An intelligent management system and platform for automobile import and export supply chain

Abstract

The application relates to the technical field of supply chain management, and particularly discloses an intelligent management system for an automobile import and export supply chain. The system comprises a compliance element real-time analysis module, a logistics network state sensing module, a risk-cost dynamic coupling engine, a multi-constraint K-optimal path solver, and a path robustness evaluation and selection module. Through deep coupling calculation of the fine compliance attributes of automobile parts and the real-time logistics network state, the risk additional cost and time are quantitatively generated; then, K optimal paths satisfying multiple constraints are solved; finally, the robustness of each path is evaluated based on Monte Carlo simulation, and the optimal scheduling instruction is generated. The application solves the fundamental defects that the compliance requirements and dynamic logistics in the cross-country circulation of automobile parts are difficult to quantitatively fuse and the path planning lacks risk elasticity evaluation, and realizes intelligent closed-loop optimization of the supply chain.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and in particular to an intelligent management system and platform for automobile import and export supply chain. Background Technology

[0002] In the globalized automotive industry, the import and export supply chain of complete vehicles and auto parts involves multiple complex links such as cross-border logistics, customs compliance, and tax calculation. Its efficient and reliable operation is crucial to automakers' inventory costs, production pace, and market responsiveness. With the volatile trade environment and increasingly complex logistics networks, traditional supply chain management systems mainly rely on experience rules and static data. They are unable to conduct real-time, quantitative cross-analysis of dynamic risks during transportation (such as port congestion and policy changes) and the static compliance attributes of the goods themselves (such as HS codes and rules of origin). This leads to logistics routes that seem optimal in terms of cost or timeliness failing in actual execution due to unforeseen compliance delays or sudden congestion, resulting in huge unplanned costs and time losses.

[0003] Existing technological solutions typically handle logistics tracking and compliance verification in isolation, or simply combine the two, lacking a mechanism that can deeply couple and calculate the unique and refined compliance elements specific to automotive parts (such as regulatory conditions corresponding to specific HS codes and rules of origin under free trade agreements) with real-time dynamic logistics network conditions (such as the real-time congestion probability of nodes). More importantly, after generating alternative routes, existing solutions lack quantitative simulations based on risk probabilities to assess the robustness of the routes in actual execution (i.e., their ability to withstand uncertainty), preventing decision-makers from making a scientific trade-off between "cost," "timeliness," and "delivery reliability." Therefore, there is an urgent need for an intelligent system that can deeply integrate compliance risks and logistics dynamics and simulate the risk resistance capabilities of planned routes to address the fundamental technological deficiency of "severe disconnect between planning and execution" in the automotive import and export supply chain.

[0004] Therefore, this invention proposes an intelligent management system and platform for the automobile import and export supply chain. Summary of the Invention

[0005] This invention provides an intelligent management system and platform for the automotive import and export supply chain, which overcomes the shortcomings of existing technologies that suffer from "disconnect between planning and execution" due to the fragmented handling of compliance attributes and logistics dynamics and the lack of risk flexibility assessment. It realizes intelligent and robust supply chain path planning based on quantitative risk coupling and simulation verification.

[0006] This invention provides an intelligent management system for the automobile import and export supply chain, comprising: The real-time compliance element analysis module is used to connect with the customs regulations database and order data. Based on the automotive parts feature library, it maps the descriptions of non-standard parts in the order into structured compliance element vectors with definite HS codes, country of origin, applicable free trade agreements and regulatory conditions in real time. The logistics network status perception module is used to access real-time operational data from ports, shipping companies, and land transportation service providers. Combined with historical data from the same period, it generates the congestion probability and expected delay time of each logistics node in terms of time windows, and outputs a network status vector. The risk-cost dynamic coupling engine is used to dynamically calculate the risk-added costs and risk-added time of each shipment on each candidate logistics route, based on compliance element vectors and network state vectors and according to a pre-set coupling rule base. The multi-constraint K-optimal path solver is used to superimpose the basic transportation cost, basic transportation time, and risk-added cost and risk-added time to construct a time-varying multi-weight transportation network graph. Based on an improved path search algorithm that avoids high-risk node overlap, it solves the top K optimal paths in the multi-weight transportation network graph that satisfy the constraints of the vehicle delivery time window and special regulatory conditions for parts. The path robustness assessment and decision-making module is used to input the top K optimal paths into a transportation robustness assessment model based on Monte Carlo simulation. It performs random extrapolation using the delay probability of each node in the optimal path as a parameter. Based on the on-time delivery guarantee rate and cost fluctuation range output by the extrapolation results, and combined with the main strategy preferences, it selects the optimal path with the highest comprehensive robustness score to generate the final scheduling instruction.

[0007] The preferred working process of the compliance element real-time analysis module includes: Construct an automotive parts feature library, in which each record is associated with a standard part number. The standard part number is bound to an authoritative HS code, country of origin information, applicable free trade agreement list, and special regulatory conditions. The system receives non-standard component description text from order data and encodes it into semantic vectors using a natural language processing model based on an attention mechanism. The semantic vector is matched with the standard information vector of all records in the automotive parts feature library for similarity. All compliant elements bound to the standard part number with the highest similarity are output as the mapping result of the current order.

[0008] Preferably, the pre-built coupling rule base in the risk-cost dynamic coupling engine includes the following rules: Tariff and Trade Agreement Rules: Based on the country of origin and HS code of the goods in the compliance element vector, combined with the currently effective free trade agreements and tariff schedules, calculate the tariffs payable or exemptions when the goods are imported along the candidate route. The tariffs payable or exemptions are included as part of the risk-added costs. Regulatory Condition Conflict Rule: Compare the special regulatory conditions of goods in the compliance element vector with the processing capacity list of each logistics node on the candidate path. If a logistics node cannot meet any special regulatory condition, a fixed cost and time penalty value representing extremely high risk will be added to the corresponding logistics node for the path. Node delay probability fusion rule: For nodes on the candidate path, obtain the current congestion probability of the node provided by the logistics network status perception module, and the probability distribution of additional customs clearance time for each category of goods at the node based on historical data statistics. The current congestion probability of the node and the probability distribution of additional customs clearance time are fused to calculate the comprehensive expected delay time. The comprehensive expected delay time is used as the risk-added time of the node.

[0009] The preferred workflow of the improved path search algorithm that avoids high-risk node overlap includes: After finding the optimal paths from the 1st to the (k-1)th path, when searching for the kth path, edges in the multi-weight transportation network graph that have been completely occupied by the first (k-1)th paths are excluded. Identify high-risk nodes marked by the risk-cost dynamic coupling engine in a multi-weighted transportation network graph; Calculate the size of the intersection between the set of high-risk nodes traversed by the current candidate path and the set of high-risk nodes traversed by each found path; When the intersection size exceeds a preset threshold, it is determined that the current candidate path and the existing path overlap excessively in risk distribution, the current candidate path is abandoned, and the search continues for the next path.

[0010] Preferred criteria for marking high-risk nodes include: The nodes reported by the logistics network status perception module whose current congestion probability exceeds the first set threshold; Customs nodes that, based on historical data statistics, have an inspection rate exceeding the second set threshold for the category to which the current goods' HS code belongs. The node is determined to be unable to process the current goods based on the regulatory condition conflict rules in the coupled rule base.

[0011] Preferably, the Monte Carlo simulation process for the transport robustness assessment model includes: For nodes in the candidate path, configure a probability distribution model representing the node delay time based on the node type and current state; In a single simulation, based on the probability distribution model of each node, the possible delay values ​​of the nodes are generated independently and randomly. The possible delay values ​​of all nodes are summed to obtain the total time of a single simulation. Calculate the total cost of a single simulation based on the basic cost of the path, the additional cost of risks, and the node delays. After repeating the simulation process a predetermined number of times, the proportion of simulations in which the total time does not exceed the order's required delivery time window is counted as the on-time delivery guarantee rate of the path. The distribution of the total cost of all simulations is also counted to assess the range of cost fluctuations.

