Intelligent Car Sales Order Management Method and System Based on Cross-border Trade

Abstract

This invention discloses an intelligent automobile sales order management method and system based on cross-border trade. The method includes acquiring a historical order dataset from cross-border trade, clustering the historical order dataset to construct a multi-dimensional user profile model, extracting user profile feature vectors from the multi-dimensional user profile model, fusing the user profile feature vectors with external market trend data and tariff policy change factors, inputting the fusion into a demand forecasting model to generate prediction results, establishing a multi-objective optimization function, solving for the optimal solution to generate a warehousing allocation and scheduling scheme, executing scheduling instructions through a blockchain smart contract, collecting real-time actual logistics data and prediction deviation values, updating user profile feature weights through an online learning mechanism, and adjusting the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm to optimize the demand forecasting model. This invention improves the operational efficiency of cross-border automobile sales by accurately predicting user behavior and thus scheduling warehousing resource allocation.

Description

Technical Field

[0001] This invention relates to the field of automobile order management technology, and in particular to an intelligent automobile sales order management method and system based on cross-border trade. Background Technology

[0002] In the cross-border automotive trade sector, the diverse types of data and complex business processes present numerous challenges in order management. Firstly, traditional demand forecasting methods often rely on single or limited-dimensional data, failing to comprehensively capture the diverse user demand characteristics in cross-border trade. This results in inaccurate predictions of future order volumes, hindering companies' ability to accurately grasp market dynamics. Secondly, in the warehousing and allocation process, previous solutions have failed to fully consider factors such as logistics costs, timeliness deviations, and inventory turnover risks. When faced with warehouse capacity limitations, cross-border transportation timeliness requirements, and customs clearance risks, they cannot achieve optimal resource allocation, often leading to cost waste and low operational efficiency. Furthermore, the lack of an effective dynamic model optimization mechanism makes it difficult to adjust in real-time based on actual logistics data and forecast deviations, causing order management solutions to gradually become disconnected from reality. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this invention provides an intelligent automobile sales order management method and system based on cross-border trade.

[0004] In a first aspect, the present invention provides an intelligent automobile sales order management method based on cross-border trade, characterized in that the method includes:

[0005] Obtain a historical order dataset for cross-border trade, cluster the historical order dataset to construct a multidimensional user profile model, and extract user profile feature vectors based on the multidimensional user profile model;

[0006] The user profile feature vector is fused with external market trend data and tariff policy change factors. The fused features are then input into a trained demand forecasting model to generate forecast results. The forecast results include the order volume forecast value within a future preset period, the vehicle model demand distribution matrix, and the transaction time window probability distribution.

[0007] A multi-objective optimization function is established based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations, and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold, and customs clearance risk coefficient as constraints; the objective function is optimized and solved to generate a warehouse allocation and scheduling scheme.

[0008] Based on the warehouse allocation and scheduling scheme, scheduling instructions are executed through blockchain smart contracts, real-time collection of actual logistics data and prediction deviation values ​​is carried out, user profile feature weights are updated through an online learning mechanism, and Bayesian optimization algorithms are used to adjust the hyperparameters of the demand forecasting model in order to optimize the demand forecasting model.

[0009] Preferably, before executing the scheduling instructions via the blockchain smart contract, the method further includes:

[0010] The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load.

[0011] When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

[0012] Preferably, the dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information.

[0013] Preferably, the output dimensions of the multi-dimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

[0014] Secondly, the present invention also provides an intelligent automobile sales order management system based on cross-border trade, the system comprising:

[0015] The data acquisition unit is used to acquire historical order datasets of cross-border trade, cluster the historical order datasets to build a multi-dimensional user profile model, and extract user profile feature vectors based on the multi-dimensional user profile model.

[0016] The demand forecasting unit is used to fuse user profile feature vectors with external market trend data and tariff policy change factors, and input the fused features into the trained demand forecasting model to generate forecast results. The forecast results include the order volume forecast value within a future preset period, the vehicle model demand distribution matrix, and the transaction time window probability distribution.

