Cross-border inventory dynamic balance method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of cross-border inventory dynamic balance, and discloses a cross-border inventory dynamic balance method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining risk data information in real time, determining a to-be-balanced overseas warehouse based on the risk data information, and storing the to-be-balanced overseas warehouse in the to-be-balanced overseas warehouse in the to-be-balanced overseas warehouse; and obtaining inventory information of each category of commodities in each to-be-balanced overseas warehouse, determining a constraint condition corresponding to each to-be-balanced overseas warehouse based on the inventory information and the risk data information, and inputting the constraint condition and the to-be-balanced overseas warehouse into a preset mixed integer programming model, and the inventory of each overseas warehouse to be balanced is balanced based on the optimization scheme. The method has the advantages that flexibility and response speed of inventory management are improved, the defect that dynamic adjustment cannot be performed according to real-time risk data in an existing method is overcome, and resource waste and stockout risks caused by static management are effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of cross-border inventory dynamic balancing technology, and in particular to a cross-border inventory dynamic balancing method, apparatus, electronic device and storage medium. Background Technology

[0002] In the current context of cross-border e-commerce, the complexity of the global market and the ever-changing risk factors (such as natural disasters, political instability, and trade policy adjustments) have placed higher demands on warehouse management.

[0003] Traditional inventory balancing methods rely primarily on manual settings and historical data analysis, failing to reflect timely market changes and real-time risk information. This static management model results in a lack of flexibility in inventory allocation and transfer decisions, making it ineffective in adapting to demand fluctuations in emerging markets. For example, when faced with sudden port strikes or extreme weather conditions, warehouse management relying on static models may miss the optimal time for rapid adjustments, leading to insufficient or excessive inventory reserves, further increasing operating costs and the risk of customer churn. Furthermore, traditional methods often fail to effectively assess the time sensitivity of different products and the demand differences between commodities and fast-moving consumer goods, resulting in resource waste and low management efficiency. Summary of the Invention

[0004] Based on this, it is necessary to propose a method, device, electronic equipment and storage medium for dynamic balancing of cross-border inventory to address the existing problem of dynamic balancing of cross-border inventory.

[0005] A method for dynamic balancing of cross-border inventory, the method comprising: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

[0006] Furthermore, before the step of determining the overseas warehouses to be balanced based on the risk data information, the method further includes: The risk data information is input into a preset risk assessment model to obtain a target risk index; Determine whether the target risk index is greater than the risk threshold; If the target risk index is greater than the risk threshold, then the step of determining the overseas warehouses to be balanced based on the risk data information is satisfied.

[0007] Furthermore, after the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimization scheme, the method further includes: Extract the dimensional information of each dimension from the optimization scheme; The information from each dimension is converted into gene segments and then encoded into chromosomes to obtain the initial chromosome; The initial chromosomes are optimized based on various dimensions to generate a Pareto front, thereby obtaining the first chromosome set. Perform a crossover mutation operation on the first chromosome set to obtain a second chromosome set; The second chromosome set is combined with the first chromosome set to generate a third chromosome set; Using the third chromosome set as the first chromosome set, the target step and the steps following the target step are repeated until the number of chromosomes in the third chromosome set is greater than or equal to a preset number; the target step is to perform a crossover mutation operation on the first chromosome set to obtain the second chromosome set. The third chromosome set whose number of chromosomes is greater than or equal to the preset number is denoted as the target third chromosome set; The fitness value of each chromosome in the target third chromosome set is calculated using a preset fitness calculation function; The chromosome with the highest fitness value is selected as the target optimization scheme. The inventory of each of the overseas warehouses to be balanced is balanced based on the target optimization scheme.

[0008] Furthermore, after the step of balancing the inventory of each of the overseas warehouses to be balanced based on the optimization scheme, the method further includes: The optimization scheme and each of the overseas warehouses to be balanced are input into a preset natural language processing model to generate sub-optimization schemes for each of the overseas warehouses to be balanced. Each of the aforementioned sub-optimization schemes is distributed to the corresponding overseas warehouses to be balanced.

[0009] Furthermore, the step of acquiring risk data information in real time includes: Multiple data sources are crawled in real time by a pre-set distributed crawler cluster to obtain multiple raw information. The original information is translated into text in a specified language using a preset model to obtain the original text data; The original text data is used to extract entities and the relationships between entities to obtain risk data information.

