Intelligent supply and demand matching dynamic optimization method, equipment and medium

Abstract

The invention discloses an intelligent supply and demand matching dynamic optimization method and device and a medium, and relates to the technical field. The method comprises the steps that demand side data and supply side data are collected, the demand side data comprise user interaction logs and real-time positions, and the supply side data comprise inventory data, logistics track data and production line load data; fusing demand side data and supply side data through a knowledge graph, inputting the demand side data and the supply side data into a graph neural network, encoding the demand side data and the supply side data into high-dimensional vectors, and generating a supply-demand matching preliminary screening candidate set; expanding a preliminary screening candidate set through a message passing mechanism of the graph neural network; and searching a non-dominated solution set by adopting a multi-objective optimization algorithm, and distributing index weights in real time so as to sort the expanded preliminary screening candidate sets. According to the method, real-time fusion of multi-source data is realized, a multi-target optimal solution is generated by using deep learning and operational research algorithms, and the strategy is continuously adjusted and optimized in combination with reinforcement learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent matching, in particular to an intelligent supply-demand matching dynamic optimization method, device and medium. BACKGROUND

[0002] In the field of fresh food e-commerce and next-day delivery, traditional supply-demand matching methods mainly rely on static rules or simple collaborative filtering algorithms. The nearest warehouse is fixedly matched according to the user area, and the matching logic is set based on historical sales data only, which cannot adapt to real-time demand fluctuations, such as users temporarily searching for low pesticide vegetables or supply-side changes; collaborative filtering algorithms mainly rely on user explicit behavior data.

[0003] On the one hand, static rules and simple collaborative filtering cannot dynamically integrate multi-source data: user real-time location, warehouse logistics track, farm production line load, resulting in coexistence of high inventory and unsalable goods and users unable to buy the required goods; on the other hand, single-objective optimization and lack of online learning mechanism make it difficult for the system to quickly adjust strategies as user behavior changes and supply fluctuations, ultimately causing high fresh food loss rate, high user churn rate and other problems.

[0004] Through the above analysis, the problems and defects of the prior art are:

[0005] The supply-demand matching method in the prior art lacks dynamic data fusion and implicit demand mining, resulting in supply-demand imbalance, coexistence of inventory accumulation and unmet user demand. SUMMARY

[0006] The embodiments of the present application provide an intelligent supply-demand matching dynamic optimization method, device and medium, which can solve the problem of lack of dynamic data fusion and implicit demand mining in the prior art supply-demand matching method, resulting in supply-demand imbalance, coexistence of inventory accumulation and unmet user demand.

[0007] In a first aspect, the embodiments of the present application provide an intelligent supply-demand matching dynamic optimization method, characterized in that the method comprises: collecting demand-side data and supply-side data, the demand-side data including user interaction logs, real-time location, and the supply-side data including inventory data, logistics track data, and production line load data; fusing the demand-side data and the supply-side data through a knowledge graph and inputting them into a graph neural network to encode them into high-dimensional vectors and generate a supply-demand matching preliminary screening candidate set; expanding the preliminary screening candidate set through a message passing mechanism of the graph neural network; and using a multi-objective optimization algorithm to search for a non-dominated solution set and real-time allocate index weights to sort the expanded preliminary screening candidate set.

[0008] In an implementation form of the present application, the demand side data and the supply side data are fused through a knowledge graph, specifically including: time and space alignment mapping of the demand side data and the supply side data; incremental learning mechanism is adopted, and subgraph reconstruction is triggered when new timestamp data is injected; abnormal data is identified and imported into an artificial review queue.

[0009] In an implementation form of the present application, and input into a graph neural network, encoded as a high-dimensional vector, to generate a supply-demand matching preliminary screening candidate set, specifically including: encoding the knowledge graph through a double-tower neural network, encoding the demand side data as a demand vector, and encoding the supply side data as a supply vector; calculating the cosine similarity of the demand vector and the supply vector and performing semantic matching, and calculating the high-dimensional vector similarity to generate a preliminary screening candidate set.

[0010] In an implementation form of the present application, the preliminary screening candidate set is expanded through the message passing mechanism of the graph neural network, specifically including: identifying action goods below a preset threshold in the user interaction log through the message passing mechanism of the graph neural network, and marking them as long-tail demand; adding the action goods to the preliminary screening candidate set to expand the coverage of the goods, so as to expand the preliminary screening candidate set.

