Multi-center warehouse joint replenishment and transportation decision method and service platform based on bi-ga

Abstract

The application relates to a multi-center warehouse joint replenishment and transportation decision method and a service platform based on Bi-GA, which comprises the following steps: determining the current demand of a plurality of target products in a plurality of execution cycles of a decision stage after the current stage of a plurality of center warehouses; encoding and initializing the center warehouses, the execution cycles, the target products and the current demand to generate a plurality of multi-dimensional encoding bodies, wherein the multi-dimensional encoding bodies comprise a plurality of replenishment decision sub-encoding bodies, and the replenishment decision sub-encoding bodies comprise a plurality of replenishment decision encoding sub-bodies for representing replenishment decisions of the corresponding center warehouses on the target products in each execution cycle; using a preset double-layer genetic algorithm to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration based on replenishment decision sub-encoding bodies corresponding to the multi-dimensional encoding bodies until a target multi-dimensional encoding body is generated, and taking all replenishment decisions and all target replenishment paths corresponding to the target multi-dimensional encoding body as a decision result.

Description

Technical Field

[0001] This application relates to the field of smart logistics and warehouse management technology, and in particular to a multi-center warehouse joint replenishment and transportation decision-making method and service platform based on Bi-GA. Background Technology

[0002] The self-operated e-commerce model refers to e-commerce companies achieving closed-loop, end-to-end management of all aspects of product selection, transportation, warehousing, and sales through independent decision-making. To improve delivery efficiency, self-operated e-commerce typically adopts a multi-level distributed warehousing model to achieve rapid response to demand within a region. In related technologies, self-operated e-commerce constructs a procurement-transportation-warehousing system consisting of multiple central warehouses, multiple suppliers, multiple products, and a fleet of transportation vehicles. The core of procurement, transportation, and inventory decisions for self-operated e-commerce lies in replenishment planning. Existing joint replenishment problems (JRPs) are usually based on deterministic demand and fixed replenishment rules. Fixed replenishment rules include fixed cycles or fixed quantities, but fixed-cycle or fixed-quantity replenishment cannot effectively cope with market fluctuations. At the same time, in the replenishment planning decisions of self-operated e-commerce, the replenishment quantity is closely coupled with the transportation routes of the vehicles. Existing replenishment planning, which separates replenishment decisions from route planning, easily leads to inventory imbalance, inventory overflow, and high total system costs.

[0003] Currently, no effective solution has been proposed for the replenishment decisions of self-operated e-commerce in related technologies, which are weak in dealing with dynamic interference and are prone to inventory imbalance and warehouse overload. Summary of the Invention

[0004] This application provides a multi-center warehouse joint replenishment and transportation decision-making method and service platform based on Bi-GA, which at least solves the replenishment decision-making of procurement, transportation and storage in self-operated e-commerce in related technologies, and addresses the problems of weak dynamic interference capability, easy to cause inventory imbalance and warehouse overload.

[0005] In a first aspect, embodiments of this application provide a multi-center warehouse joint replenishment and transportation decision-making method based on Bi-GA, comprising: determining the current demand for multiple target products to be replenished in multiple execution cycles corresponding to a decision stage following the current one, wherein the current demand is the sales volume of goods predicted by a preset sales forecasting model based on the current inventory of the multiple center warehouses, the sales forecasting model being a model trained on PSO-ALSTM, and the multiple execution cycles corresponding to the decision stage following the current one being generated by rolling over the multiple execution cycles corresponding to the previous decision stage by one execution cycle; encoding and initializing the center warehouses, the execution cycles, the target products, and the current demand according to a preset encoding method to generate multiple multi-dimensional encoding bodies, wherein the multi-dimensional encoding bodies include multiple replenishment decision sub-encodings. A replenishment decision sub-encoder is associated with a central warehouse. The replenishment decision sub-encoder includes multiple replenishment decision sub-encoders representing replenishment decisions made by the corresponding central warehouse for various target products in each execution cycle. The replenishment decisions include at least the replenishment quantity of the target products. Using a preset two-layer genetic algorithm, based on the replenishment decision sub-encoders corresponding to the multiple multidimensional encoders, an inner replenishment path planning decision iteration and an outer replenishment decision iteration are performed until a target multidimensional encoder is generated. All replenishment decisions and all target replenishment paths corresponding to the target multidimensional encoder are used as decision results. The target replenishment path is generated by using the inner genetic algorithm of the two-layer genetic algorithm to perform replenishment path planning on the replenishment decision sub-encoders of the target multidimensional encoder.

[0006] Secondly, embodiments of this application provide a service platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method as described in the first aspect.

[0007] Compared to related technologies, the multi-center warehouse joint replenishment and transportation decision-making method based on Bi-GA provided in this application involves determining the current demand for multiple target products to be replenished in multiple execution cycles corresponding to a decision stage after the current one. The central warehouse, the execution cycle, the target product, and the current demand are encoded and initialized according to a preset encoding method to generate multiple multi-dimensional encoding bodies. Each multi-dimensional encoding body includes multiple replenishment decision sub-encoding bodies, and each replenishment decision sub-encoding body is associated with one central warehouse. Each replenishment decision sub-encoding body includes multiple replenishment decision encoding sub-codes representing the replenishment decisions made by the corresponding central warehouse for multiple target products in each execution cycle. The replenishment decision includes at least the replenishment quantity for the target product. A preset two-layer genetic algorithm is used to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment based on the replenishment decision sub-encoding bodies corresponding to the multiple multi-dimensional encoding bodies. The decision-making process iterates until a target multidimensional encoding body is generated. All replenishment decisions and target replenishment paths corresponding to this target multidimensional encoding body are used as the decision results. The target replenishment path is generated by planning replenishment paths for the replenishment decision sub-encoding bodies of the target multidimensional encoding body using the inner layer genetic algorithm of the two-layer genetic algorithm. A mechanism for prediction, replenishment, and transportation decisions is constructed through fusion and collaborative optimization, combining flexible replenishment strategies and heterogeneous fleet routes to jointly optimize replenishment. Simultaneously, the complexity of solving the planning decision results and the expansion of prediction errors are reduced through the two-layer genetic algorithm and rolling time-domain method. This improves the replenishment efficiency and responsiveness of the procurement, transportation, and storage system, adapts to demand fluctuations, and achieves good replenishment and transportation path decision-making effects, reducing operating costs. This solves the problems of weak dynamic interference response, easy inventory imbalance, and warehouse overload in the replenishment decisions of self-operated e-commerce procurement, transportation, and storage systems in related technologies.

[0008] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a hardware structure block diagram of the terminal of the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method according to an embodiment of this application; Figure 2 This is a flowchart of a multi-center warehouse joint replenishment and transportation decision-making method based on Bi-GA according to an embodiment of this application; Figure 3 This is a schematic diagram of the fixed mapping relationship matrix between the target product and the target supplier in an embodiment of this application; Figure 4 This is a schematic diagram of the multidimensional code body of an embodiment of this application; Figure 5 This is a schematic diagram of the replenishment decision sub-code body according to an embodiment of this application; Figure 6 This is a schematic diagram of path segmentation of node sequences according to an embodiment of this application; Figure 7 This is a schematic diagram of the access code and the cross range in an embodiment of this application; Figure 8 This is a schematic diagram of the access code and the empty code in an embodiment of this application; Figure 9 This is a schematic diagram of the access code performing an interleaving operation according to an embodiment of this application; Figure 10 This is a schematic diagram of two first target sub-code bodies in an example of this application; Figure 11 This is a schematic diagram of the first replenishment decision sub-code body generated after product-level cross-operation in an embodiment of this application; Figure 12 This is another schematic diagram of the two second target sub-code bodies in this application example; Figure 13 This is a schematic diagram of the second replenishment decision sub-code body generated after performing a periodic crossover operation in an embodiment of this application; Figure 14 A structural block diagram of a cloud-edge collaborative multi-center warehouse joint replenishment and transportation decision-making device for self-operated e-commerce, according to an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application. Furthermore, it is understood that although the efforts made in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, modifications to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0011] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0012] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "a," "an," "an," "the," and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. "Multiple stages" used in this application refers to two or more stages. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific ordering of objects.