[0010] Furthermore, before the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimization solution, the method further includes: Multiple sets of local training data were obtained from various overseas repositories; one set of local training data included local risk data information and local optimization schemes. Based on multiple sets of local training data corresponding to each overseas warehouse, train the corresponding preset local models respectively; The training parameters of each preset local model are obtained and input into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

[0011] Furthermore, before the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimization solution, the method further includes: Acquire geopolitical and market data from multiple countries; By analyzing the geopolitical and market data of various countries using a pre-defined semantic model, the analyzed information data is obtained. Each of the aforementioned information data is input into a preset mixed integer programming initial model to obtain a preset mixed integer programming model.

[0012] A cross-border inventory dynamic balancing device, the device comprising: The first acquisition module is used to acquire risk data information in real time; The first determining module is used to determine the overseas warehouses to be balanced based on the risk data information; The second acquisition module is used to acquire inventory information of each category of goods in each of the overseas warehouses to be balanced; The second determining module is used to determine the constraints corresponding to each of the overseas warehouses to be balanced based on the inventory information and the risk data information. The input module is used to input the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization scheme; The balancing module is used to balance the inventory of each of the overseas warehouses to be balanced based on the optimization scheme.

[0013] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

[0014] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

[0015] The beneficial technical effects of this invention are as follows: By acquiring inventory information from each warehouse and calculating corresponding constraints, combined with the optimization scheme of a mixed integer programming model, enterprises can quickly make scientific decisions in complex supply chain environments, thereby optimizing inventory allocation and improving the flexibility and response speed of inventory management. This solves the shortcomings of existing methods that cannot be dynamically adjusted based on real-time risk data, and effectively reduces resource waste and stockout risks caused by static management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] in: Figure 1 This is a diagram illustrating the application environment of a cross-border inventory dynamic balancing method in one embodiment. Figure 2 Here is a flowchart of a cross-border inventory dynamic balancing method in one embodiment; Figure 3 This is a structural block diagram of a cross-border inventory dynamic balancing device in one embodiment; Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Figure 1 This is a diagram illustrating a cross-border inventory dynamic balancing application environment in one embodiment. (Refer to...) Figure 1 This cross-border inventory dynamic balancing method is applied to a cross-border inventory dynamic balancing system. The system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to acquire risk data, and the server 120 is used to generate optimization plans.

[0020] like Figure 2 As shown, in one embodiment, a method for dynamic balancing of cross-border inventory is provided. This method can be applied to both terminals and servers; this embodiment illustrates its application to terminals. The specific steps of this dynamic balancing method for cross-border inventory include: S1: Real-time acquisition of risk data and information; S2: Determine the overseas warehouses to be balanced based on the aforementioned risk data information; S3: Obtain inventory information for each category of goods in each of the overseas warehouses to be balanced; S4: Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced; S5: Input the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization scheme; S6: Balance the inventory of each of the overseas warehouses to be balanced based on the optimization scheme.

[0021] As described in step S1 above, risk data information is acquired in real time. Sources of risk data information can include dynamic discussions on social media (such as Twitter, Facebook, etc.), real-time updates from news websites, customs and government announcements, weather forecasts, and the transportation status of logistics companies. Automated data scraping tools and application programming interfaces (APIs) can be used to monitor various risk signals in real time. For example, when a port is detected to be closed due to a natural disaster, the system can immediately capture the relevant information.

[0022] As described in step S2 above, overseas warehouses to be balanced are determined based on the risk data information. After acquiring real-time risk data, in-depth analysis of the risk data information ensures that decision-makers can accurately identify affected warehouses. Typically, companies develop standards based on different risk factors (such as traffic closures, port strikes, policy changes, etc.) to determine which warehouse's risk level exceeds acceptable limits. For example, if a port in a region experiences shipping disruptions due to a strike, the warehouse's receiving and shipping capabilities will be severely affected. In this case, the warehouse will be marked as requiring inventory balancing. In a specific embodiment, a dynamic risk assessment model can be established, and the risk data information can be input into this model to identify warehouses facing potential risks, which are then marked as overseas warehouses to be balanced.

[0023] As described in step S3 above, the inventory information of each category of goods in each of the overseas warehouses to be balanced is obtained. After the overseas warehouses to be balanced are determined, the inventory information of each category of goods in these warehouses is obtained. Each overseas warehouse has its own inventory management system, which can reflect the inventory quantity, storage status and inventory turnover of each type of goods in each warehouse in real time, and therefore can be obtained directly.

