A Knowledge Graph-Based Intelligent Scheduling Method for E-commerce Commodity Warehousing and Logistics

Abstract

This invention discloses an intelligent scheduling method for e-commerce warehousing and logistics based on knowledge graphs, comprising the following steps: collecting e-commerce warehousing and logistics fulfillment data to generate a basic fulfillment dataset; constructing a snapshot sequence of the fulfillment knowledge graph to generate a set of node features and an initial set of fulfillment edges; inputting an improved EvolveGCN scheduling model to generate a set of candidate scheduling edges and resource occupancy and surplus vectors; generating a fulfillment conservation gating value, correcting neighbor message weights, and obtaining a representation of fulfillment conservation nodes; calculating the total resource occupancy and generating a fulfillment conflict residual matrix; correcting the graph convolution parameters based on the residual matrix to generate a fulfillment scheduling state representation; and generating a joint warehousing and logistics scheduling result based on the fulfillment scheduling state representation. This invention achieves joint intelligent scheduling of warehousing and logistics, improving resource matching accuracy and order fulfillment stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology for e-commerce warehousing and logistics, and in particular to an intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs. Background Technology

[0002] As the number of orders, product types, and delivery areas on e-commerce platforms continue to increase, the scheduling of goods warehousing and logistics needs to be coordinated among order fulfillment, product inventory, warehouse space occupancy, warehouse operational capacity, logistics route capacity, and delivery timeliness. Existing e-commerce warehousing and logistics scheduling methods mostly rely on inventory quantity, warehouse distance, route timeliness, and manually set rules to make phased decisions on order allocation, product location, inventory transfer, outbound sequence, and logistics routes.

[0003] Existing knowledge graph scheduling methods can establish the association structure between products, orders, warehouses, storage locations, and logistics routes. However, message passing between graph nodes mainly relies on node characteristics and edge connections. Inventory balance, storage location capacity, warehouse operation capacity, logistics route carrying capacity, and delivery timeliness resources are not directly included in the graph convolution message weight calculation process. Existing dynamic graph convolution models can handle graph structure changes within continuous time windows, but resource conflict information is mostly used as an external constraint in result filtering and is not written into the graph convolution parameter update process.

[0004] The aforementioned processing methods can easily lead to scheduling results with high matching degrees but insufficient resources. This results in problems such as multiple orders competing for the same product inventory, over-occupancy of the same storage location, overload of the same warehouse's operational capacity, insufficient capacity of the same logistics route, and conflicting delivery time windows. These issues affect the executability of the joint warehousing and logistics scheduling results and the scheduling stability within continuous time windows. Therefore, how to provide a knowledge graph-based intelligent scheduling method for e-commerce product warehousing and logistics is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs. This invention combines knowledge graphs and an improved EvolveGCN scheduling model, and incorporates fulfillment resource constraints into the graph convolution message weights and parameter update process to achieve joint scheduling of commodity storage locations, order distribution, inventory allocation, outbound priority, and logistics routes. It has the advantages of accurate resource matching, fewer scheduling conflicts, and high fulfillment stability.

[0006] According to an embodiment of the present invention, an intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs includes the following steps: Collect e-commerce warehousing and logistics fulfillment data within a continuous time window to generate a basic fulfillment dataset; A snapshot sequence of a performance knowledge graph is constructed from the performance base dataset, generating a set of node features and an initial set of performance edges; An improved EvolveGCN scheduling model is constructed, which includes a performance conservation graph convolutional layer and a residual-driven parameter evolution unit. The node feature set and the initial performance edge set are input into the improved EvolveGCN scheduling model to generate a candidate scheduling edge set, a performance resource occupancy vector, and a performance resource surplus vector. In the convolutional layer of the performance conservation graph, a performance conservation gate value is generated based on the performance resource occupancy vector and the performance resource surplus vector. The performance conservation gate value is multiplied by the original neighbor message weights of the candidate scheduling edges to generate modified neighbor message weights constrained by performance resources. The node feature set is aggregated according to the modified neighbor message weights to generate the performance conservation node representation. Candidate scheduling edges are collected according to resource ownership, the total resource occupancy is counted, and the difference between the total resource occupancy and the fulfillment resource surplus vector is calculated to generate the fulfillment conflict residual matrix. In the residual-driven parameter evolution unit, parameter correction coefficients are generated based on the performance conflict residual matrix. Based on the parameter correction coefficients, the corresponding graph convolution parameter fragments in the current time window graph convolution parameters are compressed to generate corrected graph convolution parameters. The corrected graph convolution parameters are then used to update the performance conservation node representation and generate the performance scheduling state representation. The joint scheduling results of warehousing and logistics are generated based on the performance scheduling status representation.

[0007] Optionally, the e-commerce warehousing and logistics fulfillment data includes: Order fulfillment data, product inventory data, warehouse space occupancy data, warehouse operation capacity data, logistics route transportation data, and delivery timeliness data; Order fulfillment data records the order number, product number, and delivery area number; The product inventory data records the product number, warehouse number, and available inventory balance. The storage location occupancy data records the storage location number, warehouse number, and remaining storage location capacity. Warehouse operational capacity data records warehouse number and remaining warehouse operational capacity; The logistics route data records the logistics route number, warehouse number, delivery area number, and remaining transport capacity of the logistics route; The delivery time data records the order number and remaining delivery time.

[0008] Optionally, the operation of constructing a snapshot sequence of the performance knowledge graph from the performance base dataset specifically includes: Group the basic dataset of contract performance according to continuous time windows to generate time window performance records; Extract order number, product number, warehouse number, storage location number, logistics route number, and delivery area number from the time window fulfillment record and generate entity nodes; The available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery time are written into the entity nodes to form a set of node features; Connect the order number and the product number to generate an order inclusion edge; connect the product number and the warehouse number to generate an inventory occupancy edge; connect the product number and the storage location number to generate a storage location occupancy edge; connect the warehouse number and the logistics route number to generate a route transportation edge; connect the order number and the delivery area number to generate a time constraint edge, thus forming the initial fulfillment edge set. The node feature set and the initial performance edge set are combined to form a performance knowledge graph snapshot. Several performance knowledge graph snapshots are arranged in the order of continuous time windows to generate a performance knowledge graph snapshot sequence.