[0013] Before describing the embodiments of this application, the relevant symbols and parameters involved in the embodiments of this application are defined as follows: p represents the p-th target product, p∈P; P represents the target product set, P={1,2,3,…,|P|}; s represents the s-th target supplier, s∈S; S represents the target supplier set, S={1,2,3,…,|S|}; w represents the w-th central warehouse, w∈W; W represents the central warehouse set, W={1,2,3,… |W|};r,s represent vehicle access node indices, r,s∈{w}∪S;k represents the k-th replenishment vehicle, k∈K;K represents the vehicle set;t represents the t-th execution cycle, t∈T;T represents the execution cycle set, T={1,2,3,…,|T|};T1 represents the decision cycle set for replenishment quantity and replenishment path decisions, T1={1,2,3,…,T-1};T2 represents the calculation cycle set for sales behavior, T2={1,2,3,…,T};B ps This represents the correspondence between target products and target suppliers, where p∈P and s∈S. If target product p is provided by target supplier s, then B ps=1, otherwise, B ps =0; Let w represent the predicted demand for target product p in the w-th central warehouse during the t-th execution cycle, where t∈T2, w∈W, p∈P; Let p represent the purchase price of the target product p with a price discount, where p∈P and t∈T1. ;sp p v represents the selling price of the target product p; p Let sc represent the volume of the target product p, where p∈P; w This represents the storage cost per unit volume of product in the w-th central warehouse per single execution cycle; lc represents the maximum capacity of the w-th central warehouse; p d represents the stockout penalty coefficient for target product p in the central warehouse; rs fc represents the distance between vehicle access node r and vehicle access node s, where r, s ∈ {w} ∪ S, r ≠ s; k vc represents the fixed dispatch cost of replenishment vehicle k; k This represents the unit distance travel cost of the replenishment vehicle k; This represents the maximum capacity of the replenishment vehicle k; This represents the replenishment quantity of target product p from the w-th central warehouse in the t-th execution cycle, where p∈P, w∈W, t∈T1; This indicates whether the k-th vehicle departs from the w-th central warehouse in the t-th execution cycle, traveling along the path r,s, where r,s∈{w}∪S, r≠s, w∈W, k∈K. If yes, ,otherwise, ; This indicates whether the central warehouse w is using vehicle k in the t-th execution cycle. If yes, ,otherwise, w∈W,t∈T2,k∈K; This indicates whether the vehicle departing from the central warehouse w in the t-th execution cycle accesses the s-th target supply object. If yes, ,otherwise, ,w∈W,s∈S,t∈T1; Let w represent the remaining inventory of target product p at the end of the t-th execution cycle, where w∈W, p∈P, and t∈T1∩T2; Let w represent the total volume of products replenished by the central warehouse w at the target supply object s in the t-th execution cycle, where w∈W, s∈S, and t∈T1; The total volume of products replenished by replenishment vehicle K in the central warehouse w at the target supply object s during the t-th execution cycle, w∈W, s∈S, t∈T1

[0014] The embodiments of this application are described below: The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. Taking running on a terminal as an example, Figure 1 This is a hardware structure block diagram of the terminal for the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method according to an embodiment of this application. Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. Optionally, the terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0015] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0016] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0017] This embodiment provides a Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method running on the aforementioned terminal. Figure 2This is a flowchart of a multi-center warehouse joint replenishment and transportation decision-making method based on Bi-GA according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps: Step S201: Determine the current demand for various target products to be replenished in multiple execution cycles corresponding to a decision stage following the current one. The current demand is the sales volume of goods predicted by a preset sales forecasting model based on the current inventory of multiple central warehouses. The sales forecasting model is a model trained on PSO-ALSTM. The multiple execution cycles corresponding to a decision stage following the current one are generated by rolling over the multiple execution cycles corresponding to the previous decision stage by one execution cycle.

[0018] In this embodiment, the executing entity of the decision-making method of this application includes, but is not limited to, the control terminal, control system, or cloud platform corresponding to multiple central warehouses managed by the self-operated e-commerce enterprise. In this embodiment, after the executing entity senses that multiple central warehouses have responded to the replenishment and transportation decisions of the first execution cycle of a decision-making stage, it will initiate the execution of replenishment and transportation decisions for multiple execution cycles of the next decision-making stage. That is, the replenishment and transportation decision-making method of this application adopts the approach of replanning the replenishment and transportation decisions for multiple execution cycles after running one execution cycle. At the same time, during the replanning, the demand for replenishment from the target supplier to the central warehouse within multiple execution cycles will be re-predicted. However, since the prediction is made using a prediction model, in order to reduce prediction errors and avoid over-prediction, the following measures are taken: To mitigate large errors, the system uses the available inventory in the central warehouse after it has responded to the replenishment and transportation decisions of the first execution cycle of the previous decision-making phase as input data, and the output sales volume as the corresponding current demand for the central warehouse. In essence, based on the available inventory in the central warehouse and the product demand for replenishment decided upon in the first execution cycle of the previous decision-making phase, the actual demand or outbound volume of the corresponding central warehouse in the first execution cycle of the previous decision-making phase can be determined. By inputting the corresponding outbound volume into the sales forecasting model, the sales volume of goods, which serves as the current demand, can be predicted. After obtaining the corresponding current demand, replenishment and transportation decisions for multiple central warehouses across multiple execution cycles will be made based on this current demand.

[0019] In this embodiment, the allocation volume of each target product in each node warehouse (a warehouse facing the consumer end) is calculated in each execution cycle, that is, the outbound volume of each target product in the central warehouse in each execution cycle (the execution cycle to be executed) is calculated, thereby realizing the conversion of the demand forecast results of each region (node ​​warehouse) into the replenishment volume of the central warehouse. In this embodiment, a sales forecasting model based on PSO-ALSTM is used to predict the sales volume of goods in node warehouses or stores in front of node warehouses in each future execution cycle, and the predicted sales volume of goods is converted into the replenishment demand of the central warehouse through the EOQ model. It should be understood that in the final result of the procurement decision and transportation decision, the final replenishment volume of the target product can be the same as or different from the replenishment volume corresponding to the set demand information, but it will not exceed the total demand.

[0020] In this embodiment, the sales forecasting model uses an adaptive LSTM model constructed with the particle swarm optimization (PSO) method. It is trained using historical sales data from each node warehouse and various feature parameters (e.g., seasonal and transactional attributes of the product) as the training and testing datasets. The sales forecasting model is built using the deep learning framework TensorFlow and its high-level interface Keras. The sales forecasting model is trained to output the demand for the target product in the node warehouse within the corresponding time period in the future, based on the input sales data and feature parameters of the target product.

[0021] Step S202: Encode and initialize the central warehouse, execution cycle, target product, and current demand according to a preset coding method to generate multiple multi-dimensional coding bodies. Each multi-dimensional coding body includes multiple replenishment decision sub-coding bodies. Each replenishment decision sub-coding body is associated with a central warehouse. Each replenishment decision sub-coding body includes multiple replenishment decision codes that represent the replenishment decisions made by the corresponding central warehouse for various target products in each execution cycle. The replenishment decision includes at least the replenishment quantity of the target product.