[0012] Preferably, the process by which the path robustness assessment and decision-making module calculates a comprehensive robustness score based on the subject's strategy preferences includes: When the primary strategy prioritizes timeliness, a higher weight is given to on-time delivery assurance rate in the scoring calculation; When the main strategy preference is cost-first, the scoring calculation places greater emphasis on low volatility and stability of cost distribution; The on-time delivery guarantee rate and cost assessment results of each path are weighted and combined, and the path with the highest score is selected as the final scheduling scheme.

[0013] Preferably, it also includes an adaptive optimization module, used for: Continuously collect data on actual logistics and customs clearance results executed according to the final scheduling instructions; The actual time and cost of nodes in the actual results data are compared with the predicted values ​​calculated by the risk-cost dynamic coupling engine. When there is a systematic deviation between the predicted value and the actual value for a specific type, the parameters of the corresponding rule in the coupled rule base are calibrated to reduce the error in future predictions.

[0014] Preferably, the compliance element real-time analysis module, the logistics network status perception module, and the risk-cost dynamic coupling engine are coordinated through a unified data clock and version number, including: When processing data, the compliance element real-time analysis module and the logistics network status perception module mark the output results with a unified timestamp and data processing version number; When performing calculations, the risk-cost dynamic coupling engine verifies whether the received compliance element vector and network state vector have matching timestamps and version numbers, ensuring that the data used for coupling calculations is consistent in time and processing logic.

[0015] This invention provides an intelligent management platform for the automobile import and export supply chain, including a server and at least one intelligent management system for the automobile import and export supply chain as described above, wherein the intelligent management system for the automobile import and export supply chain is communicatively connected to the server.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: By designing a collaborative computing system that includes real-time analysis of compliance elements, dynamic perception of logistics networks, quantitative coupling of risk and cost, path optimization solutions, and robustness assessment, this invention creatively solves two fundamental defects in existing automotive supply chain management technologies: First, it cannot deeply integrate and quantitatively analyze refined, static component compliance requirements (such as HS codes and rules of origin) with dynamically changing logistics network conditions (such as real-time congestion probability), leading to one-sided risk assessment; second, path planning is based solely on fixed cost and time models, lacking simulation of uncertainties and assessment of risk resistance capabilities, causing "paper-optimal" paths to fail in actual execution due to various unforeseen risks. This system can fully activate the value of industrial big data and, relying on the industrial internet platform, achieve cross-node and cross-entity data collaboration and scheduling, strengthening the overall collaborative capabilities of the supply chain. This solution realizes a closed loop from heterogeneous data to intelligent decision-making, providing a dynamic optimization solution for automobile import and export that combines economy, timeliness, and reliability.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a core module architecture diagram of an intelligent management system for automobile import and export supply chain in an embodiment of the present invention; Figure 2 This is a closed-loop diagram of the key technical processes in the embodiments of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 and Figure 2 As shown, this invention provides an embodiment of an intelligent management system for the automotive import and export supply chain, comprising: The real-time compliance element analysis module is used to connect with the customs regulations database and order data. Based on the automotive parts feature library, it maps the descriptions of non-standard parts in the order into structured compliance element vectors with definite HS codes, country of origin, applicable free trade agreements and regulatory conditions in real time. The logistics network status perception module is used to access real-time operational data from ports, shipping companies, and land transportation service providers. Combined with historical data from the same period, it generates the congestion probability and expected delay time of each logistics node in terms of time windows, and outputs a network status vector. The risk-cost dynamic coupling engine is used to dynamically calculate the risk-added costs and risk-added time of each shipment on each candidate logistics route, based on compliance element vectors and network state vectors and according to a pre-set coupling rule base. The multi-constraint K-optimal path solver is used to superimpose the basic transportation cost, basic transportation time, and risk-added cost and risk-added time to construct a time-varying multi-weight transportation network graph. Based on an improved path search algorithm that avoids high-risk node overlap, it solves the top K optimal paths in the multi-weight transportation network graph that satisfy the constraints of the vehicle delivery time window and special regulatory conditions for parts. The path robustness assessment and decision-making module is used to input the top K optimal paths into a transportation robustness assessment model based on Monte Carlo simulation. It performs random extrapolation using the delay probability of each node in the optimal path as a parameter. Based on the on-time delivery guarantee rate and cost fluctuation range output by the extrapolation results, and combined with the main strategy preferences, it selects the optimal path with the highest comprehensive robustness score to generate the final scheduling instruction.

[0022] In this embodiment, the customs regulations database refers to a collection of data, including laws and regulations, tariff codes, free trade agreement texts, commodity classification decisions, and import and export control lists, which are continuously collected and structured from official sources such as customs and commerce departments of various countries. It is an authoritative basis for determining the compliance requirements of goods.

[0023] In this embodiment, order data refers to the electronic records of commercial documents generated in automobile import and export transactions. The core fields include information on the buyer and seller, description of goods (including name, model and quantity), trade terms, delivery and receipt locations and the delivery deadline agreed in the contract.

[0024] In this embodiment, the automotive parts feature library is a pre-built standardized knowledge base, in which each record corresponds to a unique standard part, and is forcibly associated with the part's commodity code (HS code), country of origin, list of applicable preferential trade agreements, and special conditions that must be met during transportation and storage (such as temperature control and shockproofing) determined by official or industry authorities.

[0025] In this embodiment, the non-standard component description refers to the description text of goods filled in the order that does not use the standard naming in the above feature library. It is often the supplier's internal code, product common name, abbreviation, or a non-standard phrase containing specification parameters.

[0026] In this embodiment, the process of mapping non-standard parts descriptions in an order to a structured compliance element vector with definite HS codes, country of origin, applicable free trade agreements, and regulatory conditions in real time, based on the automotive parts feature library, is as follows: The system receives the natural language description text in the order and first converts it into a high-dimensional vector representing semantics (semantic vector) through a natural language processing model based on an attention mechanism; then, it calculates the similarity between this semantic vector and the high-dimensional vectors (standard information vectors) corresponding to all standard parts in the automotive parts feature library; finally, it extracts the complete compliance attributes (i.e., definite HS codes, country of origin, applicable free trade agreements, and regulatory conditions) bound to the standard parts with the highest similarity in the feature library and combines them into a structured data object, namely, the compliance element vector.

[0027] In this embodiment, determining the HS code refers to identifying the unique, globally recognized standard classification code under the Harmonized Commodity Description and Coding System (HS) for the goods in an order through the above mapping process.

[0028] In this embodiment, the real-time operational data of ports, shipping companies, and land transport service providers refers to dynamic status information obtained in real time from the information systems of relevant operators through application programming interfaces, such as the current berth occupancy rate of ports, the real-time geographical location and estimated arrival time of ships, the operating efficiency of terminal cranes, the GPS location and task status of container trucks, and the real-time inventory capacity of warehouses.

[0029] In this embodiment, historical data refers to historical performance datasets such as operational efficiency, congestion event records, and average processing time collected from the same logistics node within the same or similar time period in the past (e.g., the same week last year or the same working day last month).

[0030] In this embodiment, logistics nodes refer to key physical sites or logical processing points that goods must pass through or may stop at during cross-border circulation, including ports of departure, transshipment ports, ports of destination, border crossings, inland customs supervision stations, distribution warehouses, etc.

[0031] In this embodiment, the network state vector is a structured data output, each element of which corresponds to a monitored logistics node and contains two key prediction values: one is the probability that the node will experience operational congestion leading to delays within a specific time window in the future; the other is the average possible delay time if a delay occurs, predicted based on historical patterns.

