Multi-commodity order fulfillment method, device and equipment based on dynamic computing power distribution

Abstract

The application provides a multi-commodity order fulfillment method, device and equipment based on dynamic computing power distribution, and relates to the technical field of supply chain management.Through obtaining commodity orders and inventory information, and based on actual needs of multi-commodity order fulfillment in the order and inventory information, a multi-objective optimization algorithm model is established; then a heuristic method based on priority and inventory feasibility is used to generate an initial population; a multi-objective optimization algorithm based on dynamic computing power distribution is used to iteratively optimize the initial population to obtain a Pareto non-dominated solution set; and a final order fulfillment scheme is selected according to the Pareto non-dominated solution set. The application can effectively solve the problem of low optimization efficiency and easy falling into local optimization of multi-commodity order fulfillment in high-dimensional space, balance multiple core operating indicators under complex constraint conditions, and significantly improve the scientificity, global search efficiency and distribution fairness of decision-making.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and more specifically, to a method, apparatus, and equipment for fulfilling multi-commodity orders based on dynamic computing power allocation. Background Technology

[0002] With the rapid development of the supply chain industry, enterprises need to handle massive orders containing multiple product demands in their daily operations. Fulfilling multi-product orders involves making allocation decisions within a three-dimensional decision space comprised of orders, products, and warehouses; that is, determining whether each product in the order should be supplied by a specific warehouse and the specific quantity supplied. Since available inventory is typically distributed across multiple different warehouses, achieving optimal resource allocation has become a core issue in the field of supply chain management.

[0003] Existing methods for developing multi-product order fulfillment plans mostly focus on single-objective optimization or rely on simplified frameworks based on empirical rules. For example, they may solely prioritize cost minimization or fulfillment rate maximization, or use heuristic allocation based on a fixed fulfillment order. These methods often fail to adequately consider the fine-grained coupling relationships between multiple products across multiple warehouses and the competition between different objectives. When faced with complex scenarios involving diverse product types, widely distributed inventory, and conflicting objectives, existing technologies struggle to balance core operational metrics such as profit, order fulfillment rate, inventory utilization, and order priority. Furthermore, due to the extremely high dimensionality of the solution space, traditional algorithms are prone to inefficiency or getting trapped in local optima during optimization, making it difficult to find a high-quality global solution within a limited timeframe.

[0004] Furthermore, actual business operations require maintaining fairness in allocation under rules such as first-come, first-served, which typically necessitates the introduction of product priority based on submission time and corresponding penalty mechanisms. However, traditional formulation methods are clearly insufficient in their ability to coordinate such fairness constraints and multi-objective collaboration, making it difficult to consistently provide solutions that balance overall benefits and individual feasibility under real-time changing order and inventory data. As business scale expands and constraints intensify, existing technologies are more prone to defects such as decreased solution efficiency, shrinking feasible solutions, and solution bias, failing to continuously meet the operational requirements of modern supply chains that emphasize high profits, high order fulfillment rates, high inventory utilization, and high order priority compliance.

[0005] In view of the above, this application is hereby submitted. Summary of the Invention

[0006] The present invention aims to provide a method, apparatus and equipment for fulfilling multi-product orders based on dynamic computing power allocation, so as to solve the problems in the existing multi-product order fulfillment schemes when dealing with the three-dimensional space decision composed of orders, products and warehouses. These schemes are difficult to balance core indicators such as profit, order fulfillment rate, inventory utilization and order priority, and have defects such as low optimization efficiency, easy to get trapped in local optima and difficulty in taking into account the fairness of allocation in high-dimensional decision space.

[0007] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0008] A method for fulfilling multi-product orders based on dynamic computing power allocation includes:

[0009] S1, retrieve product order and inventory information, including the current order set, product set, and warehouse inventory set;

[0010] S2, based on the actual needs of fulfilling multi-product orders in order and inventory information, establishes a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority;

[0011] S3 uses a heuristic method based on priority and inventory feasibility to generate the initial population;

[0012] S4, the initial population is iteratively optimized using a multi-objective optimization algorithm based on dynamic computing power allocation to obtain a Pareto non-dominated solution set; the multi-objective optimization algorithm based on dynamic computing power allocation adopts a dynamic computing power allocation strategy, which allocates computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation;

[0013] S5. Select the final order fulfillment scheme based on the Pareto non-dominated solution set.

[0014] Preferably, the profit rate is the ratio of the actual total profit allocated to satisfy orders to the theoretical maximum total profit, expressed as:

[0015] ;

[0016] in, Profit margin; For orders Chinese commodities The variable determining whether the demand is fully met; For orders Chinese commodities Profits; For the current order set; A collection of goods; This indicates taking the maximum value;

[0017] The order fulfillment rate is the ratio of the number of fully fulfilled orders to the total number of orders, expressed as:

[0018] ;

[0019] in, For order fulfillment rate; The value represents the total number of items in all orders; It is the minimum value;

[0020] The inventory utilization rate is the ratio of the number of goods allocated to an order to the total available inventory, expressed as:

[0021] ;

[0022] in, For inventory utilization rate; For orders Chinese commodities Demand; For warehouse Chinese commodities Inventory levels; For warehouse inventory collection;

[0023] The degree of order priority violation is calculated in a tiered manner based on the product priority number, the identifier indicating whether the product meets the priority requirement, the product arrival time sequence number, and the sequence number of the first unassigned product that meets the priority requirement. This is used to measure the penalty value for not fulfilling early orders, and the expression is as follows:

[0024] ;

[0025] ;

[0026] in, The degree of violation of order priority; This is the theoretical maximum penalty value; This represents the theoretical minimum penalty value. for The penalty value for a group of goods; This indicates taking the minimum value;

[0027] for Priority number of the group of goods; express Products in a group Does it meet the priority? ; express Orders in a group of products Chinese commodities Arrival time sequence number; express The first unassigned product number in the group of products.