[0022] In this embodiment, after determining the current demand for each central warehouse, the corresponding encoding is performed based on the current demand, and the population is initialized to generate multiple multi-dimensional encoding bodies. This enables the use of a two-layer genetic algorithm for inner-layer replenishment path planning and outer-layer replenishment decision-making. In this embodiment, a replenishment decision sub-encoding body represents a replenishment decision scheme for a central warehouse. A replenishment decision sub-encoding body includes multiple replenishment decision encoding sub-sub ... When encoding replenishment quantity, the value of the replenishment quantity code is determined based on a range of values ​​defined by the current demand for the target product. In this embodiment, one product code is associated with one target supplier. In this embodiment, the multidimensional coding body is organized in matrix form as a three-dimensional structure of central warehouse (W) × target product (P) × execution cycle (T). A central warehouse code has a two-dimensional structure of target product (P) × execution cycle (T), with the replenishment quantity code as the core gene value. This gene value is a non-negative integer, representing the replenishment quantity of the target product for a specific central warehouse in a specific execution cycle (e.g., the first execution cycle). Before encoding the product code, a fixed mapping relationship matrix between the target product and the target supplier is established, referencing... Figure 3 As shown, when the replenishment quantity of any target product provided to a target supplier is greater than zero within a certain execution cycle, an access request to that target supplier within that execution cycle is triggered. This organically connects replenishment decisions with routing decisions. Under the premise of satisfying vehicle capacity constraints, multiple initial multidimensional codes are randomly generated, for example, containing 3 central warehouses, 5 target products, and 2 target suppliers (the corresponding mapping relationship is referenced). Figure 3 Taking 7 execution cycles as an example, a multidimensional encoding body is as follows: Figure 4 As shown, one of the replenishment decision sub-code bodies is as follows: Figure 5 As shown; in this embodiment, Figure 4 and Figure 5 A replenishment decision code consists of a target product column and a corresponding replenishment quantity code within the execution cycle. It's important to note that during encoding, the target product column is presented as a single column; however, during decoding or processing, it's presented as a replenishment decision code consisting of a target product column and a pickup quantity column. It's crucial to understand that each target product is associated with a corresponding target supplier; that is, once the target product number is determined, the corresponding target supplier is also determined (see reference). Figure 3 (As shown).

[0023] Step S203: Using a preset two-layer genetic algorithm, based on replenishment decision sub-code bodies corresponding to multiple multidimensional code bodies, perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration until a target multidimensional code body is generated. All replenishment decisions and all target replenishment paths corresponding to the target multidimensional code body are taken as decision results. The target replenishment path is generated by using the inner-layer genetic algorithm of the two-layer genetic algorithm to plan replenishment paths for the replenishment decision sub-code bodies of the target multidimensional code body.

[0024] In this embodiment, the outer layer of the two-layer genetic algorithm is responsible for determining the replenishment quantity for each target product, each central warehouse, and each execution cycle. It uses real-number encoding, where each corresponding code represents a complete replenishment plan for multiple central warehouses, and a replenishment decision sub-code represents a complete replenishment plan for one central warehouse. In this embodiment, for the outer layer's replenishment decision, a tournament selection method is used for parent selection, a two-point crossover operation is used to recombine the replenishment decision, and population diversity is maintained through vehicle capacity-constrained picking quantity encoding mutation. The fitness evaluation of the outer layer's genetic evolution operation integrates replenishment costs and transportation costs from the inner layer. In this embodiment, the fitness function... The fitness factor is to maximize revenue. The parameters affecting fitness include the sales revenue generated after the sale of the target product in the central warehouse, the cost of purchasing the target product in that demand quantity, the central warehouse storage cost, and the cost of stockouts caused by stockouts of the target product. Multiple parameters are calculated according to a preset fitness function to determine the fitness of a multi-dimensional encoding body. In this embodiment, the inner layer GA of the two-layer genetic algorithm is the replenishment path planning decision layer: it solves the capacity-constrained vehicle routing problem (CVRP) for the replenishment plan given by the outer layer. It uses integer encoding, and the encodings of the target supply objects associated with the target product of each replenishment decision encoding are arranged into a corresponding path node sequence, referencing... Figure 6 Assuming that in the current replenishment scheme, five target supply objects need to be accessed in a certain execution cycle, all of which are treated as corresponding nodes, and all nodes are arranged as integers, after the genetic operation, vehicles are allocated according to the node codes based on the constraints. Figure 6 It is known that two replenishment vehicles were allocated, thus forming two sub-paths that meet the maximum vehicle capacity constraint. The replenishment code corresponding to a multidimensional code also applies tournament selection, two-point crossover, and site mutation to optimize transportation costs under capacity and path constraints. In this embodiment, a feedback mechanism is also adopted, that is, the inner GA returns the optimized transportation cost to the outer layer, and the outer layer updates the overall fitness evaluation accordingly. This iterative process continues until the termination condition is met, thereby ensuring that replenishment and path decision are jointly optimized rather than processed in isolation.

[0025] Through steps S201 to S203, the current demand for various target products to be replenished is determined within multiple execution cycles corresponding to a decision stage following the current one. The current demand is predicted based on the current inventory levels of the multiple central warehouses using a pre-defined sales forecasting model. This sales forecasting model is based on a PSO-ALSTM-trained model. The multiple execution cycles corresponding to the decision stage following the current one are generated by rolling over the execution cycles corresponding to the previous decision stage by one cycle. The central warehouses, execution cycles, target products, and current demand are encoded and initialized according to a pre-defined encoding method, generating multiple multi-dimensional encoding bodies. Each multi-dimensional encoding body includes multiple replenishment decision sub-encoding bodies, with each replenishment decision sub-encoding body associated with a central warehouse. Each replenishment decision sub-encoding body includes multiple replenishment decision encoding sub-encoding ... This approach involves iterating through inner-layer replenishment path planning and outer-layer replenishment decisions for multiple multidimensional encoding bodies corresponding to replenishment decision sub-encoding bodies, until a target multidimensional encoding body is generated. All replenishment decisions and target replenishment paths corresponding to the target multidimensional encoding body are then used as the decision results. The target replenishment path is generated by planning replenishment paths for the replenishment decision sub-encoding bodies of the target multidimensional encoding body using the inner-layer genetic algorithm of a two-layer genetic algorithm. The approach integrates and collaboratively optimizes prediction, replenishment, and transportation decisions, constructing a prediction, decision, execution, and update mechanism. It combines flexible replenishment strategies and heterogeneous fleet routes to jointly optimize replenishment. Simultaneously, the two-layer genetic algorithm and rolling time-domain method reduce the complexity of solving the planning decision results and suppress the expansion of prediction errors, thereby improving the replenishment efficiency and responsiveness of the procurement, transportation, and storage system, adapting to demand fluctuations, and achieving good replenishment and transportation path decision-making effects. This reduces operating costs and solves the problems of weak dynamic interference response, easy inventory imbalance, and warehouse overload in the replenishment decision-making of self-operated e-commerce procurement, transportation, and storage systems in related technologies.

[0026] It should be noted that the embodiments of this application construct an integrated "prediction-decision-execution-update" solution to balance prediction accuracy, computational efficiency and system responsiveness; and manage computational complexity and adapt to demand fluctuations by designing a two-layer genetic algorithm and a rolling time-domain method.

