A warehouse commodity reorganization and path optimization management system

Abstract

The application discloses a kind of commodity reorganization and path optimization management system in warehouse, belong to warehouse management technical field, comprising: cost builds matrix module, for obtaining the basic distribution parameter of multiple warehouses to each customer, the total demand of each customer is divided into multiple sub-demand sets, and the unified cost matrix including load sensitive cost and receiving penalty cost is constructed;Demand assignment module, according to the unified cost matrix, with sub-demand set as granularity iterative execution demand distribution, after completing the distribution of each sub-demand set, update current load state, and participate in the cost calculation of next sub-demand set with updated load state;Path generation module, based on demand distribution result, constructs initial distribution path.The application can realize change traditional step-by-step independent optimization mode, can predict load nonlinear cost and customer multiple receipt cost, avoid the global comprehensive cost rising caused by local optimal, improve the rationality of multiple warehouse order distribution decision.

Description

Technical Field

[0001] This invention relates to the field of warehouse management technology, and more specifically, to a warehouse-based product reorganization and route optimization management system. Background Technology

[0002] With the rapid development of the e-commerce logistics industry, the delivery model of multiple warehouses in one location and multiple products in one order has become the mainstream logistics fulfillment form. Multi-warehouse collaborative delivery technology is widely used in core logistics links such as order sorting, resource allocation, and route planning, playing a key supporting role in improving logistics delivery efficiency and ensuring order fulfillment timeliness. At present, the mainstream multi-warehouse collaborative delivery solutions generally treat order splitting and delivery route planning as two independent links and process them sequentially, forming a fixed step-by-step operation process.

[0003] Existing technologies, in the order splitting stage, primarily allocate order demand based on the core principles of minimizing the number of splits, shortest transportation distance, and minimum cost flow. Costs are calculated solely based on the straight-line transportation distance between the warehouse and the customer, assuming a linear negative correlation between transportation distance and delivery costs. In the subsequent delivery route planning stage, route optimization is only performed based on the predetermined order splitting results. Both stages lack effective data feedback and collaborative optimization mechanisms. Splitting decisions cannot predict subsequent fluctuations in actual delivery costs, and route planning cannot correct unreasonable order splitting schemes in reverse, resulting in significant limitations in overall optimization.

[0004] The root cause of these problems lies in the structural defects of existing optimization models, which lack a unified cost measurement system that connects the two major stages. On the one hand, traditional order splitting models use only transportation distance as a single metric, completely ignoring the nonlinear impact of vehicle load on transportation costs, and cannot adapt to the actual logistics characteristics where the higher the load, the lower the marginal cost per unit distance. On the other hand, delivery route planning models only treat vehicle load as a hard capacity constraint, failing to incorporate load efficiency into the objective function optimization. Furthermore, customer receiving costs are only passively calculated as a result of the splitting, only being tallied during the route planning stage, and do not participate in the order splitting decision-making process. In actual fulfillment scenarios, traditional step-by-step optimization models are prone to the dilemma of local optima and global suboptimal. To compress single transportation distances, systems often split multi-item orders into multiple warehouses for separate delivery. While this reduces the cost per segment, it significantly increases the number of customer package deliveries and raises customer receiving costs. Simultaneously, fragmented vehicle loading leads to a sharp drop in load utilization, significantly increasing overall logistics costs. This makes it impossible to achieve globally optimal control of multi-warehouse collaborative delivery, and it is difficult to meet the needs of large-scale, refined modern logistics fulfillment. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide an in-warehouse commodity reorganization and route optimization management system, which can change the traditional step-by-step independent optimization mode, predict the nonlinear cost of load and the cost of multiple customer receipts, avoid the increase in global comprehensive cost caused by local optima, and improve the rationality of multi-warehouse order allocation decisions.

[0006] To solve the above problems, the present invention adopts the following technical solution:

[0007] A warehouse goods reorganization and route optimization management system includes: a cost matrix building module, used to obtain basic delivery parameters from multiple warehouses to each customer, divide the total demand of each customer into multiple sub-demand sets, and construct a unified cost matrix including load-sensitive cost and receiving penalty cost;

[0008] The demand assignment module iteratively performs demand allocation at the granularity of sub-demand sets based on a unified cost matrix. After completing the allocation of each sub-demand set, it updates the current load status and uses the updated load status to participate in the cost calculation of the next sub-demand set.

[0009] The route generation module constructs the initial delivery route based on the demand allocation results using a parallel clustering insertion method. During the insertion process, it senses the cumulative load in real time and dynamically adjusts the cost of subsequent insertion operations based on the cumulative load.

[0010] The operator set construction module is used to build a neighborhood search operator set, which includes operators for merging the product allocation of the same customer, optimizing the access order of the same customer, and jointly exchanging customer locations in different paths, and evaluating load utilization and receiving cost changes.

[0011] The dual-objective optimization module starts from the initial delivery route, performs tabu search, evaluates candidate solutions using a comprehensive fitness function composed of routing cost and receiving cost weights, dynamically adjusts the receiving cost weights, and accepts candidate solutions when the increase in load utilization exceeds a preset threshold.

[0012] The dynamic iteration module adjusts the parameters of the unified cost matrix and re-executes the optimization based on the deviation between the actual execution results and the planned scheme, and triggers online re-optimization when the actual state deviates from the planned state.

[0013] Furthermore, the cost-based array construction module includes:

[0014] Obtain location data for each warehouse, location data for each customer, individual item weight for each product category, inventory quantity for each warehouse, and the rated maximum load capacity of each vehicle;

[0015] Calculate the delivery distance from each warehouse to each customer based on warehouse location and customer location, and generate the original distance matrix; configure load sensitivity coefficient groups according to the rated maximum load capacity.

[0016] For each warehouse, each customer, and each product category combination, the product of delivery distance and load sensitivity coefficient is used as the load-sensitive distance cost. The receiving penalty cost is determined by querying the receiving penalty cost mapping table. The two are added together to obtain the comprehensive cost value. A unified cost matrix is ​​constructed according to the warehouse and customer dimensions.

[0017] Furthermore, the demand dispatch module includes:

[0018] Establish a current load status record table for each warehouse, which includes the allocated load and the remaining available load; use the minimum comprehensive cost value in the unified cost matrix as the minimum reachable cost of each sub-demand set, and determine the processing order of each sub-demand set in ascending order of minimum reachable cost;

[0019] Using the current set of sub-demands as the demand side, the inventory of each warehouse as the supply side, and the comprehensive cost value in the unified cost matrix as the expense, perform minimum cost flow allocation on the bipartite graph network, record the allocation results, and mark unmet demands as awaiting secondary allocation.

[0020] Furthermore, the demand dispatch module also includes:

[0021] Based on the allocation results, the allocated load of each involved warehouse is accumulated and its remaining available load is deducted. The load sensitivity coefficient is updated, and the load-sensitive distance cost of the corresponding warehouse row in the unified cost matrix is ​​refreshed.

[0022] If there is an unprocessed set of sub-requirements, process the next set of sub-requirements; if not, check the requirements to be reassigned. If they exist, merge them into a supplementary set of sub-requirements and insert them at the end of the processing sequence to re-execute the assignment, up to twice. If they do not exist, terminate the iteration.

[0023] Furthermore, the path generation module includes:

[0024] Based on the warehouse collection and allocation records, the sum of the products of the allocated quantity and corresponding weight of each customer is taken as the total weight of the customer's goods. The set of customers to be delivered and the total weight of goods in each warehouse are extracted. The sum of the total weight of goods in each warehouse is the total delivery weight. If the total delivery weight is greater than the rated maximum load, it is marked as an overloaded warehouse.

[0025] For warehouses that are not marked as exceeding limits, customers are sorted and grouped according to their azimuth angle relative to the warehouse. The number of groups is equal to the number of available vehicles, and the customer closest to the warehouse is used as the initial customer seed for each group.