[0032] In this embodiment, the process of constructing a network state vector involves accessing real-time operational data from ports, shipping companies, and land transport service providers, and combining this data with historical data from the same period in the same timeframe. This data generates the congestion probability and expected delay duration for each logistics node within a time window. The output is as follows: The system continuously accesses real-time operational data and corresponding historical data from the same period in the same timeframe as input. For each logistics node, a time series prediction model or machine learning regression model is used to analyze the operational status within a set time window. The model outputs the congestion probability and expected delay duration for that node within that time window. These two predicted values ​​for all nodes are arranged in a predetermined order to form a network state vector.

[0033] In this embodiment, the pre-set coupling rule base is a collection of multiple condition-action rules stored in the system. Each rule clearly defines how to calculate the additional costs or time delays caused when a specific cargo compliance attribute and a specific logistics node status occur in combination.

[0034] In this embodiment, based on compliance element vectors and network state vectors, and according to a pre-set coupled rule base, the system dynamically calculates the additional risk cost and additional risk time for each shipment on each candidate logistics path due to the interaction between compliance requirements and logistics status. The calculation process is as follows: For a given shipment (its compliance element vector) and a candidate logistics path consisting of several nodes, the system traverses the coupled rule base; for each node on the path, it searches for all rules triggered by the node's real-time status and shipment attributes; each triggered rule executes its internally defined calculation logic, for example, one rule may calculate the tariff difference due to discrepancies in country of origin, and another rule may calculate the additional customs clearance time caused by the superposition of node congestion and high shipment inspection rate; the cost increment and time increment calculated by the rules triggered by all nodes on the path are summed to obtain the additional risk cost and additional risk time for the shipment on that path.

[0035] In this embodiment, the construction of a time-varying multi-weighted transportation network graph by superimposing basic transportation costs, basic transportation time, risk-added costs, and risk-added time means that the system constructs a graph structure with geographical or logical transportation hubs as vertices and transportation routes as edges; each edge is assigned multiple weights, where the basic transportation cost and basic transportation time are derived from contractual or market standard rates; the dynamically calculated risk-added costs and risk-added time are accumulated onto the weights of the relevant edges according to their corresponding relationships, so that the edge weights of the entire graph can reflect the real-time, cargo-specific comprehensive transportation costs, forming a time-varying multi-weighted transportation network graph.

[0036] In this embodiment, each edge of the multi-weighted transportation network graph contains at least three weight dimensions: basic transportation cost, basic transportation time, and a dynamically calculated risk factor (which can be converted into cost increment and time increment simultaneously).

[0037] In this embodiment, based on an improved path search algorithm that avoids overlapping high-risk nodes, finding the top K optimal paths in a multi-weighted transportation network graph that satisfy the constraints of the vehicle delivery time window and special regulatory conditions for parts refers to: running an improved K-shortest path search algorithm on the constructed time-varying multi-weighted transportation network graph with comprehensive cost or comprehensive time as the optimization objective; the improvement of this algorithm lies in that, during the search process, it not only finds the path with the lowest cost, but also checks the degree of overlap between the candidate path and the found path on the set of "high-risk nodes". If the overlap exceeds a preset threshold, the candidate path is discarded to seek a solution with a more dispersed risk distribution; at the same time, the algorithm enforces two hard constraints throughout the search process: first, the total time of the entire path must be earlier than the final delivery time required by the order; second, each node on the path must have the ability to handle all special regulatory conditions for the goods; the algorithm finally outputs the top K paths that satisfy all constraints and are sorted by cost from smallest to largest.

[0038] In this embodiment, the improved path search algorithm that avoids high-risk node overlap specifically refers to the K-shortest path algorithm mentioned above, which incorporates a high-risk node overlap check and screening mechanism during the search process.

[0039] In this embodiment, satisfying the vehicle delivery time window and special regulatory conditions for parts means that any accepted transportation route must simultaneously meet two mandatory conditions: First, the estimated total transportation time from the origin to the destination must not exceed the latest delivery time stipulated in the contract; second, every transit or processing node along the route must be officially authorized or actually have the conditions to handle all special requirements for the goods (such as dangerous goods qualifications, temperature-controlled warehousing).

[0040] In this embodiment, the transportation robustness assessment model based on Monte Carlo simulation refers to a probabilistic analysis model that uses Monte Carlo random sampling to evaluate the performance of transportation routes under uncertain environments.

[0041] In this embodiment, the top K optimal paths are input into a transportation robustness assessment model based on Monte Carlo simulation. Random simulations are performed using the delay probability of each node in the optimal path as parameters. Based on the on-time delivery guarantee rate and cost fluctuation range output from the simulation results, and combined with the user's policy preferences, the optimal path with the highest comprehensive robustness score is selected to generate the final scheduling instruction. The assessment and decision-making process is as follows: The model receives K candidate paths and the delay probability distribution of each node on each path as input; it performs a large number of random simulations. In each simulation, a possible delay value is independently and randomly generated based on the probability distribution of each node, and the total transportation time for that simulation is accumulated, while the total cost is calculated simultaneously; after all simulations are completed, the proportion of simulations where the total time does not exceed the delivery deadline is calculated to obtain the on-time delivery guarantee rate; and the statistical distribution of the total cost is analyzed to obtain the cost fluctuation range; based on the user's preset policy preferences (such as cost priority), the system transforms the guarantee rate and cost performance of each path into a single comprehensive robustness score through a weighted aggregation formula; finally, the path with the highest score is selected to generate the final scheduling instruction containing its detailed route, time plan, and service requirements.

[0042] In this embodiment, the on-time delivery guarantee rate and the cost fluctuation range are two core quantitative indicators output by the above evaluation model: the on-time delivery guarantee rate is a probability value, representing the possibility that the route can still complete the transportation on time under random disturbances; the cost fluctuation range is an interval estimate, characterizing the magnitude of possible changes in the total cost of the route, usually represented by a high quantile.

[0043] In this embodiment, the subject strategy preference refers to the strategy options that are pre-set by the decision-maker to weigh the priority between time reliability and economy in the transportation plan, such as "extreme cost reduction", "ensuring timeliness" or "balanced consideration".

[0044] In this embodiment, the optimal path with the highest comprehensive robustness score refers to the transportation path that achieves the best balance between time reliability and cost controllability after Monte Carlo simulation evaluation and weighted calculation according to the set subject strategy preferences.

[0045] In this embodiment, the final scheduling instruction refers to the set of electronic instructions output by the system decision-making process that can be directly identified and processed by the downstream execution system. It clearly specifies the selected transportation route, the planned time for each node, the designated carrier, and the special requirements for cargo handling.

[0046] To address the issue of accurate and automatic mapping from descriptions of non-standard components to standard compliance information, a real-time compliance element parsing module is proposed, comprising the following steps: Construct an automotive parts feature library, in which each record is associated with a standard part number. The standard part number is bound to an authoritative HS code, country of origin information, applicable free trade agreement list, and special regulatory conditions. The system receives non-standard component description text from order data and encodes it into semantic vectors using a natural language processing model based on an attention mechanism. The semantic vector is matched with the standard information vector of all records in the automotive parts feature library for similarity. All compliant elements bound to the standard part number with the highest similarity are output as the mapping result of the current order.

[0047] In this embodiment, each record in the automotive parts feature library is associated with a standard part number. The standard part number is bound to an authoritative HS code, country of origin information, a list of applicable free trade agreements, and special regulatory conditions. This means that the feature library uses the standard part number as the primary key, and each data record is forcibly and one-to-one associated with a set of compliance attribute data recognized and published by official or industry authorities. These attribute data specifically include: a globally recognized and legally valid commodity classification code, information on the country or region of origin used to determine tariffs and trade policies, a list of effective free trade agreements that allow the commodity to enjoy tariff preferences, and special operational or management requirements that must be followed in transportation, warehousing, or customs clearance.