[0027] In some embodiments, a pre-defined two-layer genetic algorithm is used to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration based on replenishment decision sub-encoders corresponding to multiple multi-dimensional encoders, including the following steps: Step 21: Using the inner layer genetic algorithm of the two-layer genetic algorithm, replenishment path planning is performed on all replenishment decision codes corresponding to the multiple multidimensional codes participating in the current replenishment decision iteration, so as to obtain the target transportation path corresponding to each replenishment decision code.

[0028] In this embodiment, the inner layer of the two-layer genetic algorithm makes replenishment path decisions based on the replenishment plan given by the outer layer (corresponding to all replenishment decision codes). After receiving the replenishment plan from the outer layer, the inner layer genetic algorithm transforms it into the target supply object access requirements for each execution cycle, and then solves for the optimal replenishment path under the replenishment plan. Considering the independence of transportation resources in each central warehouse, a parallel optimization strategy is adopted to process the path decisions of different central warehouses. The access code body and the code of the access code in the replenishment path decision are represented by an integer sequence, which corresponds to the access order of the node codes. Each node code in the node sequence corresponds to a target supply object. The complete sub-transport path starts from the central warehouse, visits each node code in sequence, and returns to the central warehouse. In this embodiment, the codes of the target supply objects associated with the target products of each replenishment decision code are arranged into a corresponding node sequence, referring to... Figure 6 Assuming that in the current replenishment scheme, five target supply objects need to be accessed in a certain execution cycle, each acting as a corresponding node, and all node codes are arranged into integers, after the genetic operation, vehicles are allocated based on the node codes according to constraints. Figure 6 It can be seen that a total of 2 replenishment vehicles were allocated, thus forming two sub-paths (0-3-2-0 and 0-5-4-1-0) that satisfy the maximum vehicle capacity constraint.

[0029] Step 22: After determining the fitness of each multidimensional code based on the sub-fitness of each replenishment decision sub-code, the outer genetic algorithm of the two-layer genetic algorithm is used to perform at least one set outer genetic evolution operation on the multiple multidimensional codes participating in the current iteration, generating multiple multidimensional codes that have completed the current replenishment decision iteration. The sub-fitness is determined based on the replenishment transportation cost of all target transportation paths corresponding to the replenishment decision sub-code and the revenue parameter corresponding to the replenishment decision sub-code.

[0030] In this embodiment, for the replenishment decision of the outer layer, a tournament selection method is used for parent selection, and a two-point crossover operation is used to reorganize the replenishment decision. Population diversity is maintained by the voyage quantity encoding variation based on vehicle capacity constraints. The fitness evaluation of the outer layer's genetic evolution operation integrates replenishment costs and transportation costs from the inner layer. The fitness of the outer layer is to maximize revenue. The parameters affecting fitness include the sales revenue generated after the sale of the target product in the central warehouse, the cost of purchasing the target product in the required quantity, the central warehouse storage cost, and the stockout cost caused by the stockout of the target product. For multiple parameters, a preset fitness function is used to calculate the fitness corresponding to a multidimensional encoding body.

[0031] Step 23: Based on the multiple multidimensional codes that have completed the current replenishment decision iteration, repeatedly execute the steps of using the corresponding inner genetic algorithm to plan and generate the corresponding target transportation path, determine the fitness of the corresponding multidimensional code, and use the corresponding outer genetic algorithm to perform the set outer genetic evolution operation, until multiple multidimensional codes with fitness greater than the fitness threshold are generated.

[0032] Step 24: Among the multiple multidimensional code bodies with fitness greater than the fitness threshold, select the multidimensional code body with the highest fitness to obtain the target multidimensional code body.

[0033] Through steps 21 to 24 above, a two-layer genetic algorithm is used to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration based on the multiple encoded multidimensional codes. In some embodiments, the inner layer of a two-layer genetic algorithm is used to plan replenishment paths for all replenishment decision codecs corresponding to multiple multidimensional codecs participating in the current replenishment decision iteration. This is achieved through the following steps: Step 31: Obtain the replenishment decision sub-code body corresponding to each multi-dimensional code body participating in the current replenishment decision iteration, and determine all replenishment decision codes for each replenishment decision sub-code body. The replenishment decision code includes decision codes for multiple product codes and replenishment quantity codes. The replenishment quantity code is used to represent the replenishment quantity of a target product in an execution cycle. One product code is associated with one target supply object.

[0034] Step 32: For each replenishment decision code, the target supply objects associated with all product codes are used as node codes to generate a node sequence. Each node sequence is used as an access code to generate an access sub-code body corresponding to the replenishment decision code body, and a replenishment code body including multiple access sub-code bodies is generated. Each replenishment code body corresponds to a multi-dimensional code body. Based on vehicle capacity constraints, the node sequence corresponding to the access code body is path segmented to generate at least one sub-transport path. The candidate transport path corresponding to the access code body includes at least one sub-transport path. The replenishment quantity for node replenishment corresponding to each node code is determined based on the replenishment quantity codes corresponding to all product codes associated with a target supply object.

[0035] In this embodiment, the access sub-code body and the encoding of the access code for the replenishment path decision are represented by an integer sequence, corresponding to the access order of the node codes. Each node code in the node sequence corresponds to a target supply object. The complete sub-transport path starts from the central warehouse, visits each node code sequentially, and then returns to the central warehouse. In this embodiment, the codes of the target supply objects associated with the target products of each replenishment decision code are arranged into a corresponding node sequence, referring to... Figure 6 Assuming that in the current replenishment scheme, five target supply objects need to be accessed in a certain execution cycle, each acting as a corresponding node, and all node codes are arranged into integers, after the genetic operation, vehicles are allocated based on the node codes according to constraints. Figure 6 It can be seen that a total of 2 replenishment vehicles were allocated, thus forming two sub-paths (0-3-2-0 and 0-5-4-1-0) that satisfy the maximum vehicle capacity constraint.

[0036] Step 33: After determining the replenishment cost parameters corresponding to the candidate transportation paths based on the sub-replenishment cost parameters corresponding to the sub-transportation paths, the decision fitness corresponding to the replenishment code body is determined based on the replenishment cost parameters of the access sub-code bodies. Based on the decision fitness, the inner-layer genetic algorithm is used to iterate the inner-layer genetic evolution operation of the access sub-code bodies of multiple replenishment codes until at least one candidate replenishment code body with a decision fitness less than the set fitness value is generated. The inner-layer genetic evolution operation includes the following operations: tournament selection operation, two-point crossover operation, and exchange mutation operation.

[0037] In this embodiment, the fitness function calculation method of the inner layer genetic algorithm is to calculate the transportation cost under the corresponding sub-transportation path, including fixed transportation cost and variable transportation cost.

[0038] In some alternative implementations, the inner-layer genetic evolution operation for the access sub-codemas of the multiple replenishment codes in step 33 includes the following steps: Step 331: After selecting two replenishment code bodies from multiple replenishment code bodies as parent code bodies, select two access sub-code bodies located at the same sorting position from all access sub-code bodies of the two parent code bodies, and randomly select at least one access sub-code body with the same sorting position from all access sub-code bodies of the two corresponding access sub-code bodies, to obtain at least one set of parent code sub-groups, and generate an empty code sub-group corresponding to the corresponding parent code sub-group.

[0039] Step 332: Randomly select a preset number of node codes that are located at the same first coding position from all the node codes corresponding to the first parent code and the second parent code of the parent coding subgroup to obtain the first node code set and the second node code set.