[0028] Preferably, the multi-objective optimization algorithm model further includes technical constraints, which include:

[0029] Inventory allocation constraints stipulate that the total quantity of a certain item allocated to an order cannot exceed its demand. The expression is:

[0030] ;

[0031] in, To be assigned to an order A certain product The total amount; For orders Chinese commodities Demand; For the current order set; A collection of goods; For warehouse inventory collection;

[0032] Warehouse inventory capacity constraint stipulates that the total amount of goods allocated from a warehouse cannot exceed its existing inventory, expressed as:

[0033] ;

[0034] ;

[0035] in, For warehouse Chinese commodities Inventory levels; Represents an individual, with a value of 1 or 0. A value of 1 indicates an order. Goods in From warehouse Satisfy; if A value of 0 indicates an order. Goods in Not by warehouse satisfy;

[0036] Warehouse processing capacity constraint, which limits the cumulative value of decision variables in the order dimension, sets an upper limit on the number of orders each warehouse can process for each product. The expression is:

[0037] ;

[0038] in, The maximum number of orders that can be processed for each product in each warehouse.

[0039] Preferably, the specific steps for generating the initial population are as follows:

[0040] First, establish priority rules: prioritize product demand according to order arrival time, with earlier orders having higher priority; within each product group, start with the highest priority demand and try to match inventory from each warehouse in turn until the product inventory is used up or the product demand in all orders has been met; if warehouse inventory is sufficient, skip the product supply for the current order with a probability set by the perturbation factor and directly process the next priority order to increase the diversity of the population and avoid local optima;

[0041] Then, a three-dimensional binary matrix is ​​used to analyze the individuals. Encoding is performed, meaning each individual corresponds to a three-dimensional binary matrix. This is used to represent the allocation relationship between orders, goods, and warehouses. The dimension is the product of the number of orders, the number of product types, and the number of warehouses. ;

[0042] Next, following the feasibility principle, heuristic rules are used. The expression for generating the initial population is as follows:

[0043] ;

[0044] in, For the current order set; A collection of goods; For warehouse inventory collection; This is a priority rule used to determine the order in which goods are allocated; For warehouse inventory feasibility, this is used to constrain the allocated quantity from not exceeding the existing warehouse inventory. For disturbance factors;

[0045] The initial population is obtained based on the individuals in the initial population.

[0046] Preferably, the iterative optimization process of the initial population based on the multi-objective optimization algorithm with dynamic computing power allocation specifically involves:

[0047] First, the multi-objective problem is decomposed into Each sub-problem A uniformly distributed weight vector is associated with each weight; wherein the weight vector is generated through uniform sampling.

[0048] For each subproblem The nearest N weight vectors are selected based on the Euclidean distance between the weight vectors to form a neighbor set. ;

[0049] For each subproblem generate Given 0 initial decisions, each corresponding to a three-dimensional binary matrix, the total number of decisions is . ;

[0050] All initial decisions constitute the decision space, which is then divided into K mutually exclusive sub-regions.

[0051] Each decision in the decision space is expanded into a binary sequence row by row, and the binary sequence is converted into a corresponding decimal integer value, expressed as:

[0052] ;

[0053] in, The decision is flattened and converted to a decimal integer value; To flatten the decision into a string of binary numbers, the first... The value at each position; For the current order set; A collection of goods; For warehouse inventory collection;

[0054] Modulo operation is performed on the decimal integer value to map each decision to a unique sub-region among K mutually exclusive sub-regions, thus completing the division of the decision space into sub-regions. This simplifies the high-dimensional three-dimensional binary matrix into low-dimensional sub-region index numbers, as expressed in the following expression:

[0055] ;

[0056] in, For decision making Sub-region index number; The total number of sub-regions to divide the decision space; For the remainder operation;

[0057] Then iterate through For each decision, the number of decisions located in each sub-region is counted, the decision density of each sub-region is calculated, and the Shannon entropy index is introduced as a measure of the uniformity of the solution set distribution in the decision space to establish a decision density evaluation model, the expression of which is:

[0058] ;

[0059] ;

[0060] ;

[0061] in, For the first Number of decisions per sub-region; For decision-making; For decision set; For the first Decision density of each sub-region; For the uniformity of the decision distribution, i.e., Shannon entropy, when At that time, the distribution was completely uniform;

[0062] Then, for each subproblem, the maximum density value in the set of regions where the decision is made in the population is calculated as the decision density weight, expressed as:

[0063] ;

[0064] ;

[0065] in, For subproblems The set of subregions covered by the population; For subproblems The corresponding population; For the first Each decision space sub-region; For subproblems Decision density weights; Indicates existence;

[0066] Normalizing the decision density weights yields the selection probabilities of the subproblems, expressed as:

[0067] ;

[0068] in, For subproblems The probability of choosing; For subproblems Decision density weight

[0069] Based on the selection probability, the number of times dynamic computing power is allocated for each subproblem is calculated, expressed as:

[0070] ;

[0071] in, For subproblems The number of times dynamic computing power is allocated; This represents the total computing power for each generation, which is the total number of times an individual in the population exchanges information with its neighbors.

[0072] Based on the number of times the dynamic computing power is allocated, each subproblem selects parent individuals from its neighbor set for evolutionary operations to generate offspring decisions, i.e., the Pareto non-dominated solution set.

[0073] Preferably, when performing evolutionary operations on each subproblem, two parent individuals are selected from its neighbor set for crossover and mutation operations until the decision distribution becomes uniform or the preset maximum number of generations is reached.

