Foreign trade transaction demand-based supply chain intelligent matching system

Abstract

The invention relates to the technical field of supply chain management, and discloses a supply chain intelligent matching system based on foreign trade transaction demands. The system comprises a transaction demand acquisition module, a historical transaction data index module, a supply chain feature optimization module, a dynamic matching strategy generation module and an intelligent matching execution module. The transaction demand acquisition module acquires current foreign trade transaction demand information of a user; the historical transaction data indexing module is used for indexing a historical transaction characteristic data set in a historical transaction database according to a target transaction characteristic of the current demand; the supply chain feature optimization module performs optimization processing on the data set to generate an optimized data set; the dynamic matching strategy generation module calculates the remaining transaction time and generates a dynamic matching strategy according to the remaining transaction time; and the intelligent matching execution module executes supply chain matching according to a dynamic strategy based on the optimized data set and the current demand. The system can improve the precision and adaptability of supply chain matching, and meets the requirements of foreign trade transactions.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, specifically to an intelligent supply chain matching system based on foreign trade transaction needs. Background Technology

[0002] In the current globalized trade system, foreign trade transactions are characterized by diversified demands, large fluctuations in transaction cycles, and complex supply chain links. With the continuous development of business models such as cross-border e-commerce and overseas market expansion, users have placed higher demands on the efficiency and accuracy of matching supply chain partners when conducting foreign trade transactions. Traditional supply chain matching methods often rely on manual screening or fixed-rule matching. The primary problem with this model is its inability to quickly respond to dynamically changing transaction demands. For example, when users express specific foreign trade transaction needs regarding product categories, shipping timelines, and cost budgets, manual screening requires comparing supplier resources one by one. This process is easily affected by information asymmetry, leading to an inability to comprehensively obtain supply chain information that meets the requirements, thus prolonging the matching cycle and affecting transaction progress.

[0003] Traditional models lack effective utilization of historical transaction data. The vast amount of historical data accumulated in foreign trade transactions contains rich information on transaction characteristics, such as supplier fulfillment capabilities, product quality, and the stability of transportation timeliness. This information is of significant reference value for subsequent supply chain matching. However, traditional methods often fail to systematically organize and optimize this historical data, failing to extract effective characteristic information to guide current matching operations. This results in each matching process being relatively independent, making it difficult to leverage historical experience to improve matching quality. Furthermore, traditional matching strategies are usually fixed and unchanging, failing to consider the crucial factor of remaining transaction time. In actual foreign trade transactions, the requirements for matching efficiency and accuracy vary depending on the remaining transaction time. For example, when the transaction deadline is approaching, it is necessary to quickly find a supply chain partner that basically meets the requirements to ensure the transaction is completed on time; while when there is ample remaining time, more detailed screening can be used to obtain better matching results. Fixed matching strategies cannot adapt to these dynamically changing needs. Either due to overly detailed screening when time is tight, transaction opportunities are missed, or due to coarse screening when time is ample, matching results are poor.

[0004] In traditional supply chain matching processes, the matching of user-inputted current foreign trade transaction needs with supply chain resource information lacks an intelligent execution mechanism. Due to a lack of effective data processing and strategy support, information misalignment easily occurs during the matching process, leading to discrepancies between the matching results and the user's actual needs. This necessitates repeated adjustments and rematches, increasing labor and time costs and potentially reducing user satisfaction with supply chain services, thus hindering the smooth operation of foreign trade transactions. These problems render traditional supply chain matching models inadequate to meet the demands of today's rapidly developing foreign trade market. A system capable of incorporating historical data, dynamically adjusting strategies, and achieving intelligent matching is urgently needed to address these challenges. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent supply chain matching system based on foreign trade transaction needs, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a supply chain intelligent matching system based on foreign trade transaction needs, the system comprising: The transaction demand collection module is used to obtain the current foreign trade transaction demand information input by the user; The historical transaction data indexing module is used to index the historical transaction feature data set in the historical transaction database based on the target transaction features in the current foreign trade transaction demand information. The supply chain feature optimization module is used to perform feature optimization processing on the historical transaction feature data set to generate an optimized historical transaction feature dataset. The dynamic matching strategy generation module is used to calculate the remaining time of the transaction between the current time and the transaction deadline, and generate a dynamic matching strategy based on the remaining time of the transaction. The intelligent matching execution module is used to perform supply chain matching operations according to the dynamic matching strategy based on the optimized historical transaction feature dataset and the current foreign trade transaction demand information.

[0007] Preferably, the historical transaction data indexing module includes: The transaction feature classification unit is used to classify the historical transaction features within the historical transaction feature data set to obtain multiple classified transaction feature categories; The transaction indicator extraction unit is used to extract the historical transaction indicator set and the historical matching record set corresponding to the multiple categorized transaction feature categories. The matching score generation unit is used to generate multiple historical matching score sets based on the degree of deviation between the historical matching record set and the preset matching benchmark.