[0040] Step 333: Fill all the node codes of the first node code set into the first coding position corresponding to the first empty coding position of the empty coding subgroup, and fill the node codes of the second parent coding subgroup other than the first node code set into the first empty coding position in sequence. Fill all the node codes of the second node code set into the first coding position corresponding to the second empty coding position of the empty coding subgroup, and fill the node codes of the first parent coding subgroup other than the second node code set into the second empty coding position in sequence, so as to generate the child coding subgroup corresponding to the corresponding parent coding subgroup.

[0041] Step 334: Repeat the process of generating child code subgroups corresponding to the corresponding parent code subgroups to obtain child replenishment code bodies that have completed two-point crossover and correspond to the two replenishment code bodies. All corresponding child replenishment code bodies are used as replenishment code bodies that complete the current crossover operation.

[0042] refer to Figures 7 to 9 In this embodiment, a tournament selection method is used to select the parent code body. Then, for the two access code bodies at corresponding positions in the access child code bodies at the same position in the parent code body (e.g., Figure 7 The crossover operation is performed on access codecs A and B. Specifically, for access codecs A and B, a crossover range is randomly generated, such as... Figure 7 As shown in the outline, a parent-generated code subgroup corresponding to an empty code subgroup is generated, including a first empty code subgroup A' and a second empty code subgroup B'. Then, the node codes (2, 5 / 5, 4) within the intersection range of access code subgroup A and access code subgroup B are copied to the corresponding positions of the first empty code subgroup A' and the second empty code subgroup B'. Taking access code subgroup A as an example, as follows... Figure 8As shown, next, the missing positions of the first empty code A' are filled in by removing the node codes in the access code B that are the same as those in the first empty code A', and then filling the remaining node codes of the access code B into the first empty code A' in sequence, thus forming a complete child access code A', as shown. Figure 9 As shown, finally, the same method is used to generate the offspring access code B'.

[0043] In this embodiment, a swap mutation is used, randomly selecting two node codes in the sub-transportation path corresponding to the access code to swap the access order of the corresponding target supply object, thereby realizing the mutation operation of the inner replenishment code body, access sub-code body and access code.

[0044] Step 34: After selecting the candidate replenishment code body with the smallest decision fitness from at least one candidate replenishment code body as the target replenishment code body, perform path segmentation on the node sequence corresponding to each access code of the target replenishment code body, and combine at least one target sub-transport path generated accordingly into the target path corresponding to the access code body, wherein the target transportation path corresponding to the replenishment decision code body includes the target path.

[0045] Through steps 31 to 34 above, the genetic operation of replenishment route planning using the inner layer genetic algorithm of the two-layer genetic algorithm is realized to generate the corresponding replenishment transportation route.

[0046] In some embodiments, the fitness of each multidimensional code is determined based on the sub-fitness corresponding to each replenishment decision sub-code, which is achieved through the following steps: Step 41: Obtain all target transportation routes corresponding to each replenishment decision sub-code body and the first replenishment quantity corresponding to each replenishment decision sub-code body, and determine all second replenishment quantities corresponding to each replenishment decision sub-code body. The first replenishment quantity is used to represent the total replenishment quantity corresponding to all replenishment quantity codes corresponding to each replenishment decision sub-code body, and the second replenishment quantity is used to represent the total replenishment quantity corresponding to all replenishment quantity codes of each product code corresponding to the corresponding replenishment decision sub-code body in multiple execution cycles.

[0047] Step 42: After determining the first revenue parameter corresponding to each target product based on the second order quantity corresponding to each target product, determine the total sub-revenue parameter corresponding to the replenishment decision sub-code body based on the first revenue parameter. The first revenue parameter is determined based on the sales revenue parameter, procurement cost parameter, inventory holding cost parameter and stockout cost parameter corresponding to the second order quantity.

[0048] Step 43: Determine the corresponding pickup transportation cost parameters based on the path nodes corresponding to the target transportation route, and obtain the transportation cost parameters corresponding to the target transportation route based on the preset departure cost parameters and pickup transportation cost parameters, and determine the total sub-transportation cost parameters corresponding to the replenishment decision sub-code body based on the transportation cost parameters.

[0049] Step 44: Perform data processing on the total sub-revenue parameter and the total sub-transportation cost parameter. Based on the processed total sub-revenue parameter and the total sub-transportation cost parameter, use the Einstein summation convention method to solve for the corresponding cost parameter and obtain the corresponding sub-fitness. Based on the sub-fitness of all replenishment decision sub-code bodies, determine the fitness of the corresponding multi-dimensional code body. The corresponding fitness is used to characterize the difference between the total revenue parameter and the total transportation cost parameter. Data processing includes ndarray array conversion.

[0050] Through steps 41 to 44 above, the fitness of the multidimensional coding body corresponding to the outer replenishment decision is determined.

[0051] In some embodiments, the outer genetic algorithm of a two-layer genetic algorithm is used to perform at least one predetermined outer genetic evolution operation on multiple multidimensional codes participating in the current iteration, including the following steps: Step 51: From the multiple multidimensional codes participating in the current iteration, select a preset number of multidimensional codes based on fitness and a preset selection strategy to obtain candidate multidimensional codes. The selection strategy includes roulette wheel selection and elite retention strategy.

[0052] In this embodiment, a selection operation combining roulette wheel selection and elite retention strategy is adopted. The top 2% of multidimensional code bodies with the highest fitness values ​​in each generation are directly retained to the next generation, while the remaining multidimensional code bodies are selected by roulette wheel selection.

[0053] Step 52: Among all the replenishment decision sub-code bodies corresponding to the candidate multidimensional code bodies, determine the replenishment decision sub-code bodies associated with the same central warehouse, and perform cross-crossing on the replenishment decision sub-code bodies associated with the same central warehouse according to the preset outer cross-crossing operation to generate the candidate multidimensional code bodies that have completed the current cross-crossing operation, thus obtaining the candidate multidimensional code bodies.

[0054] In this embodiment, for the candidate multidimensional code bodies of the three-dimensional structure, two types of cross operations are used: product-level cross operation and execution cycle-level cross operation, to obtain candidate multidimensional code bodies, and the selection of each cross operation is based on a certain probability.

[0055] To achieve product-level cross-operation, in some optional implementations, step 52 involves cross-operating replenishment decision sub-code bodies associated with the same central warehouse according to a preset outer-layer cross-operation, including the following steps: Step 521-1: Select two replenishment decision sub-code bodies from all replenishment decision sub-code bodies of the same associated central warehouse to obtain the first target sub-code body.

[0056] Step 521-2: In the first target sub-code body, obtain the first replenishment quantity code set associated with the target product corresponding to each product code, wherein the first replenishment quantity code set includes all replenishment quantity codes of the target product in multiple execution cycles.

[0057] Step 521-3: Randomly select the first replenishment quantity code set corresponding to the same product code of two first target sub-code bodies, and exchange all the replenishment quantity codes of the two selected first replenishment quantity code sets to generate the first replenishment decision sub-code body corresponding to the replenishment decision sub-code body participating in the current outer layer cross operation. The candidate multidimensional code body includes the corresponding first replenishment decision sub-code body.

[0058] In this embodiment, for the two first target sub-code bodies (refer to...) Figure 10 (Parent 1 and Parent 2 in the data), this operation, in the same central warehouse and across all execution cycles, exchanges the allocation value of the target product, that is, exchanges the corresponding replenishment quantity code, that is, exchanges the corresponding first replenishment quantity code set (see reference). Figure 10 (The corresponding replenishment quantity code in the middle frame line) generates two corresponding first replenishment decision sub-code bodies (refer to...) Figure 11 (Child 1 and Child 2 in the middle).