[0074] Preferably, the method further includes: after each subproblem completes the current evolutionary operation to generate a descendant decision, comparing the descendant decision with the decision in the current population; if the descendant decision is better, replacing the corresponding decision in the original population to update the population.

[0075] Preferably, when selecting the final order fulfillment plan based on the Pareto non-dominated solution set, the comprehensive objective function value is calculated for each solution in the solution set, and its expression is:

[0076] ；

[0077] in, The value of the comprehensive objective function; , , These are the weighting coefficients; This is the penalty coefficient; Profit margin; For order fulfillment rate; For inventory utilization rate; The degree of violation of order priority;

[0078] Based on the comprehensive objective function value, the optimal solution is selected as the final order fulfillment plan, and the corresponding order, product, and warehouse allocation plan is output.

[0079] The present invention also provides a multi-product order fulfillment device based on dynamic computing power allocation, comprising:

[0080] The information acquisition module is used to acquire product order and inventory information, including the current order set, product set, and warehouse inventory set;

[0081] The modeling module is used to establish a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority based on the actual needs of fulfilling multi-product orders in order and inventory information.

[0082] The population generation module is used to generate an initial population using a heuristic method based on priority and inventory feasibility.

[0083] The iterative module is used to iteratively optimize the initial population using a multi-objective optimization algorithm based on dynamic computing power allocation to obtain a Pareto non-dominated solution set. The multi-objective optimization algorithm based on dynamic computing power allocation adopts a dynamic computing power allocation strategy, which allocates computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation.

[0084] The decision module is used to select the final order fulfillment plan based on the Pareto non-dominated solution set.

[0085] The present invention also provides a multi-product order fulfillment device based on dynamic computing power allocation, including a processor and a memory. The memory stores a computer program that can be executed by the processor to realize the multi-product order fulfillment method based on dynamic computing power allocation as described above.

[0086] The present invention also provides a computer-readable storage medium storing computer-readable instructions, which, when executed by a processor of the device on which the computer-readable storage medium is located, implement the multi-product order fulfillment method based on dynamic computing power allocation as described above.

[0087] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0088] First, this invention introduces a dynamic computing power allocation mechanism based on decision space sub-region partitioning and decision density evaluation. This mechanism maps the high-dimensional decision space to low-dimensional sub-regions through modular arithmetic and utilizes Shannon entropy to evaluate the uniformity of the solution set distribution, effectively solving the "curse of dimensionality" problem in high-dimensional combinatorial optimization problems. By dynamically allocating computing resources to the subproblems corresponding to high-density regions, it achieves forced dilution of the search space, avoiding the algorithm from getting trapped in local optima and significantly improving global search efficiency and the convergence quality of the solution set.

[0089] Secondly, this invention solves the technical challenge of traditional single-objective optimization schemes failing to consider multiple operational indicators by constructing a four-objective optimization model that includes profit margin, order fulfillment rate, inventory utilization rate, and order priority violation degree. The model quantifies and models each objective, and, in conjunction with constraints, finds a global equilibrium solution under complex supply chain resource limitations, thereby improving the scientific nature of order fulfillment decisions.

[0090] Third, the heuristic initial population generation rule designed in this invention combines the time series of order arrivals with the feasibility of warehouse inventory. By introducing a perturbation factor, the population is given a sufficient starting point for exploration while ensuring the quality of the initial solution. This mechanism, combined with the objective function, ensures that the solution, while pursuing efficient fulfillment, strictly adheres to the principle of first-come, first-served fairness, reducing the risk of complaints in business operations.

[0091] Fourth, this invention possesses strong scalability and universality. The method for dividing the decision space into sub-regions is independent of the absolute size of the decision space, enabling the invention to maintain stable computational performance even when facing large-scale and complex business scenarios with massive orders and multiple warehouses. Furthermore, the algorithm framework is compatible with both discrete and continuous decision variables, flexibly adapting to different types of supply chain optimization needs. Attached Figure Description

[0092] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0093] Figure 1 This is a flowchart illustrating a multi-product order fulfillment method based on dynamic computing power allocation, as provided in Example 1.

[0094] Figure 2 This is a schematic diagram of the three-dimensional binary matrix encoding of the decision variables provided in Example 1.

[0095] Figure 3 The following is a logic diagram of the dynamic computing power allocation strategy based on decision density evaluation provided in Example 1.

[0096] Figure 4 The iterative evolution diagram is shown for the decomposition-type multi-objective optimization algorithm provided in Example 1.

[0097] Figure 5 This is a schematic diagram of a multi-product order fulfillment device based on dynamic computing power allocation, provided in Embodiment 2.

[0098] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Detailed Implementation

[0099] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0100] Example 1

[0101] Embodiment 1 of the present invention provides a method for fulfilling multi-product orders based on dynamic computing power allocation, which can be implemented by a multi-product order fulfillment device based on dynamic computing power allocation (hereinafter referred to as fulfillment device), and in particular, executed by one or more processors within the fulfillment device.

[0102] In this embodiment, the fulfillment device may be an electronic device equipped with a processor, which carries a computer program for the multi-product order fulfillment method based on dynamic computing power allocation and the computer program can be executed, such as a computer, smartphone, smart tablet, workstation, etc., without limitation.

[0103] In practical applications of supply chain management and logistics, fulfilling multi-product orders presents a highly challenging resource allocation problem. With the surge in e-commerce platform business volume, companies need to process tens of thousands of orders daily, each containing multiple SKUs (Stock Keeping Units). The goods in these orders are often distributed across different warehouses nationwide or even globally. How to balance meeting customer delivery timeliness requirements with balancing profit margins, overall order fulfillment rates, inventory turnover efficiency, and fairness in order processing has become crucial for enhancing supply chain competitiveness. This invention provides a multi-product order fulfillment method based on a dynamic computing power allocation multi-objective optimization algorithm, precisely to solve this decision-making challenge in complex environments.