[0024] As described in step S4 above, constraints are determined for each of the overseas warehouses to be balanced based on the inventory information and the risk data. After obtaining the inventory information of the overseas warehouses to be balanced, the corresponding constraints are determined based on this information and real-time risk data. The constraints include the warehouse's maximum storage capacity, available transportation methods, the storage characteristics of the goods (such as temperature control requirements), sales expectations in the target market, and relevant laws and regulations. In addition, special constraints caused by risk events must also be considered, such as a reduction in processing capacity or a sudden change in transportation costs in a warehouse due to a strike.

[0025] As described in step S5 above, the constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming (MIP) model to obtain an optimization scheme. This information is then input into a preset MIP model to solve for the optimal inventory allocation scheme. MIP models are typically used to handle linear and nonlinear optimization problems; they can automatically calculate the optimal inventory allocation scheme based on different constraints. By integrating previously acquired inventory information, dynamic risk data, and constraints, and setting corresponding objective functions (such as cost minimization, service level maximization, etc.), feasible solutions are generated. In a specific embodiment, a cost-optimal solution is obtained by setting the objective function to cost minimization. Generally, a pre-defined mixed-integer nonlinear programming model is established using Pyomo or Gekko in Python, combining information from various overseas warehouses and transportation data to create an initial framework. Then, based on minimizing transportation costs, the model is transformed into a mathematical form, including only the objective function. Next, the variable type is selected, distinguishing between continuous and integer variables, and a solver is chosen. Different solvers support different algorithms and performance levels, such as Bonmin and Baron, which can be selected according to the actual situation. This completes the construction of the pre-defined mixed-integer nonlinear programming model. After obtaining the constraints, a complete objective mixed-integer nonlinear programming model is formed. This complete objective mixed-integer nonlinear programming model is used to generate optimization schemes for the overseas warehouses to be balanced. It should be noted that the warehouse information of the overseas warehouses to be balanced is pre-input into the initial framework, including their geographical location and overall inventory capacity. Therefore, simply inputting the name of the overseas warehouse to be balanced will yield the corresponding optimization scheme.

[0026] As described in step S6 above, the inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme. Based on the obtained optimization scheme, the inventory of each overseas warehouse to be balanced is actually adjusted and balanced. This process involves specific operations performed by each overseas warehouse, including the transfer of goods, adjustment of inventory records, and updating of logistics routes. The generated optimization scheme indicates which warehouses need replenishment, which warehouses need to reduce inventory, and the specific handling routes and methods. Enterprises can utilize automated warehousing equipment and transportation tools to efficiently allocate goods based on the optimization results. Furthermore, to ensure smooth implementation, enterprises should establish a comprehensive execution monitoring system to track and supervise the execution process in real time, so as to promptly address any potential problems. This can reduce inventory costs, improve supply chain responsiveness, ensure timely fulfillment of market demands, especially when facing dynamically changing emerging markets, enhance warehouse flexibility and adaptability, and achieve the goal of dynamic adjustment and balancing.

[0027] In one embodiment, before step S2 of determining the overseas warehouses to be balanced based on the risk data information, the method further includes: S101: Input the risk data information into a preset risk assessment model to obtain the target risk index; S102: Determine whether the target risk index is greater than the risk threshold; S103: If the target risk index is greater than the risk threshold, it is determined that the step of determining the overseas warehouse to be balanced based on the risk data information is satisfied.

[0028] As described in steps S101-S103 above, the real-time acquired risk data is input into a pre-designed risk assessment model to calculate the target risk index. The risk assessment model can be a statistical analysis model, machine learning algorithm, or probability theory, etc. The core function of this model is to integrate and analyze multi-dimensional risk factors to generate a quantitative indicator that intuitively reflects the level of risk in the current environment. After obtaining the target risk index, it is determined whether this target risk index exceeds a preset risk threshold. The risk threshold is a critical value set according to the company's management strategy and risk tolerance, used to distinguish whether the current risk is within an acceptable range. When the target risk index is higher than the threshold, it indicates that the current market environment or operating conditions are relatively dangerous and may affect the company's normal operation or profits, requiring warehouse adjustments. Conversely, if the target risk index is lower than the threshold, it indicates that the current situation is stable, the risk is within a controllable range, and no adjustments are needed.

