Robust two-stage cycle pick and load and path optimization method and system

Abstract

The present application relates to the technical field of logistics transportation scheduling and loading optimization, and discloses a robust two-stage cyclic pickup loading and path optimization method and system, which comprises: obtaining plan data and real-time state data; based on the plan data, solving a preset plan stage model to generate a main plan; in the execution stage, based on the real-time state data, performing execution checking on the main plan, when it is found by the checking that the container loading is not feasible or the latest factory return is late, triggering the solving of a preset minimum repair model to generate an executable repair scheme; and outputting the repair scheme or the main plan that is checked to be feasible as a final scheduling scheme. The present application constructs a plan and execution two-stage collaborative optimization closed loop, synchronously integrates three-dimensional container loading feasibility checking, opportunity cost quantification of reserved buffer space and time efficiency control based on distribution robust optimization in path planning, and significantly improves the execution reliability and economy of the cyclic pickup scheme.

Description

Technical Field

[0001] This invention relates to the field of logistics transportation scheduling and loading optimization technology, specifically to a robust two-stage cyclic loading and route optimization method and system. Background Technology

[0002] In the inbound logistics scenario of automobile manufacturing, OEMs need to replenish components such as seats, wiring harnesses, and interior parts frequently from multiple parts suppliers. To facilitate turnover and loading / unloading, components are typically packaged in standardized containers such as collapsible crates or pallet boxes, with the container serving as the basic unit for loading, unloading, and transportation. These containers generally correspond to a limited set of container types, and the demand for different components can be converted into the number of containers of the corresponding type. Transport vehicles can adopt a milk-run transportation model, departing from the OEM or consolidation center, picking up goods from multiple suppliers along a pre-planned route, and then returning to the OEM for centralized unloading. Compared to direct delivery from suppliers, this model helps reduce empty runs and waiting times, improves vehicle turnover efficiency, and allows OEMs to reduce on-site inventory and the number of loading / unloading points, thus it is quite common in automobile inbound logistics.

[0003] However, in practice, whether OEMs operate their own vehicles or collaborate with third-party logistics companies, the cyclical pickup and delivery model still relies heavily on manual experience or heuristic rules based on total volume and load capacity. Planners often determine capacity feasibility solely based on the supplier's total container volume and truck volume, rarely simulating and verifying the three-dimensional placement of containers within the truck during the planning phase. Due to differences in container dimensions, rotational and stacking constraints, even if the total volume does not exceed the truck volume limit, situations may arise where capacity is feasible at the planning level but not feasible for on-site loading. Meanwhile, with rising freight rates and stricter carbon emission controls, empty vehicle runs and insufficient truck space utilization will lead to decreased vehicle utilization, increased transportation costs, and increased carbon emissions, indicating significant room for optimization.

[0004] Several related solutions have emerged in existing research and engineering practice, such as Vehicle Routing Problem (VRP) models that only consider scalar capacity constraints such as volume or weight, and Vehicle Routing Problem (VRPTW) models with time windows; 3D bin packing problems solved using 0-1 programming or heuristic algorithms; and coupled methods that embed simplified bin packing logic into metaheuristic frameworks such as Adaptive Large Neighborhood Search (ALNS). However, these solutions typically treat bin packing as a feasibility check or local adjustment process after route planning, making it difficult to achieve coordinated optimization of loading and route costs. In industrial-scale scenarios with multiple vehicles or vehicle models, there is still a lack of systematic methods that balance loading feasibility assurance with solution stability. Furthermore, the economic impact of spare space in the vehicle compartment is mostly limited to the loading rate indicator, lacking a modeling method that explicitly quantifies it and incorporates it into the total cost target.

[0005] Based on the above background, it can be seen that the existing cyclical pickup solution still has shortcomings in the following aspects: Firstly, most methods, when designing cyclical pickup routes, often approximate vehicle loading capacity using scalar capacity constraints such as volume or load capacity. They first obtain the route and vehicle configuration through a vehicle path planning model, and then arrange the container loading within each route or rely on on-site experience. This sequential decomposition makes it difficult for the path decision-making stage to reflect the loading difficulties caused by constraints such as the placement, rotation, stacking, and load-bearing capacity of the containers in the three-dimensional space of the vehicle. It also makes it difficult to quantify and incorporate space waste during the loading process into optimization, which can easily lead to situations where the planned capacity is feasible but the actual loading is not, or where loading is possible but the loading is loose and the space is underutilized.

[0006] Secondly, existing cost characterization primarily focuses on mileage-related variable costs and vehicle activation fixed costs. Even when load factor or space utilization rate indicators are introduced, they are mostly used as auxiliary evaluations rather than optimization targets on the same scale as costs. Due to the lack of a mechanism to explicitly quantify the loss of cargo space as opportunity cost and incorporate it into the total cost target, existing solutions struggle to achieve a quantitative balance with fare calculation rules (such as linear or segmented pricing) and standardized container units. Therefore, it is difficult to make a systematic comparison of the overall cost between a solution that involves moderately detouring to consolidate pickups from more suppliers and a solution that reduces detouring but results in a greater waste of empty space.

[0007] Third, at the solution level, most publicly available technologies rely on heuristic or metaheuristic algorithms. Although they can quickly provide feasible solutions when the scale is small or the constraints are weak, they generally lack a unified optimality bound (lower bound) and a provable pruning mechanism, making it difficult to provide verifiable solution quality assurance. When further factors such as three-dimensional binning constraints, multiple vehicles, multiple vehicle models, and time constraints are introduced, the search efficiency and solution stability are prone to decline, making it difficult to simultaneously ensure feasibility and solution reliability on an industrial scale.

[0008] Fourth, existing solutions are mostly based on static planning data to formulate routes and loading plans, lacking real-time status updates and rolling remediation mechanisms for the execution process. They are difficult to cope with fluctuations in the number of boxes, order insertions, supplier operation delays, and dynamic changes in travel time caused by weather or traffic. At the same time, they lack tail risk control measures for concentrated lateness risks caused by adverse situations in JIT (Just-In-Time) scenarios, which may trigger material shortages or line stoppages in extreme cases.

[0009] Therefore, the technical problems to be solved by the present invention include at least the following: (1) Introduce three-dimensional packing feasibility assessment in the vehicle route planning stage to avoid situations where packing is feasible on paper but not feasible on site due to only volume assessment. (2) The vehicle activation cost, driving cost and opportunity cost of wasting carriage space are weighed together under the same objective function; (3) Provide a calculable rolling remediation mechanism under execution disturbances (box count fluctuations, order insertions, operation delays, weather, traffic) and define remediation boundaries; (4) To meet the Just-in-Time (JIT) requirement of returning to the factory at the latest, constraints or penalties are imposed on the tail risk of lateness. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention provides a robust two-stage cyclic loading and route optimization method and system.

[0011] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a robust two-stage cyclic loading and route optimization method, comprising: Acquire planned data and real-time status data. Planned data includes the supplier's planned number of containers, vehicle parameters, and transportation cost parameters. Real-time status data includes the number of containers executed and dynamic travel time. Based on the planned data, the pre-defined planning stage model is solved to generate the master plan. The master plan includes the activation decision of the planned vehicles, the main service relationship of the suppliers, the planned routes, and the reserve space to cope with fluctuations. During the execution phase, the master plan is checked based on real-time status data. The execution check includes packing feasibility check and arrival time propagation check based on dynamic travel time. When the execution check finds that packing is not feasible or that the latest return to the factory is delayed, the preset minimum repair model is triggered to solve. Within the resource boundaries determined by the master plan, an executable repair plan is generated by activating emergency vehicles and / or partially rearranging the planned route. Output the repair plan or the verified feasible master plan as the final scheduling plan.

[0012] In one embodiment, the dynamic driving time is obtained in the following way: A distance matrix is ​​constructed based on the shortest travel distance between nodes in the road network, and arcs are generated based on the distance matrix. Corresponding base driving time The planning period is divided into a set of discrete time buckets. and define the time mapping function. Used to convert any absolute time Mapped to the corresponding time bucket The nodes include OEM nodes and supplier nodes. For each arc and each time bucket Generate a set of micro-scenes , Let i and j represent nodes respectively, which are arcs. The two endpoints; for each micro-scenario Configure delay ratio samples and the corresponding nominal probability ; Based on the sub-Blule bar optimization method, for each arc-time bucket combination Set ambiguity radius To construct an ambiguous set containing the true probability distribution. And calculate the worst-case expected delay ratio within this ambiguity set. ; Through the time mapping function The worst-case expected delay multiplier within the time bucket. Mapped to weather delay ratio as a function of absolute time ; Represents arc In absolute time Time Bucket The worst-case expected delay ratio; Based on the aforementioned weather delay multiplier and baseline travel time, a robust travel time is generated for arrival time propagation during the planning phase. During the execution phase, based on robust driving time and the delay ratio of real-time observation or forecast. Generate the execution time for the verification. ; The dynamic driving time includes the baseline driving time. Robust driving time and execution driving time .

[0013] In one embodiment, the micro-scene is generated in the following ways: Historical sampling method: Samples are drawn from historical periods into time buckets. Inner arc Real delay rate samples And estimate the nominal probability based on historical frequency. Post-normalization; Rule-based perturbation method: Apply interval or discrete perturbations to the baseline delay ratio to generate delay ratio samples. Furthermore, the lower bound of the delay multiplier sample is truncated and verified, and the nominal probability is normalized using equal weights or according to rule weights. Combined business scenario method: Based on the actual combination of multiple weather events that may occur simultaneously in the business, generate micro-scenarios and attach event tags.

[0014] In one embodiment, the sub-Bruker optimization method is used for each arc-time bucket combination. Set ambiguity radius To construct an ambiguous set containing the true probability distribution. And calculate the worst-case expected delay ratio within this ambiguity set. Specifically, it includes: Considering the estimation error of the nominal probability corresponding to the delay ratio sample, it is necessary to consider the estimation error for each arc-time bucket combination. Constructing an Ambiguity Set It is used to characterize uncertain sets where the true distribution lies in the neighborhood of the nominal distribution, and the chi-square divergence is used to constrain the deviation of the true distribution from the nominal distribution to not exceed the radius of ambiguity. : ; The dimension is The nonnegative real vector space; It is a collection of micro-scenes The number of elements; Representing the set of ambiguities One of the candidate true probability distributions, This represents the transpose of a vector consisting entirely of 1s. This indicates satisfaction. Represents the true probability distribution of candidates In micro-scenarios The corresponding probability value.

[0015] In one embodiment, the worst-case expected delay multiple is calculated within the ambiguity set. Specifically, it includes: First calculate the nominal probability mean With variance : , ; The worst-case expected delay ratio can be obtained using either a closed-form or upper-bound form. : .