[0017] The scheduling and allocation unit is used to establish a multi-objective optimization function based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold and customs clearance risk coefficient as constraints; and optimizing the objective function to generate a warehouse allocation and scheduling scheme.

[0018] The model iteration unit is used to execute scheduling instructions through blockchain smart contracts based on the warehouse allocation and scheduling scheme, collect actual logistics data and prediction deviation values ​​in real time, update user profile feature weights through an online learning mechanism, and adjust the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm to optimize the demand forecasting model.

[0019] Preferably, the intelligent car sales order management system based on cross-border trade further includes an instruction simulation unit, used for:

[0020] The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load.

[0021] When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

[0022] Preferably, the dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information.

[0023] Preferably, the output dimensions of the multi-dimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

[0024] Thirdly, the present invention also provides an electronic device including a processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs a method as described in the first aspect above and any possible implementation thereof.

[0025] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor of an electronic device, cause the processor to perform a method as described in the first aspect above and any possible implementation thereof.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention provides an intelligent automobile sales order management method based on cross-border trade. It constructs a multi-dimensional user profile model by acquiring historical cross-border trade order datasets, and integrates its feature vectors with external market trend data and tariff policy change factors into a demand forecasting model. This fully leverages market demand information under complex environments, enabling accurate predictions of order volume, vehicle model demand distribution matrix, and transaction time window probability distribution within a preset future period, providing a solid and reliable data foundation for subsequent decision-making. Regarding warehousing allocation and scheduling, a multi-objective optimization function is established based on the prediction results, incorporating logistics costs, timeliness deviations, and inventory turnover risks. The function is optimized under constraints such as warehouse capacity, transportation timeliness, and customs clearance risks. The resulting warehousing allocation and scheduling scheme comprehensively balances various factors, achieving optimal resource allocation, significantly reducing costs, and improving operational efficiency. Furthermore, by utilizing blockchain smart contracts to execute scheduling instructions, collecting actual logistics data and prediction deviation values ​​in real time, updating user profile feature weights through an online learning mechanism, and adjusting the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm, the entire order management solution possesses dynamic optimization capabilities. This allows it to closely follow changes in the cross-border trade environment and maintain a consistently high level of order management efficiency.

[0028] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the background art, the accompanying drawings used in the embodiments of the present invention or the background art will be described below.

[0030] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the specification, serve to illustrate the technical solutions of this disclosure.

[0031] Figure 1 A flowchart illustrating an intelligent automobile sales order management method based on cross-border trade, provided as an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of the structure of an intelligent automobile sales order management system based on cross-border trade provided in an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of the structure of an intelligent automobile sales order management system based on cross-border trade, provided as another embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0035] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0036] Please see Figure 1 , Figure 1 This is a flowchart illustrating an intelligent automobile sales order management method based on cross-border trade, provided as an embodiment of the present invention. Figure 1 As shown, the method includes:

[0037] S10. Obtain the historical order dataset for cross-border trade, cluster the historical order dataset to construct a multi-dimensional user profile model, and extract user profile feature vectors based on the multi-dimensional user profile model.

[0038] User attribute data, including basic information such as age, gender, occupation, and location, is extracted from the company's internal sales management system. This data helps in understanding user characteristics. Transaction time-series data is obtained from the transaction record database, recording the order placement time, payment time, and shipping time of each order to analyze user purchase patterns. Vehicle model configuration parameters can be obtained from the product information database provided by car manufacturers, detailing engine parameters, interior configurations, and exterior colors for each model. This is crucial for analyzing user preferences for vehicle types. Cross-border logistics node information is obtained from logistics partners' systems, including timestamps and status information of goods at various cross-border logistics nodes (such as the port of origin, transit point, and destination port).