[0029] This section emphasizes the necessity of shifting from a reactive risk response process to proactive risk management. Clear assessments enable companies to make rapid decisions and respond promptly to unpredictable risk events, minimizing potential losses. Therefore, a company's supply chain management should not only rely on pre-set inventory and resource allocation plans but also possess flexible response capabilities. Furthermore, an online update mechanism for the MIP model can be set up to detect the magnitude of changes in the target risk index. When the magnitude of the change in the target risk index exceeds 10%, the MIP model is retrained to update it in real time.

[0030] In one embodiment, after step S5, which involves inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimized solution, the method further includes: S601: Extract the dimensional information of each dimension in the optimization scheme; S602: Convert information from each dimension into gene segments and encode them into chromosomes to obtain the initial chromosome; S603: Optimize the initial chromosomes based on various dimensions to generate a Pareto front, thereby obtaining the first chromosome set; S604: Perform a crossover mutation operation on the first chromosome set to obtain a second chromosome set; S605: Combine the second chromosome set with the first chromosome set to generate a third chromosome set; S606: Using the third chromosome set as the first chromosome set, repeat the target step and the steps following the target step until the number of chromosomes in the third chromosome set is greater than or equal to a preset number; the target step is to perform a crossover mutation operation on the first chromosome set to obtain the second chromosome set. S607: The third chromosome set with a number of chromosomes greater than or equal to a preset number is denoted as the target third chromosome set; S608: Calculate the fitness value of each chromosome in the target third chromosome set using a preset fitness calculation function; S609: Select the chromosome with the highest fitness value as the target optimization scheme; S610: Balance the inventory of each of the overseas warehouses to be balanced based on the target optimization scheme.

[0031] As described in steps S601-S603 above, after obtaining the optimized solution by executing the mixed-integer programming model, information from each dimension of the solution is extracted. This dimensional information typically includes multiple key parameters, such as inventory allocation quantity, warehouse location, transportation method, cost, and timeliness. By extracting this information, the system can comprehensively understand the composition of the optimized solution, laying the foundation for subsequent genetic algorithm processing. Generally, the optimized solution is represented in the form of a data table or structured array, where each dimension has a corresponding identifier and value, thus allowing direct semantic analysis extraction. In the genetic algorithm, chromosomes represent candidate solutions to the solution, and their composition is achieved through the combination of gene segments. Each gene segment corresponds to one or more dimensional information, such as the allocation quantity of a warehouse, transportation route, cost parameters, etc. From the extracted dimensional information, each parameter is encoded into a gene segment, for example, using binary encoding, decimal encoding, etc. The proportional parameters can be set according to actual needs, and specific coding rules can be pre-defined. A correspondence between dimensional information and specific codes can be established. When dimensional information is obtained, the corresponding specific code can be retrieved based on this correspondence. For example, a chromosome may include three gene segments: gene segment 1 (8 bits): warehouse selection code (e.g., 010 represents a hamburger warehouse); gene segment 2 (16 bits): inventory allocation quantity (quantified in thousands of units); gene segment 3 (4 bits): transportation mode combination (e.g., 1011 represents sea freight + rail + air freight emergency). Through these codes, decision-making schemes can be expressed in a clearly structured way, allowing the subsequent selection, crossover, and mutation processes of the genetic algorithm to proceed smoothly. The Pareto front is an optimization method that emphasizes finding the optimal trade-off between multiple objectives, ensuring that different solutions are superior to other solutions in at least one objective. The optimization process typically uses an objective function to evaluate the score of each chromosome on multidimensional objectives. For example, a chromosome may perform well in terms of cost but poorly in terms of timeliness. By calculating the score of each chromosome on each objective, non-dominated solutions are identified, i.e., a solution is superior to another solution in some dimensions but not inferior in others. Then, the Pareto front is constructed from these non-dominated solutions, forming the first chromosome set.

[0032] As described in steps S604-S607 above, after generating the first chromosome set, crossover and mutation operations are performed on these chromosomes to generate more diverse new solutions (the second chromosome set). Crossover is a mechanism in genetic algorithms that combines the genetic information of two "parent" chromosomes to generate one or more "child" chromosomes, thereby passing on desirable traits. Common crossover methods include single-point crossover and double-point crossover. This process increases the possible combinations of solutions and expands the search space. Mutation is a stochastic strategy used to introduce new gene combinations to prevent the algorithm from getting stuck in local optima. During optimization, some chromosomes undergo small-scale changes (such as rotation or value modification) to explore previously unexplored areas, increasing the diversity of solutions. After crossover and mutation operations, a second chromosome set is generated. This second chromosome set is then merged with the first chromosome set to generate a third chromosome set. This third chromosome set is then used as the first chromosome set, and the target step and subsequent steps are repeated until the number of chromosomes in the third chromosome set is greater than or equal to a preset number. This genetic algorithm process is repeated. In each iteration, the system performs repeated crossover and mutation operations, generating a new second chromosome set for the new first chromosome set, and thus a new third chromosome set. This process continues until the number of chromosome sets reaches or exceeds the preset number. When the number of chromosome sets reaches or exceeds the preset number, this chromosome set is defined as the target third chromosome set, ensuring both diversity and quality of the proposed solutions.

