A method and device for identifying risks in customs clearance of goods for import and export

Abstract

The present application provides a method and device for identifying risks in customs clearance of goods for import and export, which relate to the technical field of risk identification. A multi-source impact data set is determined according to abnormal cases of customs clearance of import and export; a multi-feature identification model is built; the model is deployed to a monitoring and identification platform to obtain the parsing features of goods orders and obtain risk identification results; a graph model is constructed and the identification results are fitted into the graph model; based on the graph model, a full-process risk fusion analysis of the risk probability of each node in the goods circulation path is performed to obtain customs clearance identification risks, which solves the technical problems of low efficiency, poor accuracy and weak adaptability in the risk identification of customs clearance of goods for import and export in the prior art, realizes intelligent identification and evaluation of the risks of customs clearance of goods for import and export, effectively improves the accuracy and efficiency of risk identification, enhances the adaptability and flexibility of freight circulation, and achieves the technical effects of improving customs clearance efficiency, reducing operating costs and ensuring trade security.

Description

Technical Field

[0001] The present invention relates to the technical field of risk identification, and in particular to a method and device for identifying risks in customs clearance of goods import and export. Background Art

[0002] In international trade, customs clearance of goods for import and export is an important link connecting domestic and foreign markets. Its efficiency and safety are directly related to the operating costs of enterprises and the trade security of the country. However, with the expansion of trade scale and the diversification of trade forms, there are many risks in the process of customs clearance of goods for import and export, such as policy changes, violations of goods types, fraud, etc. If these risks are not discovered and effectively dealt with in a timely manner, they may lead to customs clearance delays, fines, and even confiscation of goods. At present, the identification of risks in customs clearance of goods for import and export mainly relies on manual review and traditional rule matching methods. Manual review relies on the experience and judgment of professionals, and has problems such as low efficiency, strong subjectivity, and easy errors; while traditional rule matching methods are difficult to cope with complex and changing trade environments and policy changes, and are prone to omissions and false reports.

[0003] In summary, the existing technologies often have technical problems such as low efficiency, poor accuracy and weak adaptability in risk identification for import and export customs clearance of goods. Summary of the invention

[0004] The present application provides a method and device for identifying risks in customs clearance of goods for import and export, which is used to solve the technical problems of low efficiency, poor accuracy and weak adaptability in the existing technology for identifying risks in customs clearance of goods for import and export.

[0005] In view of the above problems, the present application provides a method and device for identifying risks in customs clearance of import and export goods.

[0006] In a first aspect, the present application provides a method for identifying risks in customs clearance of goods for import and export, the method comprising:

[0007] According to the abnormal cases of import and export customs clearance, a multi-source impact data set is determined, and the multi-source impact data set includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type; according to the risk characteristics, screening and combination are performed to build a multi-feature recognition model; the multi-feature recognition model is deployed to the monitoring and identification platform, and the order analysis characteristics output by the cargo import and export analysis module are obtained, and the order analysis characteristics are identified by the multi-feature recognition model to obtain the risk identification results; a graph model is constructed and the risk identification results are fitted to the graph model, wherein the graph nodes represent the key points in the cargo circulation network, the graph edges represent the circulation paths of the cargo between the graph nodes, and the weights of the graph edges correspond to the risk identification results and represent the risk values; based on the graph model, the risk probability of each node in the cargo circulation path is analyzed through the whole process risk fusion to obtain the customs clearance identification risk.

[0008] In a second aspect, the present application provides a device for identifying risks in customs clearance of goods for import and export, the device comprising:

[0009] A case analysis module is used to analyze abnormal import and export customs clearance cases and determine a multi-source impact data set, which includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type; an identification model building module is used to screen and combine according to the risk characteristics and build a multi-feature identification model; a risk identification module is used to deploy the multi-feature identification model to the monitoring and identification platform, obtain the order analysis characteristics output by the cargo import and export analysis module, identify the order analysis characteristics through the multi-feature identification model, and obtain the risk identification result; a graph model construction module is used to build a graph model and fit the risk identification result to the graph model, wherein the graph nodes represent the key points in the cargo circulation network, the graph edges represent the circulation paths of the cargo between the graph nodes, and the weights of the graph edges correspond to the risk identification results and represent the risk values; a risk fusion analysis module is used to perform a full-process risk fusion analysis of the risk probabilities of each node in the cargo circulation path based on the graph model to obtain customs clearance identification risks.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The present application provides a method for identifying risks in customs clearance of goods for import and export. The method comprises the following steps: analyzing abnormal cases of customs clearance of import and export to determine a multi-source impact data set, wherein the multi-source impact data set includes risk characteristics of multiple dimensions, such as goods type, import and export policies, and risk types; screening and combining the risk characteristics to build a multi-feature recognition model; deploying the multi-feature recognition model to a monitoring and recognition platform, obtaining order analysis features output by a goods import and export analysis module, identifying the order analysis features through the multi-feature recognition model, and obtaining risk identification results; constructing a graph model and fitting the risk identification results to the graph model, wherein graph nodes represent key points in a goods circulation network, and graph edges represent the locations of goods between graph nodes. The weight of the edge of the graph corresponds to the risk identification result and represents the risk value; based on the graph model, the risk probability of each node in the goods circulation path is analyzed through the whole process risk fusion, and the customs clearance identification risk is obtained, which solves the technical problems of low efficiency, poor accuracy and weak adaptability in the existing technology for goods import and export customs clearance risk identification. By constructing a multi-feature identification model based on a multi-source influence data set and combining the graph model for the whole process risk fusion analysis, the intelligent identification and evaluation of goods import and export customs clearance risks are realized, which effectively improves the accuracy and efficiency of risk identification, enhances the adaptability and flexibility of freight circulation, and achieves the technical effect of improving customs clearance efficiency, reducing operating costs and ensuring trade security. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flow chart of a method for identifying risks in customs clearance of goods for import and export is provided for this application.

[0013] Figure 2 A schematic diagram of the structure of a cargo import and export clearance risk identification device is provided for this application.

[0014] Explanation of the accompanying drawings: case analysis module 11, identification model building module 12, risk identification module 13, graph model construction module 14, risk fusion analysis module 15. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a method for identifying risks in customs clearance of goods for import and export, the method comprising:

[0017] Step S100: Analyze the abnormal import and export customs clearance cases to determine a multi-source impact data set, which includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type.

[0018] The risk of customs clearance for the import and export of goods refers to the need to perform complex customs clearance procedures in accordance with relevant laws and regulations when goods enter or leave the customs of another country from the customs of one country in cross-border trade. The various uncertainties and potential losses that may be faced in this process include but are not limited to policy changes, incorrect commodity classification, false price declaration, false origin, document discrepancy, substandard goods quality, transportation delays or damage, and possible fraud, etc. The importance of identifying the risks of customs clearance for the import and export of goods lies in improving customs clearance efficiency, reducing economic losses, enhancing compliance, and improving market competitiveness. Therefore, a method for identifying the risks of customs clearance for the import and export of goods is specifically proposed in this embodiment.

[0019] In the process of import and export customs clearance, due to the involvement of multiple links and multiple parties, various abnormal situations often occur, and these abnormal situations are often affected by multiple sources. In order to effectively deal with these risks, a multi-source impact data set with multi-dimensional risk characteristics is constructed to provide an effective analysis basis for subsequent risk identification. Specifically, in-depth analysis of past import and export customs clearance abnormal cases is carried out to extract the key factors that cause abnormalities. These cases may include but are not limited to abnormal types of goods, such as imported goods being misclassified, violating trade control regulations; changes in import and export policies, such as sudden policy adjustments and the implementation of new trade restrictions; document discrepancies, the declaration form is inconsistent with the actual goods information, such as quantity, name, specifications, etc.; false price declarations, such as companies underreporting or overreporting the price of goods; false reporting of origin, such as deliberately falsely reporting the origin of goods in order to enjoy preferential treatment. Through detailed analysis of these cases, different risk types and the multi-source influencing factors behind them can be identified.

[0020] Furthermore, based on the analysis results of abnormal cases, a multi-source impact data set can be constructed. The data set contains three key dimensions: goods type, import and export policies, and risk types. The goods type refers to the specific types of imported and exported goods, such as electronic products, mechanical equipment, food, etc. Different goods types may face different customs clearance risks and policy requirements. In the data set, each type of goods should be listed in detail and the risk points it may involve should be marked. For example, electronic products may involve intellectual property infringement risks, and food may face quality safety and health quarantine risks. The import and export policies refer to the laws, regulations, trade policies, etc. of the country or region where the goods are imported or exported. Policy changes will directly affect the customs clearance efficiency and cost of goods. The data set should contain the latest import and export policy information of various countries, and it is also necessary to pay attention to the trend of policy changes so that enterprises can adjust their strategies in time. The risk type refers to the specific risk factors that may lead to customs clearance anomalies, such as policy change risks, goods quality risks, document discrepancy risks, etc. The data set should list various risk types in detail, and provide specific identification methods, evaluation standards and response measures for each risk type. For example, the risk of discrepancies between documents and goods can be identified by comparing the customs declaration with the actual cargo information, and measures such as supplementary declaration or modification of the customs declaration can be taken to deal with it.

[0021] Step S200: Screening and combining according to the risk features to build a multi-feature recognition model.