[0016] In one embodiment, the step of solving a preset planning stage model based on planning data to generate a master plan includes vehicle activation decisions, primary service relationships with suppliers, planned routes, and buffer space reserved to cope with fluctuations, specifically including: Introduce the following decision variables: vehicle model Planned vehicles Enable variables , indicating supplier Whether it is based on the car model Planned vehicles The main service variable of the main service , indicating vehicle model Planned vehicles Is it in the arc? Upward travel arc selection variable Model Planned vehicles Arrival at the supplier The moment Model Planned vehicles Return to factory time Model Planned vehicles The latest number of people arriving late to the factory and car models Planned vehicles spare space Cost function during the planning phase Defined as: ; Indicates planned vehicles , Indicates model t, Represents a set of vehicle models. Indicates vehicle model The planned vehicle collection, This represents the fixed activation cost per planned vehicle corresponding to model t. Represents the set of arcs. For model In the arc The nominal transportation cost parameter on the screen, This represents the opportunity cost coefficient. This represents the penalty weighting coefficient for the latest return to the factory. Based on the planned data, constraints are configured for the planned phase model, including main service uniqueness constraints, linkage constraints, route structure constraints, lower bound constraints of free space, bin packing feasibility constraints, and arrival time propagation constraints. Main service uniqueness constraint: ; Represents the set of suppliers; Linkage constraints: ; Route structure constraints: , ; ; Indicates that the arc satisfies All nodes , This indicates that the variables satisfying the stated conditions are summed. Lower bound constraint of free space: ; ; in, For suppliers The planned number of boxes, This refers to the volume of a single box for the corresponding box type. Let t be the effective volume of the carriage. Packing feasibility constraints: Define a set of service providers for each planned vehicle k of vehicle type t. And require: ; in, Indicates supplier The set of containers under the planned number of containers. For the binning determination function, Indicates the model Under the constraint of the effective size of the carriage, there exists at least one three-dimensional placement scheme that satisfies both boundary constraints and non-overlapping constraints; Arrival time propagation constraint: when arc When selected, there are: ; Return time With the latest time to return to the factory Association, defining the late quantity : ; This indicates the absolute time at which the vehicle is scheduled to depart from node 0 of the OEM. Represents a constant; Solving the planning phase model after configuring constraints, specifically using a column generation framework: Generate a list of complete feasible routes that have passed the packing feasibility constraint verification and meet the arrival time propagation constraint; The main problem is to select several routes from the generated route list to cover all suppliers and minimize the planned cost; The pricing subproblem generates new route sequences with negative reduction costs under the guidance of dual information, and embeds bin packing increment verification and arrival time propagation pruning during route expansion.

[0017] In one embodiment, during the execution phase, the master plan is executed and verified based on real-time status data. This execution verification includes a packing feasibility verification and an arrival time propagation verification based on dynamic travel time. When the execution verification reveals that packing is infeasible or results in a late return to the factory, a preset minimum repair model is triggered, specifically including: At the rolling moment Collect real-time status data and align it with the master plan; Real-time status data includes at least: the scrolling time. Suppliers Actual deliverable box count , rolling time supplier Additional order increment Availability status and rolling schedule of planned and emergency vehicles execution vehicle Current location , rolling time execution vehicle Current absolute time , rolling time execution vehicle The already loaded container assembly Spare space, and rolling time In the arc Delay ratio of real-time observations or forecasts ; Based on the real-time status data, update parameters for the execution phase are generated: Supplier Planned number of containers Define the observation bias for the benchmark. This, in turn, forms the number of execution bins input to the execution phase. : ; The planned route already selected during the planning phase Generate trigger parameters ,when When, set the trigger parameter This triggers packing feasibility checks and arrival time propagation checks; otherwise, it causes... Skip the packing feasibility check and only perform the arrival time propagation check; Packing feasibility check: based on the number of boxes to be packed. Generate a set of boxes to be loaded, and call the box loading determination function to determine whether there is a three-dimensional placement scheme that satisfies the boundary constraints and non-overlapping constraints; Arrival time propagation verification: based on the execution travel time Arrival time propagation calculations are performed to obtain the return time of each vehicle and the latest late return time. When it is found that packing is not feasible or the latest return to the factory is delayed, the minimum repair model is triggered to perform minimum repair optimization.

[0018] In one embodiment, the step of generating an executable remediation plan by activating emergency vehicles and / or partially rearranging the planned route within the resource boundaries determined by the master plan specifically includes: Using the master plan as the resource boundary, and the set of execution bins from each vendor. Vehicle status and delay rates for real-time observations or forecasts As input, a minimal repair model is constructed, wherein the decision variables of the minimal repair model include: the vehicle performing the repair. Enable variables , indicating supplier Whether it is executed by vehicle The main service provider service allocation variable , indicating the vehicle to be executed Is it in the arc? Upward travel arc selection variable , execution vehicle Arrival at the supplier Arrival time With the execution vehicle Return to factory time The vehicles involved include planned vehicles and available emergency vehicles. Configure constraints for the minimum repair model, including resource boundary constraints, service integrity constraints, local repair constraints, route structure constraints, arrival time propagation constraints, and bin packing feasibility constraints. Solve the minimum repair model and output a repair solution. The repair solution includes vehicle activation result, supplier service allocation result, arc selection result, arrival time, return time, lateness amount, and incremental cost.

[0019] In one embodiment, the resource boundary constraint is: ; Indicates the vehicle model during the execution phase. Planned vehicles Enable or disable? Indicates the model in the planning stage Planned vehicles Enable or disable? Indicates vehicle model Planned vehicles The set, Indicates emergency vehicles The set, Indicates the first stage of execution The variables for the activation of emergency vehicles This represents the maximum number of emergency vehicles that can be deployed during the implementation phase. Service integrity constraints are: For each supplier The execution phase must be performed by one and only one execution vehicle. One service: ; For supplier aggregation, Indicates the set of vehicles to be executed; ; Local repair constraints: By limiting the adjustment range through transfer indicator variables and imposing adjustment penalties on suppliers that undergo transfer, repair schemes tend to be locally rearranged; Route structure constraints: For each executing vehicle, the route must start from OEM node 0 and return to OEM node 0. ; Each supplier being served Satisfying the one-in-one-out constraint: ; Represents the set of arcs; Arrival time propagation constraints: For any arc Adopting large Arrival time propagation constraints: ; in, For suppliers Homework time, This serves as the absolute time reference for this round of repairs. Represents a constant; based on the latest factory return time. Define the execution vehicle Late arrivals : ; Number of execution boxes used Apply packing feasibility constraints to each executing vehicle: ; For the binning determination function, This indicates that there exists a three-dimensional placement scheme that satisfies both boundary constraints and non-overlapping constraints under the effective size constraints of the carriage of a specific vehicle model; Indicates supplier Number of boxes executed The lower box assembly, This represents the union operation. Indicates the vehicle to be executed The corresponding vehicle model identifier.

[0020] In a second aspect, the present invention provides a computer system including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method of any embodiment of the first aspect.

[0021] Compared with the prior art, the beneficial technical effects of the present invention are: In the milk-run transportation scenario, the two-stage collaborative optimization technology of this invention integrates packing feasibility, spatial opportunity cost, time-varying robust travel time and late return penalty into a unified decision-making framework through integrated modeling of planning, execution verification and trigger-based minimum repair. This ensures the feasibility of the solution while achieving comprehensive optimization of cost and timeliness risks, and improves the engineering solvability in industrial-scale scenarios.

[0022] Improved loading feasibility: Packable is introduced in both the planning and execution phases, and a disabled constraint is generated for the "vehicle-service provider set" that cannot be loaded, and the solution is iteratively solved. This ensures that the output solution satisfies both vehicle capacity and three-dimensional loading physical constraints, reducing the probability of events that are feasible on paper but cannot be loaded on site.

[0023] Overall costs can be quantified and weighed by introducing variables of spare space and opportunity cost coefficients. The economic impact of spare space in the carriage is explicitly incorporated into the objective function, thereby achieving unified optimization of transportation costs (including cubic kilometers and mileage), vehicle activation costs, and space waste costs, and improving the economy and interpretability of the solution.

[0024] Robust timeliness and controllable lateness risk: By embedding time-varying delays into arrival time propagation through time bucket and weather delay multiplier, a robust travel time interface that evolves with absolute time is formed. The lateness amount is defined based on the OEM’s latest return time and is used as a trigger criterion and penalty to suppress the latest return timeout, making the timeliness assessment of the solution more robust and the lateness risk more controllable under uncertain weather or traffic conditions.

[0025] Enhanced solvability of the project: A dynamic cutting framework using column generation, branch pricing (preferred), or branch and bound with lazy constraints (compact implementation) is adopted to verify and prune complex constraints such as bin packing infeasibility and subloops as needed during the solution process, avoiding scale expansion caused by explicit expansion of geometric constraints, and improving the stability and availability of the solution in multi-supplier and multi-vehicle scenarios. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention.

[0027] Figure 2 This is a logic flowchart of the present invention.

[0028] Figure 3 This is a flowchart illustrating the time bucket division and weather delay rate calculation process of the present invention.

[0029] Figure 4 This is a flowchart illustrating the use of a column generation framework to solve the planning phase model in this invention.

[0030] Figure 5 This is a flowchart of the execution phase of the present invention.

[0031] Figure 6 This is a diagram showing the convergence of the solution in an embodiment of the present invention.

[0032] Figure 7 This is the master roadmap for the planning phase in an embodiment of the present invention.

[0033] Figure 8 This is a diagram showing the activation and repair of emergency vehicles during the execution phase in an embodiment of the present invention.

[0034] Figure 9 This is an example diagram of the packing-routes combination in an embodiment of the present invention. Detailed Implementation

[0035] A preferred embodiment of the present invention will now be described in detail with reference to the accompanying drawings.

[0036] To simultaneously ensure plan stability, execution feasibility, and timeliness risk control, this invention characterizes the milk-run transportation model with standardized packaging as a two-stage collaborative optimization problem involving planning and execution. The planning stage (Stage-1) uses the planned number of containers, vehicle parameters, transportation cost parameters, baseline travel time, and dynamic travel time during the planning stage as benchmarks to determine the reservation or activation of planned vehicles, the primary allocation relationship with suppliers, and the reserve space for fluctuations, forming a feasible master plan framework. The execution stage (Stage-2), under real-time observation or forecasting conditions, first performs container loading feasibility checks and arrival time propagation checks on the planned routes generated in the planning stage. When the checks find that container loading is infeasible or that the latest return to the factory is delayed, minimum repair optimization is triggered. This eliminates container loading infeasibility and delays by activating emergency vehicles and partially rearranging or transferring planned routes. The planned routes deemed infeasible in the execution stage are then fed back to the planning stage to avoid re-selection.

[0037] This invention proposes a method for collaborative optimization of multi-vehicle 3D packing and route costs in a milk-run scenario with standardized packaging for automotive OEMs. Starting with the OEM's Bill of Materials (BOM) data, supplier spatial distribution and packing quantity, vehicle transportation costs and freight contracts, this method employs a series of steps including data acquisition and preprocessing, scenario construction, time bucket division and weather delay rate calculation, master plan solution, and remedial plan solution. Ultimately, it generates a multi-vehicle milk-running scheme that satisfies the physical constraints of 3D packing and achieves minimum (or cost optimization) cost under the optimized model.

[0038] like Figure 1 and Figure 2 As shown, a robust two-stage cyclic loading and route optimization method of the present invention includes the following steps: S1, Data Acquisition and Preprocessing; S2, Weather Micro-Scenes and Time Bucket Data Construction; S3, Time Bucket Division and Weather Delay Multiplier Calculation; S4, master plan solution during the planning phase (planned vehicle activation, supplier master service relationship and reserved buffer). S5 performs real-time status data updates and status alignment; S6, trigger-based execution verification and minimum repair (emergency vehicle activation or partial reconfiguration) and feedback closed loop during the execution phase; S7 outputs an executable scheduling scheme and a visual loading guide.

