle completes all service tasks. In the formula, This represents the total demand from all customers.

[0081] The total order volume of a county-level distribution center must equal the sum of its vehicle loads. .

[0082] The load that a Tier 2 transport vehicle can depart with is equal to the total demand of its customer orders: .

[0083] Time window constraint: The arrival time of the second-level transport vehicle at the township transfer point must not be earlier than that of the first-level transport vehicle. .

[0084] The regulation stipulates that the departure time of second-level transport vehicles from township transfer points is zero: .

[0085] Calculate the arrival time of vehicles at each node: , Represents a node Service hours.

[0086] Ensure that the arrival times of vehicles at all levels are within the fuzzy time window: .

[0087] Decision variables 0-1 represents a vehicle From node Drive to the node , Vehicles are represented as continuous variables. From node To the node The load, Vehicles are represented as continuous variables. Reaching the node The time.

[0088] Step 2: Propose an adaptive large neighborhood search hybrid simulated annealing algorithm (HSAALNS_TC) that considers multi-vehicle collaboration and spatiotemporal distance. The algorithm, shown in Figure 3, consists of two stages: The first stage uses spatiotemporal clustering with capacity constraints to construct an initial solution, achieving preliminary vehicle matching through a greedy strategy; the second stage employs a hybrid optimization algorithm combining SA and ALNS, embedding a dynamic vehicle update mechanism during the iteration process to systematically iteratively optimize the initial solution to approximate the global optimum. The specific steps are as follows:

[0089] Step 2.1: Obtain the total number of parcels for each village collection point, which is the sum of all orders. Spatial location is represented by longitude and latitude (see Figure 4), and service time window constraints are represented by the time dimension. The spatial attributes of village collection points are determined by their geographical coordinates, and the time attributes are defined by the fuzzy time window constraints of the delivery orders.

[0090] Step 2.2: Construct the spatiotemporal distance feature matrix. The quantification of temporal distance adopts a query strategy bound to the service time window: based on the time window of the starting point, the actual road network travel time for the corresponding time period is requested from the Baidu Maps API. Spatial distance is calculated based on the actual road network data. ,in and The weighting coefficient is set to 0.5. The time distance is defined as the relative offset between the start and end points of the time window, and is then normalized.

[0091] Construct the spatiotemporal feature matrix of village collection nodes.

[0092] Step 2.3: Data processing, identifying data features, and performing principal component analysis on the normalized spatiotemporal distance matrix. Examine the spatial distribution characteristics, time window differences, and spatiotemporal interaction characteristics in the spatiotemporal matrix among village collection nodes.

[0093] Step 2.4: Spatiotemporal clustering based on genetic algorithm. Data analysis revealed that the logistics network in remote rural areas is mainly affected by uneven geographical distribution. Traditional distance metrics cannot reflect the complexity of terrain and transportation conditions. Therefore, a clustering algorithm considering spatiotemporal clustering was designed. Traditional K-means and DBSCAN algorithms produce a large number of difficult-to-identify noise points when dealing with low-density distributions. At the same time, the silhouette coefficient of the clustering results is low, resulting in clustering results that do not conform to the delivery situation, leading to a significant increase in delivery costs. Genetic algorithm provides great flexibility and can design and optimize complex objective functions. Therefore, a genetic algorithm is used for spatiotemporal clustering, with the basic objective of minimizing intra-cluster distance and maximizing inter-cluster distance.

[0094] The specific implementation process of spatiotemporal clustering is as follows:

[0095] Step 2.4.1: Determine the number of clusters belong Assume that the upper limit does not exceed the total number of candidate township transit points, and the lower limit is within a reasonable clustering range.

[0096] Step 2.4.2: Encoding scheme: The length of each chromosome is... , Indicates customer Belongs to clustering .

[0097] Step 2.4.3: Create a file containing The initial population of chromosomes, each chromosome consisting of... A candidate clustering scheme is composed of randomly selected representative customer order IDs.

[0098] Step 2.4.4: Assign the remaining orders to the nearest representative order, and sum the subproblem costs as the chromosome fitness value. The fitness function is designed as follows: ,in For the spatiotemporal distance within a cluster, The distance between clusters, To avoid division by zero, the process is executed iteratively until the maximum number of generations is reached. .