[0022] Furthermore, after determining the multi-source impact data set, it is necessary to screen and combine these risk features and build a multi-feature identification model. This process aims to improve the accuracy and comprehensiveness of risk identification by integrating information from multiple dimensions. Specifically, the risk features in the multi-source impact data set are screened. The purpose of the screening is to remove redundant, irrelevant or weakly correlated features and retain key features that have an important impact on risk assessment. The redundant features refer to repeated or highly similar features in the data set. Their contribution to risk assessment is limited and may increase the complexity of the model. Irrelevant features refer to features that are not directly related to the risk assessment target. They do not provide useful information to improve the prediction ability of the model. Weakly correlated features refer to features that have a certain correlation with the risk assessment target, but the degree of correlation is low. They may have a small impact on the prediction results of the model. The correlation coefficient (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.) can be used to evaluate the correlation between the features. For highly correlated features, one of them can be selected as a representative to reduce redundancy. At the same time, through the built-in feature importance evaluation function of machine learning algorithms (such as random forests, gradient boosting trees, etc.), it is determined which features have the greatest impact on the risk assessment results, and these features will be retained as model inputs.

[0023] After screening out the key features, these features are then combined to construct a more complex feature representation to further improve the recognition ability of the model. Multiple single features are combined into a new feature in some way (such as addition, multiplication, logical operation, etc.). The combined feature may contain richer information and help reveal hidden patterns in the data. Specifically, the Polynomial Features method can be used to perform polynomial combinations of the original features to generate new features. For example, combining the two features of cargo type and import and export policies may reveal that certain types of cargo are more likely to have customs clearance risks in specific regions. The automatic feature learning capabilities of deep learning models (such as neural networks) can also be used to combine and learn complex relationships between features through nonlinear transformations in the hidden layer. This method does not require manual specification of feature combinations, but the model automatically learns the optimal feature combination.

[0024] Finally, a multi-feature recognition model is built based on the screened and combined features, which will use these features to predict the risk of import and export customs clearance of goods. Choose a suitable machine learning or deep learning model based on the complexity of the problem and the characteristics of the data. For example, if the data volume is large and the feature dimension is high, you can choose models such as support vector machines, random forests, or gradient boosting trees; if there are complex nonlinear relationships in the data, you can choose neural networks or deep learning models. Use the training data set to train the model. During the training process, the model will learn how to predict the output risk value or risk category based on the input features. Use the test data set to evaluate the trained model. The evaluation indicators can include accuracy, recall rate, F1 score, etc. to measure the generalization ability of the model on unknown data. Finally, optimize the model based on the evaluation results, which may include adjusting model parameters, changing the feature combination method, adding regularization terms to reduce overfitting, etc.

[0025] By building a cargo import and export clearance risk assessment model based on multi-feature recognition, we can provide enterprises with accurate and comprehensive risk assessment services.

[0026] Step S300: deploy the multi-feature recognition model to the monitoring and recognition platform, obtain the order analysis features output by the goods import and export analysis module, identify the order analysis features through the multi-feature recognition model, and obtain risk identification results.

[0027] Optionally, after successfully building the multi-feature recognition model and verifying its effectiveness, deploy the model to the monitoring and identification platform to achieve real-time risk identification of goods import and export. Upload the model file (such as trained weights and model structure) to the platform, and configure the corresponding services to support online reasoning of the model. The monitoring and identification platform refers to a system that integrates data processing, model reasoning, and result display functions, and is used to monitor and identify potential risks in the process of goods import and export in real time.

[0028] After the model is deployed, the monitoring and identification platform needs to access the cargo import and export parsing module, which is responsible for extracting order-related order parsing features from import and export data. Order parsing features refer to a series of features extracted from import and export order data for risk assessment, such as cargo type, quantity, value, import and export location, mode of transportation, consignee information, etc. During the acquisition process, ensure that the data interface between the cargo import and export parsing module and the monitoring and identification platform is correctly connected and can transmit order data in real time. The parsing module extracts key feature information from the order data according to preset rules or algorithms, and formats it into the input format required by the model. When the monitoring and identification platform receives the order parsing features, it calls the deployed multi-feature recognition model for reasoning. The model calculates the risk value or risk category of the order based on the input feature information through its internal complex algorithms and logic. After the model reasoning is completed, the risk identification results are output to the monitoring and identification platform, and the platform can further save the results and perform subsequent processing.

[0029] By successfully deploying the multi-feature recognition model on the monitoring and identification platform and realizing real-time risk identification of goods import and export orders, it provides enterprises with a more accurate and efficient risk management method.

[0030] Step S400: construct a graph model and fit the risk identification result into the graph model, wherein the graph nodes represent key points in the goods circulation network, the graph edges represent the circulation paths of the goods between the graph nodes, and the weights of the graph edges correspond to the risk identification results and represent the risk values.