[0104] like Figure 1 As shown, a multi-product order fulfillment method based on dynamic computing power allocation includes steps S1 to S5.

[0105] S1 retrieves product order and inventory information, including the current order set, product set, and warehouse inventory set.

[0106] The overall execution process of this method covers the entire process from bottom-level data collection to high-level decision output.

[0107] First, the system obtains order and inventory information through a data acquisition interface (such as the enterprise's ERP (Enterprise Resource Planning) system or WMS (Warehouse Management System) interface). This information includes the current order set. Product collection and warehouse inventory collection .

[0108] These data form the basis for subsequent optimization algorithms. The order information includes the submission time of each order, the types of goods included, and the specific demand quantity for each type of goods; the inventory information includes the current available inventory of each type of goods in each warehouse and the maximum daily order processing capacity of each warehouse.

[0109] S2, based on the actual needs of fulfilling multi-product orders in order and inventory information, establishes a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority.

[0110] After obtaining the basic data, mathematical modeling is performed on the order fulfillment problem based on the actual needs of fulfilling multi-product orders in the order and inventory information.

[0111] The model established in this embodiment is a multi-objective optimization problem that includes four core operating indicators (profit margin, order fulfillment rate, inventory utilization rate, and order priority violation of fulfillment plan).

[0112] The profit rate is the ratio of the actual total profit allocated to fulfilling orders to the theoretical maximum total profit, expressed as:

[0113] ;

[0114] in, Profit margin; For orders Chinese commodities The variable determining whether the demand is fully met; For orders Chinese commodities Profits; For the current order set; A collection of goods; This indicates taking the maximum value.

[0115] The order fulfillment rate is the ratio of the number of fully fulfilled orders to the total number of orders, expressed as:

[0116] ;

[0117] in, For order fulfillment rate; The value is the total number of items in all orders; To find the minimum value, avoid having a denominator of 0.

[0118] The inventory utilization rate is the ratio of the number of goods allocated to an order to the total available inventory, expressed as:

[0119] ;

[0120] in, For inventory utilization rate; For orders Chinese commodities Demand; For warehouse Chinese commodities Inventory levels; This is a collection of warehouse inventory.

[0121] The degree of order priority violation is calculated in a tiered manner based on the product priority number, the identifier indicating whether the product meets the priority requirement, the product arrival time sequence number, and the sequence number of the first unassigned product that meets the priority requirement. This is used to measure the penalty value for not fulfilling early orders, and the expression is as follows:

[0122] ;

[0123] ;

[0124] in, The degree of violation of order priority; This is the theoretical maximum penalty value; This represents the theoretical minimum penalty value. for The penalty value for a group of goods; This indicates taking the minimum value;

[0125] for Priority number of the group of goods; express Products in a group Does it meet the priority? ; express Orders in a group of products Chinese commodities Arrival time sequence number; express The first unassigned product sequence number in a group of products, that is, the first order index number in the queue sorted by order arrival time that cannot be fully fulfilled due to stock shortage.

[0126] Profit margin reflects the economic efficiency of fulfillment plans, order fulfillment rate focuses on customer service levels, and inventory utilization rate assesses the efficiency of inventory resource turnover. Introducing the order priority violation rate as an objective is to ensure fairness in the allocation process.

[0127] In actual business operations, adhering to the "first-come, first-served" principle is the cornerstone of maintaining customer satisfaction. The degree of violation of order priority. The calculation is based on product groups. The system records the priority number of each item in each order. The table identifies the first unassigned product serial number within the group. If a later-arriving order is fulfilled, while an earlier-arriving order is not fulfilled due to insufficient inventory or algorithmic bias, the system will incur a penalty value. By normalizing the penalty values ​​for all product groups, the algorithm automatically rejects solutions that violate fairness principles during the search process, thus prioritizing early orders in the results.

[0128] In addition, the multi-objective optimization algorithm model may also have three layers of technical constraints.

[0129] The first layer is the inventory allocation constraint, which ensures that the total amount of goods allocated to a certain order will not exceed its original demand.

[0130] The second layer is the warehouse inventory capacity constraint, which mandates that inventory be taken from the warehouse. Goods transferred The quantity must be less than or equal to the actual physical inventory of the warehouse, and fictitious allocations from warehouses with zero inventory are prohibited.

[0131] The third layer is the warehouse processing capacity constraint. Considering that the manpower and equipment for packing and sorting in each warehouse are limited, the system limits the maximum number of orders that each warehouse can handle for each type of product. This effectively prevents logistics congestion caused by excessive order stacking in a particular star warehouse.

[0132] Specifically, the inventory allocation constraint stipulates that the total quantity of a certain product allocated to an order cannot exceed its demand, expressed as:

[0133] ;

[0134] in, The total quantity of a particular item allocated to an order; For orders Chinese commodities Demand; For the current order set; A collection of goods; This is a collection of warehouse inventory.

[0135] Warehouse inventory capacity constraint stipulates that the total amount of goods allocated from a warehouse cannot exceed its existing inventory, expressed as:

[0136] ;

[0137] ;

[0138] in, For warehouse Chinese commodities Inventory levels; Represents an individual, with a value of 1 or 0. A value of 1 indicates an order. Goods in From warehouse Satisfy; if A value of 0 indicates an order. Goods in Not by warehouse satisfy.