[0048] In this embodiment, the non-standard component description text in the received order data refers to the free-format text content, which is usually manually entered, extracted from the electronic order information transmitted from the outside by the system to describe the goods to be imported or exported.

[0049] In this embodiment, the non-standard component description text is encoded into a semantic vector using a natural language processing model based on an attention mechanism. This step refers to the following: the system inputs the received description text into a pre-trained natural language understanding model using an attention mechanism (such as a Transformer architecture model); the model analyzes the contextual relationships and importance of words in the text through its internal multi-layer neural network structure, and converts (encodes) the meaning of the entire text into a fixed-length high-dimensional numerical array, which is the semantic vector, and it can compactly represent the core semantic information of the text.

[0050] In this embodiment, the attention-based natural language processing model specifically refers to a type of artificial neural network model that uses an attention mechanism to endow the model with the ability to pay attention to different parts of the input text with different weights. In this invention, this model is used to understand and extract the deep semantics of non-standard component description text and encode it into a computer-processable vector form. Its construction process is as follows: the model is trained using a massive corpus of text containing standard component descriptions, so that it learns how to effectively extract key semantic features related to compliance attributes from the text. The input of the model is non-standard component description text in string form, and the output is a fixed-dimensional numerical vector representing the semantics of the input text. The model adopts a 12-layer Transformer encoder structure, containing 12 attention heads. The training data consists of two parts: (1) publicly available automotive component technical manuals and repair catalog texts; (2) hundreds of thousands of order description texts accumulated in history and manually labeled with standard component numbers. The model uses the cross-entropy loss function and is trained for 50 rounds on 8 GPUs using the Adam optimizer.

[0051] In this embodiment, the semantic vector is matched with the standard information vectors of all records in the automotive parts feature library for similarity. All compliance elements bound to the standard part number with the highest similarity are output as the mapping result of the current order. This process refers to: the system calculates the distance or angle cosine value between the semantic vector generated from the non-standard description text and the standard information vector pre-coded for each standard part in the automotive parts feature library, and uses this as a similarity measure; by comparing all similarity values, the system finds the standard part record with the largest value, i.e., the most similar one; then, the complete set of compliance attribute data bound to the most similar record is output as the unique and definite interpretation result of the non-standard description in the current order.

[0052] In this embodiment, all compliance elements refer to the set of all attribute data bound to the most similar standard component record found in the automotive parts feature library, which is used to fully define the compliance requirements for import and export of goods. Its content includes the authoritative HS code, country of origin information, applicable free trade agreement list, and special regulatory conditions bound to the record.

[0053] To quantify abstract compliance and logistics risks into cost and time increments that can be directly used for route optimization calculations, a pre-built coupling rule base in the risk-cost dynamic coupling engine is proposed, which includes the following rules: Tariff and Trade Agreement Rules: Based on the country of origin and HS code of the goods in the compliance element vector, combined with the currently effective free trade agreements and tariff schedules, calculate the tariffs payable or exemptions when the goods are imported along the candidate route. The tariffs payable or exemptions are included as part of the risk-added costs. Regulatory Condition Conflict Rule: Compare the special regulatory conditions of goods in the compliance element vector with the processing capacity list of each logistics node on the candidate path. If a logistics node cannot meet any special regulatory condition, a fixed cost and time penalty value representing extremely high risk will be added to the corresponding logistics node for the path. Node delay probability fusion rule: For nodes on the candidate path, obtain the current congestion probability of the node provided by the logistics network status perception module, and the probability distribution of additional customs clearance time for each category of goods at the node based on historical data statistics. The current congestion probability of the node and the probability distribution of additional customs clearance time are fused to calculate the comprehensive expected delay time. The comprehensive expected delay time is used as the risk-added time of the node.

[0054] In this embodiment, the currently effective free trade agreements and tariff schedules refer to the bilateral or multilateral free trade agreement documents officially issued and implemented by the importing country (or region) at the time the goods are planned to be imported, as well as the compilation of legal documents currently applicable to the customs of that country (or region) that include import tax rates, tax calculation methods, and regulatory provisions for various commodities.

[0055] In this embodiment, the tariff calculation formula is: Tariff payable = Dutiable value of goods × Applicable tariff rate + Import value-added tax (if any). The dutiable value is determined based on the transaction price, and the applicable tariff rate is obtained by real-time querying of the official tariff database. If a preferential tariff rate under a free trade agreement exists, it will be applied first. The system automatically synchronizes the free trade agreement list and tariff rate changes daily from designated data interfaces on the official websites of the World Trade Organization and its member customs agencies.

[0056] In this embodiment, based on the country of origin and HS code of the goods in the compliance element vector, and in conjunction with the currently effective free trade agreements and tariff schedules, the applicable tariffs or exemptions for goods imported along the candidate route are calculated. The applicable tariffs or exemptions are included as part of the risk-added costs. This calculation process involves: the system reading the country of origin information and HS code of the goods contained in the compliance element vector; using this as a query condition, searching the database of currently effective free trade agreements and tariff schedules to determine whether the goods from the country of origin enjoy tariff preferences and the specific preferential tariff rate; by executing the tariff calculation formula, the actual amount of tariffs payable or the amount that can be exempted when the goods are imported via the candidate route is obtained; regardless of whether the result is the amount payable or the amount exempted, this amount is included as a quantifiable financial impact directly caused by compliance requirements (rules of origin and commodity classification) in the risk-added costs of the candidate route.

[0057] In this embodiment, risk-added costs are a broad concept that covers all monetizable additional expenses directly or indirectly caused by compliance and logistics risks, mainly including: additional tariffs, container demurrage fees, port demurrage fees, freight differences caused by emergency detours, and fines for regulatory violations.

[0058] In this embodiment, the special regulatory conditions for goods refer to additional mandatory operational or management requirements that must be followed during the transportation, warehousing, or customs clearance process due to the physical and chemical characteristics of the goods themselves or national policy regulations, in addition to the regular transportation and customs clearance requirements. For example, the goods must be transported at a constant temperature throughout the process, dangerous goods require specific qualifications for transportation, products of endangered species require additional licenses, or high-value goods require armed escort.

[0059] In this embodiment, the processing capacity list of each logistics node on the candidate route refers to a list that is collected and maintained in advance for each port, airport, warehouse or customs node involved in a proposed transportation route, describing the special operation or regulatory processing capabilities of the node. For example, whether a port has a temperature-controlled warehouse, whether a customs has a fast inspection channel for dangerous goods, and whether a warehouse has explosion-proof qualifications.

[0060] In this embodiment, the system compares the special regulatory conditions of goods in the compliance element vector with the processing capacity list of each logistics node on the candidate path. If a logistics node cannot meet any special regulatory condition, a fixed cost and time penalty value representing extremely high risk is added to the corresponding logistics node of the path. The execution logic of this rule is as follows: the system traverses each logistics node on the candidate path and compares the special regulatory conditions of goods with the processing capacity list of that node one by one. If it is found that any node on the path lacks the ability to handle any special regulatory condition of goods, it is determined that the path has a fundamental defect that makes it unenforceable. The system will immediately add a preset, extremely high cost penalty value and a time penalty value to the path at the defective node to quantify the huge expected losses and delays caused by choosing this path, such as the goods being unable to be transferred, being detained, or needing to be detoured.

[0061] In this embodiment, for nodes on the candidate path, the system obtains the current congestion probability of the node provided by the logistics network status perception module, as well as the probability distribution of additional customs clearance time for each category of goods at the node, which is derived from historical data statistics. This means that for a specific node on a candidate path being evaluated, the system first obtains the predicted congestion probability value of the node in the current time period from the network status vector; at the same time, the system queries and extracts the statistical distribution of the additional time required for the standard process for the same type of goods when clearing customs at this node in the past, based on the category of the goods (determined by its HS code), which describes the probability of the additional customs clearance time taking different values.