[0033] As described in steps S608-S610 above, after generating the target third chromosome set, the fitness value of each chromosome is evaluated using a preset fitness calculation function. The fitness value quantifies the merits of each solution under the target conditions, assigning a score based on set evaluation indicators. The fitness calculation function considers multiple optimization objectives, such as inventory cost, transportation timeliness, and service capacity. Specifically, cost fitness: F1 = 1 / (actual cost / budget ceiling), where actual cost = transportation cost + warehousing holding cost + stockout penalty cost; timeliness fitness: F2 = 1 / (maximum delay days / contract allowable days), where the maximum delay days are the 95th percentile delay value calculated from historical logistics data. The fitness calculation function is F = aF1 + bF2, thus calculating the overall fitness, where a and b are preset weights. The fitness value provides a clear priority for decision-making; the highest fitness value indicates that the solution corresponding to that chromosome is in the optimal state in terms of cost, timeliness, flexibility, and other objectives. The system can quickly identify the optimal solution and clearly mark it as the target optimization scheme. Based on the obtained target optimization scheme, inventory adjustments are made to the previously identified and unbalanced overseas warehouses.

[0034] In one embodiment, after step S6 of balancing the inventory of each of the overseas warehouses to be balanced based on the optimization scheme, the method further includes: S701: Input the optimization scheme and each of the overseas warehouses to be balanced into a preset natural language processing model to generate sub-optimization schemes for each of the overseas warehouses to be balanced; S702: Distribute each of the sub-optimization schemes to the corresponding overseas warehouses to be balanced.

[0035] As described in steps S701-S702 above, after completing the inventory allocation of overseas warehouses to be balanced, the optimization plan and information of each warehouse are input into a preset Natural Language Processing (NLP) model to generate specific sub-optimization plans. The NLP model can be a GPT model or a deepseek model, converting data into easily understandable and executable natural language instructions, ensuring that relevant personnel and systems can clearly understand the inventory needs and allocation instructions for each warehouse. The NLP model can parse and transform parameters in the optimization plan (such as allocation quantity, warehouse name, transportation method, time limit, etc.) to form intuitive and clear instructions. For example, for an allocation instruction for a warehouse, the NLP model can generate the following instruction: "Transfer 500 electronic products from the Hong Kong warehouse to the Los Angeles warehouse, expected to arrive within 3 days." These plans are then distributed to the corresponding warehouses. This process involves not only information transmission but also ensuring that each warehouse receives its specific allocation instructions in a timely and accurate manner. Specifically, the generated sub-optimization plans can be sent to their respective warehouse management teams via email, real-time messaging tools (such as Slack and Teams), or a dedicated warehouse management system (WMS) interface. This ensures the timeliness and accuracy of information, avoiding operational errors due to information delays or misunderstandings.

[0036] In one embodiment, step S1 of acquiring risk data information in real time includes: S111: Multiple data sources are crawled in real time through a pre-set distributed crawler cluster to obtain multiple raw information. S112: Translate the original information into text in a specified language using a preset model to obtain the original text data; S113: Extract entities and relationships between entities from the original text data to obtain risk data information.