[0059] To implement periodic crossover operations, refer to Figure 12 and Figure 13 In some optional implementations, step 52, which involves cross-referencing replenishment decision sub-code bodies associated with the same central warehouse according to a preset outer cross-referencing operation, further includes the following steps: Step 522-1: Select two replenishment decision sub-code bodies from all replenishment decision sub-code bodies of the same associated central warehouse to obtain the second target sub-code body.

[0060] Step 522-2: In the second target sub-encoding body, obtain the replenishment decision encoding sub-encoding corresponding to each execution cycle.

[0061] Step 522-3: Randomly select two replenishment decision codes corresponding to the same execution cycle of two second target sub-code bodies, and exchange the two selected replenishment decision codes to generate the second replenishment decision code body corresponding to the replenishment decision code body participating in the current outer layer cross operation. The candidate multidimensional code body includes the corresponding second replenishment decision code body.

[0062] In this embodiment, for the two second target sub-code bodies (refer to...) Figure 12 (Parent 1 and Parent 2 in the text) This operation, on the same central warehouse and the same target product, exchanges the allocated values ​​within the same execution cycle, that is, exchanges the replenishment quantity codes within the corresponding execution cycle, which is also exchanging all replenishment quantity codes of the corresponding replenishment decision code (see reference). Figure 12 (The corresponding replenishment quantity code in the middle frame line) generates two corresponding second replenishment decision sub-code bodies (refer to...) Figure 11 (Child 1 and Child 2 in the middle).

[0063] Step 53: Encode the replenishment quantity of the candidate multidimensional code and perform random mutation based on vehicle capacity constraints to generate a multidimensional code that completes the current outer layer genetic evolution operation.

[0064] In the preferred embodiment of this application, a random mutation operation is also performed on the replenishment decision code of the candidate multidimensional code body. The mutation is carried out by gene position (corresponding to a replenishment quantity code in the replenishment decision code body). Under the premise that the total replenishment quantity of all sub-transportation paths corresponding to at least one replenishment decision code body meets the vehicle capacity constraint of the corresponding replenishment vehicle, the gene value of the gene position of the replenishment quantity code of the candidate multidimensional code body is randomly changed with a certain mutation probability (changing the code value of the replenishment quantity code, that is, changing the corresponding replenishment quantity). This improves the local search capability of the genetic algorithm, maintains the diversity of the population, and prevents premature convergence.

[0065] In some embodiments, predicting sales volume includes the following steps: Step 61: After executing the replenishment and transportation decisions of the first execution cycle of the previous decision-making stage, obtain the current inventory of multiple central warehouses and obtain the preset characteristic parameters corresponding to each central warehouse. The characteristic parameters include at least one of the following: commodity seasonal attribute parameters and commodity transaction attribute parameters.

[0066] In this embodiment, the experimental data comes from the regional historical data of six products of a self-operated e-commerce company. Each dataset is divided into a training set (80%) and a test set (20%). The seven selected features include "seasonal attribute", "number of views", "number of times added to cart", "number of times in favorites", "original price", "discount", "advertising strength" and one column of sales data.

[0067] Step 62: Determine the outbound quantity for each central warehouse based on the current inventory and historical decision demand. The historical decision demand is used to represent the product demand for replenishment goods decided by the central warehouse in the first execution cycle corresponding to the previous decision stage.

[0068] Step 63: Input the outbound quantity and corresponding feature parameters of each central warehouse into the sales forecast model, and output the sales volume of goods corresponding to the current demand of each central warehouse.

[0069] The prediction and training of the sales forecasting model in this application embodiment are described below:

[0070] In this embodiment, PSO is applied to optimize two key hyperparameters of the LSTM model: the number of hidden layers and the number of neurons.

[0071] In some preferred embodiments, PSO optimization of LSTM hyperparameters includes the following steps: Step 1: Import the preprocessed dataset and divide it into training and test sets.

[0072] Step 2: Build a variable-structure LSTM model and dynamically add the number of hidden layers using loops.

[0073] Step 3: Initialize the PSO optimization model by randomly generating the number of hidden layers and neurons for each particle to initialize the particle position.

[0074] Step 4: Pass the position value of each particle to the variable structure LSTM model to build multiple LSTM models with different structures.

[0075] Step 5: Train each LSTM model using the training dataset and calculate the loss function value of each model on the validation set.

[0076] Step 6: Pass the loss function value to the PSO model and use it as the fitness function value for each particle.

[0077] Step 7: Update pbest and gbest.

[0078] Step 8: Update the velocity and position of each particle.

[0079] Step 9: Check if the maximum number of PSO iterations has been reached. If not, return to step 4; if it has been reached, output the LSTM model with the optimal structure.

[0080] Step 10: Input the inventory level of the central warehouse into the trained optimal structure LSTM model and output the sales forecast results.

[0081] The following describes the decision results obtained from the embodiments of this application, as well as the mathematical function model constructed based on Bi-GA for multi-center warehouse joint replenishment and transportation decisions:

[0082] First, the following assumptions are made: (1) The costs of procurement and transportation occur within the product delivery time period; (2) The demand of each central warehouse is determined by sales forecasts and is therefore considered known in the model; (3) Each target supplier provides multiple target products, and each target product is provided by only one target supplier; (4) An execution cycle is 1 day; (5) The replenishment lead time is set to 1 day, that is, the time for the target product to be transported from the target supplier to the central warehouse is 1 day; (6) The procurement price of the target product adopts a three-range discount method, and the discount is set according to the different total procurement volume of each target product in all central warehouses. Different price discounts; (7) The initial inventory of each target product in the central warehouse is set to 0; (8) Stockouts are allowed in the central warehouse, and a stockout penalty cost is set; (9) Horizontal transfers between central warehouses are not considered; (10) Each target supply object can only be accessed once by a vehicle sent from any central warehouse in each execution cycle; (11) Each transport vehicle must return to the original departure warehouse after performing the pickup task, starting from the uniquely designated central warehouse, and the transport capacity resources of vehicles belonging to different warehouses operate independently; (12) Heterogeneous vehicle models are used for pickup, and different vehicle models have different start-up costs, unit transport costs and capacity limits.

[0083] In this embodiment, the objective is to maximize operational profits, which is the total sales revenue D minus the procurement cost C1, transportation cost C2, inventory holding cost C3, and stockout cost C4. Specifically, sales revenue D, Procurement cost C1, Transportation cost C2, Inventory holding cost C3, ,in, Stockout cost C4 .

[0084] In this embodiment of the application, the complete mathematical expression of the nonlinear integer programming model constructed for the Joint Replenishment Problem (JRP-T) with flexible replenishment strategy and transportation is as follows: MaxZ = D - C1 - C2 - C3 - C4.

[0085] Constraints: 1. The total purchase quantity of each target product in each cycle is constrained by the following formula: ; 2. Inventory update formula: ; 3. The total volume of the target product replenished by the central warehouse at the target supplier s during the t-th execution cycle. , ; 4. Flow balance constraint ; ; ; 5. Access limit ; 6. Vehicle quantity limit ; ; 7. MTZ constraint with capacity constraint , , , , ; 8. Consistency between replenishment and transportation activities , ; , 9. Inventory capacity constraints .

[0086] This embodiment also provides a multi-center warehouse joint replenishment and transportation decision-making device based on Bi-GA. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0087] Figure 14 This is a structural block diagram of a Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making device according to an embodiment of this application, as shown below. Figure 14 As shown, the device includes a determining module 141, an encoding module 142, and a processing module 143, wherein,

[0088] The determination module 141 is used to determine the current demand for various target products to be replenished in multiple execution cycles corresponding to a decision stage following the current one. The current demand is the sales volume of goods predicted by a preset sales forecasting model based on the current inventory of multiple central warehouses. The sales forecasting model is a model trained on PSO-ALSTM. The multiple execution cycles corresponding to the decision stage following the current one are generated by rolling over the multiple execution cycles corresponding to the previous decision stage by one execution cycle.