[0039] To improve data quality, the first step is to check the data's completeness. For records with missing values, handle them according to the data's characteristics. For example, if the missing value ratio in user attribute data is small, it can be filled by asking users or based on the characteristics of similar users. For missing values ​​in transaction time series data, if the missing value is a key time point (such as order time), consider deleting the record to avoid affecting subsequent analysis. Duplicate data is handled by comparing all fields to identify and delete identical records, ensuring data uniqueness. Data standardization is also performed, such as converting transaction time series data to a uniform time format and normalizing numerical data in car model configuration parameters (such as engine power) for subsequent analysis and modeling.

[0040] Furthermore, the historical order dataset is clustered to construct a multi-dimensional user profile model:

[0041] Preferably, the dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information. Therefore, during feature extraction, derived features are extracted from the user attribute data, such as calculating the user's potential purchasing power based on the economic development level and consumption index of the user's region. For the transaction time series data, features such as the user's purchase frequency and average purchase interval are calculated. User preference for different configurations (e.g., high-end configuration, environmentally friendly configuration) is extracted from the vehicle model configuration parameters. Combined with cross-border logistics node information, the logistics preferences of users in different regions (e.g., the choice between fast logistics and economy logistics) are analyzed.

[0042] The K-Means clustering algorithm is employed, which is simple, efficient, and suitable for large-scale datasets. First, the initial number of clusters K is determined empirically or through multiple trials. For example, one can start with K=3 and gradually increase the K value, evaluating the clustering effect by calculating metrics such as the silhouette coefficient, and selecting the optimal K value. The feature-engineered dataset is then input into the K-Means algorithm, which divides the data into K clusters based on the distance between data points (e.g., Euclidean distance). For each cluster, a user purchasing power grading index is calculated. By analyzing data such as users' income levels and the price range of cars they purchase, users' purchasing power is categorized into high, medium, and low levels. A brand preference coefficient is calculated by statistically analyzing the frequency with which users within a cluster purchase different car brands to determine the degree of preference for specific brands. For example, if the proportion of users in a cluster purchasing Tesla cars is significantly higher than that of other brands, then the user preference coefficient for Tesla in that cluster is high. The vehicle model iteration response cycle is determined by observing the time interval between purchases after the release of a new car model within the cluster, thereby evaluating the user's response speed to vehicle model iterations. For example, if a certain user group has a high purchase rate within 1-3 months of a new car model's release, then that user group has a short model iteration response cycle. Analyzing regional policy sensitivity parameters, we can study the impact of different regional policies (such as car purchase subsidies and environmental policies) on the purchasing behavior of users within the cluster. For example, if some regions have new energy vehicle purchase subsidy policies, and if the proportion of users in that region purchasing new energy vehicles increases significantly after the subsidy policy is introduced, it indicates that that user group is highly sensitive to regional policies.

[0043] Finally, user profile feature vectors are extracted based on the multi-dimensional user profile model. For each user, according to their cluster, corresponding user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters are extracted. These indicators are combined into a feature vector, for example: [user purchasing power grading (1 for high, 2 for medium, 3 for low), brand preference coefficient (e.g., Tesla brand preference coefficient is 0.8), vehicle model iteration response cycle (in months), regional policy sensitivity parameter (values ​​from 0 to 1 based on the degree of influence)]. The extracted user profile feature vectors are stored in a dedicated database for subsequent feature fusion with external market trend data and tariff policy change factors, and input into the demand forecasting model to provide strong support for automobile sales order management in cross-border trade. Preferably, the output dimensions of the multi-dimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

[0044] S20. The user profile feature vector is fused with external market trend data and tariff policy change factors. The fused features are then input into the trained demand prediction model to generate prediction results. The prediction results include the order volume prediction value, vehicle model demand distribution matrix and transaction time window probability distribution within a future preset period.

[0045] It should be noted that external market trend data includes:

[0046] Macroeconomic data: Obtain macroeconomic indicators such as GDP growth rate, inflation rate, and unemployment rate for the target market country or region from authoritative economic data publishing institutions (such as the International Monetary Fund and the World Bank). This data reflects the overall economic situation of the local area and has a significant impact on the automobile consumer market. For example, when GDP growth is high, consumer purchasing power may increase, and automobile demand may rise.