[0009] Optionally, the operations for generating the performance resource occupancy vector and the performance resource surplus vector specifically include: Read the candidate scheduling edges in the candidate scheduling edge set, and determine the commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources associated with the candidate scheduling edges; Extract the inventory occupancy, storage space occupancy, warehouse operation capacity occupancy, logistics route carrying capacity occupancy, and delivery timeliness occupancy from e-commerce warehousing and logistics fulfillment data to generate a fulfillment resource occupancy vector; Extract the available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery timeliness to generate a fulfillment resource balance vector; Bind the fulfillment resource occupancy vector and the fulfillment resource surplus vector to the candidate scheduling edge.

[0010] Optionally, the generation of the performance conservation gate value specifically includes: Read the fulfillment resource occupancy vector and fulfillment resource surplus vector bound to the candidate scheduling edge; The five resources are: commodity inventory resources, warehouse capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources. The resource occupancy and resource surplus are matched one by one. When resource usage is zero, the resource component gating value is set to one. When the resource usage is not zero and the resource reserve is not less than the resource usage, the resource component gate value is set to one. When the resource usage is not zero and the resource reserve is less than the resource usage, divide the resource reserve by the resource usage to obtain the resource component gating value. The minimum value among the five resource component gating values ​​is selected as the performance conservation gating value of the candidate scheduling edge.

[0011] Optionally, the message weight correction of the performance conservation graph convolutional layer specifically includes: According to the order of the candidate scheduling edge set, read the sending node characteristics and receiving node characteristics connected to each candidate scheduling edge; The sending node features, receiving node features, and current time window graph convolution parameters are input into the message weight calculation process of the candidate scheduling edge to generate the original neighbor message weights of the candidate scheduling edge. The original neighbor message weights are multiplied by the performance conservation gate value to generate the corrected neighbor message weights; When the performance conservation gate value is one, the original neighbor message weights are retained; When the performance conservation threshold is less than one, the weight of the original neighbor messages is compressed according to the performance conservation threshold. The features of the sending node are weighted according to the weight of the modified neighbor message to generate the modified neighbor message for the candidate scheduling edge; All corrected neighbor messages connected to the same receiving node are accumulated to generate an aggregate message for the receiving node; The aggregated message of the receiving node is concatenated with the features of the receiving node, and the feature update calculation process corresponding to the convolution parameters of the current time window graph is input to generate the performance conservation node representation of the receiving node. For all candidate scheduling edges in the candidate scheduling edge set, the following steps are executed sequentially: reading the sending node feature, reading the receiving node feature, generating the original neighbor message weight, generating the corrected neighbor message weight, generating the corrected neighbor message, generating the receiving node aggregate message, and generating the receiving node performance conservation node representation, to obtain the performance conservation node representation corresponding to the node feature set.

[0012] Optionally, the operation of aggregating candidate scheduling edges according to resource ownership and calculating the total resource usage specifically includes: According to the arrangement order of the candidate scheduling edge set, the fulfillment resource occupation vector bound to each candidate scheduling edge is split to obtain the commodity inventory occupation, storage space capacity occupation, warehouse operation capacity occupation, logistics line carrying capacity occupation and delivery time occupation. For the same candidate scheduling edge, five resource occupancy records are generated. The first resource occupancy record uses the combination of product number and warehouse number as the product inventory resource identifier and carries the product inventory occupancy amount. The second resource occupancy record uses the warehouse location number as the warehouse location capacity resource identifier and carries the warehouse location capacity occupancy amount. The third resource occupancy record uses the combination of warehouse number and current time window as the warehouse operation resource identifier and carries the warehouse operation capacity occupancy amount. The fourth resource occupancy record uses the combination of logistics line number and current time window as the logistics line transportation resource identifier and carries the logistics line transportation capacity occupancy amount. The fifth resource occupancy record uses the delivery timeliness window as the delivery timeliness resource identifier and carries the delivery timeliness occupancy amount. Sort all resource usage records according to resource type and resource identifier; Merge resource occupancy records with the same resource type and the same resource identifier, and accumulate the occupancy amount in the merged resource occupancy records to generate the total occupancy amount of commodity inventory resources, total occupancy amount of storage space capacity resources, total occupancy amount of warehouse operation resources, total occupancy amount of logistics route transportation resources, and total occupancy amount of delivery timeliness resources.

[0013] Optionally, the generation of the performance conflict residual matrix specifically includes: For the total amount of goods inventory resources occupied, the total amount of storage space capacity resources occupied, the total amount of warehouse operation resources occupied, the total amount of logistics route transportation resources occupied, and the total amount of delivery timeliness resources occupied, the resource surplus of the same type and the same resource identifier is matched from the fulfillment resource surplus vector according to the resource type and resource identifier; The resource balance corresponding to the same resource type and the same resource identifier is used as the resource balance of that resource identifier; Subtract the remaining resources from the total resource usage to generate the resource over-utilization difference; When the resource over-utilization difference is negative, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is zero, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is positive, the resource over-utilization difference is used as the corresponding resource conflict residual. Based on a fixed order of commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources, the inventory conflict residuals, storage location conflict residuals, warehouse operation conflict residuals, logistics route conflict residuals, and delivery timeliness conflict residuals are arranged into a fulfillment conflict residual matrix.

[0014] Optionally, the operation of updating the current time window graph convolution parameters based on the performance conflict residual matrix includes: The fulfillment conflict residual matrix is ​​divided and summed in the order of inventory conflict residual, storage location conflict residual, warehouse operation conflict residual, logistics route conflict residual, and delivery timeliness conflict residual to generate the total inventory conflict, total storage location conflict, total warehouse operation conflict, total logistics route conflict, and total delivery timeliness conflict. Extract the resource surplus corresponding to the total amount of various resource conflicts from the fulfillment resource surplus vector to generate the total inventory surplus, total storage location surplus, total warehouse operation surplus, total logistics route surplus, and total delivery timeliness surplus. The total amount of various resource conflicts is compared with the total amount of corresponding resource reserves to generate parameter correction coefficients; The current time window graph convolution parameters are divided into five graph convolution parameter segments according to commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources; Compress the corresponding graph convolution parameter fragments based on the parameter correction coefficients to generate corrected graph convolution parameters; The modified graph convolution parameters are input into the message weight calculation process of the performance conservation graph convolution layer. The performance conservation node representation is updated using the modified graph convolution parameters to generate the performance scheduling state representation.