[0026] Furthermore, the path generation module also includes:

[0027] For each group, starting from the initial customer seed, select customers that have not yet been inserted, use the product of the new delivery distance and the load sensitivity coefficient as the insertion cost, and add the receiving penalty cost to select the customer insertion path with the smallest overall insertion cost; after each insertion, update the cumulative load as the load sensitivity coefficient query benchmark for subsequent insertion operations;

[0028] The routes of each group are merged to generate an initial delivery route plan. If the cumulative load of a route exceeds the rated maximum load, the corresponding customer is removed and marked for reassignment.

[0029] Furthermore, the operator set construction module includes:

[0030] Traverse the initial delivery route plan to identify customers to be merged, customers whose order needs to be optimized, and candidate exchange path pairs. Among them, customers whose occurrence frequency is greater than 1 are customers to be merged, customers whose access order number difference is greater than the preset tolerance threshold are customers whose order needs to be optimized, and two paths with load utilization rates higher and lower than the average respectively are candidate exchange path pairs.

[0031] Merge each customer to be merged into the path with the lowest cumulative load in its path; adjust the latest access of each customer to be optimized to the first position of the corresponding path; swap customers in the candidate swap path pairs, and if the load utilization of the two paths is closer to the average after the swap, the swap is confirmed to be effective.

[0032] Furthermore, the dual-objective optimization module includes:

[0033] Taking the delivery route plan as the current solution, initialize the taboo table, the current optimal solution storage area and the load utilization history table, and initialize the receiving cost weight coefficient of the comprehensive fitness function composed of the weighted sum of the routing cost item and the receiving cost item.

[0034] A neighborhood solution set is generated based on the neighborhood search operator set, and the neighborhood solution with the smallest comprehensive fitness function value is selected as the candidate solution. If the move operation corresponding to the candidate solution matches the taboo record, it is rejected; otherwise, it is accepted when the load utilization improvement is greater than a preset threshold. Otherwise, the acceptance is determined by comparing the comprehensive fitness function value with the current best solution. The accepted moves are recorded in the taboo table. After each preset iteration cycle, the receiving cost weight coefficient is adjusted according to the load utilization change trend. After the iteration terminates, the current best solution is output.

[0035] Furthermore, the dynamic iteration module includes:

[0036] Obtain the actual execution results data of the final delivery route plan, use the difference between the actual load utilization rate and the planned load utilization rate of each route as the load utilization rate deviation value, and use the difference between the actual number of receipts and the planned number of receipts for each customer as the receipt number deviation value, and summarize to generate a deviation record table;

[0037] If the absolute value of the average deviation of each path in the deviation record table is greater than the system deviation detection threshold, a systemic deviation is determined and the deviation direction is identified; paths whose deviation direction is inconsistent with the systemic deviation direction are marked as sudden deviation paths; customers with a reception count deviation value greater than zero are marked as reception deviation customers; and a deviation determination result report is output.

[0038] Furthermore, the dynamic iteration module also includes:

[0039] If the deviation is systematic, the correction factor is determined according to the deviation direction and average value, the load sensitivity coefficient in the unified cost matrix is ​​corrected as a whole, and the demand allocation, path construction and tabu search are re-executed to generate a new delivery route plan; if there is a sudden deviation path or a customer with a receiving deviation, the current inventory status of the corresponding warehouse and the current load status of the vehicle are used as constraints to execute local demand allocation and local path construction to generate a locally updated delivery route plan.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] (1) This solution connects the data link between order splitting and route planning, and constructs a unified cost matrix including load-sensitive cost and receiving penalty cost for multi-warehouse collaborative delivery. It changes the traditional step-by-step independent optimization mode, can predict load nonlinear cost and customer multiple receiving costs, avoids the increase in global comprehensive cost caused by local optima, and improves the rationality of multi-warehouse order allocation decision.

[0042] (2) In the path generation stage, this solution adopts a parallel clustering insertion method. The insertion process updates the cumulative load of the path in real time and dynamically adjusts the insertion cost. This is in line with the logistics characteristic that the larger the vehicle load, the lower the marginal cost per unit distance. It balances the load level of each delivery vehicle and effectively improves the overall utilization rate of vehicle load resources in multi-warehouse fulfillment scenarios.

[0043] (3) This scheme relies on multiple types of neighborhood operators combined with hierarchical acceptance rules to carry out tabu search for dual-objective optimization, taking into account routing cost, receiving cost and load utilization constraints, dynamically adjusting the receiving cost weight to continuously balance the two types of cost objectives, reducing the situation of splitting delivery across multiple paths for the same customer, reducing customer receiving penalty costs, and optimizing the global balance of the delivery scheme.

[0044] (4) This solution establishes a closed-loop dynamic iterative optimization mechanism, quantitatively identifies systematic deviations and local sudden deviations, and adopts a differentiated re-optimization strategy to adaptively correct the parameters of the unified cost matrix or perform local path adjustments. It can continuously adapt to the dynamic working conditions on site, and allow the multi-warehouse delivery optimization model to continuously iterate and adapt with the actual performance data. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0046] Figure 1 This is a data flow diagram between the various modules of the present invention;

[0047] Figure 2 This is a data flow diagram of the cost array module of the present invention;

[0048] Figure 3 This is a data flow diagram of the path generation module of the present invention;

[0049] Figure 4 This is a data flow diagram of the operator set construction module of the present invention;

[0050] Figure 5 This is a data flow diagram of the dual-objective optimization module of the present invention;

[0051] Figure 6 This is a data flow diagram of the dynamic iteration module of the present invention. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] Please see Figures 1 to 6 A warehouse-based product reorganization and route optimization management system, comprising:

[0054] The cost matrix construction module is used to obtain basic delivery parameters from multiple warehouses to each customer, divide the total demand of each customer into multiple sub-demand sets, and construct a unified cost matrix that includes load-sensitive costs and receiving penalty costs. The specific operations are as follows:

[0055] The cost matrix building module first acquires all basic physical and operational parameters of multiple warehouses, customers, products, and delivery vehicles through standardized data collection methods. Simultaneously, it breaks down the overall product procurement needs of a single customer into several independent sub-demand sets, achieving refined granularity of demand calculation. Based on this, the cost matrix building module quantifies the load-sensitive distance cost and the customer receiving penalty cost, respectively. It then merges and superimposes these two cost indicators, ultimately constructing a standardized and unified cost matrix with warehouses and customers as the core dimensions. This allows subsequent demand allocation and path insertion cost calculations to be based on the matrix data, achieving unified cost standards for order splitting and path planning, and eliminating the disconnect in cost measurement between the two stages in the traditional step-by-step operation model.

[0056] The cost-based array building module also includes the following steps:

[0057] Step 11: Obtain the location data of each warehouse, the location data of each customer, the individual weight of each product category, the inventory quantity of each warehouse, and the rated maximum load capacity of the vehicles. The specific operations are as follows:

[0058] Location data for each warehouse and each customer can be collected through logistics site positioning equipment and customer delivery address geocoding equipment. Specifically, high-precision geographic information positioning system is used to accurately input latitude and longitude coordinates for subsequent calculation of delivery distance between warehouses and customers. The unit weight of each product category is an inherent physical parameter of the product and can be pre-calibrated and entered into the system through warehouse weighing equipment, corresponding to the standard weight value of a single unit for each type of warehoused goods. The inventory quantity of each warehouse is real-time warehouse inventory data, which is updated synchronously in real time by warehouse intelligent inventory management equipment, accurately recording the total deliverable inventory of each category of goods in each warehouse. The rated maximum load capacity of the vehicle is a fixed hardware parameter of the delivery vehicle, referring to the maximum total weight of goods that a single delivery vehicle is designed to carry. This parameter is pre-entered into the system according to the delivery vehicle model parameters, serving as a hard constraint for subsequent load segmentation, load over-limit judgment, and load coefficient configuration. All parameters collected in this step are objective, fixed, or real-time updated actual working condition data, without any subjective manual assignment, which can ensure the accuracy and objectivity of subsequent calculations.