[0039] I. Data Acquisition and Preprocessing.

[0040] During data acquisition and preprocessing, production and delivery plan data for a planning period (e.g., day, shift, or rolling window) from the OEM is accessed. Component requirements at the Bill of Materials (BOM) level are aggregated by supplier to obtain the planned pickup quantity for each supplier within the planning period. And container type information. Subsequently, based on the company's established packaging rules (correspondence between parts and container types, loading conversion coefficients), the different parts requirements are uniformly mapped to a set of container types and quantities with the packaging box as the smallest transportation unit. The validity of the container type, the non-negative integer nature of the number of boxes, and the completeness of the supplier mapping are verified, thereby providing a consistent data standard for subsequent container loading feasibility and vehicle capacity constraints.

[0041] Regarding spatial and road network parameters, this invention reads the coordinates or location codes of the OEM and its suppliers from a Geographic Information System (GIS) or enterprise geographic master data, calculates the shortest driving distance (or mileage) between nodes, and constructs a distance matrix. , For arc The distance to the road network. Based on this, combined with the vehicle type. The fare and cost structure (e.g., mileage-based pricing, segmented pricing, fixed start-up fees, and variable mileage fees) will Convert to car model In the arc On the nominal transportation cost parameter And simultaneously generate a baseline driving time matrix. Baseline driving time can be It is obtained by conversion with the reference speed coefficient or reference time coefficient, or by statistical calibration from historical driving records, and serves as the reference quantity for subsequently introducing weather delay multipliers and arrival time propagation constraints. An arc is the basic connecting unit in a path network. This indicates that the vehicle can move from the node. Drive directly to the node A transportation link.

[0042] Regarding vehicle and loading parameters, read the vehicle model set. Effective interior dimensions of the passenger compartment for each vehicle model and effective volume (Or the effective volume given in the vehicle model calibration table), and compile information such as the fixed cost, mileage cost, and maximum available quantity of planned and emergency vehicles; at the same time, load the latest return time to the factory under Just-in-Time (JIT) system. and the upper limit of emergency vehicle activation during the execution phase Regarding business constraint parameters, the time in this invention can be represented based on the relative time of the start of the planning period or a unified calendar timestamp. Finally, this invention receives weight parameters configured by the user or strategy module, including the opportunity cost coefficient of reserved space. Just-in-Time (JIT) Lateness Risk Weighting and the confidence level of the Conditional Value at Risk (CVaR) model. And so on, and complete parameter consistency and dimension verification, thereby forming the standardized input required for the two-stage optimization model.

[0043] II. Weather micro-scenarios and time bucket data construction.

[0044] like Figure 3 As shown, after data acquisition and preprocessing, this invention enters the process of constructing weather micro-scenarios and time bucket data. This process explicitly transforms common weather uncertainties during execution into a computable set of arc-time bucket-micro-scenarios, providing a unified input for robust master plan solving in the planning phase and execution verification and trigger repair in the execution phase. Firstly, based on the preprocessed data... Generate baseline driving time (For example The planning period is divided into a set of discrete time buckets. Define the time mapping function Used to express absolute time Map to the corresponding time bucket.

[0045] For each arc and each time bucket This invention generates a set of weather micro-scenes. For any micro-scenario Construct a delay ratio sample corresponding to this micro-scenario. (satisfy (and give the nominal probability) ,in and Based on this, further parameters for the sub-Bruker bar can be given (e.g., the ambiguity radius based on chi-square divergence). ), and based on this, calculate the arc in the time bucket. Worst expected delay ratio Finally, the weather delay ratio as a function of absolute time is obtained through bucket mapping. And form robust driving time : ; The above Used for arrival time propagation and late return penalty calculation during the planning phase, and also serves as the benchmark input for arrival time propagation verification during the execution phase; when the delay ratio is obtained from real-time observations or forecasts during the execution phase. This can be used to determine the execution time. Used for verification and triggering judgment.

[0046] This invention uses historical weather records, road traffic records, third-party meteorological service data, or third-party traffic service data as a basis to identify sources of disturbance that may cause travel time deviations, including but not limited to: weather events such as rainfall, snowfall, reduced visibility, icing, and strong winds, and their impact on different road types; disturbances can also be configured by business rules (e.g., multiplier ranges corresponding to different weather levels, congestion superposition coefficients for different time periods, etc.). A limited number of representative micro-scenario samples are generated according to preset rules, and the micro-scenario generation methods may include: (1) Historical sampling method: sampling from historical periods in time buckets inner arc Real delay ratio sample And estimate the nominal probability based on historical frequency. Post-normalization; (2) Rule-based perturbation method: Apply interval or discrete perturbation to the baseline delay ratio to generate And its lower bound (such as not less than 1) is truncated and verified; the nominal probability can be normalized by equal weight or by rule weight; (3) Combined business scenario method: generate micro-scenarios by combining weather events such as rainfall, fog, icing, and strong winds in a way that can occur simultaneously in business operations, and attach event tags (e.g., heavy rainfall, low visibility) so that different ambiguity radii can be used in subsequent parts. Alternatively, robustness measures can be implemented for different multiplier limits.

[0047] To control computational scale and solution time, clustering or representative sample screening can be used when there are too many candidate micro-scenes. The scope is reduced to retain a representative set of micro-scenarios covering the main delay patterns. Final output. , , And optional And based on this, , and ,in It provides a unified driving time interface input for the planning and execution phases, thereby achieving consistent data-driven processes that prioritize robust planning followed by execution verification or triggered repairs.

[0048] III. Time Bucket Division and Weather Delay Multiplier Calculation.

[0049] After completing data acquisition and preprocessing, and initializing the arc-time bucket-micro-scenario parameters, this invention proceeds to the time bucket partitioning and weather delay ratio calculation process. This maps the impact of weather on road traffic to the travel time of each arc in a manner that varies with absolute time, thereby providing a unified, robust travel time input for robust master plan solving in the planning phase and arrival time propagation verification or trigger repair in the execution phase. Generate a baseline driving time matrix based on And add weather delay multipliers to each arc at different time periods. , making the arc At the moment of departure The travel time is expressed as .in, can be Calculations show that It is a unit distance reference time coefficient, or obtained from historical driving records.

[0050] (1) Time Bucket Set Partition and mapping functions : This invention divides the planning cycle (e.g., a day's runtime) into a set of discrete time buckets. and define the time mapping function. Used to convert any absolute time It is assigned to the corresponding time bucket. Through this process, the weather and traffic performance of the same arc at different times can be expressed using a piecewise constant method, avoiding approximation by a single static coefficient.

[0051] (2) Arc – Time Bucket – Micro-Scenario Collection The establishment of: For each directed arc and each time bucket This invention establishes a set of possible weather micro-scenarios for the arc within the time bucket. For any micro-scenario Given a parameter triple: nominal probability (satisfy Delay ratio sample (Used to characterize the magnification relative to the baseline driving time) and the radius of ambiguity. (Used to characterize the uncertainty of the nominal distribution). Among them, This can be obtained based on the percentage decrease in speed under similar historical weather conditions, third-party weather and traffic correlation data, or enterprise statistical data; The value can be determined by the sample size, volatility, or strategy parameter configuration. Preferably, to avoid a denominator of zero, micro-scenarios with a nominal probability of 0 can be merged or a method can be used. Smoothing process, making .

[0052] (3) Construction of distribution ambiguity sets based on chi-square divergence: Considering nominal probability There is an estimation error, for each arc. and each time bucket Constructing a distribution ambiguity set It is used to characterize uncertain sets whose true distribution lies in the neighborhood of the nominal distribution, and the chi-square divergence is used to constrain its deviation to not exceed a certain value. Its form is: ; Through the above construction of ambiguity sets, the model can still remain robust under certain robustness when the nominal distribution is unreliable. express It is a vector in which all components are non-negative and the sum of all components is equal to 1, i.e. To define in the micro-scenario set The legal probability distribution on.

[0053] (4) Worst-case expected delay ratio Computational and engineering implementation: In the collection of ambiguities Internal calculation of the worst-case expected weather delay multiplier: ; To facilitate engineering implementation, in the implementation method, the nominal mean and variance can be calculated first: ; The worst-case expected delay ratio is obtained using either a closed-form or upper-bound form: ; The above results can be directly used to generate an arc-time bucket multiplier lookup table (worst-case expected delay multiplier lookup table) data structure. This is used for subsequent arrival time propagation and invocation.

[0054] (5) Weather delay ratio as it evolves with absolute time : By combining the bucket ratio with the mapping function, a weather delay ratio that varies with absolute time is defined: ; Therefore, the same arc At different departure times, the robustness factor of the corresponding time bucket will be automatically adopted to model the evolution of weather impact over time.

[0055] (6) Outputs and Interfaces: This section outputs the baseline driving time matrix. Time bucket set and time mapping function worst-case expected delay ratio lookup table Its time mapping form In the calculation of robust arrival time propagation constraints and latest return-to-factory time during the planning phase, arc... The travel time can be uniformly expressed as ,in This is the absolute time of the route's start. To reach the supplier The timing of arrival times is then used to embed robustness to time-varying weather into vehicle routing and the latest return-to-factory time constraint or penalty calculation. This can also be achieved during the arrival time propagation verification in the execution phase. Replace with the delay ratio of real-time observations or forecasts during the execution phase. ,form It is used for verification and triggering judgment.

[0056] IV. Solving the master plan during the planning phase.

[0057] After completing the weather micro-scenario and time bucket division and calculating the weather delay ratio, this invention enters the master plan solution process in the planning phase. This process does not directly output the final route under various disturbances in the execution phase, but generates a feasible master plan (baseline plan) within the planning period, including the activation decision of planned vehicles, the main service relationship of suppliers, the planned route (planned access order), and the spare space reserved to cope with fluctuations. This provides a stable resource boundary, allocation baseline, and route skeleton for the arrival time propagation verification and triggering minimum repair in the execution phase.

[0058] (1) Decision variables and planned cost targets: The following decision variables are introduced during the planning phase: vehicle activation variable, indicating whether a planned vehicle is booked or activated. , indicating supplier Whether it is a main service variable of the planned vehicle main service This indicates that the planned vehicle is on the arc. Upward travel arc selection variable Arrival time variable Return to factory time Number of people arriving late to the factory and planned vehicle space Cost function during the planning phase Defined as: ; This represents the vehicle activation variable, arc selection variable, available space, and latest return-to-factory delay amount for any planned vehicle.

[0059] The first item is the planned fixed operating cost of the vehicles, and the second item is the mileage or cubic kilometer transportation cost. The third item is the opportunity cost of spare space, which is used to create an adjustable trade-off between "reserving a buffer to improve execution flexibility" and "excessive vacancy leading to cost waste"; the fourth item is the late return-to-factory (JIT) penalty, which is used to suppress the risk of overtime in the plan under weather-robust driving time.

[0060] (2) Uniqueness of main service, activation linkage and route structure constraints: To ensure that each supplier has exactly one primary service vehicle at the planning level, a uniqueness constraint is imposed: .

[0061] At the same time, to ensure that the assigned vehicles must have been booked or activated, a linkage constraint is applied: .