[0015] Optionally, the operation of generating joint warehousing and logistics scheduling results based on the fulfillment scheduling status representation specifically includes: Read the candidate scheduling edge weights between commodity nodes and storage location nodes, order nodes and warehouse nodes, and warehouse nodes and logistics route nodes from the fulfillment scheduling status representation; Based on the candidate scheduling edge weights, select the corresponding storage locations for the goods and generate the goods storage location adjustment results; Based on the performance conservation threshold and the candidate scheduling edge weights, the corresponding shipping warehouses for each order are selected, and the order warehouse allocation results are generated. Based on the commodity inventory resource identifier corresponding to the inventory conflict residual, the receiving warehouse is determined, and the warehouse with the available inventory balance of the same commodity being greater than the commodity inventory occupancy is determined as the sending warehouse, and the inventory transfer result is generated. Order nodes are arranged according to delivery timeliness and warehouse operation resource threshold values ​​to generate outbound priority results; The logistics routes corresponding to the warehouses are selected based on the threshold values ​​of the logistics route's transport resources, the threshold values ​​of the delivery timeliness resources, and the weights of the candidate scheduling edges, thus generating logistics route matching results.

[0016] The beneficial effects of this invention are: (1) This invention transforms commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources and delivery timeliness resources into performance conservation gating values. The performance conservation gating values ​​are embedded in the message weight calculation process of candidate scheduling edges, so that the graph convolution message transmission is constrained by performance resources, reducing the impact of insufficient resource candidate scheduling edges on scheduling results, and improving the executability of order distribution, commodity storage location and logistics route matching.

[0017] (2) The present invention generates a performance conflict residual matrix based on the total resource occupancy and the remaining performance resources, and uses the performance conflict residual matrix to update the convolution parameters of the current time window graph, so that inventory conflict, storage location conflict, warehouse operation conflict, logistics route conflict and delivery timeliness conflict enter the model parameter update process, reducing the probability of repeated resource over-occupancy scheduling results within a continuous time window.

[0018] (3) Based on the performance scheduling status representation, the present invention generates the joint scheduling result of warehousing and logistics, so that the adjustment of commodity storage location, order distribution, inventory transfer, outbound priority and logistics route matching are completed in the same improved EvolveGCN scheduling model, reducing the decision deviation caused by the separation of warehousing-side scheduling and logistics-side scheduling, and improving the coordination and performance stability of e-commerce commodity warehousing and logistics scheduling. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of an intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs proposed in this invention. Figure 2 This is a schematic diagram of the fulfillment conservation message weight correction process for an intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs proposed in this invention. Figure 3 This is a schematic diagram of the residual-driven parameter update process for a knowledge graph-based intelligent scheduling method for e-commerce commodity warehousing and logistics proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A knowledge graph-based intelligent scheduling method for e-commerce commodity warehousing and logistics includes the following steps: Collect e-commerce warehousing and logistics fulfillment data within a continuous time window to generate a basic fulfillment dataset; A snapshot sequence of a performance knowledge graph is constructed from the performance base dataset, generating a set of node features and an initial set of performance edges; An improved EvolveGCN scheduling model is constructed, which includes a performance conservation graph convolutional layer and a residual-driven parameter evolution unit. The node feature set and the initial performance edge set are input into the improved EvolveGCN scheduling model to generate a candidate scheduling edge set, a performance resource occupancy vector, and a performance resource surplus vector. In the convolutional layer of the performance conservation graph, a performance conservation gate value is generated based on the performance resource occupancy vector and the performance resource surplus vector. The performance conservation gate value is multiplied by the original neighbor message weights of the candidate scheduling edges to generate modified neighbor message weights constrained by performance resources. The node feature set is aggregated according to the modified neighbor message weights to generate the performance conservation node representation. Candidate scheduling edges are collected according to resource ownership, the total resource occupancy is counted, and the difference between the total resource occupancy and the fulfillment resource surplus vector is calculated to generate the fulfillment conflict residual matrix. In the residual-driven parameter evolution unit, parameter correction coefficients are generated based on the performance conflict residual matrix. Based on the parameter correction coefficients, the corresponding graph convolution parameter fragments in the current time window graph convolution parameters are compressed to generate corrected graph convolution parameters. The corrected graph convolution parameters are then used to update the performance conservation node representation and generate the performance scheduling state representation. The joint scheduling results of warehousing and logistics are generated based on the performance scheduling status representation.

[0022] E-commerce warehousing and logistics fulfillment data is written into the fulfillment base dataset according to continuous time windows, with each continuous time window corresponding to one record of warehousing and logistics operation status. The node feature set in the fulfillment knowledge graph snapshot sequence is used to represent the status of orders, products, inventory, storage locations, warehouses, logistics routes, and delivery time within the current time window, and the initial fulfillment edge set is used to represent the fulfillment connections between the above entities. The improved EvolveGCN scheduling model adds a fulfillment conservation graph convolutional layer and a residual-driven parameter evolution unit to the original dynamic graph convolutional parameters that evolve over time. This allows fulfillment resource constraints to participate in the message weight calculation of candidate scheduling edges, and allows fulfillment conflict residuals to participate in the graph convolutional parameter update.

[0023] In this embodiment, e-commerce warehousing and logistics fulfillment data includes: Order fulfillment data, product inventory data, warehouse space occupancy data, warehouse operation capacity data, logistics route transportation data, and delivery timeliness data; Order fulfillment data records the order number, product number, and delivery area number; The product inventory data records the product number, warehouse number, and available inventory balance. The storage location occupancy data records the storage location number, warehouse number, and remaining storage location capacity. Warehouse operational capacity data records warehouse number and remaining warehouse operational capacity; The logistics route data records the logistics route number, warehouse number, delivery area number, and remaining transport capacity of the logistics route; The delivery time data records the order number and remaining delivery time.

[0024] Order fulfillment data is used to determine the connection between orders, goods, and delivery areas; goods inventory data is used to determine the remaining goods inventory resources; warehouse location occupancy data is used to determine the storage space for goods; warehouse operation capacity data is used to determine order processing capacity; logistics route transportation data is used to determine the transportation capacity from the warehouse to the delivery area; and delivery timeliness data is used to determine the remaining order fulfillment time. After the above data is entered into the fulfillment basic dataset, data associations are established through order number, goods number, warehouse number, warehouse location number, logistics route number, and delivery area number.