[0089] The encoding module 142, coupled to the determination module 141, is used to encode and initialize the central warehouse, execution cycle, target product, and current demand according to a preset encoding method, generating multiple multi-dimensional encoding bodies. Each multi-dimensional encoding body includes multiple replenishment decision sub-encoding bodies. Each replenishment decision sub-encoding body is associated with a central warehouse. Each replenishment decision sub-encoding body includes multiple replenishment decision encoding bodies that represent the replenishment decisions made by the corresponding central warehouse for various target products in each execution cycle. The replenishment decisions include at least the replenishment quantity of the target products.

[0090] The processing module 143, coupled to the encoding module 142, is used to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration based on replenishment decision sub-encoders corresponding to multiple multidimensional encoders using a preset two-layer genetic algorithm, until a target multidimensional encoder is generated. All replenishment decisions and all target replenishment paths corresponding to the target multidimensional encoder are used as decision results. The target replenishment path is generated by replenishment path planning of the replenishment decision sub-encoders of the target multidimensional encoder using the inner-layer genetic algorithm of the two-layer genetic algorithm.

[0091] This embodiment also provides a service platform, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0092] Optionally, the system may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0093] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0094] S1, determine the current demand for various target products to be replenished in multiple execution cycles corresponding to a decision stage following the current one. The current demand is the sales volume of goods predicted by a preset sales forecasting model based on the current inventory of multiple central warehouses. The sales forecasting model is a model trained on PSO-ALSTM. The multiple execution cycles corresponding to a decision stage following the current one are generated by rolling over the multiple execution cycles corresponding to the previous decision stage by one execution cycle.

[0095] S2 encodes and initializes the central warehouse, execution cycle, target product, and current demand according to a preset coding method, generating multiple multi-dimensional coding bodies. Each multi-dimensional coding body includes multiple replenishment decision sub-coding bodies. Each replenishment decision sub-coding body is associated with a central warehouse. Each replenishment decision sub-coding body includes multiple replenishment decision codes that represent the replenishment decisions made by the corresponding central warehouse for various target products in each execution cycle. The replenishment decision includes at least the replenishment quantity of the target product.

[0096] S3 utilizes a pre-defined two-layer genetic algorithm to perform inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration based on replenishment decision sub-code bodies corresponding to multiple multi-dimensional code bodies until a target multi-dimensional code body is generated. All replenishment decisions and all target replenishment paths corresponding to the target multi-dimensional code body are taken as decision results. The target replenishment path is generated by using the inner-layer genetic algorithm of the two-layer genetic algorithm to plan replenishment paths for the replenishment decision sub-code bodies of the target multi-dimensional code body.

[0097] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0098] Furthermore, in conjunction with the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method in the above embodiments, this application embodiment can provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making methods in the above embodiments.

[0099] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multi-center warehouse joint replenishment and transportation decision-making method based on Bi-GA, characterized in that, include: After executing the replenishment and transportation decisions of the first execution cycle of the previous decision-making stage, the current inventory of multiple central warehouses is obtained, and the preset characteristic parameters corresponding to each central warehouse are obtained, wherein the characteristic parameters include at least one of the following: commodity seasonal attribute parameters and commodity transaction attribute parameters; Based on the current inventory level and historical decision demand, the outbound quantity corresponding to each central warehouse is determined, wherein the historical decision demand is used to characterize the product demand for replenishment goods decided by the central warehouse in the first execution cycle corresponding to the previous decision stage. The outbound quantity and corresponding feature parameters of each central warehouse are input into a sales prediction model trained based on PSO-ALSTM, and the corresponding sales volume is output. The sales volume is used as the current demand for various target products to be replenished in multiple execution cycles corresponding to a decision stage after the current one for each central warehouse. The multiple execution cycles corresponding to a decision stage after the current one are generated by rolling the multiple execution cycles corresponding to the previous decision stage by one execution cycle. The central warehouse, the execution cycle, the target product, and the current demand are encoded and initialized according to a preset encoding method to generate multiple multi-dimensional encoding bodies. Each multi-dimensional encoding body includes multiple replenishment decision sub-encoding bodies. Each replenishment decision sub-encoding body is associated with a central warehouse. Each replenishment decision sub-encoding body includes multiple replenishment decision encoding bodies used to represent the replenishment decisions made by the corresponding central warehouse for various target products in each execution cycle. The replenishment decision includes at least the replenishment quantity of the target product. Using a pre-defined two-layer genetic algorithm, based on the replenishment decision sub-code bodies corresponding to multiple multidimensional code bodies, an inner layer of replenishment path planning decision iteration and an outer layer of replenishment decision iteration are performed until a target multidimensional code body is generated. All replenishment decisions and all target replenishment paths corresponding to the target multidimensional code body are taken as decision results. The target replenishment path is generated by using the inner layer genetic algorithm of the two-layer genetic algorithm to plan replenishment paths for the replenishment decision sub-code bodies of the target multidimensional code body.

2. The method according to claim 1, characterized in that, Using a pre-defined two-layer genetic algorithm, based on the replenishment decision sub-encoders corresponding to multiple multidimensional encoders, the algorithm performs inner-layer replenishment path planning decision iteration and outer-layer replenishment decision iteration, including: Using the inner layer genetic algorithm of the two-layer genetic algorithm, replenishment path planning is performed on all replenishment decision codes corresponding to the multiple multidimensional codes participating in the current replenishment decision iteration, to obtain the target transportation path corresponding to each replenishment decision code. After determining the fitness of each multidimensional code based on the sub-fitness of each replenishment decision sub-code, the outer genetic algorithm of the two-layer genetic algorithm is used to perform at least one set outer genetic evolution operation on the multiple multidimensional codes participating in the current iteration, generating multiple multidimensional codes that have completed the current replenishment decision iteration. The sub-fitness is determined based on the replenishment transportation cost of all target transportation paths corresponding to the replenishment decision sub-code and the revenue parameter corresponding to the replenishment decision sub-code. Based on the multiple multidimensional codes that have completed the current replenishment decision iteration, the steps of planning and generating the corresponding target transportation path using the corresponding inner genetic algorithm, determining the fitness of the corresponding multidimensional code, and performing the set outer genetic evolution operation using the corresponding outer genetic algorithm are repeatedly executed until multiple multidimensional codes with fitness greater than the fitness threshold are generated. Among the multiple multidimensional encoders with fitness greater than the fitness threshold, the multidimensional encoder with the highest fitness is selected to obtain the target multidimensional encoder.