[0008] Preferably, the supply chain feature optimization module includes: A benchmark feature selection unit is used to select a first benchmark matching score within a first historical matching score set among the plurality of historical matching score sets. The probability distribution construction unit is used to allocate feature distribution probabilities to form a first feature probability distribution based on the difference between other historical matching scores in the first historical matching score set and the first benchmark matching score. The dimensionality reduction optimization unit is used to perform dimensionality reduction optimization on the first historical matching score set based on the first feature probability distribution, and output the first dimensionality-reduced matching score set.

[0009] Preferably, the dynamic matching strategy generation module includes: The first strategy cycle conversion unit is used to call the preset first duration conversion rule to convert the remaining duration of the transaction into the first matching cycle. The second strategy cycle conversion unit is used to count the number of transaction feature types contained in the current foreign trade transaction demand information and call the preset second duration conversion rule to calculate the adjustment cycle. A periodic fusion unit is used to fuse the first matching period and the adjustment period to generate a second matching period; The strategy generation unit is used to generate the dynamic matching strategy based on the second matching period.

[0010] Preferably, the intelligent matching execution module includes: The prediction model invocation unit is used to invoke a pre-trained supply chain demand prediction model. The demand forecasting unit is used to control the supply chain demand forecasting model to process the optimized historical transaction feature dataset and the current foreign trade transaction demand information according to the second matching cycle, so as to generate supply and demand forecast data. The matching execution unit is used to perform supply chain matching operations based on the supply and demand forecast data.

[0011] Preferably, the dimensionality reduction optimization unit includes: The feature collaborative computation subunit is used to calculate the feature collaborative representation vector among the historical matching scores in the first historical matching score set. The feature spectrum generation subunit is used to extract statistical features from the feature collaborative representation vector to form a set of transaction feature energy distribution spectrum vectors; The distribution center calculation subunit is used to calculate the mean vector of the set of energy distribution spectrum vectors of the transaction features as the distribution center vector of the transaction features.

[0012] Preferably, the dimensionality reduction optimization unit further includes: The feature span calculation subunit is used to calculate the feature span factor between the center vector of the transaction feature distribution and the energy distribution spectrum vector of each transaction feature; The feature filtering subunit is used to filter and form the first dimensionality reduction matching score set based on the comparison result of the feature span factor and the preset threshold.

[0013] Preferably, the demand forecasting unit includes: The offset coefficient calculation subunit is used to calculate the transaction feature offset coefficient between the optimized historical transaction feature dataset and the real-time transaction feature data. The rate of change generation subunit is used to generate the rate of change of supply and demand based on the proportional relationship between the transaction feature offset coefficient and the time dimension. The forecast correction subunit is used to correct the supply and demand forecast data based on the supply and demand change rate.

[0014] Preferably, the offset coefficient calculation subunit includes: A feature response calculation component is used to calculate the feature response value between the optimized historical transaction feature dataset and the real-time transaction feature data; A convolutional coding component is used to perform multi-scale convolutional coding on the feature response values ​​to generate a feature cooperative distribution matrix; An offset generation component is used to process the feature co-distribution matrix through a sequence matching network to generate the transaction feature offset coefficients.

[0015] Preferably, the matching execution module further includes: The decision-maker invocation unit is used to invoke a preset matching strategy decision-maker; The prompt scheme generation unit is used to generate a supply chain matching prompt scheme based on the supply demand forecast data and preset matching rules, through the matching strategy decision-maker. The matching execution unit is used to perform the matching operation corresponding to the supply chain matching prompt scheme.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This intelligent supply chain matching system, based on foreign trade transaction needs, accurately acquires user-inputted information on current foreign trade transaction requirements through a transaction demand collection module. This ensures the system remains user-demand-oriented throughout the matching process, avoiding matching deviations caused by incomplete or inaccurate demand information. This module comprehensively collects user demand information regarding product categories, quantities, quality standards, transportation methods, cost budgets, and transaction deadlines. This provides complete and accurate foundational data for subsequent matching operations, enabling the matching process to closely align with user needs, reducing unnecessary adjustments and re-matching steps, and improving overall matching efficiency.

[0017] The historical transaction data indexing module enables efficient access to the historical transaction feature data set in the historical transaction database. This module indexes target transaction features based on current foreign trade transaction demand information, quickly filtering relevant historical transaction feature data from a large volume of historical transaction data. By utilizing this relevant historical data, the system can learn from past successes and failures in foreign trade transactions. For example, it can refer to supplier performance and product compatibility characteristics similar to current demands in historical transactions, providing valuable reference for current supply chain matching, avoiding repeating past mistakes, and identifying and applying advantageous practices from past matching processes, thereby improving the reliability and rationality of current matching results.