[0059] Step 12: Calculate the delivery distance from each warehouse to each customer based on warehouse location and customer location, generating the original distance matrix; configure load sensitivity coefficient groups according to rated maximum load capacity segments, the specific operation is as follows:

[0060] Based on the acquired warehouse and customer location data, both locations are represented using latitude and longitude coordinates. First, the spherical distance algorithm is used to calculate the great circle distance between the two points as a basic distance reference. Then, a geographic road network database is retrieved to match traversable roads between the two points. Road mileage is accumulated segment by segment based on road network nodes to obtain the actual delivery distance from the warehouse to the customer. The spherical distance algorithm and road network path mileage calculation are well-known geospatial distance algorithms in this field and will not be elaborated upon here. After the calculation is completed, a two-dimensional original distance matrix is ​​generated according to the correspondence between warehouse and customer identifiers. Each element in the matrix corresponds to a unique delivery distance value from a unique warehouse to a unique customer. After constructing the distance matrix, a segmented load sensitivity coefficient is configured based on the vehicle's rated maximum load parameter. Considering the real cost characteristics of logistics delivery, different vehicle load values ​​correspond to different unit distance transportation costs. Therefore, the complete load range from zero to the rated maximum load is divided into several continuous load segments. A corresponding load sensitivity coefficient is independently configured for each load segment, and all coefficients corresponding to all segments together form a load sensitivity coefficient group. The load sensitivity coefficient is a dimensionless correction coefficient used to characterize the degree of cost fluctuation per unit delivery distance under different load conditions. The closer the load is to the vehicle's rated maximum load, the lower the corresponding load sensitivity coefficient value, thus matching the real logistics conditions where the load is greater and the marginal cost per unit distance is lower.

[0061] Step 13: For each warehouse, each customer, and each product category combination, use the product of delivery distance and load sensitivity coefficient as the load-sensitive distance cost. Query the receiving penalty cost mapping table to determine the receiving penalty cost. Add the two together to obtain the comprehensive cost. Organize and construct a unified cost matrix according to warehouse and customer dimensions. The specific operations are as follows:

[0062] First, it iterates through all combinations of warehouses, customers, and product categories, calculating the load-sensitive distance cost for each independent combination. This calculation logic is abstracted based on the non-linear characteristics of actual logistics costs. By multiplying a fixed delivery distance by a sensitivity coefficient under the corresponding load condition, it corrects the calculation bias of traditional fixed-distance costs, accurately reflecting the actual transportation costs under different load conditions. The corresponding formula is: ,in, This represents the load-sensitive distance cost, where D represents the actual delivery distance from the corresponding warehouse to the customer. This represents the load sensitivity coefficient corresponding to the current load segment. The formula is derived from the non-linear relationship between load and transportation costs. It dynamically adjusts fixed delivery distances using the load sensitivity coefficient, replacing the traditional single-distance measurement method and adapting to the actual cost variations in logistics transportation.

[0063] After calculating the load-sensitive distance cost, the system retrieves the corresponding receiving penalty cost from a pre-defined receiving penalty cost mapping table. This mapping table is a standardized mapping database pre-defined by the system, linking the number of customer pickups, the number of split deliveries, and the penalty value. It is used to quantify the hidden logistics and service costs incurred by multiple customer pickups due to multi-warehouse split deliveries. For each combination of warehouse, customer, and product category, the calculated load-sensitive distance cost is added to the retrieved receiving penalty cost to obtain the comprehensive cost for that combination. The corresponding formula is: ,in, Represents comprehensive value. Represents the distance cost that is sensitive to load. The value represents the penalty cost received, which is obtained by querying the penalty cost mapping table.

[0064] The system iterates through all warehouse and customer combinations, calculates all comprehensive cost values, and then arranges all comprehensive cost values ​​in a standardized and structured manner, with warehouses as the row dimension and customers as the column dimension. This results in a unified cost matrix that covers all warehousing and distribution combinations and achieves a unified cost measurement standard for order splitting and route planning.

[0065] In a preferred embodiment of the present invention, a demand allocation module is further included. Based on a unified cost matrix, demand allocation is iteratively performed at the granularity of sub-demand sets. After the allocation of each sub-demand set is completed, the current load status is updated, and the updated load status is used to participate in the cost calculation of the next sub-demand set. The specific operations are as follows:

[0066] The demand allocation module implements iterative demand allocation operations using sub-demand sets as the smallest computational granularity. Unlike static allocation methods that complete all demand allocation at once, this module updates the corresponding warehouse's load occupancy status immediately after completing the allocation of each sub-demand set. The updated load status is then used to recalculate the cost of subsequent sub-demand sets, achieving real-time feedback on allocation costs due to load changes. This mechanism reflects the objective law of non-linear changes in transportation costs caused by continuous cargo loading in logistics scenarios, allowing demand allocation decisions to move beyond fixed cost values. The module continuously tracks warehouse resource occupancy using a load status record table and completes supply-demand matching using a bipartite graph minimum cost flow algorithm. During allocation, it continuously drives dynamic updates of the unified cost matrix. For demands that cannot be met in the initial allocation, a limited number of secondary allocation mechanisms are implemented. While fully utilizing existing warehouse storage resources and vehicle carrying capacity, it outputs demand splitting solutions that balance transportation costs and customer receiving costs, providing downstream users with a definite warehouse-to-warehouse matching relationship.

[0067] The demand assignment module also includes the following steps:

[0068] Step 21: Establish a current load status record table for each warehouse, containing the allocated load and the remaining available load; use the minimum comprehensive cost value in the unified cost matrix as the minimum reachable cost of each sub-demand set, and determine the processing order of each sub-demand set in ascending order of minimum reachable cost. The specific operations are as follows:

[0069] An independent current load status record table is created for each warehouse. This table continuously maintains two types of values: the allocated load that is already occupied in the warehouse, and the warehouse's current remaining available load. The allocated load refers to the total weight of all goods already assigned to that warehouse for shipment. The remaining available load is calculated by subtracting the allocated load from the vehicle's rated maximum load capacity. The remaining available load also constitutes a hard load constraint in subsequent allocation processes. The system traverses all pending sub-demand sets, retrieves the corresponding comprehensive cost value from the unified cost matrix, and defines this value as the minimum reachable cost for the corresponding sub-demand set. The minimum reachable cost represents the lowest comprehensive cost that the sub-demand set can achieve by allocating it to any warehouse under ideal resource conditions. All sub-demand sets are rearranged in ascending order of minimum reachable cost values, forming an ordered task processing sequence. Processing sub-demand sets in this order prioritizes matching demands with better cost potential, reducing the probability of high-cost demands crowding out quality warehousing and distribution resources later, and reducing the scale of subsequent secondary allocation demands.

[0070] Step 22: Using the current sub-demand set as the demand side, the inventory of each warehouse as the supply side, and the comprehensive cost value in the unified cost matrix as the expense, perform minimum cost flow allocation on the bipartite graph network, record the allocation results, and mark unmet demands as awaiting secondary allocation. The specific operations are as follows:

[0071] Construct a supply-demand bipartite graph for the current set of unprocessed sub-demands within the sequence, and build a bipartite graph network structure. Figure 1 The side nodes, representing the demand side, correspond to the total demand for goods contained in the current sub-demand set. The nodes on the other side of the bipartite graph, representing the supply side, correspond to the current available inventory in each warehouse. The edge weights connecting the supply and demand nodes within the bipartite graph are taken from the comprehensive cost value stored within a unified cost matrix. The minimum cost flow algorithm is a well-known supply and demand matching algorithm in operations research. Its objective is to satisfy as many goods demands as possible with the lowest comprehensive cost value within the constraints of inventory and load capacity. After the algorithm finishes, the system fully records the allocation quantity of each type of goods demand to each warehouse. Goods demands that cannot be satisfied due to warehouse inventory or remaining vehicle load capacity are marked as demands awaiting secondary allocation. The marking operation is used to distinguish between demands that were successfully allocated initially and demands that cannot be fulfilled temporarily due to insufficient resources, facilitating targeted supplementary allocation in subsequent steps.