[0062] To ensure that the primary service relationship can be translated into a concrete, executable route, this invention further establishes route structure constraints, requiring the supplier to make exactly one entry and exit when being served, for example: .

[0063] And for OEM nodes Apply the constraint "If enabled, then proceed and return": .

[0064] If necessary, sub-loop elimination constraints (SEC) or equivalent cutting planes can be added to avoid the emergence of sub-loops that are not connected to the parking lot.

[0065] (3) Definition and lower bound constraint of free space: To explicitly incorporate planning-level buffers into the optimization, this invention uses the effective volume of the carriage. The difference between the planned and actual loading volumes is used to characterize the free space, and a lower bound constraint is given: ; ; in, For suppliers The planned number of boxes, This corresponds to the volume of a single box of the box type. Because the objective function... By applying a positive penalty, the optimal solution will automatically take the minimum feasible value, thereby... Equivalent to the empty space (buffer) after the planned loading.

[0066] (4) Packing feasibility constraints and robust arrival time propagation during the planning phase: Volume difference alone is insufficient to guarantee actual loadability; therefore, three-dimensional packing feasibility constraints are further incorporated during the planning phase. A set of service providers is defined for each planned vehicle. And require: ; in Indicates supplier The set of boxes under the planned number of boxes, and the packing decision function. Indicates the model Given the effective dimensions of the carriage, there exists at least one three-dimensional placement scheme that satisfies constraints such as boundary, non-overlap, and orientation.

[0067] The bin packing decision function can be implemented in a layered manner: preferably, a fast bin packing verification process using the maximum empty space (EMS) is used to construct feasible placements; in implementations that require generating lazy constraint pruning or confirming infeasibility conclusions, a precise bin packing decision function can be further called to confirm infeasibility conclusions, and then a cutting plane can be generated accordingly.

[0068] Meanwhile, to embed the weather-robust travel time that evolves over time into the master plan, this invention is based on the weather delay multiplier output from the previous process. Compared with the baseline driving time Time constraints are established using arrival time propagation, for example, when the arc... When selected, there are: ; Return time With the latest time to return to the factory Association, defining the latest late return-to-factory quantity: ; Therefore, during the planning phase, route feasibility, packing feasibility, and the penalty for the latest return to the factory are considered simultaneously under the worst-case expected delay ratio lookup table, resulting in a feasible master plan.

[0069] (5) Solution method: Optimal column generation or branch-and-price method, compact implementation can use branch cut and lazy constraints: Since the planning phase simultaneously includes 0-1 routing decisions, arrival time propagation, and bin packing feasibility constraints, explicitly linearizing all geometric constraints would lead to increased computational complexity and decreased solution efficiency. For example... Figure 4As shown, this invention preferably employs a column generation solution framework based on Dantzig-Wolfe decomposition. The main problem is to construct a series of complete feasible routes that have passed bin packing feasibility checks and satisfy robust arrival time propagation. The main problem selects several route series from the current set to cover all suppliers and minimize planned costs. The pricing subproblem generates new route series with reduced costs under the guidance of dual information, and embeds maximum empty space (EMS) bin packing increment checks and robust arrival time propagation pruning during route expansion. When a new route series with reduced costs exists, it is added to the current set of route series, and the main problem is returned to continue iterating until no new route series with reduced costs exists. After column generation convergence, when an integer solution is required, the integer solution process begins. After obtaining integer candidate solutions, the service supplier set is extracted for each activated vehicle, and the bin packing feasibility subproblem is solved for verification. If bin packing is deemed feasible, the master plan solution that satisfies the bin packing decision function and robust arrival time propagation is output. If bin packing is deemed infeasible, an infeasible core subset is extracted from the set of infeasible service providers, and corresponding forbidden cuts are generated. These are then added to the master problem with lazy constraints, and the solution is returned to continue solving. For example, for the infeasible core subset... Add lazy constraints: ; This process removes all allocation combinations that would make bin packing infeasible, iterates until convergence, and outputs the master plan solution that satisfies the bin packing decision function and robust arrival time propagation.

[0070] 5. Perform real-time status data updates and status alignment.

[0071] like Figure 5 As shown, after the master plan is generated in the planning phase, this invention enters the real-time status data update process. This process aligns the latest on-site information with the master plan in the planning phase before vehicle departure or during route execution, forming the current status parameter set required for execution verification and trigger-based minimum repair in the execution phase. The problem addressed by this process is that the master plan is formulated in the planning phase based on the planned number of containers and the worst-case expected delay ratio, while the execution phase may encounter situations such as container number deviations, order insertions, supplier operation delays, changes in vehicle availability, and dynamic evolution of weather or traffic. Without status updates, it is difficult to accurately characterize the currently adjustable decision boundaries and the verification, repair targets, and constraint strengths in the execution phase. This process operates at the current rolling moment. (Generate state vectors using absolute time or relative time with the start of the planned cycle as zero) As input during the execution phase.

[0072] (1) Real-time status data acquisition content and data source: At any moment Read and refresh the following data from the enterprise information system, vehicle terminal, or third-party interface, and fill in missing values ​​and standardize units and definitions: Demand side: Suppliers Actual deliverable box count Additional order increments Information on supplier loading times and work delays; Vehicle side: Availability status of planned and emergency vehicles, current vehicle location or current vehicle node. Current moment 1. Loaded container collection (or container type - quantity) And available space, remaining available time, etc.; Environmental aspects: Current weather or traffic conditions and forecasts, and updated arc. In absolute time Delay ratio of real-time observations or forecasts (Obtained from real-time observation or forecast). As a preferred technical solution, when environmental information is updated, this invention retains a lookup table for the worst-case expected delay ratio during the planning phase. As a planning caliber, it also generates the delay ratio for real-time observations or forecasts. This is used for execution verification and triggering minimal repair calls during the execution phase.

[0073] (2) Map real-time information to update parameters during the execution phase: This invention is based on Define the observation bias for the benchmark. This determines the number of execution bins in the execution phase: ; If necessary, the number of execution boxes may also include Unlike updates based on scenario sets, the execution phase, under the new definition, uses... As a unified input, it is used to perform packing feasibility verification and arrival time propagation verification on the planned routes selected in the planning phase.

[0074] Simultaneously, this invention generates trigger parameters. To reduce unnecessary packing verification costs: For routes already selected during the planning phase... definition The plan allows for the inclusion of a new plan if and only if there are changes in the number of containers or container sets among the suppliers covered by the planned route, changes in container sets due to order insertions, or significant changes in remaining free space. Trigger packing feasibility check; otherwise, order The packing feasibility check can be skipped, and only the arrival time propagation check is performed.

[0075] Furthermore, this invention updates the adjustable boundaries of the execution phase: it fixes the departure status and execution prefix of dispatched vehicles (current location, current time, set of served suppliers, set of loaded containers); for planned vehicles that have not yet been dispatched, their activation status generally does not exceed the reservation boundary of the planning phase (i.e., no new unreserved planned vehicles are added during the execution phase), but allows the activation of emergency vehicles to undertake handover tasks when necessary; for vehicles that have been determined to be unavailable or unserviceable suppliers, this invention imposes a disabling constraint during the execution phase; and for emergency vehicles, the activation limit for emergency vehicles during the execution phase is applied. It is controlled by parameters such as incremental cost and deviation penalty.

[0076] (3) Determine the current absolute time base: To ensure that the multiplier evolving over time takes effect correctly during the arrival time propagation, at time... Define the absolute time reference for this round of feasibility verification and minimum repair (e.g., the vehicle's estimated departure time or the time corresponding to the current location). In the preferred technical solution, the arc in the arrival time propagation verification during the execution phase... The travel time is expressed as: .

[0077] When it is necessary to use the worst-case expected delay ratio lookup table for comparison or when the delay ratio of real-time observation or forecast is missing, the robust travel time can also be used as a fallback: .

[0078] In this way, whether the execution phase propagates the route to continue execution from the current moment or performs minimal repair propagation after triggering, it will automatically call the multiplier consistent with the current time period, instead of using the static nominal conditions.

[0079] (4) Output interface: The execution phase outputs the following to the execution phase during the real-time status data update process: a set of execution bins. Trigger parameter set The current time bucket information and the worst-case expected delay ratio can be looked up in a table. and weather delay multiplier Delay ratio of real-time observation or forecast Execution driving time Constraint parameters such as vehicle availability, execution status, available space, and remaining time, as well as baseline information of the master plan during the planning phase (e.g., (And the required correlation for deviation penalty). Based on the above state alignment results, the execution verification process enters the execution phase: first, the packing feasibility and arrival time propagation verification of the planned route in the planning phase are performed; when the verification finds that packing is not feasible or that the latest return to the factory is delayed, the minimum repair process in the execution phase is triggered to generate an executable solution and output it.

[0080] VI. Triggered execution verification and minimal repair (emergency vehicle activation or partial reconfiguration) and feedback loop during the execution phase.

[0081] After updating the real-time status data and outputting the current status parameter package, this invention enters the execution phase remedial process. Unlike multi-scenario secondary optimization based on scenario sets, the execution phase, under the new definition, preferably adopts a two-step mechanism of verification followed by triggering: first, the planned routes and main service relationships output by the master plan in the planning phase are verified (Execution Phase A); only when the verification finds that packing is infeasible or results in a late return to the factory is minimal remedial optimization triggered (Execution Phase B). This process uses the master plan in the planning phase (e.g., ...) as the basis for the remedial process. (and route skeleton) as the planning resource boundary, with the set of execution bins. Delay ratio between vehicle status and real-time observation or forecast As input, the output is a directly executable vehicle activation plan and a milk-run closed loop. The goal is to keep the plan feasible under three-dimensional packing and time constraints, and to keep transportation and adjustment costs at an acceptable level, without exceeding the resources reserved in the planning phase, through limited partial handover or rearrangement and necessary emergency vehicle activation.

[0082] (1) Decision variables and resource boundaries: Design a vehicle collection Emergency vehicle assembly Execution vehicle assembly The verification phase of execution phase A does not introduce new route decision variables; it only propagates arrival times and performs binning decisions under the planned route framework of the planning phase.

[0083] When the minimum repair (execution phase B) is triggered, this invention defines the repair decision variable: departure variable. Supplier service allocation variables Arc selection variable and arrival time With return to factory time To ensure that remedial measures do not exceed limits, planned resource boundaries and emergency resource caps are imposed: ; For planned vehicles that have already been dispatched, this invention solidifies their dispatch status and executed prefixes (current location, current time, set of served suppliers, set of loaded containers) based on real-time status, thereby narrowing the adjustable domain and ensuring continuous execution on site.

[0084] (2) Service integrity, partial handover and stability control (alternative to full redistribution): For each supplier During the execution phase, the vehicle must be served once and only once. ; And link the vehicle departure status: .

[0085] To achieve minimal repair, this invention preferably limits the scope of remedial adjustment: Let Indicates supplier For planned routes (or planned vehicles) belonging to the planning phase, adjustments allowed in execution phase B are primarily partial handovers, meaning only a small number of suppliers are transferred from their original planned routes to emergency vehicles or partially rearranged within the original route coverage set. Handover indicator variables can be used. Indicate whether a handover has occurred, and add an adjustment penalty to the objective function. This encourages remedial measures to be made with minimal changes, requiring modifications only when necessary. In one implementation, the maximum handover size can be limited by constraints. (For example Alternatively, the handover set can be limited to routes that trigger infeasibility, in order to further stabilize execution.