[0018] The supply chain feature optimization module performs feature optimization processing on the historical transaction feature dataset to generate an optimized historical transaction feature dataset. This process removes redundant and invalid information from the historical data, extracting the key features that have a significant impact on the current matching. Historical transaction data may contain some information that is irrelevant or has a minor impact on the current matching. Directly using unoptimized historical data would increase the complexity of data processing, reduce matching efficiency, and may even lead to biased matching results due to interference from invalid information. The optimized historical transaction feature dataset, however, highlights core features and reduces data redundancy, making subsequent matching calculations based on this dataset more efficient and accurate, further improving matching quality.

[0019] The dynamic matching strategy generation module calculates the remaining time between the current moment and the transaction deadline, and generates a dynamic matching strategy based on this time. This allows the system's matching strategy to adapt to the actual transaction time requirements. When the remaining time is short, the system generates an efficiency-oriented matching strategy, simplifying some non-critical screening steps and quickly identifying supply chain resources that basically meet the requirements, ensuring that the matching is completed before the transaction deadline and avoiding transaction delays due to insufficient time. When the remaining time is long, the system generates a precision-oriented matching strategy, increasing the detailed screening of characteristics of each link in the supply chain, such as a more comprehensive evaluation of the supplier's production capacity, after-sales service, and past cooperation stability, to find supply chain partners that better meet the user's deeper needs and achieve better matching results. This dynamically adjusted strategy can fully utilize the advantages of different time periods, finding the optimal balance between efficiency and precision to meet the needs of different transaction scenarios.

[0020] The intelligent matching execution module, based on optimized historical transaction feature datasets and current foreign trade transaction demand information, executes supply chain matching operations according to a dynamic matching strategy. It integrates the advantages of the preceding modules, achieving intelligent and efficient matching. This module deeply fuses optimized historical feature data with current demand information, using intelligent algorithms to perform matching calculations according to dynamically generated strategies. This avoids subjective errors and information omissions inherent in manual matching, while also quickly responding to various changes during the matching process. This ensures that the matching results not only meet the user's current needs but also fully leverage historical experience, thereby improving the overall level of supply chain matching. Attached Figure Description

[0021] Figure 1 This is a timeline diagram of the intelligent supply chain matching system based on foreign trade transaction needs as described in this invention. Figure 2 Workflow diagram for the supply chain feature optimization module; Figure 3 Workflow diagram for the dynamic matching strategy generation module; Figure 4 This is a flowchart of the demand forecasting unit. Detailed Implementation

[0022] 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.

[0023] Please see Figure 1 The present invention provides a supply chain intelligent matching system based on foreign trade transaction demand. The system includes: a transaction demand collection module, a historical transaction data indexing module, a supply chain feature optimization module, a dynamic matching strategy generation module, and an intelligent matching execution module.

[0024] The transaction demand acquisition module obtains current foreign trade transaction demand information input by the user, including elements such as traded goods, quantity, and delivery time. The historical transaction data indexing module indexes a set of historical transaction feature data from the historical transaction database based on target transaction characteristics in the current foreign trade transaction demand information, such as product type or transaction region. The supply chain feature optimization module performs feature optimization processing on the historical transaction feature data set to generate an optimized historical transaction feature dataset. The dynamic matching strategy generation module calculates the remaining transaction time between the current time and the transaction deadline, and generates a dynamic matching strategy based on the remaining time. The intelligent matching execution module executes supply chain matching operations according to the dynamic matching strategy based on the optimized historical transaction feature dataset and the current foreign trade transaction demand information, thereby completing the intelligent matching of the supply chain.

[0025] Example 1: See Figure 2 The transaction demand acquisition module obtains order information input by the user, including product type (consumer electronics), quantity (1,000 units), target market (Europe), and delivery period (30 days). These elements constitute the target transaction characteristics in the current foreign trade transaction demand information. The historical transaction data indexing module then begins its work based on these characteristics. Its transaction characteristic classification unit scans the historical transaction database, which stores a large number of foreign trade transaction records from the past few years. The classification unit filters and groups the massive historical transaction characteristics according to attributes such as product type, target region, and transaction scale. For example, it groups all historical transaction records involving consumer electronics exported to Europe into one category, records of similar products exported to other regions into other categories, and further subdivides them according to transaction amount ranges. This process generates multiple logically clear and characteristically similar classification transaction characteristic categories.

[0026] The transaction metric extraction unit operates on each of the aforementioned categorized transaction feature categories. For the category "Consumer Electronics - Europe," this unit extracts a set of relevant historical transaction metrics. These metrics may include quantitative data such as average shipping time, distribution of common payment methods, tariff cost ratio, and average return rate for this type of transaction. Simultaneously, this unit extracts multiple sets of historical matching records. These records detail which suppliers, logistics providers, and payment channels were ultimately selected for each similar transaction in history, and record the final execution results of these matches, such as on-time delivery and customer satisfaction ratings.