[0062] In this embodiment, the current congestion probability of a node is fused with the probability distribution of additional customs clearance time to calculate a comprehensive expected delay time. This comprehensive expected delay time serves as the risk-added time for the node. The fusion calculation process refers to the system using the node's current congestion probability as a weighting factor and weighting it with the probability distribution of additional customs clearance time derived from historical statistics to calculate the expected average delay time that may occur at the node, taking into account both the current congestion situation and the historical customs clearance characteristics of the goods category. This calculated time is considered as the potential delay at the node due to the interaction between the logistics status and the characteristics of the goods, and is included in the risk-added time of the candidate path.

[0063] To avoid excessive concentration of risk among multiple alternative plans and thus improve the overall resilience of the plan combination, an improved path search algorithm is proposed to avoid overlapping high-risk nodes. The workflow includes: After finding the optimal paths from the 1st to the (k-1)th path, when searching for the kth path, edges in the multi-weight transportation network graph that have been completely occupied by the first (k-1)th paths are excluded. Identify high-risk nodes marked by the risk-cost dynamic coupling engine in a multi-weighted transportation network graph; Calculate the size of the intersection between the set of high-risk nodes traversed by the current candidate path and the set of high-risk nodes traversed by each found path; When the intersection size exceeds a preset threshold, it is determined that the current candidate path and the existing path overlap excessively in risk distribution, the current candidate path is abandoned, and the search continues for the next path.

[0064] This algorithm uses weighted comprehensive cost as the path evaluation criterion. For a path, its weighted comprehensive cost C = α. (Total Cost / Cost Baseline) + β (Total Time / Time Baseline). Total cost is the sum of basic transportation cost and risk-added cost, while total time is the sum of basic transportation time and risk-added time. α and β are configurable weighting coefficients (default α=0.7, β=0.3), reflecting the emphasis on cost and timeliness, respectively. The cost baseline and time baseline are the average values ​​of all candidate paths, used for normalization. The algorithm always prioritizes searching for the path with the lowest weighted overall cost.

[0065] In this embodiment, after finding the first to (k-1)th optimal paths, when searching for the kth path, edges in the multi-weighted transportation network graph that have been completely occupied by the first (k-1)th paths are excluded. This step is one of the standard operations in the K-shortest path search algorithm, designed to ensure that the newly found path is not completely identical to the previously found paths in terms of physical route. Specifically, when searching for the kth optimal path, the algorithm temporarily removes or marks all transportation segments (edges) used by the previously found paths from the first to (k-1)th paths from the current search graph structure, thereby forcing the algorithm to explore alternative segments not used by previous paths.

[0066] In this embodiment, the risk-cost dynamic coupling engine refers to the core processing unit defined in claim 1, which is specifically responsible for performing the "risk-cost dynamic coupling" calculation function. This engine takes compliance element vectors and network state vectors as inputs, performs calculations based on a pre-set coupling rule base, and outputs risk-added costs and risk-added time.

[0067] In this embodiment, identifying high-risk nodes marked by the risk-cost dynamic coupling engine in the multi-weighted transportation network graph involves the following step: during the path search algorithm execution, the algorithm retrieves a list of logistics nodes identified as "high-risk" by the risk-cost dynamic coupling engine from its processing results. These nodes are typically determined to be highly likely to cause significant delays or generate high additional costs based on a combination of their current state (e.g., high congestion probability) and the characteristics of the goods to be transported (e.g., categories with high inspection rates) after calculation using coupling rules.

[0068] In this embodiment, the size of the intersection between the set of high-risk nodes traversed by the current candidate path and the set of high-risk nodes traversed by each found path is calculated. The calculation process is as follows: For the k-th candidate path being evaluated, the algorithm extracts all the high-risk nodes traversed by the path to form a set A; for each of the first k-1 found paths, the high-risk nodes traversed by them are extracted to form sets B1, B2, ..., B(k-1); then the number of high-risk nodes shared by set A and each set Bi (i=1 to k-1) is calculated, i.e., the size of the intersection.

[0069] In this embodiment, the preset threshold refers to a numerical standard set in advance by the system administrator or the algorithm to determine whether the degree of overlap of high-risk nodes is "excessive". This threshold can be an absolute number (e.g., a maximum of 2 overlapping nodes are allowed) or a relative proportion (e.g., the proportion of overlapping nodes to the total number of high-risk nodes in the candidate path must not exceed 50%).

[0070] In this embodiment, when the intersection size exceeds a preset threshold, it is determined that the current candidate path and the existing path excessively overlap in risk distribution. The current candidate path is then abandoned, and the search continues for the next path. This determination logic is the core of the algorithm's "avoidance of high-risk node overlap": if the number (or proportion) of overlap between the candidate path and any selected path at high-risk nodes exceeds a preset allowable range, it is considered that selecting this path would lead to excessive concentration of risk at bottleneck nodes already used by other paths, failing to effectively disperse the overall risk. Therefore, the algorithm will abandon this candidate path and instead try the next alternative path with the lowest cost in the graph.

[0071] To dynamically and accurately identify key bottlenecks that may cause delays or cost spikes to current transportation tasks, the criteria for marking high-risk nodes are proposed, including: The nodes reported by the logistics network status perception module whose current congestion probability exceeds the first set threshold; Customs nodes that, based on historical data statistics, have an inspection rate exceeding the second set threshold for the category to which the current goods' HS code belongs. The node is determined to be unable to process the current goods based on the regulatory condition conflict rules in the coupled rule base.

[0072] In this embodiment, the first threshold (congestion probability) is set to 70% by default. This value is based on an analysis of congestion events at major global ports over the past five years, finding that when the probability exceeds 70%, the actual likelihood of serious delays is greater than 85%. The second threshold (inspection rate) is set to 20% by default. This value references industry reports and the average inspection levels for various commodities published by customs authorities of major importing countries. Users can adjust these two thresholds according to their own risk tolerance in the system configuration interface.

[0073] In this embodiment, the first set threshold refers to a critical probability value, such as 70%, pre-configured by the system to determine whether the current congestion status of a logistics node has reached the "high-risk" level. When the predicted congestion probability of a node exceeds this threshold, the node will be marked as a high-risk node.

[0074] In this embodiment, the nodes reported by the logistics network status perception module as having a current congestion probability exceeding a first preset threshold refer to logistics nodes that the logistics network status perception module predicts have a higher probability of operational congestion within a specific future time window than the first preset threshold. These nodes are directly identified as high-risk points on the transportation path because they are highly likely to experience delays.

[0075] In this embodiment, the customs nodes that, based on historical data statistics, have an inspection rate exceeding a second set threshold for the category of goods to which the current HS code belongs, refer to those specific customs ports or control points where, according to historical customs clearance data statistics, the frequency of inspection of the commodity category (determined by its HS code) of the goods to be shipped is higher than the second set threshold (e.g., 30%). These nodes are identified as high-risk points that may lead to customs clearance delays because of the high probability of this category of goods being inspected in the past.

[0076] In this embodiment, nodes deemed unable to handle the current cargo by regulatory condition conflict rules in the coupling rule base refer to logistics nodes that, when the risk-cost dynamic coupling engine executes "regulatory condition conflict rules," are determined by the rule logic to lack the necessary facilities, qualifications, or processes to handle specific special regulatory conditions (such as temperature control or hazardous materials handling) for the cargo. These nodes, due to a fundamental mismatch in capabilities, are identified as extremely high-risk points that could lead to transportation disruptions or require emergency detours.