[0099] ① We employ roulette wheel selection, single-point crossover, and constrained mutation operations. Each cluster contains a transfer point and its assigned village generation clusters. We select high-quality individuals for the next generation based on their fitness values.

[0100] ② Using probability Select parent generation pairings, apply the crossover operator to generate offspring, and replace the parent generation with the offspring.

[0101] ③Probability Variations are made in the chromosomes of a population to maintain its diversity.

[0102] ④ The newly generated population enters the next iteration, and the iteration terminates when the maximum number of generations is reached.

[0103] Step 2.5: Construct the initial solution using a two-stage decimal encoding scheme. Each individual contains a first-level and a second-level cluster set and path set: the cluster set length is equal to the node set, representing service relationships; the path set represents the access order. The cluster set allows for quick extraction of nodes from the same distribution center, facilitating demand calculation and neighborhood operations. Based on the above encoding scheme, the algorithm employs a combined backward and forward initial solution construction strategy: first, construct the second-level and first-level allocation sets, and solve for the first-level path set; then, calculate vehicle arrival and departure times, and solve for the second-level path set. The specific steps for constructing the backward-forward initial solution using spatiotemporal clustering are as follows:

[0104] Step 2.5.1: Initialization and Path Segmentation. Based on the aforementioned spatiotemporal clustering results, each transit station within a cluster and all customers within its service area are considered as independent paths, and an initial set of secondary paths is constructed.

[0105] Step 2.5.2: Calculate the spatiotemporal distance savings. Based on the spatiotemporal distance calculation formula, calculate the spatiotemporal savings between nodes, generate a list of savings values ​​that integrates normalized spatial distance savings and spatiotemporal distance savings, and sort them in descending order.

[0106] Step 2.5.3: Path merging based on vehicle model adaptation logic. The savings value list is traversed sequentially using a greedy strategy, attempting to merge paths. During each merging attempt, the vehicle model adaptation logic is executed, and the state is calculated: the merged path is calculated. Cumulative load Total driving distance and the estimated completion time Feasibility screening: Traverse the set of vehicle models. Filter out those that simultaneously satisfy the capacity constraint and mileage constraints feasible vehicle subset Optimal vehicle model matching: If For non-empty data, calculate the comprehensive cost of each vehicle type in the subset and select the minimum value: For self-operated trucks, the cost is calculated as follows: Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Cost per unit distance traveled; Indicates vehicle Fixed costs; Indicates vehicle Passing through feasible arc The distance is calculated. If a match is found, the vehicle model is selected and merged; otherwise, the merge is abandoned and the list remains unchanged until the list has been traversed.

[0107] Step 2.5.4: Primary Path Construction. Based on the secondary path set, determine the set of activated transfer stations, use the nearest neighbor heuristic algorithm, and simultaneously check vehicle capacity constraints to construct the primary path. Finally, integrate the primary and secondary path sets to output the initial solution of the complete path.

[0108] Step 2.6: Simulated Annealing Outer Layer Acceptance Criterion: Algorithm convergence is controlled through simulated annealing cooling mechanism. Initial temperature is set. Minimum temperature threshold Maximum number of iterations per temperature and cooling coefficient (in When the temperature At that time, the inner loop optimization process is repeated:

[0109] Step 2.6.1: Adaptive Neighborhood Search (ALNS) for operator selection. This process adaptively selects operators for the current solution based on operator weights. The algorithm performs a perturbation by invoking the previously defined clustering neighborhood operators (such as transfer station exchange, reset, or point exchange) to exploratoryly reorganize the overall cluster structure of the solution. Then, it first selects one of the four disruptive operators from step 2.7 of this technical solution to remove some nodes, and then selects one of six repair operators (such as greedy or regret value repair) to reintegrate the removed nodes into the path network. Specifically:

[0110] ① Calculate the selection probability of the destruction operator: where To destroy the operator Generate random numbers based on the current weights. Using roulette wheel selection to choose the destruction operator: cumulative probability calculation: selecting the minimum satisfying condition. As a destruction operator .