[0037] As described in steps S111-S113 above, multiple data sources are crawled using a pre-defined distributed crawler cluster. In practice, the crawler cluster is configured to handle different data sources (such as specific websites, API interfaces, or social media platforms), and crawling tasks are managed and allocated through a scheduling framework (such as Apache Kafka or Hadoop). The crawler retrieves updated information at specific time intervals to ensure real-time changes are captured. Furthermore, each crawled data source is assigned a timestamp, and the closest data source is used for each extraction of relationships to avoid data confusion. The obtained raw information is stored in the form of unstructured text, including articles, posts, and comments. After acquiring a large amount of raw information, a pre-defined translation model translates this information into text in a specific language to obtain analyzable raw text data, ensuring the accessibility and readability of data processing. The translation model can be designed and trained based on deep learning and natural language processing technologies (such as Transformer and BERT) to handle conversions between multiple languages ​​(such as English, Spanish, and German). After translating the original text data, entities and their relationships are extracted to obtain valuable risk data. During entity extraction, entity-relationship triples can be extracted using the BERT-BiLSTM model. In practice, Named Entity Recognition (NER) and Relation Extraction methods from Natural Language Processing (NLP) are typically employed. NER identifies the parts of the text that refer to entities, such as company names, locations, dates, and events. Relation Extraction analyzes the relationships between these entities, such as the connection between a strike at a certain location and the operational impact on a specific company.

[0038] In one embodiment, before step S5, which involves inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimized solution, the method further includes: S401: Obtain multiple sets of local training data from various overseas repositories; each set of local training data includes local risk data information and local optimization schemes. S402: Train the corresponding preset local model based on the multiple sets of local training data corresponding to each overseas warehouse; S403: Obtain the training parameters of each preset local model and input them into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

[0039] As described in steps S401-S403 above, multiple sets of local training data tailored to the specific market environment of each overseas warehouse are obtained. This local training data covers the unique risk information faced by each warehouse during operation, as well as previous optimization strategies. Specifically, local risk data may include historical order data, transportation time, inventory turnover rate, response to market changes, and the impact of weather and policy changes. Furthermore, local optimization strategies involve how each warehouse has historically handled these risks to achieve optimal inventory management and resource allocation. By acquiring and merging this data, the system can ensure that the labels and input features during model training are closely related to actual preprocessing requirements. To ensure that each model fully understands the characteristics of its market environment and can provide accurate inventory optimization suggestions based on historical data, the training process typically includes data preprocessing, feature selection, model selection, and hyperparameter tuning. First, the raw data is cleaned and standardized to remove noise and redundant information, transforming it into a format suitable for model input. Then, based on the warehouse's needs and data characteristics, an appropriate algorithm (such as random forest, support vector machine, or neural network) is selected for effective training. After completing the local model training for each overseas warehouse, the training parameters obtained from these models are extracted and input into the preset mixed integer programming (MIP) initial model. The overall model (i.e. the preset mixed integer programming (MIP) initial model) is optimized through knowledge distillation to obtain the preset mixed integer programming model.

[0040] In one embodiment, before step S5, which involves inputting the constraints and the overseas warehouses to be balanced into a preset mixed-integer programming model to obtain an optimized solution, the method further includes: S411: Obtain geopolitical and market data from multiple countries; S412: Analyze the geopolitical and market data of various countries using a pre-defined semantic model to obtain the analyzed information data; S413: Input each of the aforementioned information data into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

[0041] As described in steps S411-S413 above, geopolitical and market data from multiple countries are acquired. Geopolitical data typically involves factors such as diplomatic relations, trade policies, economic sanctions, and regime stability between countries. Market data includes information on changes in consumer demand, price fluctuations, inventory levels, and supplier performance, all of which are crucial for inventory balancing decisions. Data sources can include market research reports, data released by governments and international organizations, news reports, and social media monitoring. In practice, companies can utilize web scraping technology to automatically extract the required information from public or authorized data sources. A pre-defined semantic model is then used to parse this data, transforming the raw, unstructured information into structured data for subsequent analysis and decision-making. After obtaining and parsing the geopolitical and market data from various countries, this information is input into a pre-defined mixed-integer programming (MIP) initial model to generate a complete pre-defined MIP model. During implementation, the input information typically includes multiple data items such as the parsed specific national situation, target risk index, and market demand expectations. This data needs to be mapped to the model's decision variables, objective function, and constraints. For example, changes in market demand in a particular country will affect inventory allocation decisions, while geopolitical risks may become constraints, limiting the allocation of certain resources. The system will structure this information through transformation rules, mapping it to each constraint. This allows for more effective use of local market information and time-sensitive geopolitical data during dynamic inventory management, thereby enhancing overall operational flexibility and resilience.

[0042] Reference Figure 3 The present invention also provides a cross-border inventory dynamic balancing device, the device comprising: The first acquisition module 902 is used to acquire risk data information in real time; The first determining module 904 is used to determine the overseas warehouses to be balanced based on the risk data information; The second acquisition module 906 is used to acquire inventory information of each category of goods in each of the overseas warehouses to be balanced; The second determining module 908 is used to determine the constraints corresponding to each of the overseas warehouses to be balanced based on the inventory information and the risk data information. Input module 910 is used to input the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization scheme; The balancing module 912 is used to balance the inventory of each of the overseas warehouses to be balanced based on the optimization scheme.