[0072] Step 23: Based on the allocation results, accumulate the allocated load capacity of each involved warehouse and deduct its remaining available load capacity, update the load capacity sensitivity coefficient, and refresh the load capacity sensitive distance cost of the corresponding warehouse row in the unified cost matrix. The specific operations are as follows:

[0073] Read the allocation results output from step 22. For all warehouses undertaking goods allocation tasks, add the corresponding cargo weight to the allocated load in the load status record table, and simultaneously deduct the same weight value from the remaining available load to obtain the updated load status. Based on the updated remaining available load, determine the current load segment of the warehouse, retrieve the pre-configured load sensitivity coefficient group, and complete the load sensitivity coefficient update operation. Subsequently, the system locates the entire row of data belonging to that warehouse within the unified cost matrix, recalculates the load-sensitive distance cost group by group, and completes the refresh operation of the corresponding row of data in the matrix. After the unified cost matrix is ​​updated in real time, when subsequent sub-demand sets are evaluated for cost, the updated cost value can be directly used. This ensures that the changes in transportation costs caused by the continuous increase in warehouse load take effect in real time in the allocation model, overcoming the defect of traditional static cost models that cannot perceive load changes.

[0074] Step 24: If there is an unprocessed sub-demand set, process the next sub-demand set; if not, check the demands to be reassigned. If they exist, merge them into a supplementary sub-demand set and insert it at the end of the processing sequence for reassignment, and repeat the process at most twice. If they do not exist, terminate the iteration. The specific operations are as follows:

[0075] First, the system searches the task processing sequence to determine if there are any sub-requirement sets that have not yet been allocated. If an unprocessed sub-requirement set exists, the system jumps to step 22 to continue processing the next sub-requirement set in the sequence. After all sub-requirement sets have completed one round of traversal processing, the system searches for all marked requirements awaiting secondary allocation. Once a requirement awaiting secondary allocation is found, the system integrates the scattered requirements into a single supplementary allocation sub-requirement set and places this set at the end of the original processing sequence, initiating a new round of allocation. The system sets a maximum number of secondary allocations, allowing a maximum of two rounds of supplementary allocation. Requirements that cannot be satisfied after reaching the maximum number of rounds will no longer be iterated for allocation, avoiding infinite loop operations that consume system computing power. When there are no requirements awaiting secondary allocation, or when the supplementary allocation reaches the maximum number of executions, the overall requirement allocation iteration process terminates directly, and the final determined requirement allocation result is output.

[0076] In a preferred embodiment of the present invention, a route generation module is further included. Based on the demand allocation result, an initial delivery route is constructed using a parallel clustering insertion method. During the insertion process, the cumulative load is sensed in real time and the cost of subsequent insertion operations is dynamically adjusted based on the cumulative load. The specific operations are as follows:

[0077] The path generation module, relying on the formal demand allocation results output by the demand assignment module, adopts a novel path construction logic of parallel clustering insertion, enabling the simultaneous parallel generation of multiple vehicle paths. Throughout the entire customer path insertion process, the path generation module continuously and in real-time senses the cumulative cargo load of a single delivery path. Based on the dynamically changing cumulative load and matching corresponding load sensitivity coefficients, it adjusts the insertion cost of subsequent customer nodes in real time, achieving deep adaptation between the path planning stage and the nonlinear cost characteristics of load. This working mode can accurately reproduce the objective logistical law of the unit distance delivery cost changing with load during vehicle loading, avoiding the local optimum problem of traditional path planning that only considers spatial distance and ignores load efficiency. Through the entire process of load over-limit screening, customer orientation angle clustering, dynamic cost insertion, and path compliance verification, the path generation module generates an initial delivery path scheme that takes into account routing costs, receiving penalty costs, and vehicle load constraints. This provides a compliant and effective initial iterative solution for subsequent neighborhood search, ensuring the effectiveness and accuracy of the overall system's global optimization.

[0078] The path generation module also includes the following steps:

[0079] Step 31: Based on the warehouse allocation records, sum the products of the allocated quantity and corresponding weight for each customer to obtain the total weight of the customer's goods. Extract the set of customers awaiting delivery and the total weight of goods for each warehouse. The sum of the total weight of goods for each warehouse is the total delivery weight. If the total delivery weight exceeds the rated maximum load, it is marked as an overloaded warehouse. The specific operation is as follows:

[0080] First, all warehouse product allocation records output by the demand dispatch module are collected. For each customer, the product of the quantity of products allocated to that customer in each warehouse and the corresponding individual product weight is calculated and summed to obtain the total weight of the customer's products. This calculation method can accurately summarize the overall weight of all goods to be delivered for a customer, avoiding the fragmented customer weight statistics problem caused by multi-warehouse allocation. The corresponding formula is: This formula is based on the physical measurement logic that the total weight of goods is the product of the weight of each item and the allocated quantity, and can completely summarize the weight of all goods allocated by the customer across warehouses. Represents the total weight of goods for a single customer. This represents the quantity of the i-th type of goods allocated to the customer in the corresponding warehouse. This represents the weight of a single item in the i-th category of goods.

[0081] After completing the weight statistics for all customers, the system aggregates all customers awaiting delivery for a single warehouse, using that warehouse as the statistical dimension. Simultaneously, it summarizes the total weight of goods for all customers in that warehouse, obtaining the total delivery weight for the warehouse. The system retrieves the vehicle's rated maximum load capacity as a hard threshold, comparing the total delivery weight of the warehouse with the rated maximum load capacity. If the total delivery weight exceeds the vehicle's rated maximum load capacity, it indicates that the total amount of goods awaiting delivery in that warehouse exceeds the single vehicle's carrying capacity limit and cannot be delivered in a single trip. The system marks this warehouse as an overloaded warehouse and temporarily excludes it from the subsequent routine path clustering and planning process. If the load capacity limit is not exceeded, it is determined to be a compliant warehouse and can proceed to the subsequent path generation stage. This approach avoids generating non-compliant delivery plans due to vehicle overload from the outset.

[0082] Step 32: For warehouses not marked as exceeding limits, sort and group customers according to their azimuth angle relative to the warehouse. The number of groups is equal to the number of available vehicles. Use the customer closest to the warehouse as the initial customer seed for each group. The specific operation is as follows:

[0083] Based on the latitude and longitude coordinates of unmarked compliant warehouses and customers, the system calculates the azimuth angle of each customer to be delivered relative to the corresponding warehouse. The azimuth angle is a well-known geospatial calculation parameter. Using the warehouse location as the origin and due north as the reference axis, the system rotates clockwise to the angle between the line connecting the warehouse and the customer, accurately representing the spatial distribution of customers relative to the warehouse. The system sorts and groups all customers to be delivered according to the magnitude of the azimuth angle. The number of groups is strictly equal to the number of available vehicles configured for that warehouse, ensuring that each delivery vehicle corresponds to an independent set of customers, thus enabling the basic grouping architecture for multi-vehicle parallel planning.

[0084] After completing the customer grouping operation, for each customer group, the actual delivery distance between each customer in the group and the corresponding warehouse is calculated. This distance is obtained by the road network cumulative mileage calculation method, which is consistent with the distance calculation method in the cost matrix module. Then, the customer in the group that is closest to the warehouse is selected as the initial customer seed for each group's path. The seed customer is used as the starting delivery node for each group's delivery path. This seed selection method can effectively shorten the initial route length of a single path, optimize the basic cost of the initial path, and at the same time ensure the concentration of the spatial distribution of customers in each group, reducing the probability of subsequent customers inserting routes that detour.