[0047] Industry market data: We collect information on target market car sales trends, market saturation, and competitor activities through professional automotive industry research reports and data from market research institutions. For example, understanding competitors' new model launch plans can help analyze their potential impact on demand for your own models.

[0048] Consumer behavior data: This involves using social media monitoring tools and online surveys to collect information on consumer focus areas and preference trends regarding automobiles. For example, a sustained increase in discussion about electric vehicles on social media may indicate future growth in demand.

[0049] Factors affecting tariff policy changes include:

[0050] Pay close attention to official announcements from governments and policy documents issued by customs authorities to obtain timely information on tariff rate adjustments. For example, if a country raises import tariffs on automobiles, this will directly increase the cost of importing cars, affecting car prices and demand.

[0051] Analyze the reasons and background for changes in tariff policies, and assess their impact on different vehicle models and countries of origin. For example, some countries impose lower tariffs on imported new energy vehicles while levying higher tariffs on traditional gasoline-powered vehicles in order to protect their domestic new energy vehicle industry.

[0052] Furthermore, the collected external market trend data and tariff policy change factors are cleaned and preprocessed. Missing values ​​in macroeconomic data can be filled using time series interpolation; outliers in industry market data are identified and corrected using statistical analysis methods (such as box plots). All data is formatted uniformly to make it suitable for integration with user profile feature vectors. For example, time series data are standardized to the same time intervals, and various indicator data are normalized to ensure the data are on the same scale, facilitating subsequent calculations.

[0053] In one embodiment, a concatenation and fusion approach is used to horizontally concatenate the user profile feature vector with preprocessed external market trend data and tariff policy change factors. Preferably, a weighted fusion method can also be used, assigning different weights to the user profile feature vector, external market trend data, and tariff policy change factors based on the importance of each data point to demand forecasting. For example, in some markets where tariff policy changes have a significant impact on automobile demand, the tariff policy change factor is given a higher weight, and then the fused feature vector is obtained through weighted summation.

[0054] For training the demand prediction model, various models can be selected, including but not limited to Long Short-Term Memory (LSTM) networks and Support Vector Regression (SVR) models. The fused feature vector is then input into the trained demand prediction model. Before input, it's crucial to ensure that the model's input format matches the format of the fused feature vector. For example, if the LSTM model requires input data as a three-dimensional tensor (number of samples, time step, number of features), the fused feature vector must be reshaped according to the corresponding format. The model calculates based on the input feature vector and outputs the predicted order volume for the next preset period.

[0055] For example, the model predicts that the number of car orders in a certain region will be 1,000, 1,200, and 1,500 vehicles in the next three months, respectively. Regarding the vehicle type demand distribution matrix, the model predicts the demand share of different vehicle types within a preset future period based on input features. For example, the prediction results show that in the next month, SUVs will account for 40% of demand, sedans for 35%, and MPVs for 25%. In terms of predicting the probability distribution of transaction time windows, the model calculates the probability of a user placing an order during different time periods. For example, the prediction results show that the probability of a user placing an order between 9:00 AM and 11:00 AM on weekdays is 0.2; and the probability of a user placing an order between 2:00 PM and 4:00 PM on weekends is 0.15, etc.

[0056] S30. Establish a multi-objective optimization function based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold and customs clearance risk coefficient as constraints; optimize the objective function to generate a warehouse allocation and scheduling scheme.

[0057] Step 1: Suppose there are n warehouses and m types of car models, and T is the time span of a future preset period, divided into t = 1, 2, ... T time stages.