[0139] Warehouse processing capacity constraint, which limits the cumulative value of decision variables in the order dimension, sets an upper limit on the number of orders each warehouse can process for each product. The expression is:

[0140] ;

[0141] in, The maximum number of orders that can be processed for each product in each warehouse.

[0142] S3 uses a heuristic approach based on priority and inventory feasibility to generate the initial population.

[0143] After the model is built, an initial population is generated using a heuristic method based on priority and inventory feasibility. This step is crucial for improving the algorithm's convergence speed.

[0144] Specifically, a priority rule is first established: the demand for goods is prioritized according to the order arrival time, with the earlier the order arrives, the higher the priority; in each group of goods, starting with the high-priority demand, the inventory is matched from each warehouse in turn until the inventory of the goods is used up, or the demand for goods in all orders has been met; when the warehouse inventory is sufficient, the supply of goods for the current order is skipped with a probability set by the perturbation factor, and the next priority order is processed directly, which increases the diversity of the population and avoids local optima.

[0145] like Figure 2As shown, the products are divided into three groups by type: Group 1, Product A; Group 2, Product B; Group 3, Product C. Order requests within each group are prioritized based on arrival time / business rules.

[0146] Group 1: Blue (High) → Orange → Green (Low);

[0147] Group 2: Blue (high) → Orange → Green (low);

[0148] Group 3: Orange (high) → Blue → Yellow (low).

[0149] Group 1 (Product A) violated the priority rule because lower priority requests were met while higher priority requests were not, thus violating the rule of "higher priority orders are allocated first".

[0150] Group 2 (Product B) did not violate the priority rule because Group 2 uses the highest priority requirement, which is satisfied first, in accordance with the business rule of "highest priority is allocated first".

[0151] Group 3 (product C) violated priority because the low-priority demand of Group 3 was met while the higher-priority medium-priority demand was not met, thus disrupting the priority order.

[0152] like Figure 3 As shown, based on the input current order set Warehouse inventory collection By commodity Grouping, resulting in product groups Then, based on order arrival time / priority rules... Calculate product priority; orders that arrive earlier have higher priority. Then proceed with the initialization process.

[0153] The initialization process will first attempt to retrieve data from repository number 1 (…). The system begins providing inventory. If warehouse 1 runs out of inventory, it automatically moves to warehouse 2, and so on, until demand is met or the entire network's inventory is depleted. To overcome the local optima limitation caused by this greedy allocation, this invention introduces a perturbation factor during initialization. If we assume its value range is 0.01~0.05, preferably 0.025, this means that during initialization, even if a warehouse has sufficient inventory, the system still has a 2.5% probability of deliberately skipping the allocation of that order, leaving the opportunity for subsequent orders or subsequent evolutionary processes, thereby significantly enhancing the diversity of the initial population.

[0154] Then, a three-dimensional binary matrix is ​​used to analyze the individuals. Encoding is performed, meaning each individual corresponds to a three-dimensional binary matrix. This is used to represent the allocation relationship between orders, goods, and warehouses. The dimension is the product of the number of orders, the number of product types, and the number of warehouses. .

[0155] In a three-dimensional binary matrix, if the coordinates A value of 1 at this location indicates an order. Goods in Determined by warehouse Provide physical supply; if the value is 0, it means that the warehouse does not participate in the fulfillment of the goods under this order.

[0156] This encoding method transforms the complex discrete allocation problem into a binary search space that is easy for computers to process.

[0157] Next, following the feasibility principle and considering relevant constraints, heuristic rules are used. The expression for generating the initial population is as follows:

[0158] ;

[0159] in, For the current order set; A collection of goods; For warehouse inventory collection; This is a priority rule used to determine the order in which goods are allocated; For warehouse inventory feasibility, this is used to constrain the allocated quantity from not exceeding the existing warehouse inventory. This is the disturbance factor.

[0160] Until the product The inventory in all warehouses has been exhausted, that is... Or all orders for goods All of our needs have been met, that is... Thus, the initial population was obtained.

[0161] S4, based on a multi-objective optimization algorithm with dynamic computing power allocation, iteratively optimizes the initial population to obtain a Pareto non-dominated solution set; the multi-objective optimization algorithm with dynamic computing power allocation adopts a dynamic computing power allocation strategy, which allocates computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation.

[0162] When traditional optimization algorithms deal with high-dimensional three-dimensional matrix decision spaces, the search space is extremely large, which often leads to a serious "search clustering" phenomenon in certain regions. That is, a large amount of computational resources are wasted on solution regions that have been fully explored, while potential optimal solution regions are left unexplored.

[0163] To address this issue, the present invention implements a dynamic computing power allocation strategy.

[0164] like Figure 4 As shown, the multi-objective optimization algorithm based on dynamic computing power allocation (MOEA / D-DDEA, decomposition-type multi-objective optimization algorithm + decision density evaluation allocation strategy) first inputs the current order set. Warehouse inventory collection By commodity Grouping, resulting in product groups .

[0165] Then based on order arrival time / priority rules Product priority is calculated, and the order priority within the group is determined to provide a business priority basis for subsequent heuristic initialization and constraint satisfaction.

[0166] The initial population is then iteratively optimized, and the process specifically includes initialization, decision space partitioning, decision density evaluation and sub-problem computing power allocation, offspring generation, overall update and termination conditions.

[0167] (1) In the initialization phase, the multi-objective problem is decomposed into Each sub-problem A uniformly distributed weight vector is associated with each weight vector; wherein the weight vector is generated by uniform sampling; for example, the weight vector can be uniformly generated by Latin hypercube sampling, covering the entire Pareto front.

[0168] For each subproblem The nearest N weight vectors are selected based on the Euclidean distance between the weight vectors to form a neighbor set. ;

[0169] For each subproblem generate Given 0 initial decisions, each corresponding to a three-dimensional binary matrix, the total number of decisions is . .