[0111] ② Similarly, select the repair operator: generate random numbers. Select the repair operator Generate neighborhood solutions.

[0112] ③ Solution evaluation and acceptance judgment: Step 2.6.2.1 Calculate the cost of the objective function;

[0113] ④ Check constraints in the model such as time windows, loading, and integrated passenger, freight, and mail bus routes;

[0114] ⑤ Calculation cost difference Define the probability of acceptance Generate random numbers ,like Therefore, according to the Metropolis criterion, this inferior solution is also accepted. If so, then update the current solution; at the same time, if Then update the global optimal solution.

[0115] ⑥ Update operator statistics: Regardless of acceptance, score the operator's performance according to the three-level scoring rule: find the global optimal solution. By dividing the neighborhood, a better solution can be found. The solution that is not improved is accepted. Points. Algorithm iterations. Then, it will be based on the scores of each operator during this period. and number of times used The weights are dynamically adjusted using a weight update formula: Its weights are dynamically adjusted, among which The reaction factor is used. Subsequently, during operator selection, a roulette wheel strategy is adopted, and a probabilistic selection is made from two independent pools of destroy and repair operators based on the updated weights.

[0116] Step 2.7: Acceptance Decision: Calculate the cost difference and decide whether to accept the new solution based on the Metropolis criterion: Step 2.7.1 Comparison of Global Optimal Solutions: Record the number of iterations and temperature at which the optimal solution is found. Step 2.7.2 Solution diversity check: Calculate the similarity of solutions to avoid getting trapped in local optima.

[0117] Step 2.8: Cooling Termination: After completing the iteration of the current temperature, cool down until the termination condition is reached.

[0118] Step 2.9: Output the optimal clustering scheme and delivery route planning results. The optimal clustering scheme determines the service range and spatiotemporal distribution characteristics of each transfer point, providing a decision-making basis for the actual delivery network layout. The delivery route planning results include detailed arrangements of primary routes (county to township) and secondary routes (township to village).

[0119] Step 2.10: Generate a multi-vehicle collaborative delivery scheduling plan. This plan clarifies the specific details such as task allocation, route arrangement, and time scheduling for different vehicle types, ensuring the coordinated cooperation of various transportation resources and maximizing overall delivery efficiency.