[0058] x ijt : Represents the quantity of vehicle type j allocated to warehouse i in stage t, where i = 1, ..., n; j = 1, ..., m;

[0059] y ijt : Represents the number of j-th vehicle types actually shipped from the i-th warehouse in stage t;

[0060] D jt : This indicates the predicted order volume for the j-th vehicle type at stage t, derived from the demand forecasting model;

[0061] C ij : Represents the unit logistics cost of storing the j-th type of vehicle in the i-th warehouse;

[0062] S it : Represents the inventory turnover risk coefficient of the i-th warehouse in stage t, which can be estimated based on historical inventory data and sales data;

[0063] L it : Represents the actual inventory level of the i-th warehouse at stage t;

[0064] This represents the maximum capacity of the i-th warehouse;

[0065] T ij : Indicates the transportation time for transporting the j-th type of vehicle from the i-th warehouse to its destination;

[0066] Indicates the timeliness threshold for cross-border transportation;

[0067] R i : Represents the customs clearance risk coefficient of goods in the i-th warehouse, which is mainly assessed based on historical clearance data and current customs policies.

[0068] Step 2: Establish the objective function:

[0069] 1) Logistics cost objective function Z1:

[0070] Logistics costs are the sum of the products of the unit logistics cost of storing different vehicle models in each warehouse at each stage and the quantity stored:

[0071]

[0072] 2) Objective function for timeliness deviation Z2:

[0073] Timeliness deviation is measured by the difference between the actual transportation time and the timeliness threshold, let w ij Let the timeliness weight be the transportation time of the j-th type of vehicle from the i-th warehouse, then we have:

[0074]

[0075] 3) Inventory turnover risk objective function Z3:

[0076] Inventory turnover risk is related to the inventory level and inventory turnover risk coefficient of each warehouse at each stage. Let α be the inventory turnover risk weighting coefficient:

[0077]

[0078] Considering the three objective functions above, a multi-objective optimization function Z is constructed. The three objectives are combined using a weighted summation method. Let δ1, δ2, and δ3 be the weights of logistics cost, timeliness deviation, and inventory turnover risk, respectively, and satisfy δ1 + δ2 + δ3 = 1. Therefore, the multi-objective optimization function Z is:

[0079] Z = δ1Z1 + δ2Z2 + δ3Z3

[0080] Step 3: Determine the constraints:

[0081] Order satisfaction constraint: The sum of the quantities allocated to each vehicle model by each warehouse at each stage should satisfy the predicted order quantity for that vehicle model at the corresponding stage.

[0082]

[0083] Warehouse capacity constraint: The inventory level of each warehouse at each stage cannot exceed its maximum capacity.

[0084]

[0085] Cross-border transportation time constraints: The time required for transport vehicles from each warehouse to the destination must not exceed the time limit threshold.

[0086]

[0087] Customs clearance risk constraints: An overall acceptable level of customs clearance risk, R, can be set. max This ensures that the total customs clearance risk factor for all warehouses is within an acceptable range.

[0088]

[0089] Step 4: Find the optimal solution:

[0090] The non-dominated sorting genetic algorithm is used to solve the above multi-objective optimization problem, specifically including:

[0091] Population initialization: Randomly generate a set of initial solutions that satisfy all constraints, i.e., a series of x. ijt The value combination is used as the initial population.

[0092] Fitness calculation: For each individual in the population (i.e., a storage allocation and scheduling scheme), calculate its logistics cost Z1, timeliness deviation Z2, and inventory turnover risk Z3 according to the above objective function, and then obtain the comprehensive objective function value Z. The smaller this value, the better the fitness of the individual.

[0093] Genetic manipulation: New populations are generated through genetic operations such as selection, crossover, and mutation. In selection, methods such as tournament selection are used to choose individuals with better fitness for the next generation; crossover involves exchanging partial genes, i.e., x-genes, between two individuals. ijt Values ​​are used to generate new individuals; mutation operations randomly change certain genes of an individual to increase the diversity of the population.

[0094] Iterative optimization: The fitness and genetic operation steps are repeatedly calculated. After multiple generations of iteration, the population gradually moves closer to the optimal solution. When certain termination conditions are met (such as the number of iterations reaching a set value, or the convergence degree of the population meeting the requirements), the iteration stops, and the optimal solution is output, thus obtaining a warehouse allocation and scheduling scheme that meets the requirements of multi-objective optimization.