[0043] In one embodiment, the cross-border inventory dynamic balancing device further includes: The risk data information input module is used to input the risk data information into a preset risk assessment model to obtain a target risk index; The target risk index determination module is used to determine whether the target risk index is greater than a risk threshold. The determination module is used to determine whether the target risk index is greater than the risk threshold, and thus determine whether the step of determining the overseas warehouse to be balanced based on the risk data information is met.

[0044] In one embodiment, the cross-border inventory dynamic balancing device further includes: The dimension information extraction module is used to extract the dimension information of each dimension in the optimization scheme; The gene segment conversion module is used to convert information from various dimensions into gene segments and encode them into chromosomes to obtain the initial chromosome. The chromosome optimization module is used to optimize the initial chromosome based on various dimensions to generate a Pareto front, thereby obtaining the first chromosome set; The mutation module is used to perform a crossover mutation operation on the first chromosome set to obtain a second chromosome set; The combination module is used to combine the second chromosome set with the first chromosome set to generate a third chromosome set; An iterative module is used to repeatedly execute the target step and the steps following the target step, using the third chromosome set as the first chromosome set, until the number of chromosomes in the third chromosome set is greater than or equal to a preset number; the target step is to perform a crossover mutation operation on the first chromosome set to obtain the second chromosome set; The tagging module is used to mark the third chromosome set whose number of chromosomes is greater than or equal to a preset number as the target third chromosome set; The fitness value calculation module is used to calculate the fitness value of each chromosome in the target third chromosome set through a preset fitness calculation function; The chromosome selection module is used to select the chromosome with the highest fitness value as the target optimization scheme; The inventory balancing module is used to balance the inventory of each of the overseas warehouses to be balanced based on the target optimization scheme.

[0045] In one embodiment, the cross-border inventory dynamic balancing device further includes: The sub-optimization scheme generation module is used to input the optimization scheme and each of the overseas warehouses to be balanced into a preset natural language processing model to generate sub-optimization schemes for each of the overseas warehouses to be balanced. The sub-optimization scheme distribution module is used to distribute each of the sub-optimization schemes to the corresponding overseas warehouses to be balanced.

[0046] In one embodiment, the first acquisition module 902 includes: The crawling submodule is used to crawl multiple data sources in real time through a pre-set distributed crawler cluster to obtain multiple raw information. The translation submodule is used to translate the original information into text in a specified language using a preset model, thereby obtaining the original text data; The extraction submodule is used to extract entities and the relationships between entities from the original text data, thereby obtaining risk data information.

[0047] In one embodiment, the cross-border inventory dynamic balancing device further includes: The local training data acquisition module is used to acquire multiple sets of local training data from various overseas repositories; each set of local training data includes local risk data information and local optimization schemes. The local model training module is used to train the corresponding preset local models based on multiple sets of local training data from each overseas repository. The training parameter acquisition module is used to acquire the training parameters of each preset local model and input them into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

[0048] In one embodiment, the cross-border inventory dynamic balancing device further includes: The data acquisition module is used to acquire geopolitical and market data from multiple countries. The data parsing module is used to parse geopolitical and market data of various countries using a preset semantic model to obtain parsed information data. The data input module is used to input the various information data into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

[0049] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a cross-border inventory dynamic balancing method. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement the cross-border inventory dynamic balancing method. Those skilled in the art will understand that... Figure 4The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0050] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

[0051] By acquiring inventory information from each warehouse and calculating corresponding constraints, and combining the optimization scheme of a mixed-integer programming model, enterprises can quickly make scientific decisions in complex supply chain environments, thereby optimizing inventory allocation and improving the flexibility and responsiveness of inventory management. This solves the shortcomings of existing methods that cannot be dynamically adjusted based on real-time risk data, and effectively reduces resource waste and stockout risks caused by static management.

[0052] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

[0053] By acquiring inventory information from each warehouse and calculating corresponding constraints, and combining the optimization scheme of a mixed-integer programming model, enterprises can quickly make scientific decisions in complex supply chain environments, thereby optimizing inventory allocation and improving the flexibility and responsiveness of inventory management. This solves the shortcomings of existing methods that cannot be dynamically adjusted based on real-time risk data, and effectively reduces resource waste and stockout risks caused by static management.