[0085] Step 33: For each group, using the initial customer seed as the starting point of the path, select customers that have not yet been inserted. Use the product of the new delivery distance and the load sensitivity coefficient as the insertion cost, and add the receiving penalty cost to select the customer insertion path with the lowest overall insertion cost. After each insertion, update the cumulative load as the load sensitivity coefficient query benchmark for subsequent insertion operations. The specific operations are as follows:

[0086] Using the initial customer seed preset for each group as the fixed starting point of a single delivery route, all customer nodes within the group that have not yet been inserted into the route are traversed, and the comprehensive insertion cost for each customer to be inserted is calculated. The single customer insertion cost consists of the superposition of two core costs: dynamic load-sensitive distance cost and customer receiving penalty cost. The load-sensitive distance cost is obtained by multiplying the new delivery distance by the load sensitivity coefficient corresponding to the current load status, and the corresponding formula is: ,in, Representing the overall cost of insertion for the client, This represents the additional delivery distance generated by inserting this customer. This represents the load sensitivity coefficient corresponding to the cumulative load of the current path. The value represents the penalty cost received, which is obtained by querying the penalty cost mapping table.

[0087] After calculating the comprehensive insertion cost for all uninserted customers, the customer with the lowest cost is selected as the optimal insertion node, and this customer is added to the current delivery route. Each time a customer insertion operation is completed, the total weight of the customer's goods is immediately accumulated, updating the latest cumulative load of the current route. This updated cumulative load is used as the benchmark for the load sensitivity coefficient query in the next round of customer insertion operations, re-matching the sensitivity coefficient of the corresponding load segment, thus achieving dynamic updates to the subsequent insertion cost. This iterative insertion method can adapt to the non-linear cost changes during vehicle loading, ensuring that every customer insertion decision is based on real-time load conditions, significantly improving the load adaptability and cost rationality of the initial route plan.

[0088] Step 34: Merge the routes of each group to generate an initial delivery route plan. If the cumulative load of a route exceeds the rated maximum load, remove the corresponding customer and mark it for reallocation. The specific operations are as follows:

[0089] The system aggregates all independent delivery routes generated iteratively by vehicle groups, merging them into a complete initial warehouse delivery route plan. This enables the overall output of multi-vehicle delivery routes within a single warehouse. After route merging, a load compliance check is performed on each independent delivery route. The total weight of goods for all customers along a single route is extracted to obtain the route's cumulative load. This cumulative load is then compared to the vehicle's rated maximum load. If the cumulative load is less than or equal to the rated maximum load, the route is deemed compliant and retained in the initial route plan. If the cumulative load exceeds the rated maximum load, it indicates that the total amount of goods on the route exceeds the vehicle's carrying capacity, posing a risk of delivery violations. The system automatically removes the corresponding customer node from the route and marks the removed customer requests as pending reallocation, awaiting subsequent secondary optimization. This final load check completely eliminates out-of-limit routes generated during the initial route generation process, ensuring that the output initial delivery route plan fully complies with vehicle hardware constraints. It also accurately identifies abnormal requests, providing complete and compliant initial plan data for subsequent global optimization iterations.

[0090] In a preferred embodiment of the present invention, an operator set construction module is further included, which is used to construct a neighborhood search operator set. The set includes operators for merging the product allocation of the same customer, optimizing the access order of the same customer, and jointly exchanging customer locations in different paths, and evaluating load utilization and changes in receiving costs. The specific operations are as follows:

[0091] Traditional logistics route optimization neighborhood operators only adjust the customer access order, which cannot adapt to specific problems such as multiple customer receptions and uneven route load caused by multi-warehouse split delivery. This module designs three types of differentiated optimization operators to achieve optimization capabilities such as merging multiple route allocations for the same customer, optimizing the access order of the same customer, and jointly exchanging customer locations on different routes. The execution effect of all operators is evaluated based on both load utilization and receiving cost. The operator set construction module completes the optimization object identification and effective operator selection based on the real route data of the initial delivery route plan. The constructed neighborhood search operator set can continuously correct problems such as scattered customer delivery, unreasonable access time, and unbalanced route load in the initial route plan without violating vehicle load constraints and warehouse inventory constraints. It provides a sufficient, effective, and business-constrained neighborhood solution space for the dual-objective optimization module, ensuring the convergence and optimality of the global optimization process.

[0092] The operator subset construction module also includes the following steps:

[0093] Step 41: Traverse the initial delivery route plan to identify customers to be merged, customers to be optimized in order, and candidate exchange path pairs. Among them: customers appearing more than 1 times are customers to be merged; customers to be merged whose access sequence number difference is greater than a preset tolerance threshold are customers to be optimized in order; and two paths with load utilization rates higher and lower than the average, respectively, are candidate exchange path pairs. The specific operations are as follows:

[0094] First, the initial delivery route plan output by the route generation module is fully traversed, and the frequency of each customer appearing on all delivery routes is counted. Customers that appear repeatedly on multiple different delivery routes, with more than one appearance, are defined as customers to be merged. These customers correspond to the multi-warehouse split delivery scenario, which directly increases the number of customer receptions and generates additional reception penalties. They are the primary optimization targets that reduce the overall rationality of the delivery plan. For the identified customers to be merged, their access sequence number in each delivery route is further extracted. The access sequence number is the numerical value of the sequential arrangement of customers within the route, used to characterize the delivery order of customers in a single delivery route. The system presets a fixed tolerance threshold, which is a pre-configured benchmark parameter for judging timing deviations. This threshold is used to quantify the degree of irrationality in customer access timing. When the difference in the access sequence number of the same customer to be merged on different routes exceeds this tolerance threshold, it is determined that the customer has an access timing disorder problem and is marked as a customer whose order needs to be optimized, so that the customer's delivery timing can be uniformly adjusted in the future, reducing invalid route detours.

[0095] After identifying the target customers, the load utilization rate of all delivery routes in the initial plan is calculated and statistically analyzed. The load utilization rate is the ratio of the actual cumulative load of a single route to the vehicle's rated maximum load, used to characterize the saturation of vehicle loading resources. The average load utilization rate of all delivery routes is calculated. Using this average as a balancing benchmark, high-load routes with load utilization rates above the average and low-load routes with load utilization rates below the average are selected. The corresponding high-load and low-load route pairs are defined as candidate exchange route pairs, providing precise operational objects for subsequent route load balancing optimization and solving the problem of uneven distribution of load resources in multi-vehicle delivery processes.

[0096] Step 42: Merge each customer to be merged into the path with the lowest cumulative load in its path; adjust the latest access of each customer to be optimized to the first position of its corresponding path; swap customers in candidate swap path pairs. If the load utilization of both paths is closer to the average after the swap, the swap is confirmed to be effective. The specific operations are as follows:

[0097] This step sequentially executes the corresponding neighborhood optimization operations for the three types of optimization objects identified in step 41, and configures the corresponding validity judgment rules to complete the generation of the entire set of neighborhood search operators. For all customers to be merged, a customer merging optimization operation is performed, retrieving all delivery paths that the customer has appeared on, extracting the current cumulative load value of each corresponding path, and selecting the path with the smallest cumulative load as the target merging path. All delivery tasks of the customer scattered on other paths are merged into the target path. This merging method can reduce the number of customer pickups and reduce the receiving penalty cost without increasing the risk of vehicle overload. At the same time, it reduces the low-load transportation conditions caused by scattered goods and improves the overall load utilization efficiency.