[0095] S40. Based on the warehouse allocation and scheduling scheme, scheduling instructions are executed through blockchain smart contracts, real-time collection of actual logistics data and prediction deviation values ​​is carried out, user profile feature weights are updated through an online learning mechanism, and Bayesian optimization algorithm is used to adjust the hyperparameters of the demand forecasting model in order to optimize the demand forecasting model.

[0096] The smart contract code must cover key information in the warehouse allocation and scheduling plan, including detailed instructions such as which warehouse to allocate which type of vehicle, the quantity allocated, and the transportation destination. The written smart contract is compiled and tested to ensure its logical correctness and the absence of vulnerabilities and errors. Then, the smart contract is deployed to the blockchain network, making it effective on the distributed nodes of the blockchain. When the conditions for executing the scheduling instructions are triggered (e.g., reaching a preset delivery time or receiving a specific order confirmation signal), the system automatically invokes the deployed blockchain smart contract. The smart contract sends accurate scheduling instructions to relevant logistics participants (such as warehouse management systems, transportation company systems, etc.) according to the specific arrangements in the warehouse allocation and scheduling plan. These instructions are propagated through the decentralized network of the blockchain, ensuring the immutability and accurate transmission of the instructions. After receiving the instructions, the logistics participants execute the corresponding operations as required, such as retrieving vehicles from designated warehouses and arranging transportation. Simultaneously, the logistics participants feed back confirmation information of the operation execution to the blockchain; the smart contract records this feedback information, achieving transparency and traceability of the entire scheduling process.

[0097] Deploy data collection devices or connect to relevant system data interfaces at key nodes in logistics and transportation (such as warehouse shipping points, important transit stations, and destination receiving points). For example, during warehouse shipment, the warehouse management system automatically collects data such as the actual vehicle type, quantity, and time of shipment; during transportation, IoT devices (such as vehicle-mounted GPS devices) collect real-time data on vehicle location and speed, which can be used to calculate actual delivery time. In the order processing stage, collect information such as actual transaction time and customer feedback. For example, record the difference between the actual order placement time and the predicted transaction time window, as well as the difference between the actual customer demand for vehicle types and the predicted demand distribution.

[0098] At the data processing center, the real-time collected actual logistics data is compared with the prediction results generated by the previous demand forecasting model.

[0099] For order volume, calculate the difference between the actual order volume and the predicted order volume, i.e., deviation value = actual order volume - predicted order volume, and take the absolute value. For vehicle type demand distribution, calculate the difference between the actual demand share and the predicted demand share for each vehicle type. For example, if the actual demand share for SUVs is 45% and the predicted share is 40%, then the difference is 45% - 40% = 5%. For transaction time windows, calculate the time difference between the actual transaction time and the predicted transaction time window, such as if the actual transaction time is 2 hours later than the predicted time. Organize and store these calculated prediction deviation values ​​for subsequent model optimization. When updating user profile feature weights through an online learning mechanism, the stochastic gradient descent (SGD) algorithm can be used as the basic algorithm for the online learning mechanism. The SGD algorithm can update the model parameters each time new data samples are received (i.e., real-time collected actual logistics data and prediction deviation values), making it suitable for online learning scenarios handling large-scale data. Finally, the hyperparameters of the demand forecasting model are adjusted using a Bayesian optimization algorithm.

[0100] In summary, the intelligent automobile sales order management method based on cross-border trade provided by this invention constructs a multi-dimensional user profile model by acquiring historical order datasets from cross-border trade. This model's feature vectors are then integrated with external market trend data and tariff policy change factors and input into a demand forecasting model. This allows for the full mining of market demand information under complex environments, enabling accurate predictions of order volume, vehicle model demand distribution matrix, and transaction time window probability distribution within a preset future period, providing a solid and reliable data foundation for subsequent decision-making. Regarding warehousing allocation and scheduling, a multi-objective optimization function is established based on the prediction results, incorporating logistics costs, timeliness deviations, and inventory turnover risks. This function is optimized under constraints such as warehouse capacity, transportation timeliness, and customs clearance risks. The resulting warehousing allocation and scheduling scheme comprehensively balances various factors, achieving optimal resource allocation, significantly reducing costs, and improving operational efficiency. Furthermore, by utilizing blockchain smart contracts to execute scheduling instructions, collecting actual logistics data and prediction deviation values ​​in real time, updating user profile feature weights through an online learning mechanism, and adjusting the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm, the entire order management solution possesses dynamic optimization capabilities. This allows it to closely follow changes in the cross-border trade environment and maintain a consistently high level of order management efficiency.