[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0055] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0056] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for dynamic balancing of cross-border inventory, characterized in that, The method includes: Real-time acquisition of risk data and information; Based on the aforementioned risk data, overseas warehouses to be balanced were identified. Obtain inventory information for each category of goods in each of the aforementioned overseas warehouses to be balanced; Based on the inventory information and the risk data information, determine the constraints corresponding to each of the overseas warehouses to be balanced. The constraints and the overseas warehouses to be balanced are input into a preset mixed integer programming model to obtain an optimization scheme; The inventory of each of the overseas warehouses to be balanced is balanced based on the optimization scheme.

2. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, Before the step of determining the overseas warehouses to be balanced based on the risk data information, the method further includes: The risk data information is input into a preset risk assessment model to obtain a target risk index; Determine whether the target risk index is greater than the risk threshold; If the target risk index is greater than the risk threshold, then the step of determining the overseas warehouses to be balanced based on the risk data information is satisfied.

3. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, After the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization scheme, the method further includes: Extract the dimensional information of each dimension from the optimization scheme; The information from each dimension is converted into gene segments and then encoded into chromosomes to obtain the initial chromosome; The initial chromosomes are optimized based on various dimensions to generate a Pareto front, thereby obtaining the first chromosome set. Perform a crossover mutation operation on the first chromosome set to obtain a second chromosome set; The second chromosome set is combined with the first chromosome set to generate a third chromosome set; Using the third chromosome set as the first chromosome set, the target step and the steps following the target step are repeated until the number of chromosomes in the third chromosome set is greater than or equal to a preset number; the target step is to perform a crossover mutation operation on the first chromosome set to obtain the second chromosome set. The third chromosome set whose number of chromosomes is greater than or equal to the preset number is denoted as the target third chromosome set; The fitness value of each chromosome in the target third chromosome set is calculated using a preset fitness calculation function; The chromosome with the highest fitness value is selected as the target optimization scheme. The inventory of each of the overseas warehouses to be balanced is balanced based on the target optimization scheme.

4. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, After the step of balancing the inventory of each of the overseas warehouses to be balanced based on the optimization scheme, the method further includes: The optimization scheme and each of the overseas warehouses to be balanced are input into a preset natural language processing model to generate sub-optimization schemes for each of the overseas warehouses to be balanced. Each of the aforementioned sub-optimization schemes is distributed to the corresponding overseas warehouses to be balanced.

5. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, The steps for acquiring risk data information in real time include: Multiple data sources are crawled in real time by a pre-set distributed crawler cluster to obtain multiple raw information. The original information is translated into text in a specified language using a preset model to obtain the original text data; The original text data is used to extract entities and the relationships between entities to obtain risk data information.

6. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, Before the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization solution, the method further includes: Multiple sets of local training data were obtained from various overseas repositories; one set of local training data included local risk data information and local optimization schemes. Based on multiple sets of local training data corresponding to each overseas warehouse, train the corresponding preset local models respectively; The training parameters of each preset local model are obtained and input into the preset mixed integer programming initial model to obtain the preset mixed integer programming model.

7. The cross-border inventory dynamic balancing method according to claim 1, characterized in that, Before the step of inputting the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization solution, the method further includes: Acquire geopolitical and market data from multiple countries; By analyzing the geopolitical and market data of various countries using a pre-defined semantic model, the analyzed information data is obtained. Each of the aforementioned information data is input into a preset mixed integer programming initial model to obtain a preset mixed integer programming model.

8. A cross-border inventory dynamic balancing device, characterized in that, The device includes: The first acquisition module is used to acquire risk data information in real time; The first determining module is used to determine the overseas warehouses to be balanced based on the risk data information; The second acquisition module is used to acquire inventory information of each category of goods in each of the overseas warehouses to be balanced; The second determining module is used to determine the constraints corresponding to each of the overseas warehouses to be balanced based on the inventory information and the risk data information. The input module is used to input the constraints and the overseas warehouses to be balanced into a preset mixed integer programming model to obtain an optimization scheme; The balancing module is used to balance the inventory of each of the overseas warehouses to be balanced based on the optimization scheme.

9. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, causes the processor to perform the steps of the cross-border inventory dynamic balancing method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the cross-border inventory dynamic balancing method as described in any one of claims 1 to 7.

Images