[0086] (3) Route Restructuring: Departure to Factory, Service Consistency, Flow Conservation and Sub-loop Elimination: For each vehicle, the route needs to start from the OEM node. Depart and return : ; To ensure consistency between service and access, for each supplier Apply an input-output constraint: ; Sub-loops are avoided by using MTZ constraints or cutting planes; preferably, sub-loop cutting can be dynamically added during the solution process using lazy constraints to improve efficiency.

[0087] To ensure that partial rescheduling does not cross the responsibility boundaries of the planning phase, in implementations that allow partial rescheduling of planned routes, restrictions may be imposed: if vehicles The set of planning service providers corresponding to the planning phase is as follows In execution phase B, the vehicles in this plan are only allowed to serve. Suppliers within the country (excluding those transferred out), i.e., those... Apply .

[0088] (4) Time-of-arrival propagation of embedded time-varying magnification (priority given to delay magnification of real-time observations or forecasts): During the execution phase, the arrival time and return time are calculated simultaneously with the route determination. For any arc... Adopting large Arrival time propagation constraints: ; in, For homework time, Based on the driving time, The delay factor for real-time observations or forecasts during the execution phase. This serves as the absolute time reference for the current round of verification and minimum repair.

[0089] When the delay factor for real-time observations or forecasts is missing or a robust control is required, the planned robust factor can be used as a fallback. The return-to-factory time can be determined by... Defined in the same way, and based on the latest return time to the factory. Define the amount of lateness: ; Late arrivals Used for the repair target penalty in execution phase B or the verification trigger judgment in execution phase A (when) (Triggering a fix).

[0090] (5) Feasibility of bin packing during the execution phase (by number of execution bins) and iterative verification: Number of execution bins used during the execution phase Apply packing feasibility constraints to each executing vehicle: ; Packable(·) reusable packing verification process during the planning phase: EMS is preferred for rapid feasible placement construction; when a vehicle service combination is determined to be unpackable, a disabled cut is generated and added in the form of lazy constraints, for example, for the infeasible core set during the execution phase. Add lazy constraints: ; The process continues until all selected combinations pass the binning check, thus ensuring that the output solution is feasible in the field.

[0091] (6) Solution and Engineering Acceleration (Replacing Multi-Scenario Parallelism, Filtering and Reweighting): For the minimum repair model in the execution phase B, branch and bound are used, combined with lazy constraints to dynamically add binning infeasible cuts and sub-circle cuts; and the following strategies can be adopted to improve real-time efficiency: Warm-start: Using the planned route from the planning phase as the initial solution; Status solidification: Fix the set of vehicles that have been dispatched, executed prefixes, and loaded containers; Triggered solution: Only when check A in execution phase finds bin packing infeasible or Phase B will only be initiated at that time; Localized repair: Only local rearrangements are allowed within planned routes that trigger infeasibility, and the scale of handover and the number of emergency vehicles activated are limited to reduce the search space.

[0092] (7) Output: The execution phase outputs include vehicle (planned vehicle, emergency vehicle) activation results, supplier service allocation, arc selection results, arrival time and return time, and execution phase delays. Including incremental costs and other results, and feeding back unfeasible plans and service combinations determined during the execution phase to the planning phase when necessary to avoid re-selection, thereby forming a closed-loop consistency between planning and execution.

[0093] VII. Risk assessment and control using the Just-in-Time-Condition Value at Risk (JIT-CvaR) model.

[0094] To adapt to the highly asymmetrical nature of Just-In-Time (JIT) supply in terms of lateness costs, this invention uses the latest return time of the OEM (Original Equipment Manufacturer) to the factory. As a key timeliness benchmark, and within the framework of two-stage trigger-based execution verification and minimum repair, the lateness of return to the factory is explicitly constrained and penalized to suppress the risk of production disruption caused by the latest return to the factory timeout.

[0095] This invention is based on the execution phase in the current execution state (rolling time). The vehicle return time obtained from the downpropagation Introducing late arrival data and apply: ; in, Equivalent means that the vehicle's return time exceeds the latest return time. The timeout amount can be used as one of the triggering conditions: if the execution phase A check finds that any vehicle is late. If this is triggered, the minimum repair optimization in execution phase B will be activated, and delays will be eliminated or reduced through methods such as emergency vehicle activation, partial handover, and partial rescheduling.

[0096] This invention can achieve timeliness control in two ways: First, by... By weight The objective function of execution phase B is incorporated to form an adjustable trade-off between transportation costs and lateness penalties, enabling minimum repairs to proactively reduce return-to-factory timeouts while controlling incremental costs; secondly, it ensures on-time delivery by setting a threshold. Set a late arrival limit constraint, for example: ; Alternatively, stricter limits could be imposed on critical vehicles and schedules to minimize costs while still meeting on-time performance requirements.

[0097] Through the above mechanism, during the execution phase, when emergency vehicles are activated, partially handed over, and partially rearranged after the execution verification is triggered, the exposure of return timeout will be automatically reduced, avoiding concentrated timeouts under adverse weather or congestion conditions, thereby generating a robust inbound cyclical pickup solution for Just-In-Time (JIT) systems.

[0098] 8. Feasibility verification of packing.

[0099] After the master plan is generated during the planning phase and during the execution phase, triggering execution verification and minimum repair processes, this invention performs a packing feasibility check on the container set carried by each actual execution vehicle to ensure that the model output is executable in a physical loading sense. Specifically, for the execution vehicle... Effective interior dimensions of the passenger compartment for the corresponding vehicle model The set of containers to be loaded is obtained by combining the services provided by the suppliers. (The box size parameters and number of boxes are determined), and the packing determination function is called. The system determines whether a 3D placement scheme that satisfies both boundary constraints and non-overlapping constraints exists. The packing decision function employs an Empty Maximal Space (EMS) check: using the carriage as the initial empty space set, the boxes are sorted according to preset rules and placed one by one. If placement is successful, the empty space set is updated until all boxes are placed, at which point packing is considered possible. If a box cannot be placed in any of the empty spaces, it is considered impossible to pack. When an impossible-to-pack situation occurs, the invention applies a disabling constraint (e.g., disallowing the combination from recurring) to the current vehicle-service provider combination and triggers a minimum repair re-solution during the execution phase. If the impossible-to-pack situation occurs during the candidate solution verification phase of the planning phase, the combination is written back to the planning phase solution process using lazy constraints or a cutting plane approach for pruning, thereby iteratively obtaining the final scheme that satisfies Packable=1.

[0100] 9. Output executable scheduling schemes and visualized loading instructions.

[0101] Once the requirements for container loading feasibility and latest return-to-factory timeliness are met, this invention outputs a cyclical pickup scheme that can be used for scheduling and execution. This includes: vehicle activation results u, main service allocation results y, and planned routes (arc selection) x during the planning phase; and actual vehicle departure decisions (planned vehicles, emergency vehicles) in the current state during the execution phase, supplier service allocation, access order of each vehicle (OEM, supplier sequence, return-to-factory), arrival time, return-to-factory time, and corresponding cost and timeliness indicators (including the latest return-to-factory delay amount, trigger verification flag, emergency vehicle activation status, and minimum repair incremental cost). The cyclical pickup scheme can be provided in structured data or report form to support route allocation, vehicle organization, and execution review. Optionally, this invention can also output the container placement results obtained from container loading verification for loading guidance, but this is not a necessary limitation for implementing this invention.

[0102] This invention focuses on the milk-run scenario under uniform packaging conditions. Addressing common issues in existing solutions such as inconsistencies between planned feasibility and on-site loading limitations, difficulty in quantifying and balancing cargo space availability, challenges in quickly remedying execution disturbances, and difficulty in effectively suppressing late return-to-factory timeouts, this invention proposes a technical solution centered on two-stage collaborative optimization. This solution integrates three-dimensional loading feasibility, cost per square kilometer, opportunity cost of cargo space, real-time state alignment, time-varying robust travel time, and triggered execution verification with minimal repair. The invention presents a holistic framework for path and three-dimensional packing collaborative optimization. By combining path travel cost, vehicle activation cost, and opportunity cost of cargo space availability in the objective function, it forms an integrated optimization model of path-packing-cost. This transforms wasted cargo space into economic losses and explicitly considers them during path planning. The optimization process no longer merely pursues shorter mileage or fewer vehicles but actively seeks the optimal balance between mileage, vehicle usage, and space utilization.

[0103] Two-stage collaborative optimization and execution boundary control: This invention constructs a two-stage collaborative framework of planning and execution. The planning stage outputs the planned vehicle activation, supplier master service relationship, planned route skeleton and buffer parameters. The execution stage performs triggered execution verification based on real-time status alignment results at rolling moments. Minimum repair optimization is only initiated when the verification finds that packing is infeasible or that the latest return to the factory is delayed. Minimum repair achieves feasible recovery without exceeding the planned resource boundary through emergency resource upper limit constraints, partial handover and partial rearrangement. Infeasible combinations identified in the execution stage are fed back to the planning stage to form closed-loop consistency.

[0104] Unified Packaging Box Modeling and 3D Loading Feasibility Assurance: This invention summarizes the BOM and delivery plan by supplier and maps them to a set of boxes according to unified packaging rules; in both the planning and execution phases, loading feasibility judgment constraints ensure that the 3D physical constraints such as the boundary of the carriage, non-overlapping of boxes, and preset orientation are met, and feasible solutions are achieved by disabling cuts (lazy constraints, cutting planes) for infeasible combinations; in the execution phase, the box loading feasibility verification takes the set of boxes generated by the number of boxes executed as input to ensure that the output solution is feasible for on-site loading.

[0105] Unified characterization of comprehensive cost objectives (pricing per cubic kilometer, opportunity cost): This invention uses a unit price per cubic kilometer to include the product of arc load volume and arc mileage in the transportation cost, and embeds the cost per cubic kilometer into the optimization model by linearizing the load volume variable and arc load variable; at the same time, it introduces a spare space variable and includes it in the objective with an opportunity cost coefficient, so as to achieve a unified trade-off between vehicle activation cost, transportation cost and loading buffer, and supports evaluating the cost of minimum repair during the execution phase with incremental cost caliber.

[0106] Time-varying robust driving time modeling and late return-to-factory penalty control: This invention is based on time bucket partitioning and arc-time bucket-micro-scenario set construction. It uses the sub-Bruker method to calculate the robustness ratio and form a driving time interface that evolves with absolute time for consistent arrival time propagation between the planning and execution phases; using the OEM's latest return-to-factory time as the benchmark. The late arrival amount is constructed as a benchmark and used as a trigger condition for execution verification and a penalty item (or optional upper limit constraint) in the minimum repair target, thereby suppressing the latest return timeout and improving timeliness robustness under weather or traffic dynamic evolution conditions.

[0107] In summary, this invention uses a two-stage collaborative optimization approach to jointly decide on packing feasibility, overall cost, and timeliness robustness within a unified framework. It also improves the loading feasibility of the cyclic pickup scheme, reduces overall transportation costs, and enhances timeliness robustness and controllability of the risk of late return to the factory under weather or traffic disturbance conditions through a closed-loop mechanism of "trigger-based execution verification - minimum repair - return-to-factory feedback".