[0077] To scientifically assess the on-time delivery probability and cost fluctuation range of each route under uncertainties in actual implementation through probabilistic simulation, a Monte Carlo simulation process for a transportation robustness assessment model is proposed, including: For nodes in the candidate path, configure a probability distribution model representing the node delay time based on the node type and current state; In a single simulation, based on the probability distribution model of each node, the possible delay values ​​of the nodes are generated independently and randomly. The possible delay values ​​of all nodes are summed to obtain the total time of a single simulation. Calculate the total cost of a single simulation based on the basic cost of the path, the additional cost of risks, and the node delays. After repeating the simulation process a predetermined number of times, the proportion of simulations in which the total time does not exceed the order's required delivery time window is counted as the on-time delivery guarantee rate of the path. The distribution of the total cost of all simulations is also counted to assess the range of cost fluctuations.

[0078] In this embodiment, for transportation nodes such as ports and airports, the probability distribution model for delay duration adopts a gamma distribution. The shape parameter k and scale parameter θ are obtained by fitting historical data of delay events occurring during the same working hours (e.g., day / night) within the past 30 days at this node. For customs and other clearance nodes, the probability distribution model for additional clearance time adopts a log-normal distribution. The parameters (μ, σ) are obtained by statistically analyzing the difference between the actual clearance time and the standard clearance time for the first 6 digits of the same HS code for goods at this node over the past year.

[0079] In this embodiment, for nodes in the candidate path, a probability distribution model representing the delay duration of each node is configured based on the node type and current state. Specifically, for each node (e.g., a port, customs) on a transportation path being evaluated, the system matches or calculates an uncertainty model describing the potential delay duration based on the node's type (e.g., hub port, inland port) and its current perceived state (e.g., current congestion probability). This model is used to generate random delay values ​​in the simulation.

[0080] In this embodiment, the probability distribution model characterizing the node delay duration is a mathematical function or data structure that describes the likelihood of the node generating various values ​​of additional delay time under given conditions (such as node type and current state). For example, it can be a normal distribution (describing random fluctuations around a mean) or an exponential distribution (describing the time interval between random events), whose parameters (such as mean and variance) are determined by the node's current state and historical data.

[0081] In this embodiment, in a single simulation, the possible delay values ​​of each node are independently and randomly generated according to the probability distribution model of each node. The possible delay values ​​of all nodes are summed to obtain the total time of a single simulation. This step is the core sampling process of Monte Carlo simulation: the simulator randomly selects a delay time value for the first node on the path according to its corresponding probability distribution model; then it independently selects a value for the second node, and so on, until the last node; finally, the random delay values ​​(and basic transportation time) of all nodes are summed to obtain the total time required for the path to complete transportation under this random experiment.

[0082] In this embodiment, the total cost of a single simulation is calculated based on the basic cost of the path, the additional risk cost, and the node delay situation. This means that while completing a time sampling, the system calculates the total cost required to execute the path in this random scenario based on the specific node delay situation generated in the simulation (such as whether a specific rule is triggered), combined with the fixed transportation cost (basic cost) preset for the path and the dynamically calculated additional risk cost.

[0083] In this embodiment, the predetermined number of times refers to the total number of times the "single simulation" process described above needs to be repeated, set before the Monte Carlo simulation begins, such as 10,000 times. This number is large enough to ensure the statistical stability of the simulation results.

[0084] In this embodiment, after repeating the simulation process a predetermined number of times, the proportion of simulations in which the total time does not exceed the order's required delivery time window is counted as the on-time delivery guarantee rate for the path. The distribution of the total cost of all simulations is also analyzed to assess the cost fluctuation range. This process refers to the system performing result analysis after completing a predetermined number of independent random simulations (e.g., 10,000). First, the number of times the total path time is earlier than or equal to the customer's required delivery deadline is counted. This number is divided by the total number of simulations to obtain a percentage, which is the on-time delivery guarantee rate. Simultaneously, the system collects all total cost values ​​generated from these 10,000 simulations and analyzes the distribution of these values, such as calculating their maximum, minimum, average, standard deviation, or specific quantiles (e.g., the 90th percentile), to quantify the potential fluctuation range of the path's cost.

[0085] In this embodiment, the order delivery time window refers to the latest time point or time period that is explicitly specified in the commercial order, and the goods must be delivered to the destination.

[0086] In this embodiment, the simulation count ratio refers to the ratio of the number of simulations that meet a specific condition (such as "total time does not exceed the delivery time window") to the total number of simulations in a Monte Carlo simulation. It is usually used to estimate the probability of the event occurring.

[0087] In this embodiment, the cost fluctuation range is a quantitative measure of the uncertainty of the total cost during the execution of a path. It is usually obtained by analyzing the total cost sample data generated by multiple simulations. For example, it can be expressed as "90% of the simulation results show that the total cost is between X yuan and Y yuan".

[0088] In this embodiment, the distribution of all simulated total costs refers to the statistical characteristics of the set of all total cost values ​​obtained after completing a predetermined number of Monte Carlo simulations, such as which range these values ​​are mainly concentrated in, the frequency of extreme values, etc., which are usually described by histograms or statistics (such as quantiles).

[0089] To select the optimal solution that best suits the strategic preferences of different users from multiple high-quality paths based on their business priorities (such as timeliness or cost), a path robustness assessment and selection module is proposed. This module calculates a comprehensive robustness score based on the user's strategic preferences, including: When the primary strategy prioritizes timeliness, a higher weight is given to on-time delivery assurance rate in the scoring calculation; When the main strategy preference is cost-first, the scoring calculation places greater emphasis on low volatility and stability of cost distribution; The on-time delivery guarantee rate and cost assessment results of each path are weighted and combined, and the path with the highest score is selected as the final scheduling scheme.

[0090] In this embodiment, the comprehensive robustness score is calculated using a weighted summation formula. The score equals the on-time delivery guarantee rate multiplied by a first weight, plus 1 minus the cost volatility coefficient, multiplied by a second weight. The on-time delivery guarantee rate is a value between 0 and 1. The cost volatility coefficient is defined as the standard deviation of the total cost distribution obtained from multiple simulations divided by its average, used to quantify cost uncertainty. The specific values ​​of the first and second weights are dynamically set according to the subject's strategy preference: when the preference is timeliness priority, the first weight is 0.8 and the second weight is 0.2; when the preference is cost priority, the first weight is 0.3 and the second weight is 0.7.

[0091] In this embodiment, when the main strategy preference is timeliness priority, the on-time delivery guarantee rate is given a higher weight in the scoring calculation. This means that when calculating the comprehensive robustness score of the path, if the system determines that the current strategy is "timeliness priority", the coefficient of the on-time delivery guarantee rate in the scoring formula will be set to be larger than the coefficient of the cost-related indicators, so that the scoring result is more sensitive to the reliability of time.

[0092] In this embodiment, the main strategy preference is cost priority, which means that when making decisions, system users (such as cargo owners or logistics managers) will prioritize minimizing the total transportation cost, and are relatively willing to accept a certain degree of fluctuation or delay risk in transportation time.

[0093] In this embodiment, when the primary strategy preference is cost-first, the scoring calculation places greater emphasis on the low volatility and stability of the cost distribution. This means that when calculating the comprehensive robustness score of a path, if the system determines that a "cost-first" strategy is currently being adopted, the consideration of cost assessment results will dominate in the scoring formula. This consideration not only focuses on the expected value of the cost (average cost), but also emphasizes the "low volatility and stability" of the cost. For example, it assigns greater negative impact weights to higher quantiles of the cost distribution (such as the 90th percentile, representing the possible upper limit of cost) or cost variance (representing the degree of cost volatility), thus tending to select the path with the most controllable cost and the lowest risk of overspending.

[0094] In this embodiment, the on-time delivery guarantee rate and cost assessment results of each path are weighted and combined, and the path with the highest score is selected as the final scheduling scheme. This decision-making step refers to: the system determining a set of weight coefficients based on the main strategy preference; for each path to be evaluated, the on-time delivery guarantee rate value obtained by Monte Carlo simulation and the cost assessment results (such as expected cost value, high quantile, etc.) are multiplied by the corresponding weights respectively, and summed (or weighted averaged) to calculate a comprehensive score; after comparing the comprehensive scores of all paths, the path with the largest value will be selected by the system as the final recommended transportation scheme.