[0120] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A delivery method based on spatiotemporal clustering under a shared delivery model in remote rural areas, characterized in that, The steps include: Step 1: For the two-level joint distribution route planning problem, construct a mixed integer nonlinear programming model; the joint distribution mode consists of 1 county-level distribution center, Township transit points The bus routes, which integrate passenger, freight, and mail services, are composed of these lines. The village collection points are operated uniformly by the express delivery alliance. County-level distribution centers publish cargo information through a platform, while crowdsourced vehicles publish their own route information. After acquiring information on these three types of vehicles, the county-level distribution center rationally selects transportation methods and plans routes, pays service fees to crowdsourced vehicles, and sets up a time penalty mechanism. Passenger-freight-mail integrated public transport and self-operated freight trucks depart from and return to the county-level distribution center; crowdsourced vehicles have no fixed origin or destination and can detour to pick up goods without returning. The specific model design is as follows: self-operated freight truck transportation considers fixed vehicle costs; passenger-freight-mail integrated public transport considers the unit product freight cost of entrusted transportation; crowdsourced vehicles adopt an open-loop transportation model and introduce entrusted transportation costs based on unit product and distance, as well as time-related penalties for vehicle violations. The objective function is to minimize the total delivery cost, as follows: In the formula, Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Passing through feasible arc distance, Indicates vehicle Cost per unit distance traveled Indicates vehicle Unit cargo transportation cost Represents village nodes The number of orders, Indicates vehicle Fixed costs. It is a 0-1 variable, representing when the vehicle From node Drive to the node The value is 1 if the time is right and 0 otherwise. To balance delivery efficiency and customer satisfaction, the model introduces a fuzzy time window constraint: requiring that the arrival time of vehicles at each node must be within the acceptable time range for that node. Within the system, for the order collection points in each village, the most urgent orders are selected and their on-time arrival times are recorded. This serves as the optimal service time window for that point. In the time penalty function, let... Penalty cost for arriving early per unit time and The objective function is the penalty cost per unit time for being late. The formula for the time-related penalty for vehicle violations is as follows: The specific constraints are as follows: Node constraints: Each vehicle must return to this point after departing from the county-level distribution center, and each vehicle can use this point at most once. In the formula, the first-level transport vehicle set （ Intercity freight trains, A transportation network integrating passenger, freight, and mail services (public transportation) from county-level distribution centers to township transfer points; Second-level transport vehicle collection （ Township trucks Integrated passenger, freight, and mail public transport (Crowdsourced delivery vehicles) serve the transition from township transfer points to village collection points. Among them, the passenger, freight and mail integrated bus relies on urban and rural bus routes to connect county-level distribution centers, township transfer points and village collection points. Represents the entire set of vertices; These represent the county-level distribution center node set, the village collection node set, and the township transfer network node set, respectively. Indicates the first A collection of vehicles of all grades This represents the collection of all vehicles. 。 express A subset of serviceable nodes for vehicles of the same class; express A subset of village-level collection nodes; express Hierarchical path; ensuring each customer is served by only one vehicle: The number of times the same type of vehicle can be served at each transfer point is limited to no more than once. Vehicle-related constraints: Ensure that the vehicle load does not exceed the capacity limit. ， Indicates when vehicle From node To the node The load, Indicates vehicle Maximum capacity; maximum travel time per vehicle: In the formula, Vehicles are represented as continuous variables. Reaching the node Time, Indicates the maximum routing duration; Customer demand-related constraints: The load after a vehicle accesses a node is equal to the load before access minus the node's demand. ， Representative node Order requirements at the location; ensure vehicles complete their routes empty: In the formula, This represents the total order demand from all customers; it requires that the total cargo volume of the county-level distribution center equals the sum of customer order demands. The departure load of a Level 2 transport vehicle is limited to equal the total order demand of its service customers. The requirement is that second-level transport vehicles arrive at township transfer points no earlier than first-level vehicles. The regulations stipulate that the departure time of first-level transport vehicles from the county-level distribution center is zero. ; Calculate the arrival time of vehicles at each node: ， Indicates vehicle Passing through feasible arc Travel time, Represents a node Service time; ensure arrival time is within the fuzzy soft time window: 。 2. The delivery method based on spatiotemporal clustering under the shared delivery model in remote rural areas as described in claim 1, characterized in that, Step 2, specifically, is as follows: Step 2.1: Data definition and preprocessing; acquire logistics network data, as well as multi-vehicle set and attributes; for crowdsourced vehicles, set mileage limits. And timeliness response capability; obtain order data for each village collection point, and construct a three-dimensional spatiotemporal coordinate system with latitude and longitude as the spatial axis and service time window as the time axis; Step 2.2: Construct a spatiotemporal distance feature matrix; A query strategy tied to service time windows is adopted to request actual road network travel time from the Baidu Maps API, and the spatiotemporal distance is calculated by combining the order time window distance difference; wherein, the calculation of spatiotemporal distance includes: And normalize it using the formula: ; where the weight coefficients α and β are 0.5; the time distance is defined as the relative offset of the start and end points of the time window and normalized to construct the spatiotemporal feature matrix of the village collection nodes; Step 2.3: Data feature identification; perform principal component analysis on the normalized spatiotemporal distance matrix constructed in Step 2.2 to identify the spatial distribution characteristics, time window differences and spatiotemporal interaction characteristics in the spatiotemporal matrix between village collection nodes; Step 2.4: Spatiotemporal clustering based on genetic algorithm; use genetic algorithm to perform spatiotemporal clustering with the minimum intra-cluster distance and the maximum inter-cluster distance as the basic objective; Step 2.5: construct the initial solution using a decimal two-stage encoding scheme; the algorithm adopts a backward and forward combination of initial solution construction strategy: first construct the secondary and primary allocation sets, solve the primary path set; then calculate the vehicle arrival and departure times, solve the secondary path set; Step 2.6: Neighborhood structure design; through the clustering neighborhood search operator in the second stage of the algorithm, randomly select one of them to generate the neighborhood solution after each iteration of spatiotemporal clustering to construct the initial solution; specifically including: (1) (1) Transfer point exchange operator: randomly select two activated township transfer points and exchange the subset of customers they serve; (2) Transfer point reset operator: randomly close an activated transfer point and activate one from the remaining transfer points, and redistribute the customers served by the original transfer point; (3) Customer point exchange operator: randomly select two customers belonging to different transfer points and exchange their affiliation; Step 2.7: ALNS operator design; including the following four types of destruction operators and six types of repair operators: (1) Random removal operator, from the path Random selection (2) Importance removal operator, based on the importance index of customer points. The selection is made based on this indicator, calculated using the following formula: ， Indicates the quantity demanded. Indicates the width of the time window. (3) Worst-case removal operator, removing items via random factor [0.8, 1.2] The node with the worst cost contribution after disturbance; (4) Correlation removal operator, remove the node with the highest correlation with the selected node in turn, and prioritize the node that is not on the same path; Repair operator: (1) Random repair operator: randomly select one of all feasible paths that meet the capacity constraints and insert the customer to be assigned into a random position; (2) Greedy repair operator: traverse all feasible insertion positions and calculate the cost increment caused by the insertion operation. The cost calculation requires re-execution of the vehicle model adaptation process to determine the optimal vehicle model required after the load increase and its corresponding fixed cost and variable cost. The customer point insertion position is selected based on the minimum cost increase after insertion; (3) Random greedy repair operator: increase the cost increment when calculating transportation cost. Random perturbation to ensure the diversity of solutions obtained in the repair phase; (4) Importance repair operator: After sorting according to customer importance, customers with high importance are inserted into the path first according to the greedy principle; (5) Maximum regret repair operator: Calculate the cost difference between inserting a customer into the optimal position and the second-best position, and prioritize inserting the customer with the largest cost difference; (6) First-level path repair operator: Use a greedy strategy to insert the transfer station into the optimal position of the first-level path; Step 2.8: The algorithm introduces an adaptive weight adjustment mechanism; In the early stage of the search, all destruction and repair operators are given the same initial weight; During the iteration process, the algorithm tracks the performance of each operator and scores it according to the three-level scoring rule: Find the global optimal solution By dividing the neighborhood, a better solution can be found. The solution that is not improved is accepted. Points; each iteration of the algorithm Then, it will be based on the scores of each operator during this period. and number of times used The weights are dynamically adjusted using a weight update formula: In the formula The reaction factor is used to adjust the influence of historical experience and current performance on the weight update, so as to realize the adaptive evolution of the search strategy; Step 2.9: Iterative convergence determination and global optimal solution output; Perform simulated annealing cooling and termination condition determination to control the convergence process of the algorithm and output the final delivery plan.