[0025] In this embodiment, the operation of constructing a snapshot sequence of the performance knowledge graph from the performance foundation dataset specifically includes: Group the basic dataset of contract performance according to continuous time windows to generate time window performance records; Extract order number, product number, warehouse number, storage location number, logistics route number, and delivery area number from the time window fulfillment record and generate entity nodes; The available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery time are written into the entity nodes to form a set of node features; Connect the order number and the product number to generate an order inclusion edge; connect the product number and the warehouse number to generate an inventory occupancy edge; connect the product number and the storage location number to generate a storage location occupancy edge; connect the warehouse number and the logistics route number to generate a route transportation edge; connect the order number and the delivery area number to generate a time constraint edge, thus forming the initial fulfillment edge set. The node feature set and the initial performance edge set are combined to form a performance knowledge graph snapshot. Several performance knowledge graph snapshots are arranged in the order of continuous time windows to generate a performance knowledge graph snapshot sequence.

[0026] The fulfillment knowledge graph snapshots are generated sequentially according to continuous time windows. Each time window's fulfillment record forms a fulfillment knowledge graph snapshot, and several snapshots are arranged in chronological order to form a sequence of fulfillment knowledge graph snapshots. The numerical features in the node feature set come from the available inventory of goods, remaining storage space capacity, remaining warehouse operating capacity, remaining transport capacity of logistics routes, and remaining delivery timeliness. The edges in the initial fulfillment edge set are used to provide the basis for generating candidate scheduling edges for the improved EvolveGCN scheduling model.

[0027] In this embodiment, the operations for generating the performance resource occupancy vector and the performance resource surplus vector specifically include: Read the candidate scheduling edges in the candidate scheduling edge set, and determine the commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources associated with the candidate scheduling edges; Extract the inventory occupancy, storage space occupancy, warehouse operation capacity occupancy, logistics route carrying capacity occupancy, and delivery timeliness occupancy from e-commerce warehousing and logistics fulfillment data to generate a fulfillment resource occupancy vector; Extract the available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery timeliness to generate a fulfillment resource balance vector; Bind the fulfillment resource occupancy vector and the fulfillment resource surplus vector to the candidate scheduling edge.

[0028] Candidate scheduling edges represent candidate fulfillment combinations among orders, products, storage locations, warehouses, and logistics routes. The fulfillment resource occupancy vector records the five resources required for the execution of a candidate scheduling edge, while the fulfillment resource surplus vector records the remaining available quantity of the corresponding five resources within the current time window. The fulfillment resource occupancy vector and the fulfillment resource surplus vector are bound to the same candidate scheduling edge in the same resource order, enabling subsequent fulfillment conservation gating values ​​to be calculated item by item.

[0029] In this embodiment, the generation of the performance conservation threshold value specifically includes: Read the fulfillment resource occupancy vector and fulfillment resource surplus vector bound to the candidate scheduling edge; The five resources are: commodity inventory resources, warehouse capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources. The resource occupancy and resource surplus are matched one by one. When resource usage is zero, the resource component gating value is set to one. When the resource usage is not zero and the resource reserve is not less than the resource usage, the resource component gate value is set to one. When the resource usage is not zero and the resource reserve is less than the resource usage, divide the resource reserve by the resource usage to obtain the resource component gating value. The minimum value among the five resource component gating values ​​is selected as the performance conservation gating value of the candidate scheduling edge.

[0030] The performance conservation gating value is used to represent the transitivity strength of candidate scheduling edges under five resource constraints. A resource component gating value is generated for each resource. The resource component gating value is set to one when the resource occupancy is zero, one when the resource reserve meets the resource occupancy, and the resource component gating value is obtained by dividing the resource reserve by the resource occupancy when the resource reserve is less than the resource occupancy. The minimum value among the five resource component gating values ​​is used as the performance conservation gating value for the candidate scheduling edge, ensuring that the most strained resource controls the message transit strength of the candidate scheduling edge.

[0031] In this embodiment, the message weight correction of the performance conservation graph convolutional layer specifically includes: According to the order of the candidate scheduling edge set, read the sending node characteristics and receiving node characteristics connected to each candidate scheduling edge; The sending node features, receiving node features, and current time window graph convolution parameters are input into the message weight calculation process of the candidate scheduling edge to generate the original neighbor message weights of the candidate scheduling edge. The original neighbor message weights are multiplied by the performance conservation gate value to generate the corrected neighbor message weights; When the performance conservation gate value is one, the original neighbor message weights are retained; When the performance conservation threshold is less than one, the weight of the original neighbor messages is compressed according to the performance conservation threshold. The features of the sending node are weighted according to the weight of the modified neighbor message to generate the modified neighbor message for the candidate scheduling edge; All corrected neighbor messages connected to the same receiving node are accumulated to generate an aggregate message for the receiving node; The aggregated message of the receiving node is concatenated with the features of the receiving node, and the feature update calculation process corresponding to the convolution parameters of the current time window graph is input to generate the performance conservation node representation of the receiving node. For all candidate scheduling edges in the candidate scheduling edge set, the following steps are executed sequentially: reading the sending node feature, reading the receiving node feature, generating the original neighbor message weight, generating the corrected neighbor message weight, generating the corrected neighbor message, generating the receiving node aggregate message, and generating the receiving node performance conservation node representation, to obtain the performance conservation node representation corresponding to the node feature set.

[0032] The performance conservation graph convolutional layer performs message weight correction on candidate scheduling edges. The sending node features, receiving node features, and current time window graph convolution parameters of the candidate scheduling edge generate the original neighbor message weights. The performance conservation threshold value is multiplied by the original neighbor message weights to generate the corrected neighbor message weights. When resources are sufficient, the performance conservation threshold value is one, and the original neighbor message weights remain unchanged; when resources are insufficient, the performance conservation threshold value is less than one, and the original neighbor message weights are compressed. The corrected neighbor message weights participate in node feature aggregation to generate a performance conservation node representation.