[0011] In an implementation form of the present application, a multi-objective optimization algorithm is used to search for a non-dominated solution set, and the index weight is allocated in real time to sort the expanded preliminary screening candidate set, specifically including: defining a cost index, a timeliness index and a user satisfaction conflict index, and using a multi-objective optimization algorithm to screen a non-dominated solution set; dynamically allocating the weights of the cost index, the timeliness index and the user satisfaction conflict index according to the current business target, calculating the scores of the non-dominated solution set, and sorting; using the penalty function method to make the non-dominated solution set satisfy the inventory upper limit and the service radius limit condition.

[0012] In an implementation form of the present application, after using a multi-objective optimization algorithm to search for a non-dominated solution set, and allocating the index weight in real time to sort the expanded preliminary screening candidate set, the method further includes: grouping the user interaction log according to a preset frequency, calculating the KL divergence of the user features of each group and the baseline distribution; when the KL divergence of any subgroup exceeds a preset divergence threshold, triggering data pulling, feature reconstruction and shadow deployment verification.

[0013] In an implementation form of the present application, the abnormal data is identified and imported into an artificial review queue, specifically including: when the real-time location of the demand side data is missing, the IP address is located to supplement; abnormal delivery time in the logistics track data is identified through the Z-score algorithm, marked as suspicious data and triggered for artificial review.

[0014] In an implementation form of the application, the method further comprises: dividing the real-time location by a geographic fence to obtain a divided area; defining a user with a registration time less than a preset number of days and without interaction data as a cold start user; identifying a high-frequency commodity higher than a preset threshold in the user interaction log to obtain the high-frequency commodity; and extracting the high-frequency commodity in the divided area as an initial candidate set of the cold start user.

[0015] In a second aspect, the embodiments of the application further provide an intelligent supply-demand matching dynamic optimization device, which comprises at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: collect demand-side data and supply-side data, wherein the demand-side data comprises user interaction logs and real-time locations, and the supply-side data comprises inventory data, logistics track data and production line load data; fuse the demand-side data and the supply-side data by a knowledge graph, and input the fused data into a graph neural network to encode the data into a high-dimensional vector, thereby generating a supply-demand matching preliminary screening candidate set; expand the preliminary screening candidate set by a message passing mechanism of the graph neural network; and search a non-dominated solution set by a multi-objective optimization algorithm, and real-time assign index weights to sort the expanded preliminary screening candidate set.

[0016] In a third aspect, the embodiments of the application further provide an intelligent supply-demand matching dynamic optimization nonvolatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are configured to: collect demand-side data and supply-side data, wherein the demand-side data comprises user interaction logs and real-time locations, and the supply-side data comprises inventory data, logistics track data and production line load data; fuse the demand-side data and the supply-side data by a knowledge graph, and input the fused data into a graph neural network to encode the data into a high-dimensional vector, thereby generating a supply-demand matching preliminary screening candidate set; expand the preliminary screening candidate set by a message passing mechanism of the graph neural network; and search a non-dominated solution set by a multi-objective optimization algorithm, and real-time assign index weights to sort the expanded preliminary screening candidate set.

[0017] The intelligent supply-demand matching dynamic optimization method, device and medium provided by the embodiments of the application realize the deep correlation of explicit demand of users and implicit demand of users, social relationship chain preference, static data (inventory) of a supply end and dynamic data (logistics track) by real-time collection of multi-source data and fusion of a knowledge graph; the long-tail demand and social association demand missed by traditional methods are covered by expansion of a candidate set by a graph neural network; the conflict of cost, timeliness and user satisfaction is dynamically balanced by multi-objective optimization sorting and an online learning mechanism, and a penalty function constraint is combined to ensure inventory and service radius limitation; finally, the loss rate of fresh food and the user churn rate are reduced, the inventory turnover rate and user satisfaction are improved, and the supply-demand matching efficiency and operation benefit of fresh food e-commerce are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0019] Figure 1 A flowchart of an intelligent supply-demand matching dynamic optimization method provided for an embodiment of the application;

[0020] Figure 2 An internal structure schematic diagram of an intelligent supply-demand matching dynamic optimization device provided for an embodiment of the application. DETAILED DESCRIPTION

[0021] To make the objects, technical solutions and advantages of the application clearer, the technical solutions of the application will be described below in detail with reference to the embodiments of the application and the corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the application.