3. The method according to claim 2, characterized in that, Using the inner layer genetic algorithm of the two-layer genetic algorithm, replenishment path planning is performed on all replenishment decision codecs corresponding to the multiple multidimensional codecs participating in the current replenishment decision iteration, including: Obtain the replenishment decision sub-encoder corresponding to each of the multi-dimensional encoding bodies participating in the current replenishment decision iteration, and determine all the replenishment decision encoding sub-encoders of each of the replenishment decision sub-encoders, wherein the replenishment decision encoding sub-encoders include multiple product codes and replenishment quantity codes, the replenishment quantity code is used to characterize the replenishment quantity of a target product in one execution cycle, and one product code is associated with one target supply object; Each replenishment decision code uses the target supply object associated with all product codes as node codes to generate a node sequence. Each node sequence is then used as an access code to generate an access sub-code body corresponding to the replenishment decision code body. A replenishment code body comprising multiple access sub-code bodies is also generated, where each replenishment code body corresponds to one multi-dimensional code body. Based on vehicle capacity constraints, the node sequence corresponding to the access code body is path-segmented to generate at least one sub-transport path. Each candidate transport path corresponding to the access code body includes at least one sub-transport path. The replenishment quantity for node replenishment at each node code is determined based on the replenishment quantity codes corresponding to all product codes associated with a target supply object. After determining the replenishment cost parameters corresponding to the candidate transportation path based on the sub-replenishment cost parameters corresponding to the sub-transportation path, the decision fitness corresponding to the replenishment code body is determined based on the replenishment cost parameters of the access sub-code body. Based on the decision fitness, an inner-layer genetic algorithm is used to iterate the access sub-code bodies of multiple replenishment codes with a set inner-layer genetic evolution operation until at least one candidate replenishment code body with a decision fitness less than a set fitness value is generated. The inner-layer genetic evolution operation includes the following operations: tournament selection operation, two-point crossover operation, and exchange mutation operation. After selecting the candidate replenishment code with the lowest decision fitness among at least one candidate replenishment code as the target replenishment code, the node sequence corresponding to each access code of the target replenishment code is path segmented, and at least one target sub-transport path generated accordingly is combined into the target path corresponding to the access code, wherein the target transport path corresponding to the replenishment decision code includes the target path.

4. The method according to claim 3, characterized in that, Performing a defined inner-layer genetic evolution operation on the access sub-codemas of the plurality of replenishment codemas, including: After selecting two replenishment codes from a plurality of replenishment codes as parent codes, two access sub-codes located at the same sorting position are selected from all access sub-codes of the two parent codes, and at least one access code with the same sorting position is randomly selected from all access codes of the two corresponding access sub-codes, to obtain at least one set of parent code sub-groups, and to generate an empty code sub-group corresponding to the corresponding parent code sub-group; From all the node codes corresponding to the first parent code and the second parent code of the parent code subgroup, a preset number of node codes located at the same first coding position are randomly selected to obtain the first node code set and the second node code set. All the node codes of the first node code set are respectively filled into the first code bit corresponding to the first empty code bit of the empty code subgroup, and the node codes of the second parent code subgroup other than the first node code set are sequentially filled into the first empty code bit. All the node codes of the second node code set are respectively filled into the first code bit corresponding to the second empty code bit of the empty code subgroup, and the node codes of the first parent code subgroup other than the second node code set are sequentially filled into the second empty code bit, so as to generate a child code subgroup corresponding to the corresponding parent code subgroup. Repeatedly generate the child code subgroup corresponding to the corresponding parent code subgroup to obtain the child replenishment code body that completes the two-point cross and corresponds to the two replenishment code bodies, and use all the corresponding child replenishment code bodies as the replenishment code body that completes the current cross operation.

5. The method according to claim 3, characterized in that, Based on the sub-fitness corresponding to each replenishment decision sub-code body, determine the fitness corresponding to each multidimensional code body, including: Obtain all target transportation routes corresponding to each replenishment decision sub-code body and the first replenishment quantity corresponding to each replenishment decision sub-code body, and determine all second replenishment quantities corresponding to each replenishment decision sub-code body. The first replenishment quantity is used to represent the total replenishment quantity corresponding to all replenishment quantity codes corresponding to each replenishment decision sub-code body, and the second replenishment quantity is used to represent the total replenishment quantity corresponding to all replenishment quantity codes of each product code corresponding to the corresponding replenishment decision sub-code body in multiple execution cycles. After determining the first revenue parameter corresponding to each target product based on the second replenishment quantity corresponding to each target product, the total sub-revenue parameter corresponding to the replenishment decision sub-code body is determined based on the first revenue parameter, wherein the first revenue parameter is determined based on the sales revenue parameter, procurement cost parameter, inventory holding cost parameter and stockout cost parameter corresponding to the second replenishment quantity; Based on the path node corresponding to the target transportation route, determine the corresponding pickup transportation cost parameter, and based on the preset departure cost parameter and the pickup transportation cost parameter, obtain the transportation cost parameter corresponding to the target transportation route, and based on the transportation cost parameter, determine the total sub-transportation cost parameter corresponding to the replenishment decision sub-code body; The total sub-revenue parameter and the total sub-transportation cost parameter are processed, and based on the processed total sub-revenue parameter and the total sub-transportation cost parameter, the corresponding cost parameter is solved using the Einstein summation convention method to obtain the corresponding sub-fitness. Based on the sub-fitness of all the replenishment decision sub-encoders, the fitness of the corresponding multidimensional code is determined. The corresponding fitness is used to characterize the difference between the total sub-revenue parameter and the total sub-transportation cost parameter. The data processing includes ndarray array conversion.

6. The method according to claim 2, characterized in that, Using the outer layer genetic algorithm of the two-layer genetic algorithm, at least one predetermined outer layer genetic evolution operation is performed on the multiple multidimensional codes participating in the current iteration, including: From the multiple multidimensional encoders participating in the current iteration, a preset number of multidimensional encoders are selected based on the fitness and a preset selection strategy to obtain candidate multidimensional encoders, wherein the selection strategy includes roulette wheel selection and elite retention strategy. Among all the replenishment decision sub-code bodies corresponding to the candidate multidimensional code bodies, the replenishment decision sub-code bodies associated with the same central warehouse are determined, and the replenishment decision sub-code bodies associated with the same central warehouse are cross-crossed according to the preset outer cross-cross operation to generate the candidate multidimensional code bodies that have completed the current cross-cross operation, thus obtaining the candidate multidimensional code bodies. The replenishment quantity code of the candidate multidimensional code is subjected to random mutation based on vehicle capacity constraints to generate the multidimensional code that completes the current outer layer genetic evolution operation.

7. The method according to claim 6, characterized in that, For the replenishment decision sub-code bodies associated with the same central warehouse, cross-linking is performed according to a preset outer cross-linking operation, including: From all the replenishment decision sub-code bodies associated with the same central warehouse, select two replenishment decision sub-code bodies to obtain the first target sub-code body; In the first target sub-encoding body, a first replenishment quantity encoding set associated with the target product corresponding to each product code is obtained, wherein the first replenishment quantity encoding set includes all the replenishment quantity codes of the target product in multiple execution cycles; Randomly select two first replenishment quantity code sets corresponding to the same product code of the first target sub-code bodies, and exchange all the replenishment quantity codes of the two selected first replenishment quantity code sets to generate a first replenishment decision sub-code body corresponding to the replenishment decision sub-code body participating in the current outer layer cross operation, wherein the candidate multidimensional code body includes the corresponding first replenishment decision sub-code body.

8. The method according to claim 6, characterized in that, For the replenishment decision sub-code bodies associated with the same central warehouse, cross-linking is performed according to a preset outer cross-linking operation, including: From all the replenishment decision sub-code bodies associated with the same central warehouse, select two replenishment decision sub-code bodies to obtain the second target sub-code body; In the second target sub-encoding body, the replenishment decision encoding sub-encoding corresponding to each execution cycle is obtained; Two replenishment decision codes corresponding to the same execution cycle of two second target sub-code bodies are randomly selected, and the two selected replenishment decision codes are exchanged to generate a second replenishment decision code body corresponding to the replenishment decision code body participating in the current outer layer cross operation, wherein the candidate multidimensional code body includes the corresponding second replenishment decision code body.

9. A service platform, comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the steps of the Bi-GA-based multi-center warehouse joint replenishment and transportation decision-making method as described in any one of claims 1 to 8.

Images