[0108] Example: This embodiment selects a logistics company's segmented pricing model converted to a cubic kilometer basis, 15 suppliers, a 9.6-meter wing-opening van as both planned and emergency vehicles, two types of standard collapsible boxes, and experimental examples of time bucket division and weather micro-scenario decomposition using Blue Bar Optimization Disturbance (DRO) to illustrate the implementation process of the method of this invention. The implementation uses instance data objects, mixed integer programming, a column-generated solution framework, and callback cut planes (bin infeasible cuts and sub-circle cuts) for execution. The specific steps are as follows.

[0109] I. Instantiation and Basic Parameter Construction.

[0110] S101: Read the basic data of the example and complete the instantiation.

[0111] Read the instance file and complete the instantiation, marking the OEM node as 0 and the supplier node set as [missing information]. Read the coordinates or location codes of each node (for calculating distance or road network distance), and read the service time of each supplier within the planned period. Planned number of boxes And the planned volume obtained from the box type conversion. At the same time, it reads the latest time the OEM returned to the factory. (For the latest return-to-factory time penalty and trigger determination in the planning and execution phases); If the enterprise's business has supplier-side delivery and loading time limit parameters, they can also be read and written as constraints or penalties, but they are not necessary restrictions for implementing this invention.

[0112] S102: Read vehicle parameters and construct a planned vehicle set and an emergency vehicle set.

[0113] Read vehicle parameters and build a vehicle set: Planned vehicle limit With the upper limit of emergency vehicles In this embodiment, a 9.6m double-wing opening van is selected as the vehicle type. The transport vehicle has the following external dimensions: (m), effective internal dimensions of the carriage (Given from the vehicle calibration table) and effective volume Write the vehicle parameter table and the planned fixed activation cost of the vehicle. variable cost per unit mileage (Or unit price per square kilometer parameter). Simultaneously, corresponding fixed costs and variable costs per unit mileage are written for emergency vehicles to facilitate economic trade-offs when minimum repairs are triggered during the execution phase, and an upper limit for emergency vehicle activation during the execution phase is set. To constrain on-site resources and scheduling complexity.

[0114] S103: Read the uniform packaging box specifications and construct a collection of box objects.

[0115] Read the standardized packaging box specifications and construct a set of box objects. In this embodiment, standard collapsible boxes (preferably two types) are used. For example, the outer dimensions of box type A are... (m), the outer dimensions of box type B can be given according to the enterprise's turnover equipment library. Instantiate each packaging box as a 3D axis-aligned box object. and build a set of planning boxes for each supplier. (Depend on (Mapped and used for binning verification during the planning phase); Meanwhile, an execution bin generation interface is reserved for real-time updates during the execution phase. Used during scrolling Actual number of deliverable boxes (Incremental orders may be included) are mapped to a set of containers, and used for trigger-based containerization feasibility verification and minimum repair calls during the execution phase. To avoid introducing a scenario set definition, this embodiment does not construct a multi-scenario container set.

[0116] S104: Load the transportation pricing contract of this embodiment and construct the cost function per square kilometer.

[0117] Load the transportation pricing contract of this embodiment and construct the cubic kilometer pricing function. Assume the vehicle type... The unit price per square kilometer is (Unit: Yuan / ( If km), then for any executing vehicle For a closed route formed under the current execution (or planning) status, the transportation cost is calculated as follows: transportation volume. Transportation mileage unit price.

[0118] To align with the physical process of gradually increasing load during cyclic pickup, an arc load per square kilometer is preferred: assuming the executing vehicle... Leave node The internal load volume of the vehicle is (unit: If so, the vehicle will be executed on the arc. The cost per square kilometer generated is : ; in, For arc Road network distance (km) For the execution of vehicles Along the arc The volume of the arc load during travel can be obtained by propagating the cumulative load volume upon arrival at or departure from the node.

[0119] When using the arc load caliber, the total transportation cost will be... Written as ; If the contract stipulates a minimum billing volume or minimum billing mileage This can be written into the cost interface, for example, by replacing the payload with... Or replace mileage with However, this is not a limiting condition necessary for implementing the present invention.

[0120] S105: Construct a distance matrix and an arc set, and map the cost per square kilometer to a linear cost expression that can be optimized.

[0121] Construct a distance matrix based on node coordinates or road network data. And generate an arc set accordingly. To calculate the cost per square kilometer and maintain solvability in mixed-integer programming, the following variables are introduced and a linearization mapping is performed (this mapping can be used both for solving the master plan in the planning phase and for calculating the incremental cost of minimum repairs in the execution phase): (1) Load volume change: For each execution vehicle With nodes Introducing load volume variables This indicates that the vehicle has left the node. The total volume already loaded inside the vehicle ( Initial values ​​are set for the OEM node: ; And apply an upper limit to the capacity (consistent with the vehicle's volume): ; Among them, vehicle activation variables Indicates the vehicle to be executed Whether the train will actually depart or be put into operation; this can be considered during the planning phase. Variables of vehicle activation during the planning phase Equivalence or linkage, in the minimum repair phase of execution, is given by the enable decision.

[0122] (2) Arc load variables (key to linearization): For each execution vehicle With arc Introducing arc load variables And use standard linearization constraints to associate it with whether it travels a specific arc and the load leaving the node: ; in Indicates the vehicle to be executed Is the driving arc , The load is adaptively given by the upper limit of the vehicle volume. (For example The load M is a sufficiently large constant.

[0123] (3) Load propagation (load increase due to picking up goods): For supplier nodes Let its pickup volume be... (Planning phase by) Obtained by mapping with the box type; the execution phase is determined by the number of execution boxes. (obtained by mapping), and used with large The formal implementation of the propagation constraint that "the load increases with the arc travel" is as follows: ; in, It can be given by the upper limit of the vehicle volume (e.g.) ), This represents a large capacity constant in the load propagation constraints. Therefore, the total cost per square kilometer can be calculated within a unified linear framework. The total transportation cost is given for any solution round (minimum repair in the planning or execution phase). Written as: ; When incremental costs need to be calculated for minimum repairs during the execution phase, the above costs can be subtracted from the baseline costs during the planning phase, or a combination of emergency vehicle fixed costs, per-kilometer costs, and adjustment penalties can be directly added to the target.

[0124] S106: Read the opportunity cost coefficient of the spare space and form the opportunity cost calculation method.

[0125] The opportunity cost factor for reading or setting the free space by the user. For each planned vehicle During the planning phase, the effective volume of the vehicle's cargo compartment should be recorded simultaneously. With the planned loading volume and with spare space This represents the buffer amount available for executing volatility, with the opportunity cost written as... This is used to curb two extremes: excessive empty space leading to waste and overfilling with inflexible materials. Among them, The lower bound can be defined as follows and will automatically take the minimum value under the penalty of the objective function: .

[0126] II. Weather micro-scenarios and time bucket data construction.

[0127] S107: Constructing Arc – Time Bucket – Collecting Micro-Scenes and Instantiating Micro-Scene Objects.

[0128] Time is discretized into a set of time buckets based on the planning cycle. and for each arc With Time Bucket Constructing a collection of weather micro-scenes Instantiate micro-context objects Each The corresponding micro-context objects include: (1) Nominal probability ,satisfy ; (2) Delay ratio sample Used to characterize travel time relative to a baseline. Magnification; (3) radius of ambiguity , used to characterize the uncertainty of the nominal distribution; (4) The worst-case expected delay ratio calculated from the above parameters and its lookup table structure .

[0129] In this embodiment, micro-scenario samples can be sampled from historical weather and road condition data or generated according to business rules; when there are too many candidate micro-scenarios, clustering and filtering can be used. This is scaled down to control the computational size. Further, this is achieved through a time mapping function. Define the weather delay ratio as a function of absolute time. To form robust driving time This is used for unified arrival time propagation calls in the subsequent planning and execution phases.

[0130] III. Construction and solution of robust master program model in the planning phase.

[0131] S108: Establish decision variables for the planning phase and construct the objective function.

[0132] Establish decision variables for the planning phase: Planned vehicle activation variables Main service variables Arc selection variable Arrival time variable Return to factory time Number of people arriving late to the factory and spare space The planning phase objectives are: fixed operating cost of planned vehicles, transportation cost per cubic kilometer, opportunity cost of spare space, and penalty for late return to the factory. ; in, The fixed start-up cost for planned vehicle k, It can be expressed as a linearized cost per square kilometer of S105 or calculated from the cost per unit mileage. For nominal transportation cost parameters, This indicates the planned vehicle limit.

[0133] S109: Apply the uniqueness of the main service during the planning phase, and enable linkage and route structure constraints.

[0134] Apply a primary service uniqueness constraint to ensure that each supplier has exactly one planned vehicle primary service at the planning level: ; Apply activation linkage constraints to ensure that "assigned vehicles must be activated": ; To ensure that the primary service relationship can be implemented as an executable path, further "service-access-consistency" and in-degree / out-degree constraints are imposed (example): ; And apply the "If enabled, depart and return to the field" option to OEM node 0: ; Optionally, sub-loop elimination constraints (SEC) or cutting planes can be used to avoid sub-loops. If necessary, lazy constraints can be dynamically added to improve efficiency.

[0135] S110: Apply lower bound and capacity consistency constraints to the planned phase free space.

[0136] Describe the buffer volume using the planned volume, and apply: ; Indicates planned vehicles The upper limit of the effective volume of the passenger compartment for the corresponding vehicle model is preferred. , This is the planned volume obtained from the box-type conversion; To avoid outliers caused by relaxation, a capacity consistency constraint can be optionally added: .

[0137] S111: Incorporate the robustness of time-varying weather into the arrival time propagation and the definition of the latest late return to the factory during the planning phase.

[0138] Robust driving time based on the output of the second part , for arc Apply arrival time propagation constraints (example is large) form): ; in, Use the absolute time reference for the start of the planned route. Service time for each supplier within the planned period This represents the Big-M time constant in the arrival time propagation constraint. Further, the return-to-work time is defined. And construct the latest number of late arrivals to the factory. : .

[0139] This allows for penalty control of the latest return-to-factory timeout in the objective function of the planning phase, and provides a consistent timeliness criterion for the trigger-based verification in the execution phase.

[0140] S112: For large The parameters are adaptively tightened to improve solution efficiency.

[0141] Automated calculations based on instance data ensure security, but with stricter controls. Parameters such as arrival time propagation, load propagation, and linearization constraints are used in large-scale operations. This approach ensures that feasible solutions are not removed while reducing numerical relaxation, thereby improving the solution efficiency of branch and bound and column generation frameworks.

[0142] S113: Encapsulate the feasibility of packing into a determination process in the form of a packing determination interface.

[0143] The 3D bin packing constraints are encapsulated into a bin packing decision interface, with the 3D dimensions of the vehicle as the input. The algorithm, along with the bin set, outputs the bin packing feasibility determination result and, optionally, the infeasible core set. In one implementation, EMS rules are preferably used for fast verification; in implementations where confirmation of infeasibility is required, the precise bin packing determiner can be further invoked to confirm the infeasibility conclusion, and the determined combinations can be cached to avoid repeated solutions.

[0144] S114: During the planning phase of the solution process, perform binning checks on integer candidate solutions and dynamically add disabled cuts using lazy constraints.