[0022] The embodiments of the application provide an intelligent supply-demand matching dynamic optimization method, device and medium, and solve the problem that the supply-demand matching method in the prior art lacks dynamic data fusion and implicit demand mining, resulting in a coexistence of supply-demand imbalance, inventory backlog and unmet user demand.

[0023] The technical solutions of the embodiments of the application will be described in detail below with reference to the drawings.

[0024] Figure 1 A flowchart of an intelligent supply-demand matching dynamic optimization method provided for an embodiment of the application. As shown in the figure, Figure 1 The intelligent supply-demand matching dynamic optimization method provided by the embodiments of the application specifically includes the following steps:

[0025] Step 10: Collecting demand-side data and supply-side data, the demand-side data including user interaction logs and real-time positions, and the supply-side data including inventory data, logistics track data and production line load data;

[0026] Step 20: Fusing the demand-side data and the supply-side data through a knowledge graph, and inputting the data into a graph neural network to encode into a high-dimensional vector and generate a supply-demand matching preliminary screening candidate set;

[0027] As an optional embodiment, the demand side data and the supply side data are fused through a knowledge graph, and specifically can include: step 201: time and space alignment mapping of the demand side data and the supply side data; step 202: incremental learning mechanism is adopted, and when new timestamp data is injected, subgraph reconstruction is triggered; step 203: abnormal data is identified and imported into an artificial review queue.

[0028] In this step, the demand side data is aligned with the supply side data according to the user ID and timestamp, the commodity SKU ID, the warehouse ID and the timestamp, and a knowledge graph is constructed through the user location-warehouse location-time window triplets; the entities of the knowledge graph include users, commodity SKUs, warehouses and farms, and the relationships include user-commodity preference, warehouse-commodity inventory, farm-warehouse supply and user-user social closeness.

[0029] As an optional embodiment, the abnormal data is identified and imported into an artificial review queue, and specifically can include: step 2031: when the real-time location of the demand side data is missing, the IP address is located to supplement; step 2032: the Z-score algorithm is used to identify abnormal delivery time in the logistics track data, and the abnormal data is marked as suspicious data and triggers artificial review.

[0030] In this step, the IP address location supplement solves the missing or abnormal problems caused by device failure or network delay in multi-source data collection, ensuring the accuracy of subsequent processing.

[0031] As an optional embodiment, the input graph neural network is encoded into a high-dimensional vector to generate a supply-demand matching preliminary screening candidate set, and specifically can include: step 204: the knowledge graph is encoded through a double-tower neural network, the demand side data is encoded into a demand vector, and the supply side data is encoded into a supply vector; step 205: the cosine similarity of the demand vector and the supply vector is calculated and semantic matching is performed, and the high-dimensional vector similarity is calculated to generate a preliminary screening candidate set.

[0032] In this step, the demand tower inputs user attributes such as age, region and behavior characteristics such as search keywords in the last 7 days, clicked commodity categories and high-frequency commodities of social friends, and outputs a demand vector; the supply tower inputs commodity attributes such as categories, freshness levels, supply capacity warehouse inventory, farm daily replenishment capacity, warehouse-to-user delivery time and commodity loss rate, and introduces a time decay factor, such as a weight of 1.2 for a picked commodity on the same day and a weight of 0.8 for a commodity on the next day, and outputs a supply vector; the cosine similarity of the demand vector and the supply vector is calculated, the text similarity of the user search words and the commodity labels is combined, such as the semantic similarity of low pesticide and organic is 0.9, the candidate pairs with a similarity greater than or equal to 0.8 are screened, and a preliminary screening candidate set is generated.

[0033] Step 30: The preliminary screening candidate set is expanded through the message passing mechanism of the graph neural network.

[0034] As an optional embodiment, the initial screening candidate set is expanded through the message passing mechanism of the graph neural network, which can specifically include the following steps: step 301: identifying action goods below a preset threshold in the user interaction log through the message passing mechanism of the graph neural network, and marking the action goods as long-tail demand; step 302: adding the action goods to the initial screening candidate set to expand the coverage of the goods and expand the initial screening candidate set.

[0035] In this step, the attention weight of the user to each good is calculated through the GAT attention mechanism of the graph neural network, weight = historical purchase times x a + search times x b + social friend purchase times x c, and the goods with an attention weight less than or equal to 0.2 are marked as long-tail demand; a user-user graph is constructed, the edge weight = number of common friends x 0.5 + number of common group purchase times x 0.5, and for a long-tail demand user, high-frequency goods with an attention weight greater than or equal to 0.6 in the social friends of the user are extracted and added to the candidate set.