[0027] The matching score generation unit is responsible for evaluating the quality of these historical matching records. The system pre-defines a matching benchmark, which defines various criteria for an ideal match, such as an expected on-time delivery rate of over 98% and cost control within a specific range. This unit calculates the degree of deviation of each historical matching record's metrics from the pre-determined matching benchmark. Records with small deviations are considered high-quality matches and receive higher scores; records with large deviations receive lower scores. By performing this evaluation on all historical matching records within a categorized transaction feature class, the generation unit ultimately outputs a set of historical matching scores corresponding to that class. This set contains multiple score values, each representing a quality score for a historical match.

[0028] The supply chain feature optimization module begins processing the generated set of historical matching scores to extract more valuable information. The benchmark feature selection unit selects a score from the set of historical matching scores corresponding to the "Consumer Electronics - Europe" category (i.e., the first set of historical matching scores) as the first benchmark matching score. This benchmark score may be the highest score in the set, representing the best historical matching level; or it may be the calculated average score, representing the average historical matching level.

[0029] The probability distribution construction unit uses the selected first benchmark matching score as a reference and calculates the magnitude of the difference between all other historical matching scores in the set and this benchmark score. Scores with larger difference magnitudes indicate that their matching effect differs significantly from the benchmark level, and their distribution probability will be assigned a lower probability; scores with smaller difference magnitudes have a higher distribution probability. Based on the difference magnitudes of all scores, this unit constructs a first feature probability distribution, which describes the likelihood of different score values ​​occurring.

[0030] The dimensionality reduction optimization unit utilizes this first feature probability distribution to optimize the original first historical matching score set. The principle is to retain scores corresponding to high-probability regions in the probability distribution; these scores represent matching records closer to the baseline level and contain more reliable and effective matching features. Conversely, scores corresponding to low-probability regions are filtered out, as these scores may represent abnormal or ineffective matching attempts. Through this probability distribution-based filtering, a large amount of redundant or low-value historical data is removed, achieving data dimensionality reduction and outputting a more refined and higher-quality first dimensionality-reduced matching score set. This optimized dataset focuses on the most successful or typical matching cases in history, providing higher-quality data input for subsequent intelligent matching execution. The entire process is automated, requiring no manual intervention. Through the classification, scoring, and probabilistic filtering of historical data, the original data is gradually transformed into an optimized feature set that can drive high-quality decision-making.

[0031] Example 2: See Figure 3The dynamic matching strategy generation module begins processing the current foreign trade transaction demand information. The deadline for this medical supplies order is 72 hours from the current time. The first strategy cycle conversion unit calls the system's preset first duration conversion rule. This rule is not a simple time division, but a non-linear mapping relationship derived from historical data analysis, taking into account the efficiency changes of matching operations under time pressure. It converts the remaining 72 hours of transaction time into a specific first matching cycle value, which represents the initially suggested matching operation time interval in a purely time dimension.

[0032] Specific methods for historical data analysis: A sliding window method (window duration set to 90 days, step size 15 days) is used to extract sample pairs of remaining transaction duration versus matching completion time from historical transaction data. Valid samples are then selected (samples with failed matches or missing data are removed) to form a sample set. ,in For the first The remaining time (in hours) for a historical transaction. The actual matching completion time for this transaction (in hours).

[0033] The nonlinear mapping relationship is constructed using the hyperbolic tangent function, with the first matching period... The calculation formula is:

[0034] in, The remaining duration of the current transaction (in hours); The amplitude coefficient (range 1.2-1.8, calibrated based on the fluctuation range of historical sample matching time; 1.8 is used when the sample fluctuation range is >30%, and 1.2 is used when the fluctuation range is <10%). The attenuation coefficient (range 0.005-0.012, calibrated according to time pressure sensitivity; 0.012 for high timeliness requirements and 0.008 for ordinary requirements). The offset coefficient (ranging from -50 to -20, calibrated based on the average remaining time of historical transactions; -50 is used when the average remaining time is >100 hours, and -20 is used when the average remaining time is <50 hours). The baseline period (the value is the minimum time for historical matching to complete, in hours).

[0035] when When under time pressure, Approaching 1, Approaching (Shorten the matching cycle to improve efficiency); when When time pressure is low, Approaching 0.3, Approaching (Extend the matching cycle to improve accuracy) to achieve dynamic adaptation between time pressure and matching efficiency.

[0036] The second strategy cycle conversion unit analyzes the number of transaction feature types contained in the current foreign trade transaction demand information. This medical supplies order involves relatively complex feature types, including special temperature-controlled transportation requirements, urgent customs clearance procedures, multimodal transport arrangements, and packaging that meets medical standards. The system counts the total number of these feature types and calls the preset second duration conversion rule. This rule defines the relationship between feature complexity and time adjustment; the more numerous and specialized the features, the more frequent the matching adjustments and monitoring are typically required. Based on the statistical results and the conversion rule, this unit calculates an adjustment cycle designed to address the additional matching needs arising from the complexity of the transaction.

[0037] Count the number of transaction characteristic types included in the current foreign trade transaction demand information. Transaction characteristics are categorized into core characteristics (such as product category, port of delivery, and payment method, denoted as...). ) and additional features (such as packaging requirements, insurance terms, after-sales guarantees, denoted as ), Statistical results Simultaneously, the average value of the number of features minus the adjustment period in similar historical transactions is calculated. To form a statistical data set .