[0098] For all customers whose order needs to be optimized, the access order adjustment operation is performed. The latest access time position of the customer in the corresponding delivery path is extracted, and the delivery node of the customer is uniformly adjusted to the first access position of the corresponding path. Regularizing the customer access order can avoid the path backtracking and mileage redundancy problems caused by the same customer's delayed delivery in multiple routes, stabilize the routing efficiency of a single path, and realize local path optimization in the time dimension.

[0099] For all candidate swap path pairs, a joint customer location swap operation is performed. Two swappable customer nodes from paired paths are randomly selected to swap locations. After the swap, the updated load utilization rates of the two paths are recalculated. The system determines whether the load utilization rates of the two swapped paths are closer to the average load utilization rate of all paths compared to before the swap. If this condition is met, the customer swap operation is considered valid, and the corresponding swap optimization operator is retained. If this condition is not met, the swap is considered invalid, and the corresponding operation is discarded. This determination rule strictly ensures that path swap operations always optimize towards load balancing, avoiding invalid swaps from disrupting the existing path load balance. Finally, all valid merging operations, sequence optimization operations, and path swap operations are integrated to form a complete set of neighborhood search operators.

[0100] In a preferred embodiment of the present invention, a dual-objective optimization module is further included. Starting from the initial delivery path, a tabu search is performed, and a comprehensive fitness function composed of routing cost and receiving cost is used to evaluate candidate solutions. The receiving cost weight is dynamically adjusted, and a candidate solution is accepted when the increase in load utilization is greater than a preset threshold. The specific operations are as follows:

[0101] The dual-objective optimization module uses the initial delivery route scheme output by the path generation module as the starting solution for iteration. Relying on the iterative mechanism of the tabu search algorithm, it completes the global optimization of the delivery route, achieving bidirectional collaborative optimization of routing costs and customer receiving costs. Unlike traditional single-objective path optimization methods, the dual-objective optimization module constructs a two-dimensional comprehensive fitness function to evaluate the merits of candidate solutions. It weights and integrates the routing costs generated during the delivery process with the receiving costs incurred by the customer from multiple pickups, forming a unified quantitative evaluation standard. Simultaneously, the dual-objective optimization module introduces a load utilization constraint mechanism and a dynamic weight adjustment mechanism. During the iterative optimization process, it no longer judges the merits of solutions solely based on cost values, but prioritizes optimization schemes that improve load utilization. Furthermore, it dynamically adjusts the weight ratio of receiving costs based on the overall trend of load utilization during iteration, achieving a dynamic balance between vehicle loading efficiency and fulfillment service costs. The dual-objective optimization module avoids repeated and ineffective searches during the iteration process through a tabu mechanism, preventing the algorithm from getting stuck in local loops and local optima. It completes the acceptance and rejection of candidate solutions through multi-condition hierarchical judgment rules, continuously iterates and updates the current optimal delivery route solution, and finally outputs a globally optimal delivery solution that takes into account load utilization, routing economy and customer receiving cost.

[0102] The dual-objective optimization module also includes the following steps:

[0103] Step 51: Taking the delivery route plan as the current solution, initialize the taboo table, the current optimal solution storage area, and the load utilization history table. Initialize the receiving cost weight coefficient of the comprehensive fitness function, which is composed of the weighted sum of the routing cost item and the receiving cost item. The specific operations are as follows:

[0104] The compliant initial delivery route scheme output by the route generation module is set as the initial current solution for algorithm iteration. This initial current solution serves as the benchmark scheme for all subsequent neighborhood iterations, carrying all operational information such as route arrangement, customer allocation, and vehicle load distribution in the initial stage. The system simultaneously completes the initial configuration of three types of data storage structures: a taboo table, a current optimal solution storage area, and a load utilization history table. The taboo table records neighborhood movement operations that have been performed during the iteration process to avoid duplicate searches. The current optimal solution storage area is used to retain the delivery route scheme with the best overall performance throughout the iteration process. The load utilization history table records the overall path load utilization value corresponding to each iteration, providing data traceability for subsequent dynamic weight adjustments.

[0105] This step also initializes the parameters of the comprehensive fitness function. The comprehensive fitness function is the sole quantitative standard for evaluating the merits of a delivery route plan. It is obtained by a weighted sum of the routing cost and receiving cost, and the corresponding formula is: This formula balances the proportion of routing hardware transportation costs and implicit customer reception costs using weighting coefficients, achieving a unified quantitative comparison of these two different cost dimensions. Here, F represents the overall fitness function value; a smaller function value indicates better overall path solution performance. This represents the routing cost, which is the sum of the dynamic load-sensitive distance costs for all delivery routes; This represents the receiving cost item, which is the sum of the receiving penalty costs for all customers. This represents the receiving cost weighting coefficient, used to adjust the proportion of receiving cost in the overall evaluation system. An initial fixed value is configured for the receiving cost weighting coefficient before the iteration starts to ensure that the first round of neighborhood solution evaluation has a unified judgment benchmark.

[0106] Step 52: Generate a neighborhood solution set based on the neighborhood search operator set, and select the neighborhood solution with the smallest comprehensive fitness function value as the candidate solution; if the move operation corresponding to the candidate solution matches the taboo record, it is rejected; otherwise, it is accepted when the load utilization improvement is greater than a preset threshold; otherwise, it is determined whether to accept it by comparing the comprehensive fitness function value with the current best solution; the accepted move is recorded in the taboo table, and the receiving cost weight coefficient is adjusted according to the load utilization change trend after each preset iteration cycle. After the iteration terminates, the current best solution is output. The specific operation is as follows:

[0107] The complete set of neighborhood search operators generated by the operator set construction module is invoked to perform various local optimization operations on the current solution, generating multiple sets of differentiated neighborhood solutions in batches to form a neighborhood solution set. Each set of neighborhood solutions corresponds to an optimization scheme such as customer merging, access order adjustment, or cross-path customer exchange. The comprehensive fitness function value of all neighborhood solutions is calculated one by one, and the neighborhood solution with the smallest function value is selected as the only candidate solution for this round of iteration. This candidate solution is the optimization scheme with the best comprehensive cost in the current neighborhood space.

[0108] A hierarchical acceptance logic is applied to the selected candidate solutions. First, it checks whether the neighborhood move operation corresponding to the candidate solution is recorded in the taboo list. If the move operation exists in the taboo list, the candidate solution is discarded, and the current optimization iteration is prohibited. If the move operation is not taboo, the overall path load utilization improvement after optimization is further calculated. This improvement is the difference between the load utilization of the current candidate solution and the load utilization of the current solution before the iteration. This value is compared with a preset load improvement threshold. When the load utilization improvement is greater than the preset threshold, the candidate solution is directly accepted as the new current solution without comparing the comprehensive fitness value, prioritizing the optimization benefits of vehicle loading efficiency. When the load utilization improvement is less than or equal to the preset threshold, the determination is made based on the value of the comprehensive fitness function. If the comprehensive fitness value of the candidate solution is better than the current optimal solution, the candidate solution is accepted and the current optimal solution is updated; otherwise, the current optimization result is discarded, and the original path scheme is retained.

[0109] All approved and successfully accepted movement operations are recorded and added to a taboo list to prevent the repeated execution of the same local optimization actions in subsequent iterations, thus improving the algorithm's search breadth. Weight updates are performed according to a preset fixed iteration cycle. After each complete iteration cycle, accumulated data from the load utilization history table is retrieved to analyze the overall trend of load utilization during the iteration process. Based on the trend characteristics, the value of the receiving cost weight coefficient is dynamically adjusted. When load utilization continues to improve, the receiving cost weight can be appropriately reduced to further explore loading efficiency potential; when load utilization tends to stabilize, the receiving cost weight can be appropriately increased to optimize the customer fulfillment experience. Iteration continues until the algorithm's preset iteration termination condition is met. The iteration process then stops, and the final optimal delivery route plan is retrieved and output from the optimal solution storage area.