[0036] Step 40: using a multi-objective optimization algorithm to search for a non-dominated solution set, and assigning weights to the indicators in real time to sort the expanded initial screening candidate set.

[0037] As an optional embodiment, a multi-objective optimization algorithm is used to search for a non-dominated solution set, and weights are assigned to the indicators in real time to sort the expanded initial screening candidate set, which can specifically include the following steps: step 401: defining a cost indicator, a timeliness indicator, and a user satisfaction conflict indicator, and using a multi-objective optimization algorithm to filter a non-dominated solution set; step 402: dynamically assigning weights to the cost indicator, the timeliness indicator, and the user satisfaction conflict indicator according to the current business target, calculating the scores of the non-dominated solution set, and sorting; step 403: using a penalty function method to make the non-dominated solution set satisfy the upper limit of inventory and the service radius limit condition.

[0038] In this step, the cost indicator is: logistics cost (distance from warehouse to user x unit transportation cost) + inventory holding cost (part of excess safety inventory x storage rate); the timeliness indicator is: order response time (from ordering to warehouse shipping time) + delivery time (from shipping to user signing time); and the user satisfaction indicator is: 1-(historical return rate + rate of negative reviews).

[0039] Further, the NSGA-II algorithm is used to search for a non-dominated solution set, 100 groups of candidate sorting strategies are randomly generated, each group of strategies corresponds to a combination of weights of each indicator; iterative optimization: generating offspring through crossover (swapping the weights of two groups of strategies) and mutation (randomly adjusting a weight ± 5%); calculating the cost, timeliness, and satisfaction values of the offspring; and screening non-dominated solutions: retaining strategies that are superior to other solutions in at least one indicator and are not completely dominated by other solutions.

[0040] Further, dynamically allocate index weights according to business objectives, such as prioritizing timeliness during big promotions, with a weight ratio of 60%; prioritizing cost during routine operations, with a weight ratio of 50%; and precise sorting: based on the strategy in the non-dominated solution set, calculate the comprehensive score of each candidate pair (score = cost x weight 1 + timeliness x weight 2 + satisfaction x weight 3) according to real-time weights, and sort in descending order of score.

[0041] Define constraints: for example, inventory upper limit: the demand of the candidate pair ≤ the real-time inventory of the warehouse; service radius: the distance between the user's location and the warehouse ≤ the maximum service radius of the warehouse. If the candidate pair violates any constraint, subtract the penalty term (penalty value = violation degree x penalty coefficient, such as deducting 10 x 0.5 if 10 items are overstocked) from the comprehensive score to ensure that the final sorting result only retains candidates that meet the constraints.

[0042] As an optional embodiment, after using a multi-objective optimization algorithm to search for a non-dominated solution set, real-time allocation of index weights, and sorting of the expanded preliminary candidate set, the method can further include: grouping user interaction logs according to a preset frequency, calculating the KL divergence of the user features of each group and the baseline distribution; when the KL divergence of any subgroup exceeds the preset divergence threshold, triggering data pulling, feature reconstruction, and shadow deployment verification.

[0043] In this step, for example, user traffic is allocated to the current main strategy, the newly trained strategy, and the historically optimal strategy in the ratio of 7:2:1, respectively labeled as A / B / C groups; for discrete features, such as the city where the user is located, calculate the probability difference sum of squares of the current distribution and the baseline distribution; for continuous features, such as the user's monthly consumption amount, use kernel density estimation to fit the probability distribution, and then calculate the KL divergence; through historical data statistics, take the KL divergence value at the 95% quantile as the threshold;

[0044] Feature pipeline reconstruction: including feature cleaning, feature transformation, taking the logarithm of consumption amount for normalization, feature selection, and filtering features with importance ≥ 0.2 through random forest; feature drift monitoring: real-time calculation of user features: age distribution, product preference, and baseline distribution, such as the KL divergence of the stable distribution at the beginning of online; when the KL divergence of any subgroup exceeds the threshold, trigger: data pulling: backtrack the user behavior data of the abnormal subgroup in the past 7 days; feature reconstruction: reselect key features, such as adding live streaming viewing time as a new feature; shadow deployment verification: train the strategy with new features in an offline environment to verify its matching effect, and replace the strategy of the abnormal subgroup.