[0095] In this embodiment, the on-time delivery guarantee rate and cost assessment result for each path refer to two core performance indicators calculated for each candidate path through the aforementioned Monte Carlo simulation process: the on-time delivery guarantee rate is a probability value (e.g., 95%), and the cost assessment result can be a specific cost value (e.g., average cost) or a statistic describing the cost distribution (e.g., the 90th percentile of cost is 10,000 yuan). These two are the input data for path scoring.

[0096] In this embodiment, the highest-scoring path refers to the candidate transportation path that obtains the highest comprehensive score after weighted comprehensive scoring. This path is considered to have achieved the best balance between time reliability and economy under the currently set main strategy preference (time priority or cost priority).

[0097] To enable the system's risk prediction model to learn and continuously optimize based on actual execution feedback, thereby improving long-term decision-making accuracy, an adaptive optimization module is also proposed, which is used for: Continuously collect data on actual logistics and customs clearance results executed according to the final scheduling instructions; The actual time and cost of nodes in the actual results data are compared with the predicted values ​​calculated by the risk-cost dynamic coupling engine. When there is a systematic deviation between the predicted value and the actual value for a specific type, the parameters of the corresponding rule in the coupled rule base are calibrated to reduce the error in future predictions.

[0098] In this embodiment, continuously collecting actual logistics and customs clearance results data executed according to the final scheduling instruction means that after the system generates the final scheduling instruction and issues it for execution, it continuously tracks and obtains the data records of the entire flow of the goods in the real world through interfaces or manual input.

[0099] In this embodiment, actual result data refers to a set of data that reflects the real situation during the actual transportation and customs clearance of goods. The core content includes: the actual arrival time, actual departure time, actual stay time at each logistics node (i.e., actual time spent at the node), and various expenses actually incurred at the node or in the entire shipment (i.e., actual costs).

[0100] In this embodiment, the actual time and cost of nodes in the actual result data are compared with the predicted values ​​calculated by the risk-cost dynamic coupling engine. This means that the system extracts the actual time and actual cost of each node from the actual result data, and then compares these measured values ​​one-to-one with the risk-added time value and risk-added cost value predicted by the risk-cost dynamic coupling engine for the same node and the same goods in the previous (when formulating the scheduling plan), and calculates the difference between the two.

[0101] In this embodiment, the predicted values ​​calculated by the risk-cost dynamic coupling engine refer to the risk-added time value and risk-added cost value calculated by the risk-cost dynamic coupling engine for a specific node and for a specific cargo during the route planning stage, based on the input data and coupling rule base at that time. These values ​​are estimates of future risks before decision-making.

[0102] In this embodiment, a specific type of predicted value refers to a group of risk predicted values ​​that are classified according to a certain classification standard (e.g., a delay time prediction calculated by the "node delay probability fusion rule", a penalty cost prediction calculated by the "regulatory condition conflict rule", or a prediction for a specific category or a specific node type).

[0103] In this embodiment, the actual value refers to the real measurement result extracted from the actual result data that corresponds to the "predicted value of a specific type" mentioned above, such as the actual delay time of a certain node or the actual additional cost incurred.

[0104] In this embodiment, a systematic deviation between predicted and actual values ​​for a specific type refers to a situation where, after accumulating a certain number of similar scenarios (such as goods of the same type passing through the same type of node), statistical analysis reveals that the risk-cost dynamic coupling engine's predicted value for a certain type of risk (e.g., delay time) is consistently higher or lower than the actual value, rather than fluctuating randomly around the actual value. This indicates that the model's estimation of this type of risk has an inherent, directional error. The criterion for determining systematic deviation is: for 10 consecutive predictions of the same rule and the same scenario (such as goods of the same type and the same node), the mean absolute percentage error (MAPE) between the predicted and actual values ​​consistently exceeds 15%. When calibration is triggered, the system uses a recursive least squares method with a forgetting factor. It uses newly generated actual data streams to iteratively update the linear weight parameters within the rule that triggered the deviation online. The forgetting factor is set to 0.95 by default to balance adaptation to new data and retention of historical experience.

[0105] In this embodiment, when there is a systematic deviation between the predicted value and the actual value for a specific type, the parameters of the corresponding rule in the coupled rule base are calibrated to reduce the error of future predictions. This closed-loop optimization process refers to the following: when the system detects a systematic deviation in a certain type of prediction, a calibration process is automatically initiated; this process, for the specific coupled rule that generates this type of prediction (e.g., the "node delay probability fusion rule"), uses newly accumulated actual result data and corresponding historical input data to recalculate or adjust the parameters inside the rule (e.g., the weight coefficients and penalty values ​​in the probability fusion formula); after the parameter adjustment is completed, the corresponding rule in the coupled rule base is updated, so that the engine can produce more accurate results when predicting similar scenarios in the future.

[0106] To ensure strict synchronization of cross-module data in terms of processing timeliness and logical version, and to avoid risk calculation distortion caused by data latency or version mismatch, a unified data clock and version number are proposed to coordinate the compliance element real-time analysis module, the logistics network status perception module, and the risk-cost dynamic coupling engine. This includes: When processing data, the compliance element real-time analysis module and the logistics network status perception module mark the output results with a unified timestamp and data processing version number; When performing calculations, the risk-cost dynamic coupling engine verifies whether the received compliance element vector and network state vector have matching timestamps and version numbers, ensuring that the data used for coupling calculations is consistent in time and processing logic.

[0107] In this embodiment, the compliance element real-time analysis module and the logistics network status perception module, when processing data, label the output results with a unified timestamp and data processing version number. This step means that when the compliance element real-time analysis module completes the mapping of an order and outputs a compliance element vector, it attaches two metadata tags to the vector data: one is a timestamp based on the system's unified clock, indicating the time when the vector was generated (e.g., "2023-10-27 14:30:00UTC"); the other is the number of the data processing logic or model version used to generate the vector (e.g., "model version V2.1" or "rule version R1.5"). The logistics network status perception module performs the exact same operation when generating network status vectors, attaching a timestamp from the same clock source and its own processing version number. This ensures that each output vector can be uniquely identified by its "birth" time and "genes".

[0108] In this embodiment, the risk-cost dynamic coupling engine verifies whether the received compliance element vector and network state vector have matching timestamps and version numbers during calculations. This ensures that the data used for coupling calculations is consistent in time and processing logic. This verification process means that when the risk-cost dynamic coupling engine is triggered to calculate the risk of a shipment, it simultaneously receives compliance element vectors and network state vectors from upstream modules. Before performing any calculations, the engine first checks the timestamps carried by these two vectors to ensure that the data they represent was generated at the same or very close points in time (e.g., within the same minute), thus avoiding mismatched coupling calculations using outdated "old" compliance data and "new" network state data (or vice versa). Simultaneously, the engine checks the data processing version numbers of the two vectors to ensure they are generated based on currently recognized and compatible data processing logic, avoiding logical conflicts caused by asynchronous upgrades of upstream modules. Only after passing these two verifications will the engine use this pair of vectors for subsequent risk calculations, thereby fundamentally guaranteeing the spatiotemporal and logical consistency of the input data. Timestamps use the ISO 8601 standard format, accurate to milliseconds. All modules ensure clock synchronization by accessing a unified network time protocol service. Data processing version numbers use the format 'major version number.minor version number.revision number'. When the algorithm logic, model, or rule base of any module changes and is tested, its version number must be upgraded according to the semantic versioning specification and broadcast to the registry center at startup. The risk-cost dynamic coupling engine will refuse to process input vectors with timestamp differences exceeding 5 seconds or mismatched major version numbers.