[0038] Adjustment cycle The calculation formula is:

[0039] in, For statistical data sets Number of features in the middle The corresponding historical average adjustment duration (in hours), if Since there is no direct corresponding data, linear interpolation is used for calculation (e.g., hour, ); For the excess compensation coefficient of the number of features, when The time compensation coefficient is 0, the number of features is moderate, and no additional time adjustment is required. For each additional feature added, the adjustment time increases by 10% to address the matching adjustment needs arising from the increased feature complexity.

[0040] The cycle fusion unit receives two inputs: a first matching cycle and an adjustment cycle. Its fusion calculation is not a simple averaging or minimizing, but rather employs a weighted algorithm. The weight coefficients of this algorithm are dynamically adjusted based on the fundamental attributes of the traded goods. For time-sensitive and demanding goods like medical supplies, the algorithm assigns a greater weight to the adjustment cycle, resulting in a shorter second matching cycle than the first matching cycle, which is based solely on time calculations. This means the system will perform matching evaluations and adjustments more frequently. The strategy generation unit ultimately generates a dynamic matching strategy based on this second matching cycle. This strategy clearly defines the specific timeframes for when the system initiates matching analysis, executes matching operations, and conducts strategy reviews.

[0041] Second matching period The weighted summation formula is used for calculation, specifically:

[0042] And satisfy ,in, The weight for the first matching period (value range 0.3-0.7); To adjust the cycle weight (value range 0.3-0.7); the weight allocation rule is based on the priority of transaction timeliness. ( The values ​​range from 1 to 5, with 1 being the lowest priority and 5 being the highest priority. The specific correspondence is shown in Table 1 below.

[0043] Table 1: Correspondence Table of Transaction Timeliness Priority and Weight Allocation

[0044] When transaction timeliness priority hour, And adjustment cycle satisfy (Based on historical high-priority transaction statistics), substituting into the formula, we get: That is, when the adjustment cycle is given a greater weight ( During the second matching period Significantly smaller than the first matching period .

[0045] The intelligent matching execution module then operates according to the aforementioned dynamic matching strategy. The prediction model invocation unit first invokes the pre-trained supply chain demand prediction model. This model is a specialized model trained using machine learning methods based on a large amount of historical foreign trade transaction data, capable of understanding the complex relationship between different transaction characteristics and supply chain resource demands.

[0046] The demand forecasting unit strictly follows the second matching cycle specified in the dynamic matching strategy to control the supply chain demand forecasting model's calculations. The model simultaneously processes optimized historical transaction feature datasets and real-time demand information from current medical supply orders. The optimized historical dataset provides refined historical matching patterns similar to the current situation, while the real-time demand information incorporates specific parameters from the current transaction. Through its internal algorithmic structure, the model integrates these two pieces of information to generate forecast data about supply chain demand over a future period. This data may predict tight cargo space on specific shipping routes or extended customs clearance times at certain ports.

[0047] The matching execution unit ultimately performs substantive supply chain matching operations based on supply and demand forecast data. This unit connects to multiple supply chain resource databases, including freight forwarders, airlines, shipping companies, and warehousing service providers. Based on potential bottlenecks and opportunities identified by the forecast data, the unit automatically filters eligible potential service providers, generates matching solutions, and can automatically initiate bookings or push recommended solutions to operators according to system permission settings. Throughout the remaining time of the transaction, the system periodically repeats this forecasting and matching process. This second matching cycle ensures that the evaluation frequency matches the urgency and complexity of the transaction, thereby dynamically adjusting the matching strategy to cope with constantly changing market conditions. The entire process demonstrates the system's fully automated closed-loop processing capability, from generating time-based strategies to executing physical operations.

[0048] Example 3: After the dimensionality reduction optimization unit is started, it is first processed by the feature collaborative computation subunit. This subunit does not simply compare the numerical size of the scores, but rather deeply analyzes the intrinsic connections and collaborative relationships between the historical matching scores within the first set of historical matching scores. Each historical matching score corresponds to a set of complex transaction feature vectors, which encode multi-dimensional information of the original transaction. This subunit generates a feature collaborative representation vector by calculating the correlation strength between the feature vectors behind these scores. This vector quantifies the similarity or difference between the transaction cases represented by any two historical matching scores in the feature space. Its calculation process involves complex vector operations and similarity measurements, aiming to capture the deep, non-linear dependencies between scores.

[0049] The feature spectrum generation subunit receives the feature co-representation vector output by the feature co-computation subunit and extracts key statistical features from it. These statistical features are not simply the mean or variance, but higher-order features that characterize the distribution shape and energy concentration of the vector, such as statistics reflecting the distribution skewness or indicators representing the distribution kurtosis. These extracted statistical features are systematically organized to form a set of transaction feature energy distribution spectrum vectors. Each vector in this set uniquely represents the distribution characteristics of the original first historical matching score set from a specific perspective, much like observing the same set of data through multiple different spectral lenses, thus obtaining its comprehensive "energy distribution spectrum".