[0170] (2) In the decision space partitioning stage, all initial decisions constitute the decision space, and the decision space is divided into K mutually exclusive sub-regions.

[0171] Each decision in the decision space is expanded into a binary sequence row by row, and the binary sequence is converted into a corresponding decimal integer value, expressed as:

[0172] ;

[0173] in, The decision is flattened and converted to a decimal integer value; To flatten the decision into a string of binary numbers, the first... The value at each position; For the current order set; A collection of goods; This is a collection of warehouse inventory.

[0174] Modulo operation is performed on the decimal integer value to map each decision to a unique sub-region among K mutually exclusive sub-regions, thus completing the division of the decision space into sub-regions. This simplifies the high-dimensional three-dimensional binary matrix into low-dimensional sub-region index numbers, as expressed in the following expression:

[0175] ;

[0176] in, For decision making Sub-region index number; The total number of sub-regions to divide the decision space; For the remainder operation.

[0177] This step provides a spatial partitioning basis for subsequent decision density evaluation and computing power allocation.

[0178] (3) In the decision density evaluation stage, traverse For each decision, the number of decisions located in each sub-region is counted, the decision density of each sub-region is calculated, and the Shannon entropy index is introduced as a measure of the uniformity of the solution set distribution in the decision space to establish a decision density evaluation model, the expression of which is:

[0179] ;

[0180] ;

[0181] ;

[0182] in, For the first Number of decisions per sub-region; For decision-making; For decision set; For the first Decision density of each sub-region; For the uniformity of the decision distribution, i.e., Shannon entropy, when At that time, the distribution was completely uniform.

[0183] Then, for each subproblem, the maximum density value in the set of regions where the decision is made in the population is calculated as the decision density weight, expressed as:

[0184] ;

[0185] ;

[0186] in, For subproblems The set of subregions covered by the population; For subproblems The corresponding population; For the first Each decision space sub-region; For subproblems Decision density weights; It indicates that it exists.

[0187] (4) In the subproblem computing power allocation stage, the decision density weights are first normalized to obtain the selection probability of the subproblem, expressed as:

[0188] ;

[0189] in, For subproblems The probability of choosing; For subproblems Decision density weights;

[0190] Then, based on the selection probability, the number of times dynamic computing power is allocated for each subproblem is calculated, as expressed by:

[0191] ;

[0192] in, For subproblems The number of times dynamic computing power is allocated; This represents the total computing power per generation, which is the total number of times an individual in the population exchanges information with its neighbors.

[0193] (5) In the offspring generation stage: according to the number of times the dynamic computing power is allocated, each subproblem selects parent individuals from its neighbor set to perform evolutionary operations and generate offspring decisions, i.e., Pareto non-dominated solution sets.

[0194] (6) Termination condition: When performing evolutionary operations on each subproblem, from its neighbor set Two parent individuals are selected for crossover and mutation until the termination condition is met.

[0195] The termination conditions can be set in two dimensions: first, the decision distribution uniformity tends to be uniform, that is, the decision distribution uniformity reaches the preset saturation threshold; second, the preset maximum number of generations is reached.

[0196] (7) Overall update phase: After each subproblem completes the current evolutionary operation and generates the offspring decision, the offspring decision is compared with the decision in the current population. If the offspring decision is better, the corresponding decision in the original population is replaced to update the population.

[0197] Once the iteration stops, the algorithm will output a Pareto non-dominated solution set. This solution set represents a series of optimal solutions that achieve different balances between profit, satisfaction rate, inventory utilization, and priority.

[0198] S5. Select the final order fulfillment scheme based on the Pareto non-dominated solution set.

[0199] When selecting a final order fulfillment plan from the Pareto non-dominated solution set based on the decision-maker's preferences, the comprehensive objective function value can be calculated for each solution in the solution set, and its expression is as follows:

[0200] ；

[0201] in, The value of the comprehensive objective function; , , These are the weighting coefficients; This is the penalty coefficient; Profit margin; For order fulfillment rate; For inventory utilization rate; The degree of violation of order priority.

[0202] Based on the comprehensive objective function value, the optimal solution is selected as the final order fulfillment plan, and the corresponding order, product, and warehouse allocation plan is output.

[0203] For example, during major sales events like "Double Eleven," businesses may place more emphasis on order fulfillment rates. Then you can adjust it to higher. During periods of low sales volume, companies may prioritize profits and thus adjust prices accordingly. The system is based on the weighted... The system automatically selects the solution with the highest score as the final execution plan. This plan clearly records which warehouse will ship each item for each order. These instructions are then sent to the WMS system to execute physical picking and outbound processing.

[0204] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0205] This invention effectively solves technical challenges in supply chain order fulfillment, such as multi-objective conflicts, search space explosion, and unfair allocation, by constructing a refined four-objective optimization model and innovatively introducing a dynamic computing power allocation mechanism based on decision space sub-region partitioning. This method not only improves enterprise operational efficiency but also ensures fairness in order processing through technical means, providing an efficient and reliable decision support tool for modern smart logistics and digital supply chains. In practical applications, this invention can be widely used in complex allocation scenarios with limited resources, such as e-commerce logistics, pharmaceutical delivery, and parts allocation in manufacturing. Through continuously iteratively optimized Pareto solutions, enterprises can flexibly respond to rapidly changing market demands and achieve globally optimal output with limited inventory resources.

[0206] Example 2

[0207] like Figure 5 As shown, the second embodiment of the present invention also provides a multi-product order fulfillment device based on dynamic computing power allocation, comprising:

[0208] The information acquisition module is used to acquire product order and inventory information, including the current order set, product set, and warehouse inventory set;

[0209] The modeling module is used to establish a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority based on the actual needs of fulfilling multi-product orders in order and inventory information.