[0110] In a preferred embodiment of the present invention, a dynamic iteration module is further included, which corrects the parameters of the unified cost matrix based on the deviation between the actual execution result and the planned scheme and re-executes the optimization, and triggers online re-optimization when the actual state deviates from the planned state. The specific operations are as follows:

[0111] The aforementioned demand allocation, route generation, and dual-objective optimization modules all perform static global optimization based on preset model parameters. This fails to adapt to dynamic operational disturbances such as vehicle loading fluctuations, delivery deviations, and changes in receiving status that occur during actual field operations. The dynamic iteration module uses the actual delivery operation data as a benchmark, compares it with the optimized planned delivery scheme data, quantifies and calculates two core deviation indicators: load utilization rate and customer receiving frequency. It distinguishes between systemic overall deviations and local sudden deviations, employing differentiated re-optimization strategies. For systemic overall operational deviations, the dynamic iteration module performs overall calibration of the load sensitivity coefficient of the unified cost matrix by solving for correction factors, achieving adaptive iterative updates of the model cost system and restarting the entire optimization process to obtain a new delivery scheme adapted to the current operational conditions. For single-point local abnormal disturbances, the dynamic iteration module performs local redistribution and local route reconstruction under real-time inventory and vehicle load status constraints. This ensures the scheme conforms to the actual field conditions while reducing iterative computation overhead, achieving continuous dynamic correction and long-term global adaptation for multi-warehouse delivery management.

[0112] The dynamic iteration module also includes the following steps:

[0113] Step 61: Obtain the actual execution results data of the final delivery route plan. Use the difference between the actual load utilization rate and the planned load utilization rate for each route as the load utilization rate deviation value, and use the difference between the actual number of receipts and the planned number of receipts for each customer as the receipt number deviation value. Summarize and generate a deviation record table. The specific operations are as follows:

[0114] The system reads the planned data corresponding to the final delivery route scheme output and executed by the dual-objective optimization module. Simultaneously, it collects actual execution result data from the warehouse scheduling system, vehicle-mounted terminal equipment, and customer receiving registration equipment. This provides a complete picture of the actual loaded weight, vehicle rated load capacity, actual customer receiving frequency, planned load utilization rate, and planned customer receiving frequency for each delivery route. The vehicle-mounted terminal equipment records the actual load weight of each vehicle and the delivery task list in real time. The warehouse scheduling system synchronously records the actual outbound allocation data for each warehouse, and the customer receiving registration equipment records single receiving fulfillment records. All data provides objective and traceable raw data for system deviation calculation.

[0115] Using a single delivery route as the calculation unit, the difference between the actual load utilization rate and the planned load utilization rate of the route is calculated to obtain the load utilization rate deviation value. The corresponding formula is: This formula precisely quantifies the deviation of vehicle loading efficiency by subtracting actual operating parameters from planned model parameters; among which, This represents the deviation value of load utilization rate. The actual load utilization rate of the route is calculated by dividing the total actual load weight of the route by the vehicle's rated maximum load capacity. The target load utilization rate is the pre-set target load of the scheme.

[0116] Using a single customer as the calculation unit, the difference between the actual number of receptions and the planned number of receptions is calculated to obtain the reception deviation value. The corresponding formula is: This formula is used to quantify the deviation between the customer's receiving experience and the planned outcome. Among other things, This represents the deviation value of the number of receptions. This represents the actual number of times a customer receives goods. This represents the number of customer plans to receive items as preset in the optimization plan; it iterates through all delivery routes and all service customers, summarizes all load utilization deviation values ​​and receiving number deviation values, organizes and stores them according to route dimension and customer dimension, generates a complete structured deviation record table, and realizes comprehensive quantitative retention of operational deviations.

[0117] Step 62: If the absolute value of the average deviation of each path in the deviation record table is greater than the system deviation detection threshold, then a systematic deviation is determined and the deviation direction is identified; paths whose deviation direction is inconsistent with the systematic deviation direction are marked as sudden deviation paths; customers with a reception count deviation value greater than zero are marked as reception deviation customers; output a deviation determination result report. The specific operations are as follows:

[0118] First, extract the load utilization deviation values ​​for all paths in the deviation record table. Calculate the arithmetic mean of all path deviation values ​​to obtain the global path deviation average. Then, calculate the absolute value of this average and compare it with the system's preset deviation detection threshold. The deviation detection threshold is a fixed judgment parameter preset by the system, used to define whether the operational condition deviation reaches the degree of overall model mismatch. If the absolute value of the global path deviation average is greater than the deviation detection threshold, it proves that the overall load utilization status of the delivery operation has deviated from the model's preset status. It is determined that there is a systematic deviation in the current operation. The direction of the overall deviation is determined based on the positive or negative value of the deviation average. A positive value indicates that the overall actual load utilization rate is higher than the planned level, and a negative value indicates that the overall actual load utilization rate is lower than the planned level.

[0119] Based on the determination of systematic deviations, the consistency between the deviation direction of a single path and the global systematic deviation direction is further compared path by path. Paths whose single-path deviation trend is opposite to the overall deviation trend are defined as sudden deviation paths. These paths are not affected by the overall operating condition offset and belong to abnormal paths caused by single-point random operation disturbances. At the same time, the system compares the receiving frequency deviation value for each customer. Customers with a receiving frequency deviation value greater than zero are marked as receiving deviation customers, indicating that the actual number of receiving times for these customers is more than the planned preset number, resulting in additional receiving penalty costs. The conclusions of the systematic deviation determination, the list of sudden deviation paths, and the list of receiving deviation customers are summarized and a standardized deviation determination result report is output, providing an accurate basis for anomaly classification for subsequent hierarchical re-optimization strategies.

[0120] Step 63: If the deviation is systematic, determine the correction factor based on the deviation direction and average value, correct the load sensitivity coefficient in the unified cost matrix, and re-execute demand allocation, route construction, and tabu search to generate a new delivery route plan. If there are sudden deviations in routes or customers, use the current inventory status of the corresponding warehouse and the current load status of the vehicle as constraints to perform local demand allocation and local route construction, generating a locally updated delivery route plan. The specific operations are as follows:

[0121] If the system determines that a systematic deviation exists, it indicates that the load sensitivity coefficient of the original unified cost matrix is ​​no longer suitable for the current overall logistics conditions, and there is a global mismatch in the original model parameters. The system calculates the corresponding parameter correction factor based on the determined deviation direction and the global deviation average. The magnitude of the correction factor is positively correlated with the overall deviation offset, ensuring that the parameter correction magnitude matches the degree of operational condition offset. The system uses the correction factor to perform a global correction and update of the load sensitivity coefficients corresponding to all warehouses within the unified cost matrix, completing the global parameter calibration of the cost model. After the parameter correction is completed, the entire process optimization link is restarted, and the dual-objective optimization of demand assignment, initial path construction, and tabu search is executed iteratively to generate a new global delivery route plan that adapts to the latest operational conditions, eliminating the plan deviation caused by the systematic model mismatch.

[0122] If the system detects a sudden deviation in the delivery path or a customer with a delivery deviation, it determines the current deviation as a local single-point anomaly. The overall model parameters remain adapted to the mainstream operating conditions, and no global parameter correction or full re-optimization is required. The system reads the current remaining inventory status of the warehouse corresponding to the abnormal path and the current load occupancy status of the corresponding delivery vehicle in real time. Using real-time inventory and load constraints as hard boundary conditions, it performs local demand reallocation and local path reconstruction for the demand range corresponding to the abnormal path and the deviating customer. Only the delivery allocation relationship and path node arrangement corresponding to the abnormal point are updated, while the original optimization results of other normal paths are retained. A locally updated delivery route plan is generated through local updates. While correcting local fulfillment deviations, it reduces the consumption of redundant calculation resources, achieving rapid online correction of dynamic operating condition disturbances. This ensures that the system continuously outputs compliant and balanced delivery execution plans under dynamically changing warehouse and vehicle conditions.