[0050] Building upon this, the distribution center calculation subunit integrates the set of transaction feature energy distribution spectrum vectors. It calculates the mean vector of all vectors in this set, using it as the transaction feature distribution center vector. This center vector is not a simple arithmetic mean, but a comprehensive representation after considering the weights of each distribution spectrum vector. It represents the core and central trend of the entire first historical matching score set after feature coordination and energy distribution analysis, serving as the benchmark anchor for subsequent screening operations.

[0051] The feature span calculation subunit performs the crucial computational steps, quantitatively evaluating the degree of difference between each individual vector in the set of trading feature energy distribution spectral vectors and the trading feature distribution center vector obtained in the previous step. This degree of difference is measured by a quantity called the feature span factor. The calculation of this factor involves a distance metric in the vector space, and its formula is as follows:

[0052] In this formula: Indicates the first The larger the value of the feature span factor corresponding to the energy distribution spectrum vector of each transaction feature, the higher the degree of deviation of the vector from the distribution center. The first in the set of energy distribution spectrum vectors representing transaction features A vector. This represents the previously calculated center vector of the transaction feature distribution. The L2 norm of a vector, also known as the Euclidean distance, is used to calculate the vector's L2 norm. With the center The absolute distance between them. This represents the total number of vectors in the set of energy distribution spectrum vectors representing trading features. The denominator calculates the standard deviation of the distances from all vectors to the center vector, which is used to standardize the absolute distances, making the feature span factor a relative value, facilitating comparisons between different datasets.

[0053] The feature filtering subunit filters the original first historical matching score set based on the comparison between the feature span factor and the system's preset threshold. Historical matching scores corresponding to the energy distribution spectrum vectors of trading features with a feature span factor less than or equal to the preset threshold are retained. This means that the trading cases they represent maintain a high degree of consistency with the central trend of the overall set in terms of feature distribution, and are considered more representative and reliable. Scores corresponding to vectors with a feature span factor significantly greater than the preset threshold are considered outliers or minor cases and are filtered out. Through this series of collaborative calculations, spectrum generation, center positioning, span measurement, and threshold filtering, the dimensionality reduction optimization unit ultimately outputs a first dimensionality-reduced matching score set that retains core distribution information while significantly reducing the amount of data, providing higher-quality and more focused input data for subsequent matching strategy generation. The entire process achieves a balance between data simplification and information preservation through rigorous mathematical calculations and logical judgments.

[0054] Example 4: See Figure 4 Suppose the system is processing an export order for a batch of fresh fruit, targeting overseas markets. The demand forecasting unit first receives an optimized historical transaction feature dataset from the upstream module. This dataset contains refined data on similar past fruit export transactions, such as historical prices, shipping volume patterns, transportation time, and typical spoilage rates. Simultaneously, the system continuously acquires real-time transaction feature data related to this order through a data interface. This data may include the latest daily purchase price at the production site, weather forecasts for the coming week (affecting harvesting and transportation), real-time freight rate changes in the international shipping market, and the real-time inventory depletion rate at the target sales location.

[0055] The offset coefficient calculation subunit begins processing these two sets of data. Its internal feature response calculation component does not simply compare numerical differences, but rather analyzes the correlation and response patterns between the two sets of data across different feature dimensions. For example, it analyzes the correlation curve between price and shipment volume in historical data and compares it with the new price and actual shipment volume presented in real-time data, calculating a feature response value that characterizes the dynamic response relationship between the two. This value quantifies the intensity of change and behavioral pattern differences in real-time data relative to historical patterns.

[0056] The convolutional coding component processes the feature response value, scanning and encoding it using analysis windows of multiple different scales (analogous to different time granularities or data aggregation granularities). This multi-scale processing can simultaneously capture short-term sudden fluctuations and long-term trend changes. Through this process, the complex response relationship is transformed into a structured, multi-dimensional feature co-distribution matrix. The rows and columns of this matrix may represent different feature dimensions and time scales, while the element values ​​characterize the degree of co-occurrence or divergence between historical patterns and real-time conditions at that specific dimension and scale.

[0057] The offset generation component is responsible for parsing this feature co-distribution matrix. This component employs a sequence matching network, an algorithmic structure capable of processing serialized data and recognizing patterns, to perform in-depth analysis of the sequence information contained in the matrix. It identifies pattern sequences within the matrix, such as a continuous small increase in freight rates accompanied by a steady decrease in shipment volume, and matches and compares these with similar historical sequence patterns, ultimately generating a comprehensive, quantitative trading feature offset coefficient. This coefficient is a comprehensive indicator reflecting the overall degree of deviation of the current real-time trading environment from historical benchmark patterns.