[0145] During the integerization process of branch delimitation or column generation, when integer candidate solutions are obtained, for each activated planned vehicle... Extract its service provider set And call the box packing determination interface to verify the box set. Can it be installed? If not, from the conflict set. Generate a disabled cut and add it as a lazy constraint: ; This prevents subsequent searches from selecting the same unpackable combination again, iterating until the master plan solution that satisfies bin packing feasibility is obtained in the planning phase.

[0146] When the number of suppliers is large, it is preferable to use Dantzig-Wolfe decomposition to generate columns by using complete feasible routes that have passed packing verification and meet robust arrival time propagation, and to embed packing increment verification and pruning in the pricing process; when integer solutions are required, a branch pricing framework is formed by combining branching strategies.

[0147] S115: The output of the master plan during the planning phase serves as the baseline for comparison with the planning resource boundaries during the execution phase.

[0148] Output the master plan results for the planning phase ( ), where vehicle activation variable Used to define the availability boundaries of planned vehicles, main service variable Arc selection variable Used to provide planned service relationships and route framework, and planned vehicle free space. Used to characterize buffer size and opportunity cost, arrival time variable Return to factory time Number of people arriving late to the factory Used to provide a time baseline under the planning scope and support the arrival time propagation verification and trigger determination in the execution phase; when the execution phase finds that a combination of a planned vehicle-service provider set or a planned vehicle-route sequence is infeasible in the execution state, the infeasible combination is written back as a disabled constraint to form a closed loop and avoid being selected again in the next planning phase.

[0149] Constructing a set of scenarios (In this embodiment, for example, take...) ), and for each scenario Instantiate the context object Each scenario includes: (1) Scenario requirements of each supplier With scene volume ; (2) Scenario disturbance factors for each arc (Used to describe the impact of weather or traffic on arc travel time in this scenario); (3) Collection of scenario boxes for each supplier (Used for packing verification during the execution phase).

[0150] Optionally, It can be synthesized by looking up the time bucket and the worst expected delay ratio in a table, and then written into the scenario data table to realize the discretization of time-varying impact.

[0151] IV. Triggered execution verification and minimal repair (emergency vehicle activation or partial reconfiguration) and feedback loop during the execution phase.

[0152] S201: Establish execution phase A to perform verification inputs and construct trigger criteria.

[0153] At the rolling moment Read the state vector output during the real-time state data update process. This includes: the set of execution bins. The vehicle's current location, current time, executed prefixes, loaded container set, vehicle availability, and the real-time observed or predicted delay ratio. (or execution driving time) Baseline information output during the planning phase. As a reference to the planned resource boundary and baseline, a trigger criterion is constructed: when loading containers is infeasible for any planned vehicle under the specified number of containers, or when the latest return-to-factory delay is obtained based on the delay ratio propagation from real-time observations or forecasts. If the condition is met, the minimum repair in phase B is triggered; otherwise, the baseline plan in phase B is output directly and execution continues.

[0154] S202: Perform feasibility verification of packing in Phase A (by number of packings to be executed).

[0155] For each of the planned routes (or service providers for the planned vehicles) selected during the planning phase, a packing feasibility check is performed. For any planned vehicle... Extract the set of suppliers that still need to perform services. (Excluding executed prefixes), construct the box set. It then calls the boxing decision interface to determine whether the item can be packed.

[0156] If binning is not feasible, it is marked as binning triggered, and the set of infeasible cores is recorded. This serves as the basis for disabling cuts and reverting to the ban in the subsequent execution phase B.

[0157] S203: Perform the arrival time propagation verification and calculate the delay amount in execution phase A.

[0158] Under the premise of a fixed route skeleton in the fixed planning phase (or a fixed current execution prefix), based on the execution travel time Arrival time propagation is performed to obtain the return time of each vehicle to the depot. And calculate the number of people arriving late to return to the factory: ; When it exists When this occurs, it is recorded as a time-limited trigger. Boxing trigger and time-limited trigger can be used as sufficient conditions for triggering minimum repair.

[0159] S204: Determine whether to trigger and freeze the executed prefix, forming an adjustable boundary for execution phase B.

[0160] If execution phase A is not triggered, the instruction to continue execution according to the baseline plan is output and the current execution phase ends; if it is triggered, the executed prefixes (current location, current time, set of served suppliers, set of loaded containers) of the dispatched vehicles are frozen, and the activation boundary of the undeparted planned vehicles is restricted to... (Without exceeding the planned pre-booked resources); simultaneously, an emergency vehicle fleet will be activated. As a fallback resource, the minimum repair model is constructed in the execution phase B.

[0161] S205: Establish the minimum repair decision variables and incremental cost targets for the execution phase B.

[0162] Establish decision variables for the minimum repair triggered in this round: vehicle activation variable Supplier service allocation variables Arc selection variable Arrival time Return to factory time Late arrivals and optional handover indicator variables The handover indicator variable indicates whether the supplier has transferred responsibility from the original planned set of responsibilities to emergency vehicles or whether a partial adjustment has occurred.

[0163] With the goal of minimizing incremental costs: while retaining the original transportation costs (including linearization costs per square kilometer) and fixed vehicle usage costs, adjustment penalties and lateness penalties are introduced, resulting in: ; in The linearized expression of S105 can be used directly; if a strict minimum increment is required, it can also be written as a difference between the repair scheme cost and the baseline cost, but this is not a necessary limitation for implementing the present invention.

[0164] S206: Apply execution phase B service integrity, resource boundary, and localized repair constraints.

[0165] Apply service integrity: ; in This refers to the set of suppliers that have not yet been served in this round; and it applies dispatch linkage. .

[0166] Apply planned resource boundaries and emergency limits: ; To minimize repairs, the adjustable domain of the planned vehicles is preferentially limited: planned vehicles The original plan was to include the following responsibilities: Vehicles under repair are only allowed to be serviced. Internal suppliers (excluding those transferred out), for Apply ; Changes in supplier responsibility sets through This is achieved through emergency vehicle services, thereby ensuring the stability of the plan and its feasibility on-site.

[0167] S207: Apply path structure constraints, eliminate sub-loops and propagate arrival time (delay factor of real-time observation or forecast).

[0168] Selecting variables by arc Construct a closed-loop milk-run route and impose OEM departure-to-factory constraints on each executing vehicle: ; And apply an inbound-outbound constraint to each supplier node to ensure service-access consistency: ; To avoid subcycles, a subcycle elimination constraint (SEC) cutting plane strategy is adopted: user cuts are optionally generated at relaxed solutions, and the set of subcycles is separated based on connectivity at integer candidate solutions. And add it with lazy constraints: ; Propagation arrival time according to the delay factor of real-time observation or forecast: for any arc Apply ; And obtained by propagation from the return arc And then calculate .

[0169] S208: The number of late returns to the factory is incorporated as a penalty into the minimum repair target (replacing JIT-CVaR).

[0170] Instead of constructing a Conditional Value at Risk (CvaR) model based on a set of scenarios, it is obtained by propagating from the current execution state. Define lateness quantity based on the basic definition and by weight Incorporate it into the objective function of execution phase B to achieve an adjustable trade-off between transportation costs, adjustment costs, and lateness penalties.

[0171] In one implementation, a lateness limit can also be set for critical vehicles. This serves as a hard constraint to ensure on-time assurance levels, but is not a necessary limitation for implementing this invention.

[0172] S209: During the execution phase B solution process, perform binning checks on integer candidate solutions and add disabled cuts using lazy constraints.

[0173] During the branch and bound process, when integer candidate solutions are obtained At that time, for each executing vehicle Extraction service provider set Call the box packing check interface to verify the box set. Can it be installed? If not, from the conflict set. Generate a disabled cut and add it as a lazy constraint: ; This process involves eliminating service combinations that make binning infeasible, iterating until all executed vehicles meet binning feasibility requirements.

[0174] S210: Perform a warm-start and localized acceleration to improve real-time minimum repair efficiency (optional).

[0175] Optional warm-start: Uses the planned routes from the planning phase as the initial solution, and combines state solidification (fixing dispatched vehicles and executed prefixes) with localized repair (allowing adjustments only to route families that trigger infeasibility, and limiting the handover scale). This reduces the search space, thereby shortening the time to obtain the first feasible fix solution.

[0176] S211: Completes the closed loop of solving, separating, cutting, and solving again through a callback mechanism, and executes the feedback output scheme of blocking.

[0177] Enable the solver's lazy constraints mode and perform optimization: during the solution process, continuously separate sub-loop elimination constraints (SEC) sub-loop cuts and packable box cuts, and combine Big-M adaptive tightening and warm-start acceleration to finally output the multi-vehicle cyclic pickup scheme of this embodiment.

[0178] Simultaneously, for the infeasible vehicle-service provider sets or vehicle-route sequences identified in Execution Phase A and Execution Phase B, a feedback loop is executed: the infeasible core set (such as...) is removed from the list. , The infeasible route list (or infeasible route list identifier) ​​is written back as a disabled constraint in the planning phase, so that subsequent plan solutions can automatically avoid similar infeasible combinations, thereby forming a plan-execution closed-loop consistency.

[0179] V. Output Results.

[0180] S212: Outputs a dispatchable scheduling scheme and optional visualization results.

[0181] Output an executable cyclic pickup scheduling plan, including: vehicle activation results during the planning phase, main service allocation results, planned route skeleton (arc selection), and available space. The system outputs the planned return-to-factory time and the latest return-to-factory delay amount, as well as the actual vehicle (planned vehicle, emergency vehicle) activation decision, supplier final service allocation, access order of each vehicle (OEM-supplier sequence-return-to-factory), arrival time, return-to-factory time, latest return-to-factory delay amount, trigger verification flag, emergency vehicle activation status, minimum repair adjustment, handover information, and corresponding cost indicators (including transportation cost and incremental cost). It can also output the container placement coordinates and orientation returned by the container determination unit for loading guidance and visualization.

[0182] like Figure 6 As shown, as the search node or iteration process progresses, the optimal deviation value gradually decreases and tends to stabilize, indicating that the solution framework for the planning phase described in this invention can converge to a stable solution in a relatively short time. This figure is used to illustrate the solution convergence characteristics under the combined effect of column generation, branch and bound, and lazy constraints, and does not impose any limitations on specific values.

[0183] like Figure 7 As shown, during the planning phase, based on the spatial distribution of suppliers, vehicle activation results, and main service allocation relationships, a cyclical pickup route corresponding to each planned vehicle is generated. The OEM serves as the unified departure and return node, with different routes covering their respective supplier sets, forming the baseline scheme and resource boundaries for the execution phase.

[0184] like Figure 8 As shown, when the execution phase verifies the original master plan based on real-time status data and finds that packing is not feasible or there is a risk of late return to the factory, this invention, without exceeding the boundaries of the planned resources, activates emergency vehicles to transfer the tasks of some suppliers, thereby forming a repaired executable route. Figure 8 The dashed or highlighted lines represent newly activated emergency vehicle routes, illustrating that the local repair mechanism of this invention can restore the feasibility of execution while maintaining the stability of the overall solution.

[0185] Figure 9 Figure (a) shows the arrangement of the boxes inside the carriage of a certain vehicle. Figure 9 Figure (b) shows the supplier access order corresponding to the packing result. This joint graph can simultaneously reflect the consistency among the vehicle service set, path order, and carriage placement scheme, indicating that the present invention not only meets the path optimization requirements in the planning stage, but also meets the three-dimensional packing feasibility requirements.