[0033] In this embodiment, the operation of aggregating candidate scheduling edges according to resource ownership and calculating the total resource usage specifically includes: According to the arrangement order of the candidate scheduling edge set, the fulfillment resource occupation vector bound to each candidate scheduling edge is split to obtain the commodity inventory occupation, storage space capacity occupation, warehouse operation capacity occupation, logistics line carrying capacity occupation and delivery time occupation. For the same candidate scheduling edge, five resource occupancy records are generated. The first resource occupancy record uses the combination of product number and warehouse number as the product inventory resource identifier and carries the product inventory occupancy amount. The second resource occupancy record uses the warehouse location number as the warehouse location capacity resource identifier and carries the warehouse location capacity occupancy amount. The third resource occupancy record uses the combination of warehouse number and current time window as the warehouse operation resource identifier and carries the warehouse operation capacity occupancy amount. The fourth resource occupancy record uses the combination of logistics line number and current time window as the logistics line transportation resource identifier and carries the logistics line transportation capacity occupancy amount. The fifth resource occupancy record uses the delivery timeliness window as the delivery timeliness resource identifier and carries the delivery timeliness occupancy amount. Sort all resource usage records according to resource type and resource identifier; Merge resource occupancy records with the same resource type and the same resource identifier, and accumulate the occupancy amount in the merged resource occupancy records to generate the total occupancy amount of commodity inventory resources, total occupancy amount of storage space capacity resources, total occupancy amount of warehouse operation resources, total occupancy amount of logistics route transportation resources, and total occupancy amount of delivery timeliness resources.

[0034] The total resource usage is obtained by merging resource usage records. The fulfillment resource usage vector of each candidate scheduling edge is split into five types of resource usage and converted into five resource usage records. Each resource usage record carries the resource type, resource identifier, and usage amount. Resource usage records with the same resource type and resource identifier are merged, and the merged usage amounts are added together to obtain the corresponding total resource usage. The above process transforms the competition relationship of several candidate scheduling edges for the same resource into a computable total usage amount.

[0035] In this embodiment, the generation of the performance conflict residual matrix specifically includes: For the total amount of goods inventory resources occupied, the total amount of storage space capacity resources occupied, the total amount of warehouse operation resources occupied, the total amount of logistics route transportation resources occupied, and the total amount of delivery timeliness resources occupied, the resource surplus of the same type and the same resource identifier is matched from the fulfillment resource surplus vector according to the resource type and resource identifier; The resource balance corresponding to the same resource type and the same resource identifier is used as the resource balance of that resource identifier; Subtract the remaining resources from the total resource usage to generate the resource over-utilization difference; When the resource over-utilization difference is negative, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is zero, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is positive, the resource over-utilization difference is used as the corresponding resource conflict residual. Based on a fixed order of commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources, the inventory conflict residuals, storage location conflict residuals, warehouse operation conflict residuals, logistics route conflict residuals, and delivery timeliness conflict residuals are arranged into a fulfillment conflict residual matrix.

[0036] The fulfillment conflict residual matrix records the over-occupancy status of five resources within the current time window. The total resource occupancy is matched with the corresponding resource surplus in the fulfillment resource surplus vector, and then the difference is calculated. When the total resource occupancy does not exceed the resource surplus, the resource conflict residual is set to zero; when the total resource occupancy exceeds the resource surplus, the excess portion is written into the corresponding resource conflict residual. Inventory conflict residuals, storage location conflict residuals, warehouse operation conflict residuals, logistics route conflict residuals, and delivery timeliness conflict residuals are arranged in a fixed resource order to form the fulfillment conflict residual matrix.

[0037] In this embodiment, the operation of updating the current time window graph convolution parameters based on the performance conflict residual matrix includes: The fulfillment conflict residual matrix is ​​divided and summed in the order of inventory conflict residual, storage location conflict residual, warehouse operation conflict residual, logistics route conflict residual, and delivery timeliness conflict residual to generate the total inventory conflict, total storage location conflict, total warehouse operation conflict, total logistics route conflict, and total delivery timeliness conflict. Extract the resource surplus corresponding to the total amount of various resource conflicts from the fulfillment resource surplus vector to generate the total inventory surplus, total storage location surplus, total warehouse operation surplus, total logistics route surplus, and total delivery timeliness surplus. The total amount of various resource conflicts is compared with the total amount of corresponding resource reserves to generate parameter correction coefficients; The current time window graph convolution parameters are divided into five graph convolution parameter segments according to commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources; Compress the corresponding graph convolution parameter fragments based on the parameter correction coefficients to generate corrected graph convolution parameters; The modified graph convolution parameters are input into the message weight calculation process of the performance conservation graph convolution layer. The performance conservation node representation is updated using the modified graph convolution parameters to generate the performance scheduling state representation.

[0038] The residual-driven parameter evolution unit reads the performance conflict residual matrix, sums the residuals of the five types of resource conflicts separately, and generates the total amount of the five types of resource conflicts. The total amount of the five types of resource conflicts is compared with the corresponding total resource surplus to generate parameter correction coefficients. The graph convolution parameters of the current time window are divided into five graph convolution parameter segments according to the five resources. The parameter correction coefficients are used to compress the corresponding graph convolution parameter segments. After compression, corrected graph convolution parameters are formed. These corrected graph convolution parameters are re-inputted into the performance conservation graph convolution layer to update the performance conservation node representation and generate the performance scheduling state representation.

[0039] In this embodiment, the operation of generating joint warehousing and logistics scheduling results based on the fulfillment scheduling status representation specifically includes: Read the candidate scheduling edge weights between commodity nodes and storage location nodes, order nodes and warehouse nodes, and warehouse nodes and logistics route nodes from the fulfillment scheduling status representation; Based on the candidate scheduling edge weights, select the corresponding storage locations for the goods and generate the goods storage location adjustment results; Based on the performance conservation threshold and the candidate scheduling edge weights, the corresponding shipping warehouses for each order are selected, and the order warehouse allocation results are generated. Based on the commodity inventory resource identifier corresponding to the inventory conflict residual, the receiving warehouse is determined, and the warehouse with the available inventory balance of the same commodity being greater than the commodity inventory occupancy is determined as the sending warehouse, and the inventory transfer result is generated. Order nodes are arranged according to delivery timeliness and warehouse operation resource threshold values ​​to generate outbound priority results; The logistics routes corresponding to the warehouses are selected based on the threshold values ​​of the logistics route's transport resources, the threshold values ​​of the delivery timeliness resources, and the weights of the candidate scheduling edges, thus generating logistics route matching results.