[0101] In one embodiment, prior to executing the scheduling instructions via a blockchain smart contract, the method further includes:

[0102] The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load.

[0103] When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

[0104] Specifically, the predicted order volume, vehicle type demand distribution matrix, and transaction time window probability distribution generated by the demand forecasting model are accurately imported into the digital twin simulation system according to the data format required by the system. Within the digital twin simulation system, a virtual model highly similar to the actual logistics network is constructed, encompassing key elements such as warehouses, transportation routes, and transportation vehicles. Based on the input prediction results, the system simulates the flow of goods in the logistics network under different warehousing allocation schemes, thereby calculating the congestion probability of key network load nodes (such as transportation hubs and busy warehouses).

[0105] Furthermore, a preset threshold is set (e.g., a congestion probability of 30%). When the congestion probability simulated by the system exceeds this preset threshold, the system automatically triggers an adjustment mechanism. The algorithm replans transportation routes, filters out alternative routes, and identifies emergency warehouses. For example, if a regular transportation route has an excessively high congestion probability in the simulation, the algorithm can search for and recommend other feasible routes based on historical traffic data and real-time road condition information. Emergency warehouses are selected based on factors such as geographical location, warehouse capacity, and distance from key nodes. Based on these new plans, a new scheduling scheme is generated, incorporating alternative transportation routes and emergency warehouses, leading to new scheduling instructions.

[0106] Therefore, the above embodiments, through advance simulation and adjustment, avoid potential congestion at key nodes of the logistics network, ensuring smoother and more efficient transportation of goods, reducing transportation delays, and improving overall logistics efficiency. When potential congestion risks arise, alternative transportation routes and emergency warehouses are activated promptly, enhancing the logistics system's ability to respond to emergencies and improving the stability and reliability of the entire system. Simultaneously, it effectively reduces additional transportation costs caused by congestion, such as increased fuel consumption due to increased vehicle waiting time and rising labor costs, while also making rational use of emergency warehouse resources to avoid unnecessary resource waste, thereby reducing the company's operating costs.

[0107] See also Figure 2 In one embodiment of the present invention, an intelligent automobile sales order management system based on cross-border trade is also provided, the system comprising:

[0108] The data acquisition unit 100 is used to acquire cross-border trade historical order datasets, cluster the historical order datasets to construct a multi-dimensional user profile model, and extract user profile feature vectors based on the multi-dimensional user profile model.

[0109] The demand forecasting unit 200 is used to fuse user profile feature vectors with external market trend data and tariff policy change factors, and input the fused features into a trained demand forecasting model to generate forecast results. The forecast results include the order volume forecast value, vehicle model demand distribution matrix and transaction time window probability distribution within a future preset period.

[0110] The scheduling and allocation unit 300 is used to establish a multi-objective optimization function based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold and customs clearance risk coefficient as constraints; and optimizing the objective function to generate a warehouse allocation and scheduling scheme.

[0111] Model iteration unit 400 is used to execute scheduling instructions through blockchain smart contracts based on the warehouse allocation and scheduling scheme, collect actual logistics data and prediction deviation values ​​in real time, update user profile feature weights through an online learning mechanism, and adjust the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm to optimize the demand forecasting model.

[0112] See also Figure 3 In one embodiment, the intelligent car sales order management system based on cross-border trade further includes an instruction simulation order 500, used for:

[0113] The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load.

[0114] When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

[0115] Preferably, the dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information.

[0116] Preferably, the output dimensions of the multi-dimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

[0117] It is understood that the system provided in this embodiment has functions or includes modules that can be used to execute the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0118] The present invention also provides an electronic device including a processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs a method as described in any of the above possible implementations.