3. The delivery method based on spatiotemporal clustering under a shared delivery model in remote rural areas as described in claim 2, characterized in that, Step 2.4 is specifically operated as follows: Step 2.4.1: Determine the number of clusters. If the upper limit does not exceed the total number of candidate township transit points, and the lower limit is within a reasonable clustering range, then iterate... Select appropriate clustering units; Step 2.4.2: Calculate the spatiotemporal distance matrix; Step 2.4.3: Population initialization; Create a population containing... The initial population of chromosomes, each chromosome consisting of... A candidate clustering scheme is composed of randomly selected representative customer order IDs. Step 2.4.4: Fitness assessment and capacity constraint check; allocate the remaining orders to the nearest representative order, and summarize the sub-problem costs as the chromosome fitness value; the fitness function is designed as follows: ,in For the spatiotemporal distance within a cluster, The distance between clusters, To avoid division by zero, a validity check is performed before calculating fitness: Calculate the fitness of each cluster. Total demand ； like If the clustering requirement exceeds the load limit of the largest vehicle in the fleet, then that individual is deemed infeasible, and the process is iteratively executed until the maximum number of generations is reached. Step 2.4.5: Perform selection, crossover, and mutation operations; use roulette wheel selection, single-point crossover, and constrained mutation operations. Each cluster contains a transfer point and its assigned set of village collection points. Select high-quality individuals to enter the next generation based on fitness values. With probability Select parent generation pairs, apply the crossover operator to generate offspring, and replace parent generation with offspring; probability. The chromosomes in the population are mutated to maintain the diversity of the population; the newly generated population enters the next round of iteration, returns to step 2.4.4 to evaluate fitness, and terminates the iteration when the maximum number of generations is reached.