[0058] The rate of change generation subunit receives this transaction feature offset coefficient. Instead of using the coefficient directly, it examines it over time, analyzing its trajectory and rate of change. For example, it might observe that the coefficient has been continuously increasing over the past few analysis periods, and the rate of increase is accelerating. By calculating the ratio of the change in the offset coefficient to the change over time, this subunit generates a supply-demand change rate. This rate of change is a dynamic indicator that predicts the possible future direction and speed of changes in supply and demand. See Table 2, which shows a segment of real-time transaction feature data used for the analysis at a specific moment.

[0059] Table 2: Sample Table of Real-Time Transaction Characteristics of Fruit Export Orders

[0060] The final correction subunit performs the correction operation. It receives the initial supply and demand forecast data generated by the supply chain demand forecasting model, which is primarily based on optimized historical transaction feature datasets. Simultaneously, it receives the supply and demand change rates output by the change rate generation subunit. The correction logic involves incorporating the future trend indicated by the change rate into the static historical forecast baseline. For example, if the initial forecast data shows stable future demand, but the supply and demand change rate indicates a clear trend of declining demand (as implied by decreasing inventory absorption and increasing costs in the table above), the correction subunit will adjust the final forecast data output downwards. Conversely, if the change rate indicates a positive trend, the forecast data will be adjusted upwards accordingly. This entire process allows the system's forecasting capabilities to go beyond simple extrapolation of historical patterns, enabling it to adapt to real-time market dynamics and providing subsequent matching execution units with a more realistic decision-making basis.

[0061] Example 5: In a specific application scenario, the system is processing a batch of export orders for industrial machinery parts. The demand forecasting unit has generated supply and demand forecast data, which may indicate that in the near future, the supply of shipping space will tend to be tight, and the demand for packaging materials for certain specifications of parts will increase. At this time, the decision-maker invocation unit is activated. This unit invokes the system's preset matching strategy decision-maker. This decision-maker is not a simple rule base, but a composite intelligent agent integrating multiple decision-making logics. It may internally encapsulate a rule-based inference engine for handling decisions with explicit constraints (e.g., prioritizing suppliers with certain qualifications); simultaneously, it may also integrate a scorer based on a lightweight machine learning model for quantitatively evaluating the potential risks or overall benefits of candidate solutions. The configuration parameters and logical weights of the decision-maker can be preset according to different trade categories and company strategies.

[0062] The solution generation unit combines the received supply and demand forecast data with preset matching rules and processes it through the aforementioned matching strategy decision-maker. The preset matching rules are a set of business constraints and objectives, such as: a cost minimization rule, requiring that the total cost of the selected solution cannot exceed a certain threshold; a timeliness priority rule, stipulating that under certain conditions, delivery speed takes precedence over cost considerations; and a risk diversification rule, requiring avoidance of over-reliance on a single supplier or a single transportation route. The decision-maker uses the supply and demand forecast data as the expected future market scenario and the preset matching rules as the boundary conditions and optimization objectives for decision-making, comprehensively evaluating and calculating solution combinations for all currently available supply chain resources (including suppliers, logistics providers, customs brokers, etc.).

[0063] This process is not a simple selection, but a multi-objective optimization problem. The decision-maker may generate multiple non-dominant feasible solutions. For example, for the aforementioned mechanical parts order, based on the predicted tight space and increased demand for packaging materials, the decision-maker might output three alternative solutions: Solution A suggests immediately booking currently available but slightly more expensive space and signing an emergency supply agreement with a recommended packaging material supplier to secure resources; Solution B suggests adopting a segmented transportation strategy, first transporting the goods by land to another port city and then utilizing the more abundant space resources there, while simultaneously initiating the inquiry process for alternative packaging suppliers; Solution C suggests negotiating with the core supplier to slightly postpone the delivery date of some orders to avoid peak transportation periods and using standard packaging materials instead of special-specification materials to reduce supply risks. Each alternative solution includes specific operational steps, a list of expected resources, cost estimates, and a risk description.

[0064] The matching execution unit is ultimately responsible for executing the final selected or optimized supply chain matching suggestion plan. This unit has an integration interface with external supply chain management systems and databases. Based on the specific instructions of the suggestion plan, the matching execution unit can automatically perform a series of operations. For example, if Plan A is adopted, the unit may automatically send a booking request to the selected shipping company's booking system, generate a purchase order, place an order with the designated packaging material supplier, and simultaneously update all these operations in the order's tracking log. The entire execution process is highly automated, requiring no manual intervention at each step. The system monitors key nodes in the execution process, such as whether space confirmation is successfully obtained and whether the supplier accepts the order, and triggers warnings or automatically activates backup plans when obstacles are encountered. For situations requiring final manual confirmation or complex negotiation, the matching execution unit pushes the complete suggestion plan to relevant operators, along with the decision-maker's recommendation reasons and key data support, to assist personnel in decision-making. From invoking the decision-maker to generating the suggestion plan, and then to final execution, this implementation method completes a closed loop from data prediction to actual action, demonstrating the system's intelligent decision-making and automated execution capabilities.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A supply chain intelligent matching system based on foreign trade transaction needs, characterized in that, The system includes: The transaction demand collection module is used to obtain the current foreign trade transaction demand information input by the user; The historical transaction data indexing module is used to index the historical transaction feature data set in the historical transaction database based on the target transaction features in the current foreign trade transaction demand information. The supply chain feature optimization module is used to perform feature optimization processing on the historical transaction feature data set to generate an optimized historical transaction feature dataset. The dynamic matching strategy generation module is used to calculate the remaining time of the transaction between the current time and the transaction deadline, and generate a dynamic matching strategy based on the remaining time of the transaction. The intelligent matching execution module is used to perform supply chain matching operations according to the dynamic matching strategy based on the optimized historical transaction feature dataset and the current foreign trade transaction demand information.

2. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 1, characterized in that, The historical transaction data index module includes: The transaction feature classification unit is used to classify the historical transaction features within the historical transaction feature data set to obtain multiple classified transaction feature categories; The transaction indicator extraction unit is used to extract the historical transaction indicator set and the historical matching record set corresponding to the multiple categorized transaction feature categories. The matching score generation unit is used to generate multiple historical matching score sets based on the degree of deviation between the historical matching record set and the preset matching benchmark.

3. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 2, characterized in that, The supply chain feature optimization module includes: A benchmark feature selection unit is used to select a first benchmark matching score within a first historical matching score set among the plurality of historical matching score sets. The probability distribution construction unit is used to allocate feature distribution probabilities to form a first feature probability distribution based on the difference between other historical matching scores in the first historical matching score set and the first benchmark matching score. The dimensionality reduction optimization unit is used to perform dimensionality reduction optimization on the first historical matching score set based on the first feature probability distribution, and output the first dimensionality-reduced matching score set.

4. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 3, characterized in that, The dynamic matching strategy generation module includes: The first strategy cycle conversion unit is used to call the preset first duration conversion rule to convert the remaining duration of the transaction into the first matching cycle. The second strategy cycle conversion unit is used to count the number of transaction feature types contained in the current foreign trade transaction demand information and call the preset second duration conversion rule to calculate the adjustment cycle. A periodic fusion unit is used to fuse the first matching period and the adjustment period to generate a second matching period; The strategy generation unit is used to generate the dynamic matching strategy based on the second matching period.

5. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 4, characterized in that, The intelligent matching execution module includes: The prediction model invocation unit is used to invoke a pre-trained supply chain demand prediction model. The demand forecasting unit is used to control the supply chain demand forecasting model to process the optimized historical transaction feature dataset and the current foreign trade transaction demand information according to the second matching cycle, so as to generate supply and demand forecast data. The matching execution unit is used to perform supply chain matching operations based on the supply and demand forecast data.

6. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 3, characterized in that, The dimensionality reduction optimization unit includes: The feature collaborative computation subunit is used to calculate the feature collaborative representation vector among the historical matching scores in the first historical matching score set. The feature spectrum generation subunit is used to extract statistical features from the feature collaborative representation vector to form a set of transaction feature energy distribution spectrum vectors; The distribution center calculation subunit is used to calculate the mean vector of the set of energy distribution spectrum vectors of the transaction features as the distribution center vector of the transaction features.

7. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 6, characterized in that, The dimensionality reduction optimization unit also includes: The feature span calculation subunit is used to calculate the feature span factor between the center vector of the transaction feature distribution and the energy distribution spectrum vector of each transaction feature; The feature filtering subunit is used to filter and form the first dimensionality reduction matching score set based on the comparison result of the feature span factor and the preset threshold.

8. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 5, characterized in that, The demand forecasting unit includes: The offset coefficient calculation subunit is used to calculate the transaction feature offset coefficient between the optimized historical transaction feature dataset and the real-time transaction feature data. The rate of change generation subunit is used to generate the rate of change of supply and demand based on the proportional relationship between the transaction feature offset coefficient and the time dimension. The forecast correction subunit is used to correct the supply and demand forecast data based on the supply and demand change rate.

9. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 8, characterized in that, The offset coefficient calculation subunit includes: A feature response calculation component is used to calculate the feature response value between the optimized historical transaction feature dataset and the real-time transaction feature data; A convolutional coding component is used to perform multi-scale convolutional coding on the feature response values ​​to generate a feature cooperative distribution matrix; An offset generation component is used to process the feature co-distribution matrix through a sequence matching network to generate the transaction feature offset coefficients.

10. The intelligent supply chain matching system based on foreign trade transaction demand according to claim 5, characterized in that, The matching execution module further includes: The decision-maker invocation unit is used to invoke a preset matching strategy decision-maker; The prompt scheme generation unit is used to generate a supply chain matching prompt scheme based on the supply demand forecast data and preset matching rules, through the matching strategy decision-maker. The matching execution unit is used to perform the matching operation corresponding to the supply chain matching prompt scheme.

Images