[0109] This invention provides an intelligent management platform for the automobile import and export supply chain, including a server and at least one intelligent management system for the automobile import and export supply chain as described above, wherein the intelligent management system for the automobile import and export supply chain is communicatively connected to the server.

[0110] In this embodiment, the server refers to one or more centrally deployed computer devices with data storage, computing, and network communication capabilities. In this embodiment, the communication connection between the intelligent management system for the automotive import and export supply chain and the server refers to multiple instances of the aforementioned intelligent management system deployed in different physical locations (such as automotive company headquarters, logistics companies, and customs brokers), establishing a bidirectional data link with the central server through network protocols (such as HTTP / HTTPS, message queues). In this embodiment, the intelligent management platform for the automotive import and export supply chain is a distributed system architecture. Its core is that the central server acts as a hub, responsible for uniformly receiving decision results (such as final scheduling instructions), execution feedback data, and key status information from various independent intelligent management systems, and performing aggregation, persistent storage, global monitoring, and cross-system coordination. Simultaneously, the server can issue unified configuration updates, model parameters, or strategy instructions to each subsystem. Each individual intelligent management system maintains its local decision-making capabilities, but its connection to the server makes it a node in the platform network, achieving a combination of local intelligent decision-making and global data aggregation and management.

[0111] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent management system for the automobile import and export supply chain, characterized in that, include: The real-time compliance element analysis module is used to connect with the customs regulations database and order data. Based on the automotive parts feature library, it maps the descriptions of non-standard parts in the order into structured compliance element vectors with definite HS codes, country of origin, applicable free trade agreements and regulatory conditions in real time. The logistics network status perception module is used to access real-time operational data from ports, shipping companies, and land transportation service providers. Combined with historical data from the same period, it generates the congestion probability and expected delay time of each logistics node in terms of time windows, and outputs a network status vector. The risk-cost dynamic coupling engine is used to dynamically calculate the risk-added costs and risk-added time of each shipment on each candidate logistics route, based on compliance element vectors and network state vectors and according to a pre-set coupling rule base. The multi-constraint K-optimal path solver is used to superimpose the basic transportation cost, basic transportation time, and risk-added cost and risk-added time to construct a time-varying multi-weight transportation network graph. Based on an improved path search algorithm that avoids high-risk node overlap, it solves the top K optimal paths in the multi-weight transportation network graph that satisfy the constraints of the vehicle delivery time window and special regulatory conditions for parts. The path robustness assessment and decision-making module is used to input the top K optimal paths into a transportation robustness assessment model based on Monte Carlo simulation. It performs random extrapolation using the delay probability of each node in the optimal path as a parameter. Based on the on-time delivery guarantee rate and cost fluctuation range output by the extrapolation results, and combined with the main strategy preferences, it selects the optimal path with the highest comprehensive robustness score to generate the final scheduling instruction.

2. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, The working process of the real-time compliance element analysis module includes: Construct an automotive parts feature library, in which each record is associated with a standard part number. The standard part number is bound to an authoritative HS code, country of origin information, applicable free trade agreement list, and special regulatory conditions. The system receives non-standard component description text from order data and encodes it into semantic vectors using a natural language processing model based on an attention mechanism. The semantic vector is matched with the standard information vector of all records in the automotive parts feature library for similarity. All compliant elements bound to the standard part number with the highest similarity are output as the mapping result of the current order.

3. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, The pre-built coupling rule base in the risk-cost dynamic coupling engine includes the following rules: Tariff and Trade Agreement Rules: Based on the country of origin and HS code of the goods in the compliance element vector, combined with the currently effective free trade agreements and tariff schedules, calculate the tariffs payable or exemptions when the goods are imported along the candidate route. The tariffs payable or exemptions are included as part of the risk-added costs. Regulatory Condition Conflict Rule: Compare the special regulatory conditions of goods in the compliance element vector with the processing capacity list of each logistics node on the candidate path. If a logistics node cannot meet any special regulatory condition, a fixed cost and time penalty value representing extremely high risk will be added to the corresponding logistics node for the path. Node delay probability fusion rule: For nodes on the candidate path, obtain the current congestion probability of the node provided by the logistics network status perception module, and the probability distribution of additional customs clearance time for each category of goods at the node based on historical data statistics. The current congestion probability of the node and the probability distribution of additional customs clearance time are fused to calculate the comprehensive expected delay time. The comprehensive expected delay time is used as the risk-added time of the node.

4. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, The workflow of the improved path search algorithm that avoids high-risk node overlap includes: After finding the optimal paths from the 1st to the (k-1)th path, when searching for the kth path, edges in the multi-weight transportation network graph that have been completely occupied by the first (k-1)th paths are excluded. Identify high-risk nodes marked by the risk-cost dynamic coupling engine in a multi-weighted transportation network graph; Calculate the size of the intersection between the set of high-risk nodes traversed by the current candidate path and the set of high-risk nodes traversed by each found path; When the intersection size exceeds a preset threshold, it is determined that the current candidate path and the existing path overlap excessively in risk distribution, the current candidate path is abandoned, and the search continues for the next path.

5. The intelligent management system for automobile import and export supply chain according to claim 4, characterized in that, The criteria for marking high-risk nodes include: The nodes reported by the logistics network status perception module whose current congestion probability exceeds the first set threshold; Customs nodes that, based on historical data statistics, have an inspection rate exceeding the second set threshold for the category to which the current goods' HS code belongs. The node is determined to be unable to process the current goods based on the regulatory condition conflict rules in the coupled rule base.

6. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, The Monte Carlo simulation process for the transportation robustness assessment model includes: For nodes in the candidate path, configure a probability distribution model representing the node delay time based on the node type and current state; In a single simulation, based on the probability distribution model of each node, the possible delay values ​​of the nodes are generated independently and randomly. The possible delay values ​​of all nodes are summed to obtain the total time of a single simulation. Calculate the total cost of a single simulation based on the basic cost of the path, the additional cost of risks, and the node delays. After repeating the simulation process a predetermined number of times, the proportion of simulations in which the total time does not exceed the order's required delivery time window is counted as the on-time delivery guarantee rate of the path. The distribution of the total cost of all simulations is also counted to assess the range of cost fluctuations.

7. The intelligent management system for automobile import and export supply chain according to claim 6, characterized in that, The path robustness assessment and decision-making module calculates a comprehensive robustness score based on the subject's strategy preferences, including: When the primary strategy prioritizes timeliness, a higher weight is given to on-time delivery assurance rate in the scoring calculation; When the main strategy preference is cost-first, the scoring calculation places greater emphasis on low volatility and stability of cost distribution; The on-time delivery guarantee rate and cost assessment results of each path are weighted and combined, and the path with the highest score is selected as the final scheduling scheme.

8. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, It also includes an adaptive optimization module, used for: Continuously collect data on actual logistics and customs clearance results executed according to the final scheduling instructions; The actual time and cost of nodes in the actual results data are compared with the predicted values ​​calculated by the risk-cost dynamic coupling engine. When there is a systematic deviation between the predicted value and the actual value for a specific type, the parameters of the corresponding rule in the coupled rule base are calibrated to reduce the error in future predictions.

9. The intelligent management system for automobile import and export supply chain according to claim 1, characterized in that, The compliance element real-time analysis module, the logistics network status perception module, and the risk-cost dynamic coupling engine are coordinated through a unified data clock and version number, including: When processing data, the compliance element real-time analysis module and the logistics network status perception module mark the output results with a unified timestamp and data processing version number; When performing calculations, the risk-cost dynamic coupling engine verifies whether the received compliance element vector and network state vector have matching timestamps and version numbers, ensuring that the data used for coupling calculations is consistent in time and processing logic.

10. An intelligent management platform for the automobile import and export supply chain, characterized in that, It includes a server and at least one intelligent management system for automobile import and export supply chain as described in any one of claims 1-9, wherein the intelligent management system for automobile import and export supply chain is communicatively connected to the server.

Images