[0210] The population generation module is used to generate an initial population using a heuristic method based on priority and inventory feasibility.

[0211] The iterative module is used to iteratively optimize the initial population using a multi-objective optimization algorithm based on dynamic computing power allocation to obtain a Pareto non-dominated solution set. The multi-objective optimization algorithm based on dynamic computing power allocation adopts a dynamic computing power allocation strategy, which allocates computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation.

[0212] The decision module is used to select the final order fulfillment plan based on the Pareto non-dominated solution set.

[0213] These modules do not exist in isolation, but interact through a high-speed data bus. The information acquisition module is responsible for converting external sensor data and database records into tensor formats that the algorithm can recognize; the modeling module constructs a computational graph of the objective function in memory; the population generation module fills the initial video memory using a random number generator and heuristic logic; the iteration module utilizes the processor's parallel computing capabilities to perform large-scale matrix operations and density evaluations; and the decision-making module receives human instructions through an interactive interface to finalize the solution.

[0214] Example 3

[0215] The third embodiment of the present invention also provides a multi-product order fulfillment device based on dynamic computing power allocation, which includes a memory and a processor. The memory stores a computer program, which can be executed by the processor to realize the multi-product order fulfillment method based on dynamic computing power allocation as described above.

[0216] Example 4

[0217] The fourth embodiment of the present invention also provides a computer-readable storage medium storing computer-readable instructions. When the computer-readable instructions are executed by the processor of the device where the computer-readable storage medium is located, the computer-readable instructions implement the multi-product order fulfillment method based on dynamic computing power allocation as described above.

[0218] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-commodity order fulfillment method based on dynamic computing power allocation, characterized in that, include: S1, retrieve product order and inventory information, including the current order set, product set, and warehouse inventory set; S2, based on the actual needs of fulfilling multi-product orders in order and inventory information, establishes a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority; S3 uses a heuristic method based on priority and inventory feasibility to generate the initial population; S4, the initial population is iteratively optimized using a multi-objective optimization algorithm based on dynamic computing power allocation to obtain a Pareto non-dominated solution set; the multi-objective optimization algorithm based on dynamic computing power allocation employs a dynamic computing power allocation strategy, allocating computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation; the specific process of iteratively optimizing the initial population using the multi-objective optimization algorithm based on dynamic computing power allocation is as follows: First, the multi-objective problem is decomposed into sub-problems, each of which corresponds to an associated uniformly distributed weight vector; wherein the weight vector is generated by uniform sampling; For each sub-problem , the nearest N weight vectors are selected according to the Euclidean distance of the weight vectors to form a neighbor set ; For each sub-problem generate initial decisions, each decision corresponding to a three-dimensional binary matrix, the total number of decisions being ; All initial decisions constitute the decision space, which is then divided into K mutually exclusive sub-regions. Each decision in the decision space is expanded into a binary sequence row by row, and the binary sequence is converted into a corresponding decimal integer value, expressed as: ； wherein, is the decision flattened and converted to a decimal integer value; is the value in the position after the decision has been flattened into a string of binary digits; is the current order collection; is the item collection; is the warehouse inventory collection; Modulo operation is performed on the decimal integer value to map each decision to a unique sub-region among K mutually exclusive sub-regions, thus completing the division of the decision space into sub-regions. This simplifies the high-dimensional three-dimensional binary matrix into low-dimensional sub-region index numbers, as expressed in the following expression: ； wherein, for decision the sub-region index number; for decision space division sub-region total number; for remainder operation; Then iterate The number of decisions in each sub-region is counted, the decision density of each sub-region is calculated, and the Shannon entropy index is introduced as a measure of the uniformity of the distribution of the solution set in the decision space to establish a decision density evaluation model, which is expressed as: ； ； ； wherein, is the number of decisions for the th sub-region; is the decision; is the set of decisions; is the decision density for the th sub-region; is the decision distribution uniformity, i.e. Shannon entropy, when the distribution is perfectly uniform. Then, for each subproblem, the maximum density value in the set of regions where the decision is made in the population is calculated as the decision density weight, expressed as: ； ； wherein, is a set of sub-regions covered by the population of sub-problems ; is a corresponding population of sub-problems ; is the th decision space sub-region; is a decision density weight for the sub-problem ; denotes existence; Normalizing the decision density weights yields the selection probabilities of the subproblems, expressed as: ； wherein, is the selection probability for the sub-problem ; is the decision density weight for the sub-problem ; Based on the selection probability, the number of times dynamic computing power is allocated for each subproblem is calculated, expressed as: ； wherein, is the dynamic number of times of work distribution for the sub-problems is the dynamic number of times of work distribution for the sub-problems is the total number of times of work distribution for each generation, i.e. the total number of times of information exchange between population individuals and neighbors Based on the number of times the dynamic computing power is allocated, each subproblem selects parent individuals from its neighbor set for evolutionary operations to generate offspring decisions, i.e., the Pareto non-dominated solution set; S5. Select the final order fulfillment scheme based on the Pareto non-dominated solution set.