[0045] As an optional embodiment, the method can further comprise: dividing the real-time location by a geo-fence to obtain a divided area; defining a user with a registration time less than a preset number of days and without interaction data as a cold-start user; identifying a high-frequency commodity higher than a preset threshold in the user interaction log to obtain a high-frequency commodity; and extracting the high-frequency commodity in the divided area as an initial candidate set of the cold-start user.

[0046] In this step, the generation of the candidate set for the new user without behavior data is solved by the regional group preference and weak social relationship migration.

[0047] The above is the method embodiment of the present application. Based on the same inventive concept, the present application also provides an intelligent supply-demand matching dynamic optimization device, the structure of which is shown in Figure 2 .

[0048] Figure 2 An intelligent supply-demand matching dynamic optimization device provided by an embodiment of the present application is shown in Figure 2 . The device comprises:

[0049] at least one processor 201;

[0050] and a memory 202 in communication connection with the at least one processor;

[0051] The memory 202 stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: collect demand-side data and supply-side data, the demand-side data comprising a user interaction log and a real-time location, and the supply-side data comprising inventory data, logistics track data and production line load data; fuse the demand-side data and the supply-side data by a knowledge graph and input into a graph neural network to encode into a high-dimensional vector to generate a supply-demand matching preliminary screening candidate set; expand the preliminary screening candidate set by a message passing mechanism of the graph neural network; and search a non-dominated solution set by a multi-objective optimization algorithm, and real-time assign an index weight to sort the expanded preliminary screening candidate set.

[0052] Some embodiments of the present application provide a corresponding Figure 1The application discloses a nonvolatile computer storage medium for intelligently matching supply and demand and dynamically optimizing the nonvolatile computer storage medium, and the nonvolatile computer storage medium stores computer executable instructions, and the computer executable instructions are configured to: collect demand side data and supply side data, the demand side data includes user interaction logs and real-time positions, and the supply side data includes inventory data, logistics track data and production line load data; the demand side data and the supply side data are fused through a knowledge graph, and are input into a graph neural network to be encoded into a high-dimensional vector to generate a supply and demand matching preliminary screening candidate set; the preliminary screening candidate set is expanded through a message passing mechanism of the graph neural network; and a multi-objective optimization algorithm is used to search for a non-dominated solution set, and index weights are allocated in real time to sort the expanded preliminary screening candidate set.

[0053] Each of the embodiments in the application is described in a progressive manner, and the same or similar parts of each of the embodiments can be referred to each other, and each of the embodiments mainly describes differences from other embodiments. In particular, for the Internet of Things device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the related parts can be referred to the part of the method embodiments.

[0054] The system and medium provided by the embodiments of the application are one-to-one corresponding to the method, so the system and medium also have similar beneficial technical effects to the method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be described here.

[0055] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0056] The application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks can be implemented by computer program instructions. The computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The device for implementing the functions specified in one block or multiple blocks.

[0057] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0058] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0059] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0060] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0061] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to computing devices. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0062] It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0063] The above description is merely illustrative of the application, and not restrictive. Various modifications and changes can become apparent to those skilled in the art. Incorporating any modification, equivalent substitution, improvement, etc. within the spirit and principle of the application, shall be included in the scope of the claims of the application.