[0186] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0187] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0188] In one embodiment, a computer system is provided, which may be a server. The computer system includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data used in the methods described above. The network interface communicates with external terminals via a network connection. The computer program is executed by the processor to implement the methods described above.

[0189] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0190] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0191] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A robust two-stage cyclic loading and route optimization method, characterized in that, include: Acquire planned data and real-time status data. Planned data includes the supplier's planned number of containers, vehicle parameters, and transportation cost parameters. Real-time status data includes the number of containers executed and dynamic travel time. Based on the planned data, the pre-defined planning stage model is solved to generate the master plan. The master plan includes the activation decision of the planned vehicles, the main service relationship of the suppliers, the planned routes, and the reserve space to cope with fluctuations. During the execution phase, the master plan is checked based on real-time status data. The execution check includes packing feasibility check and arrival time propagation check based on dynamic travel time. When the execution check finds that packing is not feasible or that the latest return to the factory is delayed, the preset minimum repair model is triggered to solve. Within the resource boundaries determined by the master plan, an executable repair plan is generated by activating emergency vehicles and / or partially rearranging the planned route. Output the repair plan or the verified feasible master plan as the final scheduling plan; The dynamic driving time is obtained in the following way: A distance matrix is ​​constructed based on the shortest travel distance between nodes in the road network, and arcs are generated based on the distance matrix. Corresponding base driving time The planning period is divided into a set of discrete time buckets. and define the time mapping function. Used to convert any absolute time Mapped to the corresponding time bucket The nodes include OEM nodes and supplier nodes; the supplier nodes constitute a supplier set. The OEM node is designated as node 0; For each arc and each time bucket Generate a set of micro-scenes , Let i and j represent nodes respectively, which are arcs. The two endpoints; for each micro-scenario Configure delay ratio samples and the corresponding nominal probability ; Based on the sub-Blule bar optimization method, for each arc-time bucket combination Set ambiguity radius To construct an ambiguous set containing the true probability distribution. And calculate the worst-case expected delay ratio within this ambiguity set. ; Through the time mapping function The worst-case expected delay multiplier within the time bucket. Mapped to weather delay ratio as a function of absolute time ; Represents arc In absolute time Time Bucket The worst-case expected delay ratio; Based on the aforementioned weather delay multiplier and baseline travel time, a robust travel time is generated for arrival time propagation during the planning phase. ; During the execution phase, based on robust driving time and the delay ratio of real-time observation or forecast. Generate the execution time for the verification. ; The dynamic driving time includes the baseline driving time. Robust driving time and execution driving time ; Micro-scenes can be generated in the following ways: Historical sampling method: Samples are drawn from historical periods into time buckets. Inner arc Real delay rate samples And estimate the nominal probability based on historical frequency. Post-normalization; Rule-based perturbation method: Apply interval or discrete perturbations to the baseline delay ratio to generate delay ratio samples. Furthermore, the lower bound of the delay multiplier sample is truncated and verified, and the nominal probability is normalized using equal weights or according to rule weights. Combined business scenario approach: Based on the actual combination of multiple weather events that may occur simultaneously in business operations, generate micro-scenarios and attach event tags; The method based on the sub-Blule bar optimization is used for each arc-time bucket combination. Set ambiguity radius To construct an ambiguous set containing the true probability distribution. And calculate the worst-case expected delay ratio within this ambiguity set. Specifically, it includes: Considering the estimation error of the nominal probability corresponding to the delay ratio sample, it is necessary to consider the estimation error for each arc-time bucket combination. Constructing an Ambiguity Set It is used to characterize uncertain sets where the true distribution lies in the neighborhood of the nominal distribution, and the chi-square divergence is used to constrain the deviation of the true distribution from the nominal distribution to not exceed the radius of ambiguity. : ； The dimension is The nonnegative real vector space; It is a collection of micro-scenes The number of elements; Representing the set of ambiguities One of the candidate true probability distributions, This represents the transpose of a vector consisting entirely of 1s. This indicates satisfaction. Represents the true probability distribution of candidates In micro-scenarios The corresponding probability value; The worst-case expected delay ratio is calculated within this ambiguity set. Specifically, it includes: First calculate the nominal probability mean With variance : ， ； The worst-case expected delay ratio can be obtained using either a closed-form or upper-bound form. : ； Based on the planned data, a pre-defined planning stage model is solved to generate a master plan. The master plan includes vehicle activation decisions, primary service relationships with suppliers, planned routes, and contingency space reserved to cope with fluctuations. Specifically, it includes: Introduce the following decision variables: vehicle model Planned vehicles Enable variables , indicating supplier Whether it is based on the car model Planned vehicles The main service variable of the main service , indicating vehicle model Planned vehicles Is it in the arc? Upward travel arc selection variable Model Planned vehicles Arrival at the supplier The moment Model Planned vehicles Return to factory time Model Planned vehicles The latest number of people arriving late to the factory and car models Planned vehicles spare space Cost function during the planning phase Defined as: ； Indicates planned vehicles , Indicates model t, Represents a set of vehicle models. Indicates vehicle model The planned vehicle collection, This represents the fixed activation cost per planned vehicle corresponding to model t. Represents the set of arcs. For model In the arc The nominal transportation cost parameter on the screen, This represents the opportunity cost coefficient. This represents the penalty weighting coefficient for the latest return to the factory; Based on the planned data, constraints are configured for the planned phase model, including main service uniqueness constraints, linkage constraints, route structure constraints, lower bound constraints of free space constraints, bin packing feasibility constraints, and arrival time propagation constraints. Main service uniqueness constraint: ; Represents the set of suppliers; Linkage constraints: ; Route structure constraints: ， ； ； Indicates that the arc satisfies All nodes , This indicates that the variables satisfying the stated conditions are summed. Lower bound constraint of free space: ； ； in, For suppliers The planned number of boxes, This refers to the volume of a single box for the corresponding box type. Let t be the effective volume of the carriage. Packing feasibility constraints: Define a set of service providers for each planned vehicle k of vehicle type t. And require: ； in, Indicates supplier The set of containers under the planned number of containers. For the binning determination function, Indicates the model Under the constraint of the effective size of the carriage, there exists at least one three-dimensional placement scheme that satisfies both boundary constraints and non-overlapping constraints; Arrival time propagation constraint: when arc When selected, there are: ； Return time With the latest time to return to the factory Association, defining the late quantity : ； This indicates the absolute time at which the vehicle is scheduled to depart from node 0 of the OEM. Represents a constant; Solving the planning phase model after configuring constraints, specifically using a column generation framework: Generate a list of complete feasible routes that have passed the packing feasibility constraint verification and meet the arrival time propagation constraint; The main problem is to select several routes from the generated route list to cover all suppliers and minimize the planned cost; The pricing subproblem generates a new route sequence with reduced costs under the guidance of dual information, and embeds bin packing increment verification and arrival time propagation pruning during the route expansion process; During the execution phase, the master plan is checked based on real-time status data. This check includes a packing feasibility check and an arrival time propagation check based on dynamic travel time. When the execution check reveals that packing is infeasible or results in a late return to the factory, a preset minimum repair model is triggered, specifically including: At the rolling moment Collect real-time status data and align it with the master plan; Real-time status data includes at least: the scrolling time. Suppliers Actual deliverable box count , rolling time supplier Additional order increment Availability status and rolling schedule of planned and emergency vehicles execution vehicle Current location , rolling time execution vehicle Current absolute time , rolling time execution vehicle The already loaded container assembly Spare space, and rolling time In the arc Delay ratio of real-time observations or forecasts ; The update parameters for the execution phase are generated based on the real-time status data: Supplier Planned number of containers Define the observation bias for the benchmark. This, in turn, forms the number of execution bins input to the execution phase. : ； The planned route already selected during the planning phase Generate trigger parameters ,when When, set the trigger parameter This triggers packing feasibility checks and arrival time propagation checks; otherwise, it causes... Skip the packing feasibility check and only perform the arrival time propagation check; Packing feasibility check: based on the number of boxes to be packed. Generate a set of boxes to be loaded, and call the box loading determination function to determine whether there is a three-dimensional placement scheme that satisfies the boundary constraints and non-overlapping constraints; Arrival time propagation verification: based on the execution travel time Arrival time propagation calculations are performed to obtain the return time of each vehicle and the latest late return time. When it is found that packing is not feasible or the latest return to the factory is delayed, the minimum repair model is triggered to perform minimum repair optimization. Within the resource boundaries defined by the master plan, an executable remediation plan is generated by activating emergency vehicles and / or partially rearranging the planned routes. Specifically, this includes: Using the master plan as the resource boundary, and the set of execution bins from each vendor. Vehicle status and delay rates for real-time observations or forecasts As input, a minimal repair model is constructed, wherein the decision variables of the minimal repair model include: the vehicle performing the repair. Enable variables , indicating supplier Whether it is executed by vehicle The main service provider service allocation variable , indicating the vehicle to be executed Is it in the arc? Upward travel arc selection variable , execution vehicle Arrival at the supplier Arrival time With the execution vehicle Return to factory time The vehicles involved include planned vehicles and available emergency vehicles. Configure constraints for the minimum repair model, including resource boundary constraints, service integrity constraints, local repair constraints, route structure constraints, arrival time propagation constraints, and bin packing feasibility constraints. Solve the minimum repair model and output a repair solution. The repair solution includes vehicle activation result, supplier service allocation result, arc selection result, arrival time, return time, lateness amount, and incremental cost.

2. The robust two-stage cyclic loading and route optimization method according to claim 1, characterized in that, Resource boundary constraints are: ； Indicates the vehicle model during the execution phase. Planned vehicles Enable or disable? Indicates the model in the planning stage Planned vehicles Enable or disable? Indicates vehicle model Planned vehicles The set, Indicates emergency vehicles The set, Indicates the first stage of execution The variables for the activation of emergency vehicles, This represents the maximum number of emergency vehicles that can be deployed during the implementation phase. Service integrity constraints are: For each supplier The execution phase must be performed by one and only one execution vehicle. One service: ； For supplier aggregation, Indicates the set of vehicles to be executed; ； Local repair constraints: By limiting the adjustment range through transfer indicator variables and imposing adjustment penalties on suppliers that undergo transfer, repair schemes tend to be locally rearranged; Route structure constraints: For each executing vehicle, the route must start from OEM node 0 and return to OEM node 0. ； Each supplier being served Satisfying the one-in-one-out constraint: ； Represents the set of arcs; Arrival time propagation constraints: For any arc Adopting large Arrival time propagation constraints: ； in, For suppliers Homework time, This serves as the absolute time reference for this round of repairs. Represents a constant; based on the latest factory return time. Define the execution vehicle Late arrivals : ； Number of execution boxes used Apply packing feasibility constraints to each executing vehicle: ； For the binning determination function, This indicates that there exists a three-dimensional placement scheme that satisfies both boundary constraints and non-overlapping constraints under the effective size constraints of the carriage of a specific vehicle model; Indicates supplier Number of execution boxes The lower box assembly, This represents the union operation. Indicates the vehicle to be executed The corresponding vehicle model identifier.

3. A computer system comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 2.

Images