[0123] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A warehouse-based product reorganization and route optimization management system, characterized in that, include: The cost matrix building module is used to obtain basic delivery parameters from multiple warehouses to each customer, divide the total demand of each customer into multiple sub-demand sets, and construct a unified cost matrix that includes load-sensitive costs and receiving penalty costs. The demand assignment module iteratively performs demand allocation at the granularity of sub-demand sets based on a unified cost matrix. After completing the allocation of each sub-demand set, it updates the current load status and uses the updated load status to participate in the cost calculation of the next sub-demand set. The route generation module constructs the initial delivery route based on the demand allocation results using a parallel clustering insertion method. During the insertion process, it senses the cumulative load in real time and dynamically adjusts the cost of subsequent insertion operations based on the cumulative load. The operator set construction module is used to build a neighborhood search operator set, which includes operators for merging the product allocation of the same customer, optimizing the access order of the same customer, and jointly exchanging customer locations in different paths, and evaluating load utilization and receiving cost changes. The dual-objective optimization module starts from the initial delivery route, performs tabu search, evaluates candidate solutions using a comprehensive fitness function composed of routing cost and receiving cost weights, dynamically adjusts the receiving cost weights, and accepts candidate solutions when the increase in load utilization exceeds a preset threshold. The dynamic iteration module adjusts the parameters of the unified cost matrix and re-executes the optimization based on the deviation between the actual execution results and the planned scheme, and triggers online re-optimization when the actual state deviates from the planned state.

2. The warehouse product reorganization and route optimization management system according to claim 1, characterized in that, The cost-based array building module includes: Obtain location data for each warehouse, location data for each customer, individual item weight for each product category, inventory quantity for each warehouse, and the rated maximum load capacity of each vehicle; Calculate the delivery distance from each warehouse to each customer based on warehouse location and customer location, and generate the original distance matrix; configure load sensitivity coefficient groups according to the rated maximum load capacity. For each warehouse, each customer, and each product category combination, the product of delivery distance and load sensitivity coefficient is used as the load sensitivity distance cost. The receiving penalty cost is determined by querying the receiving penalty cost mapping table. The two are added together to obtain the comprehensive cost value. A unified cost matrix is ​​constructed according to the warehouse and customer dimensions.

3. The warehouse product reorganization and route optimization management system according to claim 2, characterized in that, The demand assignment module includes: Establish a current load status record table for each warehouse, which includes the allocated load and the remaining available load; use the minimum comprehensive cost value in the unified cost matrix as the minimum reachable cost of each sub-demand set, and determine the processing order of each sub-demand set in ascending order of minimum reachable cost; Using the current set of sub-demands as the demand side, the inventory of each warehouse as the supply side, and the comprehensive cost value in the unified cost matrix as the expense, perform minimum cost flow allocation on the bipartite graph network, record the allocation results, and mark unmet demands as awaiting secondary allocation.

4. The warehouse product reorganization and route optimization management system according to claim 3, characterized in that, The demand assignment module also includes: Based on the allocation results, the allocated load of each involved warehouse is accumulated and its remaining available load is deducted. The load sensitivity coefficient is updated, and the load-sensitive distance cost of the corresponding warehouse row in the unified cost matrix is ​​refreshed. If there is an unprocessed set of sub-requirements, process the next set of sub-requirements; if not, check the requirements to be reassigned. If they exist, merge them into a supplementary set of sub-requirements and insert them at the end of the processing sequence to re-execute the assignment, up to twice. If they do not exist, terminate the iteration.

5. The warehouse product reorganization and route optimization management system according to claim 4, characterized in that, The path generation module includes: Based on the warehouse collection and allocation records, the sum of the products of the allocated quantity and corresponding weight of each customer is taken as the total weight of the customer's goods. The set of customers to be delivered and the total weight of goods in each warehouse are extracted. The sum of the total weight of goods in each warehouse is the total delivery weight. If the total delivery weight is greater than the rated maximum load, it is marked as an overloaded warehouse. For warehouses that are not marked as exceeding limits, customers are sorted and grouped according to their azimuth angle relative to the warehouse. The number of groups is equal to the number of available vehicles, and the customer closest to the warehouse is used as the initial customer seed for each group.

6. The warehouse product reorganization and route optimization management system according to claim 5, characterized in that, The path generation module also includes: For each group, starting from the initial customer seed, select customers that have not yet been inserted, use the product of the new delivery distance and the load sensitivity coefficient as the insertion cost, and add the receiving penalty cost to select the customer insertion path with the smallest overall insertion cost; after each insertion, update the cumulative load as the load sensitivity coefficient query benchmark for subsequent insertion operations; The routes of each group are merged to generate an initial delivery route plan. If the cumulative load of a route exceeds the rated maximum load, the corresponding customer is removed and marked for reallocation.

7. The warehouse product reorganization and route optimization management system according to claim 6, characterized in that, The operator set construction module includes: Traverse the initial delivery route plan to identify customers to be merged, customers whose order needs to be optimized, and candidate exchange path pairs. Among them, customers whose occurrence frequency is greater than 1 are customers to be merged, customers whose access order number difference is greater than the preset tolerance threshold are customers whose order needs to be optimized, and two paths with load utilization rates higher and lower than the average respectively are candidate exchange path pairs. Merge each customer to be merged into the path with the lowest cumulative load in its path; adjust the latest access of each customer to be optimized to the first position of the corresponding path; swap customers in the candidate swap path pairs, and if the load utilization of the two paths is closer to the average after the swap, the swap is confirmed to be effective.

8. The warehouse product reorganization and route optimization management system according to claim 7, characterized in that, The dual-objective optimization module includes: Taking the delivery route plan as the current solution, initialize the taboo table, the current optimal solution storage area and the load utilization history table, and initialize the receiving cost weight coefficient of the comprehensive fitness function composed of the weighted sum of the routing cost item and the receiving cost item. A neighborhood solution set is generated based on the neighborhood search operator set, and the neighborhood solution with the smallest comprehensive fitness function value is selected as the candidate solution. If the move operation corresponding to the candidate solution matches the taboo record, it is rejected; otherwise, it is accepted when the load utilization improvement is greater than a preset threshold. Otherwise, the acceptance is determined by comparing the comprehensive fitness function value with the current best solution. The accepted moves are recorded in the taboo table. After each preset iteration cycle, the receiving cost weight coefficient is adjusted according to the load utilization change trend. After the iteration terminates, the current best solution is output.

9. The warehouse product reorganization and route optimization management system according to claim 8, characterized in that, The dynamic iteration module includes: Obtain the actual execution results data of the final delivery route plan, use the difference between the actual load utilization rate and the planned load utilization rate of each route as the load utilization rate deviation value, and use the difference between the actual number of receipts and the planned number of receipts for each customer as the receipt number deviation value, and summarize to generate a deviation record table; If the absolute value of the average deviation of each path in the deviation record table is greater than the system deviation detection threshold, a systemic deviation is determined and the deviation direction is identified; paths whose deviation direction is inconsistent with the systemic deviation direction are marked as sudden deviation paths; customers with a reception count deviation value greater than zero are marked as reception deviation customers; and a deviation determination result report is output.

10. The warehouse commodity reorganization and route optimization management system according to claim 8, characterized in that, The dynamic iteration module also includes: If the deviation is systematic, the correction factor is determined according to the deviation direction and average value, the load sensitivity coefficient in the unified cost matrix is ​​corrected as a whole, and the demand allocation, path construction and tabu search are re-executed to generate a new delivery route plan; if there is a sudden deviation path or a customer with a receiving deviation, the current inventory status of the corresponding warehouse and the current load status of the vehicle are used as constraints to execute local demand allocation and local path construction to generate a locally updated delivery route plan.

Images