[0119] The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0120] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

Claims

1. An intelligent automobile sales order management method based on cross-border trade, characterized in that, The method comprises: Obtain a historical order dataset for cross-border trade, cluster the historical order dataset to construct a multidimensional user profile model, and extract user profile feature vectors based on the multidimensional user profile model; The user profile feature vector is fused with external market trend data and tariff policy change factors. The fused features are then input into a trained demand forecasting model to generate forecast results. The forecast results include the order volume forecast value within a future preset period, the vehicle model demand distribution matrix, and the transaction time window probability distribution. A multi-objective optimization function is established based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations, and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold, and customs clearance risk coefficient as constraints; the objective function is optimized and solved to generate a warehouse allocation and scheduling scheme. Based on the warehouse allocation and scheduling scheme, scheduling instructions are executed through blockchain smart contracts, real-time collection of actual logistics data and prediction deviation values ​​is carried out, user profile feature weights are updated through an online learning mechanism, and Bayesian optimization algorithms are used to adjust the hyperparameters of the demand forecasting model in order to optimize the demand forecasting model.

2. The intelligent automobile sales order management method based on cross-border trade according to claim 1, characterized in that, Before executing the scheduling instructions via the blockchain smart contract, the following is also included: The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load. When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

3. The intelligent automobile sales order management method based on cross-border trade according to claim 1, characterized in that, The dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information.

4. The intelligent automobile sales order management method based on cross-border trade according to claim 1, characterized in that, The output dimensions of the multidimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

5. An intelligent automobile sales order management system based on cross-border trade, characterized in that, The system includes: The data acquisition unit is used to acquire historical order datasets of cross-border trade, cluster the historical order datasets to build a multi-dimensional user profile model, and extract user profile feature vectors based on the multi-dimensional user profile model. The demand forecasting unit is used to fuse user profile feature vectors with external market trend data and tariff policy change factors, and input the fused features into the trained demand forecasting model to generate forecast results. The forecast results include the order volume forecast value within a future preset period, the vehicle model demand distribution matrix, and the transaction time window probability distribution. The scheduling and allocation unit is used to establish a multi-objective optimization function based on the prediction results, including establishing an objective function based on logistics costs, timeliness deviations and inventory turnover risks, with the maximum capacity of each warehouse, cross-border transportation timeliness threshold and customs clearance risk coefficient as constraints; and optimizing the objective function to generate a warehouse allocation and scheduling scheme. The model iteration unit is used to execute scheduling instructions through blockchain smart contracts based on the warehouse allocation and scheduling scheme, collect actual logistics data and prediction deviation values ​​in real time, update user profile feature weights through an online learning mechanism, and adjust the hyperparameters of the demand forecasting model using a Bayesian optimization algorithm to optimize the demand forecasting model.

6. The intelligent automobile sales order management system based on cross-border trade according to claim 5, characterized in that, It also includes an instruction emulation unit, used for: The prediction results output by the demand forecasting model are input into the digital twin simulation system to simulate the logistics network load status under different warehousing allocation schemes and obtain the congestion probability of key nodes in the network load. When the congestion probability exceeds a preset threshold, a new scheduling scheme containing alternative transportation routes and emergency warehouses is generated to generate new scheduling instructions.

7. The intelligent automobile sales order management system based on cross-border trade according to claim 5, characterized in that, The dataset includes user attribute data, transaction time series data, vehicle model configuration parameters, and cross-border logistics node information.

8. The intelligent automobile sales order management system based on cross-border trade according to claim 5, characterized in that, The output dimensions of the multidimensional user profile model include user purchasing power grading indicators, brand preference coefficients, vehicle model iteration response cycles, and regional policy sensitivity parameters.

9. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein when the processor executes the computer instructions, the electronic device performs the intelligent automobile sales order management method based on cross-border trade as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform the intelligent automobile sales order management method based on cross-border trade as described in any one of claims 1 to 4.

Images