[0040] The fulfillment scheduling status representation includes the weight information of candidate scheduling edges after evolution by fulfillment conservation gating and residual driving parameters. The weights of candidate scheduling edges between product nodes and storage location nodes are used to generate product storage location adjustment results; the weights of candidate scheduling edges between order nodes and warehouse nodes and the fulfillment conservation gating value are used to generate order distribution results; inventory conflict residuals and available inventory balances are used to generate inventory transfer results; delivery timeliness balances and warehouse operation resource component gating values ​​are used to generate outbound priority results; and the weights of candidate scheduling edges between warehouse nodes and logistics route nodes, the logistics route carrier resource component gating value, and the delivery timeliness resource component gating value are used to generate logistics route matching results.

[0041] Example 1: To verify the feasibility of this invention in practice, it was applied to an e-commerce product warehousing and logistics fulfillment scenario. This scenario includes multiple product categories, multiple warehouse distributions, multiple storage locations, multiple logistics routes, and fulfillment needs across multiple delivery areas. During operation, popular products are locked by multiple orders, resulting in insufficient storage capacity in some locations, some warehouses operating at near-capacity, and some logistics routes having insufficient capacity. Traditional rule-based scheduling and ordinary knowledge graph scheduling mainly rely on inventory balance, warehouse distance, route timeliness, and node connectivity for matching, which easily leads to scheduling results with insufficient fulfillment resources.

[0042] In this scenario, order fulfillment data, product inventory data, warehouse space occupancy data, warehouse operation capacity data, logistics route transportation data, and delivery timeliness data are collected within a continuous time window to form a basic fulfillment dataset. Then, a snapshot sequence of the fulfillment knowledge graph is constructed according to the continuous time window, converting orders, products, warehouses, warehouse spaces, logistics routes, and delivery areas into entity nodes, and writing the remaining available inventory of products, the remaining capacity of warehouse spaces, the remaining operation capacity of warehouses, the remaining transportation capacity of logistics routes, and the remaining delivery timeliness into the node feature set.

[0043] The improved EvolveGCN scheduling model receives the node feature set and the initial fulfillment edge set, generates a candidate scheduling edge set, and binds a fulfillment resource occupancy vector and a fulfillment resource surplus vector to each candidate scheduling edge. The fulfillment resource occupancy vector records the occupancy of commodity inventory, storage space capacity, warehouse operation capacity, logistics route carrying capacity, and delivery timeliness, while the fulfillment resource surplus vector records the corresponding resource surplus. The fulfillment conservation graph convolutional layer generates a fulfillment conservation gating value based on the fulfillment resource occupancy vector and the fulfillment resource surplus vector, and multiplies the fulfillment conservation gating value by the original neighbor message weights of the candidate scheduling edges to form modified neighbor message weights. Candidate scheduling edges with insufficient resources are compressed during the graph convolutional message passing stage, while candidate scheduling edges with sufficient resources retain their original neighbor message weights. Candidate scheduling edges generate resource occupancy records according to resource ownership relationships. Resource occupancy records with the same resource type and the same resource identifier are merged to form the total resource occupancy. The difference between the total resource occupancy and the corresponding resource surplus is calculated to form a fulfillment conflict residual matrix. The residual-driven parameter evolution unit generates parameter correction coefficients based on the performance conflict residual matrix and compresses the corresponding graph convolution parameter fragments in the current time window graph convolution parameters to obtain the corrected graph convolution parameters. The corrected graph convolution parameters continue to participate in the message weight calculation of the performance conservation graph convolution layer to generate a performance scheduling state representation.

[0044] The fulfillment scheduling status is used to generate joint warehousing and logistics scheduling results, including product location adjustment results, order distribution results, inventory transfer results, outbound priority results, and logistics route matching results. The traditional rule-based scheduling, ordinary knowledge graph scheduling, and the method of this invention are compared using the same fulfillment sample. The sample includes 12,000 orders, 1,600 types of goods, eight warehouses, 12,000 storage locations, 24 logistics routes, and 36 delivery areas. The comparison results are shown in the table below: Table 1: Comparison of Contract Performance Scheduling Effects

[0045] Analysis of the table data shows that the order fulfillment success rate under traditional rule-based scheduling is 91.8%, with resource conflict orders accounting for 11.9%, 186 instances of inventory over-occupancy, 143 instances of warehouse space over-occupancy, 97 instances of warehouse operation overload, 116 instances of logistics route overload, 132 instances of delivery timeliness conflicts, and an order splitting rate of 13.4%. These results indicate that when sorting only by inventory balance, warehouse distance, and route timeliness, the scheduling results are prone to resource competition among inventory, warehouse space, warehouse operations, logistics routes, and delivery timeliness in scenarios with multiple orders being fulfilled simultaneously, thus impacting the fulfillment success rate.

[0046] The order fulfillment success rate of ordinary knowledge graph scheduling increased to 94.2%, the proportion of resource conflict orders decreased to 7.8%, the average fulfillment time was shortened from 34.7 hours to 31.5 hours, and the average outbound waiting time was shortened from 5.8 hours to 4.6 hours. These results indicate that knowledge graphs can improve scheduling matching by utilizing the connections between orders, goods, warehouses, storage locations, and logistics routes. However, because fulfillment resource occupancy and remaining fulfillment resources are not directly written into the message weight calculation process, there are still 121 instances of inventory over-occupancy, 61 instances of warehouse operation overload, 74 instances of logistics route overload, and 85 instances of delivery timeliness conflicts.

[0047] The method of this invention achieves an order fulfillment success rate of 97.6%, which is 5.8 percentage points higher than traditional rule-based scheduling. The proportion of resource-conflicting orders decreases to 2.4%, a reduction of 9.5 percentage points compared to traditional rule-based scheduling. The number of times inventory is over-occupied decreases to 34 times, storage space over-occupied decreases to 27 times, warehouse operation overload decreases to 19 times, logistics route overload decreases to 21 times, and delivery timeliness conflicts decrease to 26 times. These results demonstrate that the fulfillment conservation gating value can compress message weights for candidate scheduling edges with insufficient resources based on the satisfaction levels of commodity inventory resources, storage space capacity resources, warehouse operation resources, logistics route carrying resources, and delivery timeliness resources, thereby weakening resource-insufficient paths during the graph convolution propagation stage.