2. The multi-commodity order fulfillment method based on dynamic computing power distribution according to claim 1, characterized in that The profit rate is the ratio of the actual total profit allocated to fulfilling orders to the theoretical maximum total profit, expressed as: ； wherein, is the profit margin; is the order of the goods whether the demand for the goods in the order is fully met; is the profit of the goods in the order is the current set of orders; is the set of goods; denotes taking the maximum value; The order fulfillment rate is the ratio of the number of fully fulfilled orders to the total number of orders, expressed as: ； wherein, is the order fill rate; the value represents the total number of items for all orders; is a minimum value; The inventory utilization rate is the ratio of the number of goods allocated to an order to the total available inventory, expressed as: ； in, For inventory utilization rate; For orders Chinese commodities Demand; For warehouse Chinese commodities Inventory levels; For warehouse inventory collection; The severity of order priority violation is calculated in a tiered manner based on the product priority number, the identifier indicating whether the product meets the priority requirement, the product arrival time sequence number, and the sequence number of the first unassigned product to meet the priority requirement. This is used to measure the penalty for not fulfilling early orders, and the expression is as follows: ； ； wherein is the order priority violation degree; is the theoretical maximum penalty value; is the theoretical minimum penalty value; is the penalty value of the group item; denotes taking the minimum value; For priority number of group commodity; Indicates commodity in group commodity whether to meet priority ; Indicates order in group commodity commodity arrival time sequence number; Indicates the first unassigned meeting commodity sequence number in group commodity.

3. The multi-product order fulfillment method based on dynamic computing power allocation according to claim 1, characterized in that... The multi-objective optimization algorithm model also includes technical constraints, which include: Inventory allocation constraints stipulate that the total quantity of a certain item allocated to an order cannot exceed its demand. The expression is: ； in, To be assigned to an order A certain product The total amount; For orders Chinese commodities Demand; For the current order set; A collection of goods; For warehouse inventory collection; Warehouse inventory capacity constraint stipulates that the total amount of goods allocated from a warehouse cannot exceed its existing inventory, expressed as: ； ； in, For warehouse Chinese commodities Inventory levels; Represents an individual, with a value of 1 or 0. A value of 1 indicates an order. Goods in From warehouse Satisfy; if A value of 0 indicates an order. Goods in Not by warehouse satisfy; Warehouse processing capacity constraint, which limits the cumulative value of decision variables in the order dimension, sets an upper limit on the number of orders each warehouse can process for each product. The expression is: ； in, The maximum number of orders that can be processed for each product in each warehouse.

4. The multi-product order fulfillment method based on dynamic computing power allocation according to claim 3, characterized in that... The specific steps for generating the initial population are as follows: First, establish priority rules: prioritize product demand according to order arrival time, with earlier orders having higher priority; within each product group, start with the highest priority demand and try to match inventory from each warehouse in turn until the product inventory is used up or the product demand in all orders has been met; if warehouse inventory is sufficient, skip the product supply for the current order with a probability set by the perturbation factor and directly process the next priority order to increase the diversity of the population and avoid local optima; Then, a three-dimensional binary matrix is ​​used to analyze the individuals. Encoding is performed, meaning each individual corresponds to a three-dimensional binary matrix. This is used to represent the allocation relationship between orders, goods, and warehouses. The dimension is the product of the number of orders, the number of product types, and the number of warehouses. ; Next, following the feasibility principle, heuristic rules are used. The expression for generating the initial population is as follows: ； in, For the current order set; A collection of goods; For warehouse inventory collection; This is a priority rule used to determine the order in which goods are allocated; For warehouse inventory feasibility, this is used to constrain the allocated quantity from not exceeding the existing warehouse inventory. For disturbance factors; The initial population is obtained based on the individuals in the initial population.

5. A multi-product order fulfillment method based on dynamic computing power allocation according to claim 1, characterized in that... When performing evolutionary operations on each subproblem, two parent individuals are selected from its neighbor set for crossover and mutation operations until the decision distribution becomes uniform or the preset maximum number of generations is reached.

6. The multi-product order fulfillment method based on dynamic computing power allocation according to claim 1, characterized in that... It also includes: after each subproblem completes the current evolutionary operation and generates the offspring decision, the offspring decision is compared with the decision in the current population. If the offspring decision is better, the corresponding decision in the original population is replaced to update the population.

7. A multi-product order fulfillment method based on dynamic computing power allocation according to claim 2, characterized in that... When selecting the final order fulfillment plan based on the Pareto non-dominated solution set, the comprehensive objective function value is calculated for each solution in the solution set, and its expression is: ； in, The value of the comprehensive objective function; , , These are the weighting coefficients; This is the penalty coefficient; Profit margin; For order fulfillment rate; For inventory utilization rate; The degree of violation of order priority; Based on the comprehensive objective function value, the optimal solution is selected as the final order fulfillment plan, and the corresponding order, product, and warehouse allocation plan is output.

8. A multi-product order fulfillment device based on dynamic computing power allocation, used to implement the multi-product order fulfillment method based on dynamic computing power allocation as described in any one of claims 1-7, characterized in that... include: The information acquisition module is used to acquire product order and inventory information, including the current order set, product set, and warehouse inventory set; The modeling module is used to establish a multi-objective optimization algorithm model that includes profit margin, order fulfillment rate, inventory utilization rate, and fulfillment plan violation of order priority based on the actual needs of fulfilling multi-product orders in order and inventory information. The population generation module is used to generate an initial population using a heuristic method based on priority and inventory feasibility. The iterative module is used to iteratively optimize the initial population using a multi-objective optimization algorithm based on dynamic computing power allocation to obtain a Pareto non-dominated solution set. The multi-objective optimization algorithm based on dynamic computing power allocation adopts a dynamic computing power allocation strategy, which allocates computing resources to each sub-problem through decision space sub-region partitioning and decision density evaluation. The decision module is used to select the final order fulfillment plan based on the Pareto non-dominated solution set.

9. A multi-commodity order fulfillment device based on dynamic computing power allocation, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that can be executed by the processor to implement a multi-product order fulfillment method based on dynamic computing power allocation as described in any one of claims 1-7.

Images