4. The delivery method based on spatiotemporal clustering under a shared delivery model in remote rural areas as described in claim 2, characterized in that, Step 2.5 is specifically implemented as follows: Step 2.5.1: Initialization and path segmentation; Based on the clustering scheme determined in Step 2.4, each transit station within the cluster and all customers within its service range are considered as independent paths, and an initial set of secondary paths is constructed; Step 2.5.2: Calculation of spatiotemporal distance savings; According to the spatiotemporal distance calculation formula, the spatiotemporal savings between nodes are calculated, and a list of savings values ​​that integrates normalized spatial distance savings and spatiotemporal distance savings is generated and arranged in descending order; Step 2.5.3: Path merging based on vehicle model adaptation logic; traverse the savings value list sequentially using a greedy strategy, attempting to merge paths; during each merging attempt, execute the vehicle model adaptation logic and calculate the state: calculate the merged path. Cumulative load Total driving distance and the estimated completion time Feasibility screening: Traversing the vehicle model set Filter out those that simultaneously meet the capacity constraint ( ) and mileage constraints ( Fable vehicle subset Optimal vehicle matching: If For non-empty data, calculate the comprehensive cost of each vehicle type in the subset and select the minimum value: For self-operated trucks, the cost is calculated as follows: Represents the total cost of multiple transportation modes in rural areas. Indicates vehicle Passing through feasible arc Travel time, Indicates vehicle Cost per unit time travel distance Indicates vehicle Cost per unit distance traveled; Indicates vehicle Fixed costs; Indicates vehicle Passing through feasible arc The distance; the cost of freight buses is Crowdsourced car service is The fare for passenger, freight, and mail integrated public transportation is calculated based on the unit volume of goods transported. All of them have time penalties added. ； If a match is found, select the vehicle model and perform the merge; otherwise, abandon the merge and keep the original state until the list has been traversed. Step 2.5.4: Constructing the first-level path; The set of activated transfer stations is determined based on the secondary path set and used as the demand points. The primary path is constructed using the nearest neighbor heuristic algorithm, and the vehicle model adaptation logic in step 2.5.3 is applied to match the optimal vehicle model for the primary path. Finally, the primary and secondary path sets are integrated to output an initial solution containing complete vehicle model and path information.

5. The delivery method based on spatiotemporal clustering under a shared delivery model in remote rural areas as described in claim 2, characterized in that, Step 2.9 is specifically performed as follows: Step 2.9.1: Accept the judgment; calculate the cost difference. Define the probability of acceptance Generate random numbers ,like Therefore, according to the Metropolis criterion, this inferior solution is also accepted. If so, then update the current solution; Update the global optimal solution; regardless of acceptance, update the operator's weight in ALNS based on its performance in this operation; Step 2.9.2: Cooling termination; complete the current temperature... After the second internal circulation, cooling and temperature reduction are performed: Then return to step 2.9.1 to proceed to the next temperature iteration; if If the condition is met, the algorithm terminates and outputs the final globally optimal solution.

6. The delivery method based on spatiotemporal clustering under a shared delivery model in remote rural areas as described in claim 2, characterized in that, In step 2.7, the relevance removal operator removes a customer randomly from the customer set and adds it to the set to be removed. Subsequently, according to the formula Calculate the remaining customers and selected customers in the customer set. The correlation, On behalf of clients With customers The distance between, Indicates customer The maximum distance between the remaining customers and the customer cluster. This is a binary variable used to determine if two clients are on the same path. If the client... With customers If the path is the same, select 1; otherwise, select 0. Finally, if the set to be removed is not full... For each customer, select the customer with the highest relevance and add them to the set, then treat them as the new current customer and repeat the previous step; otherwise, terminate.

Images