[0048] Based on the continuous time window operation results, the recurring conflict rate of the method of this invention decreased to 1.7%, a reduction of 6.9 percentage points compared to traditional rule-based scheduling; the manual intervention rate of scheduling results decreased to 3.2%, a reduction of 8.9 percentage points compared to traditional rule-based scheduling; the warehouse location matching accuracy reached 96.1%, and the logistics route matching accuracy reached 95.4%. These results indicate that the fulfillment conflict residual matrix can record the over-occupancy difference formed when the same resource is jointly occupied by multiple candidate scheduling edges, and the residual-driven parameter evolution unit can reduce the recurrence of resource conflicts in subsequent continuous time windows based on the convolution parameters of the resource conflict residual correction graph.

[0049] Overall, the method of this invention outperforms traditional rule-based scheduling and ordinary knowledge graph scheduling in terms of order fulfillment success rate, average fulfillment time, proportion of resource conflicting orders, order splitting rate, warehouse location matching accuracy, and logistics route matching accuracy. This demonstrates that the present invention can improve the executability and fulfillment stability of the joint scheduling results for e-commerce commodity warehousing and logistics. The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based intelligent scheduling method for e-commerce commodity warehousing and logistics, characterized in that, Includes the following steps: Collect e-commerce warehousing and logistics fulfillment data within a continuous time window to generate a basic fulfillment dataset; A snapshot sequence of a performance knowledge graph is constructed from the performance base dataset, generating a set of node features and an initial set of performance edges; An improved EvolveGCN scheduling model is constructed, which includes a performance conservation graph convolutional layer and a residual-driven parameter evolution unit. The node feature set and the initial performance edge set are input into the improved EvolveGCN scheduling model to generate a candidate scheduling edge set, a performance resource occupancy vector, and a performance resource surplus vector. In the convolutional layer of the performance conservation graph, a performance conservation gate value is generated based on the performance resource occupancy vector and the performance resource surplus vector. The performance conservation gate value is multiplied by the original neighbor message weights of the candidate scheduling edges to generate modified neighbor message weights constrained by performance resources. The node feature set is aggregated according to the modified neighbor message weights to generate the performance conservation node representation. Candidate scheduling edges are collected according to resource ownership, the total resource occupancy is counted, and the difference between the total resource occupancy and the fulfillment resource surplus vector is calculated to generate the fulfillment conflict residual matrix. In the residual-driven parameter evolution unit, parameter correction coefficients are generated based on the performance conflict residual matrix. Based on the parameter correction coefficients, the corresponding graph convolution parameter fragments in the current time window graph convolution parameters are compressed to generate corrected graph convolution parameters. The corrected graph convolution parameters are then used to update the performance conservation node representation and generate the performance scheduling state representation. The joint scheduling results of warehousing and logistics are generated based on the performance scheduling status representation.

2. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 1, characterized in that, The e-commerce warehousing and logistics fulfillment data includes: Order fulfillment data, product inventory data, warehouse space occupancy data, warehouse operation capacity data, logistics route transportation data, and delivery timeliness data; Order fulfillment data records the order number, product number, and delivery area number; The product inventory data records the product number, warehouse number, and available inventory balance. The storage location occupancy data records the storage location number, warehouse number, and remaining storage location capacity. Warehouse operational capacity data records warehouse number and remaining warehouse operational capacity; The logistics route data records the logistics route number, warehouse number, delivery area number, and remaining transport capacity of the logistics route; The delivery time data records the order number and remaining delivery time.

3. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 2, characterized in that, The specific operations for constructing a snapshot sequence of the performance knowledge graph from the performance foundation dataset include: Group the basic dataset of contract performance according to continuous time windows to generate time window performance records; Extract order number, product number, warehouse number, storage location number, logistics route number, and delivery area number from the time window fulfillment record and generate entity nodes; The available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery time are written into the entity nodes to form a set of node features; Connect the order number and the product number to generate an order inclusion edge; connect the product number and the warehouse number to generate an inventory occupancy edge; connect the product number and the storage location number to generate a storage location occupancy edge; connect the warehouse number and the logistics route number to generate a route transportation edge; connect the order number and the delivery area number to generate a time constraint edge, thus forming the initial fulfillment edge set. The node feature set and the initial performance edge set are combined to form a performance knowledge graph snapshot. Several performance knowledge graph snapshots are arranged in the order of continuous time windows to generate a performance knowledge graph snapshot sequence.

4. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 3, characterized in that, The specific operations for generating the performance resource occupancy vector and the performance resource surplus vector include: Read the candidate scheduling edges in the candidate scheduling edge set, and determine the commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources associated with the candidate scheduling edges; Extract the inventory occupancy, storage space occupancy, warehouse operation capacity occupancy, logistics route carrying capacity occupancy, and delivery timeliness occupancy from e-commerce warehousing and logistics fulfillment data to generate a fulfillment resource occupancy vector; Extract the available inventory of goods, the remaining capacity of storage locations, the remaining operational capacity of warehouses, the remaining carrying capacity of logistics routes, and the remaining delivery timeliness to generate a fulfillment resource balance vector; Bind the fulfillment resource occupancy vector and the fulfillment resource surplus vector to the candidate scheduling edge.

5. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 4, characterized in that, The generation of the performance conservation threshold value specifically includes: Read the fulfillment resource occupancy vector and fulfillment resource surplus vector bound to the candidate scheduling edge; The five resources are: commodity inventory resources, warehouse capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources. The resource occupancy and resource surplus are matched one by one. When resource usage is zero, the resource component gating value is set to one. When the resource usage is not zero and the resource reserve is not less than the resource usage, the resource component gate value is set to one. When the resource usage is not zero and the resource reserve is less than the resource usage, divide the resource reserve by the resource usage to obtain the resource component gating value. The minimum value among the five resource component gating values ​​is selected as the performance conservation gating value of the candidate scheduling edge.

6. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 5, characterized in that, The message weight correction of the performance conservation graph convolutional layer specifically includes: According to the order of the candidate scheduling edge set, read the sending node characteristics and receiving node characteristics connected to each candidate scheduling edge; The sending node features, receiving node features, and current time window graph convolution parameters are input into the message weight calculation process of the candidate scheduling edge to generate the original neighbor message weights of the candidate scheduling edge. The original neighbor message weights are multiplied by the performance conservation gate value to generate the corrected neighbor message weights; When the performance conservation gate value is one, the original neighbor message weights are retained; When the performance conservation threshold is less than one, the weight of the original neighbor messages is compressed according to the performance conservation threshold. The features of the sending node are weighted according to the weight of the modified neighbor message to generate the modified neighbor message for the candidate scheduling edge; All corrected neighbor messages connected to the same receiving node are accumulated to generate an aggregate message for the receiving node; The aggregated message of the receiving node is concatenated with the features of the receiving node, and the feature update calculation process corresponding to the convolution parameters of the current time window graph is input to generate the performance conservation node representation of the receiving node. For all candidate scheduling edges in the candidate scheduling edge set, the following steps are executed sequentially: reading the sending node feature, reading the receiving node feature, generating the original neighbor message weight, generating the corrected neighbor message weight, generating the corrected neighbor message, generating the receiving node aggregate message, and generating the receiving node performance conservation node representation, to obtain the performance conservation node representation corresponding to the node feature set.

7. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 6, characterized in that, The operation of collecting candidate scheduling edges according to resource ownership and calculating the total resource usage specifically includes: According to the arrangement order of the candidate scheduling edge set, the fulfillment resource occupation vector bound to each candidate scheduling edge is split to obtain the commodity inventory occupation, storage space capacity occupation, warehouse operation capacity occupation, logistics line carrying capacity occupation and delivery time occupation. For the same candidate scheduling edge, five resource occupancy records are generated. The first resource occupancy record uses the combination of product number and warehouse number as the product inventory resource identifier and carries the product inventory occupancy amount. The second resource occupancy record uses the warehouse location number as the warehouse location capacity resource identifier and carries the warehouse location capacity occupancy amount. The third resource occupancy record uses the combination of warehouse number and current time window as the warehouse operation resource identifier and carries the warehouse operation capacity occupancy amount. The fourth resource occupancy record uses the combination of logistics line number and current time window as the logistics line transportation resource identifier and carries the logistics line transportation capacity occupancy amount. The fifth resource occupancy record uses the delivery timeliness window as the delivery timeliness resource identifier and carries the delivery timeliness occupancy amount. Sort all resource usage records according to resource type and resource identifier; Merge resource occupancy records with the same resource type and the same resource identifier, and accumulate the occupancy amount in the merged resource occupancy records to generate the total occupancy amount of commodity inventory resources, total occupancy amount of storage space capacity resources, total occupancy amount of warehouse operation resources, total occupancy amount of logistics route transportation resources, and total occupancy amount of delivery timeliness resources.

8. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 7, characterized in that, The generation of the performance conflict residual matrix specifically includes: For the total amount of goods inventory resources occupied, the total amount of storage space capacity resources occupied, the total amount of warehouse operation resources occupied, the total amount of logistics route transportation resources occupied, and the total amount of delivery timeliness resources occupied, the resource surplus of the same type and the same resource identifier is matched from the fulfillment resource surplus vector according to the resource type and resource identifier; The resource balance corresponding to the same resource type and the same resource identifier is used as the resource balance of that resource identifier; Subtract the remaining resources from the total resource usage to generate the resource over-utilization difference; When the resource over-utilization difference is negative, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is zero, the corresponding resource conflict residual is set to zero; When the resource over-utilization difference is positive, the resource over-utilization difference is used as the corresponding resource conflict residual. Based on a fixed order of commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources, the inventory conflict residuals, storage location conflict residuals, warehouse operation conflict residuals, logistics route conflict residuals, and delivery timeliness conflict residuals are arranged into a fulfillment conflict residual matrix.

9. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 8, characterized in that, The operation of updating the current time window graph convolution parameters based on the performance conflict residual matrix includes: The fulfillment conflict residual matrix is ​​divided and summed in the order of inventory conflict residual, storage location conflict residual, warehouse operation conflict residual, logistics route conflict residual, and delivery timeliness conflict residual to generate the total inventory conflict, total storage location conflict, total warehouse operation conflict, total logistics route conflict, and total delivery timeliness conflict. Extract the resource surplus corresponding to the total amount of various resource conflicts from the fulfillment resource surplus vector to generate the total inventory surplus, total storage location surplus, total warehouse operation surplus, total logistics route surplus, and total delivery timeliness surplus. The total amount of various resource conflicts is compared with the total amount of corresponding resource reserves to generate parameter correction coefficients; The current time window graph convolution parameters are divided into five graph convolution parameter segments according to commodity inventory resources, storage capacity resources, warehouse operation resources, logistics route transportation resources, and delivery timeliness resources; Compress the corresponding graph convolution parameter fragments based on the parameter correction coefficients to generate corrected graph convolution parameters; The modified graph convolution parameters are input into the message weight calculation process of the performance conservation graph convolution layer. The performance conservation node representation is updated using the modified graph convolution parameters to generate the performance scheduling state representation.

10. The intelligent scheduling method for e-commerce commodity warehousing and logistics based on knowledge graphs according to claim 9, characterized in that, The operation of generating joint warehousing and logistics scheduling results based on the performance scheduling status representation specifically includes: Read the candidate scheduling edge weights between commodity nodes and storage location nodes, order nodes and warehouse nodes, and warehouse nodes and logistics route nodes from the fulfillment scheduling status representation; Based on the candidate scheduling edge weights, select the corresponding storage locations for the goods and generate the goods storage location adjustment results; Based on the performance conservation threshold and the candidate scheduling edge weights, the corresponding shipping warehouses for each order are selected, and the order warehouse allocation results are generated. Based on the commodity inventory resource identifier corresponding to the inventory conflict residual, the receiving warehouse is determined, and the warehouse with the available inventory balance of the same commodity being greater than the commodity inventory occupancy is determined as the sending warehouse, and the inventory transfer result is generated. Order nodes are arranged according to delivery timeliness and warehouse operation resource threshold values ​​to generate outbound priority results; The logistics routes corresponding to the warehouses are selected based on the threshold values ​​of the logistics route's transport resources, the threshold values ​​of the delivery timeliness resources, and the weights of the candidate scheduling edges, thus generating logistics route matching results.

Images