Claims

1. An intelligent supply-demand matching dynamic optimization method, characterized in that, The method comprises: Collecting demand-side data and supply-side data, wherein the demand-side data comprises user interaction logs and real-time positions, and the supply-side data comprises inventory data, logistics track data, and production line load data; Fusing the demand-side data and the supply-side data through a knowledge graph, and inputting the data into a graph neural network to encode the data into a high-dimensional vector and generate a supply-demand matching preliminary screening candidate set; Expanding the preliminary screening candidate set through a message passing mechanism of the graph neural network; Using a multi-objective optimization algorithm to search for a non-dominated solution set, and real-time distribution of index weights to sort the expanded preliminary screening candidate set. 2.The intelligent supply-demand matching dynamic optimization method of claim 1, wherein, The fusing of the demand-side data and the supply-side data through the knowledge graph specifically comprises: Performing spatio-temporal alignment mapping on the demand-side data and the supply-side data; Using an incremental learning mechanism to trigger subgraph reconstruction when new timestamp data is injected; Identifying abnormal data and importing the data into an artificial review queue. 3.The intelligent supply-demand matching dynamic optimization method of claim 1, wherein, The inputting of the data into the graph neural network to encode the data into a high-dimensional vector and generate a supply-demand matching preliminary screening candidate set specifically comprises: Encoding the knowledge graph through a double-tower neural network, encoding the demand-side data into a demand vector, and encoding the supply-side data into a supply vector; Calculating the cosine similarity of the demand vector and the supply vector and performing semantic matching, calculating the high-dimensional vector similarity to generate a preliminary screening candidate set. 4.The intelligent supply-demand matching dynamic optimization method of claim 1, wherein, The expanding of the preliminary screening candidate set through the message passing mechanism of the graph neural network specifically comprises: Identifying action goods in the user interaction logs that are below a preset threshold through the message passing mechanism of the graph neural network, and marking the goods as long-tail demand; Adding the action goods to the preliminary screening candidate set to expand the coverage of goods, so as to expand the preliminary screening candidate set.

5. The intelligent supply-demand matching dynamic optimization method according to claim 1, characterized in that, The using of the multi-objective optimization algorithm to search for a non-dominated solution set, and real-time distribution of index weights to sort the expanded preliminary screening candidate set specifically comprises: Defining a cost index, a timeliness index, and a user satisfaction conflict index, and using a multi-objective optimization algorithm to screen the non-dominated solution set; Dynamically distributing the weights of the cost index, the timeliness index, and the user satisfaction conflict index according to the current business target, calculating the scores of the non-dominated solution set, and sorting the non-dominated solution set; Using a penalty function method to make the non-dominated solution set satisfy the upper limit of inventory and the service radius limit condition.

6. The intelligent supply-demand matching dynamic optimization method according to claim 1, characterized in that, After the using of the multi-objective optimization algorithm to search for a non-dominated solution set, and real-time distribution of index weights to sort the expanded preliminary screening candidate set, the method further comprises: Grouping the user interaction logs according to a preset frequency, and calculating the KL divergence of the user features of each group and the benchmark distribution; When the KL divergence of any subgroup exceeds a preset divergence threshold, triggering data pulling, feature reconstruction, and shadow deployment verification.

7. The intelligent supply-demand matching dynamic optimization method according to claim 2, characterized in that, The identifying of abnormal data and the importing of the data into an artificial review queue specifically comprises: When the real-time position of the demand-side data is missing, supplementing the position through an IP address; Identifying abnormal delivery time in the logistics track data through a Z-score algorithm, marking the data as suspicious data, and triggering artificial review. 8.The intelligent supply-demand matching dynamic optimization method of claim 1, wherein, The method further comprises: Dividing the real-time position through a geographic fence to obtain a divided area; Defining a user whose registration time is less than a preset number of days and who has no interaction data as a cold start user; Identify the action commodity in the user interaction log higher than the preset threshold, and obtain the high-frequency commodity; Extract the high-frequency commodity in the divided area as the initial candidate set of the cold start user.

9. An intelligent supply-demand matching dynamic optimization device, characterized in that, The device comprises: at least one processor; and a memory connected in communication with the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Collect demand side data and supply side data, the demand side data including user interaction logs, order history, real-time location, the supply side data including inventory data, logistics track data, production line load data; Fuse the demand side data and the supply side data through the knowledge graph, and input into the graph neural network to encode into a high-dimensional vector to generate a supply-demand matching preliminary screening candidate set; Expand the preliminary screening candidate set through the message passing mechanism of the graph neural network; Use a multi-objective optimization algorithm to search for a non-dominated solution set, and real-time assign index weights to sort the expanded preliminary screening candidate set.

10. A non-volatile computer storage medium for intelligent supply and demand matching and dynamic optimization, storing computer-executable instructions, characterized in that, The computer executable instructions are set to: Collect demand side data and supply side data, the demand side data including user interaction logs, order history, real-time location, the supply side data including inventory data, logistics track data, production line load data; Fuse the demand side data and the supply side data through the knowledge graph, and input into the graph neural network to encode into a high-dimensional vector to generate a supply-demand matching preliminary screening candidate set; Expand the preliminary screening candidate set through the message passing mechanism of the graph neural network; Use a multi-objective optimization algorithm to search for a non-dominated solution set, and real-time assign index weights to sort the expanded preliminary screening candidate set.

Images