[0031] For example, these risks can be further intuitively displayed and analyzed by constructing a graph model. A graph model can be constructed based on the actual situation of the goods circulation network to represent this complex system. In mathematics and computer science, a graph is a structure composed of nodes (or vertices) and edges connecting these nodes. In a graph model, nodes and edges can be used to represent different entities and the relationships between them. The goods circulation network refers to a network composed of a series of locations and paths that goods pass through from the place of production to the place of consumption. Identify key points in the goods circulation network, such as production plants, warehouses, ports, customs, distributors, etc., and use them as nodes of the graph model. At the same time, define the edges of the graph model based on the actual circulation paths of goods between these nodes, and each edge represents the flow of goods from one node to another. In a graph model, the weight of an edge is usually used to represent a certain measure or importance. The risk identification results can be used as the weight of the graph edge to represent the risk value of the goods on the circulation path.

[0032] Furthermore, the risk identification results obtained by the multi-feature identification model are fitted into the constructed graph model. The risk identification results of each order or cargo circulation path are mapped to the corresponding graph edge, and the risk value (which may be a continuous value or a classification label) is assigned to each edge. The risk values ​​can be appropriately adjusted or standardized according to actual needs to ensure that they are comparable between different paths or nodes. After the graph model of risk values ​​is constructed and fitted, it can be used for further analysis and application. In the graph model, by analyzing the risk values ​​on different paths, high-risk paths and low-risk paths can be identified. According to the results of the path analysis, the circulation path of the goods can be adjusted and the path with lower risk can be selected for transportation.

[0033] By constructing a graph model that reflects the actual situation of the goods circulation network and effectively fitting the risk identification results into the model, it provides adaptability and accuracy of risk identification, provides enterprises with a powerful risk management tool, and helps to better understand and deal with potential risks in the import and export of goods.

[0034] Step S500: Based on the graph model, a full-process risk fusion analysis is performed on the risk probability of each node in the goods circulation path to obtain customs clearance identification risks.

[0035] Specifically, the risk probability of each node in the graph model is evaluated, which is usually based on a variety of factors, including but not limited to historical risk records, current regulatory policies, cargo characteristics, etc. The historical risk data, policy change information, and cargo characteristics data related to each node are collected, and the risk probability of each node is calculated based on the collected data using statistical models, machine learning models, or expert systems. The risk probability refers to the possibility of a risk occurring at a certain node or path in a certain period of time in the future. It is a value between 0 and 1, indicating the relative frequency or possibility of the risk occurring.

[0036] Next, a risk fusion analysis is performed, that is, the risk probability of each node is comprehensively considered to evaluate the risk level of the entire goods circulation path. The risk information from different nodes or different sources is integrated to form a comprehensive assessment of the overall risk. The whole process risk analysis refers to the continuous and systematic analysis of the risks of each stage and each node in the entire circulation process from the starting point of the goods to the destination. Specifically, according to the actual path of the goods circulation, all relevant nodes are traversed in the graph model. The risk probability of each node is transferred to its adjacent nodes according to the connection relationship between the nodes and the risk propagation mechanism (such as conditional probability, transfer probability, etc.). During the traversal process, the risk value of each path is continuously accumulated and updated, and finally the customs clearance identification risk of the entire goods circulation path is obtained. The final customs clearance identification risk, that is, the overall risk level that the goods may encounter during the customs clearance process, can be obtained through the risk fusion analysis. It reflects the probability of customs clearance obstacles or delays that may occur due to various factors in the entire circulation process from the starting point to the destination.

[0037] Through the above steps, the whole process risk fusion analysis of the risk probability of each node in the goods circulation path is realized, and the customs clearance identification risk is obtained. This process not only improves the accuracy and comprehensiveness of risk assessment, but also provides strong support for subsequent risk management and decision-making.

[0038] Furthermore, the analysis is performed based on the abnormal import and export customs clearance cases to determine the multi-source impact data set. Step S100 of this application also includes:

[0039] Step S110: configuring case collection channels, including a main data channel and a supplementary extension channel, wherein the main data channel includes customs abnormal case data, and the supplementary extension channel includes an international trade database, a logistics company, and a weather data platform.

[0040] Step S120: Integrate the case data acquired by the main data channel and the supplementary extension channel, and align the cases based on timestamps.

[0041] Step S130: classify and analyze the integrated import and export customs clearance exception cases to obtain the multi-source impact data set.

[0042] Optionally, in order to fully capture all kinds of abnormal cases that affect import and export customs clearance, a data collection channel is designed to collect case data. The data collection channel includes a main data channel and a supplementary extension channel. The main data channel refers to a channel that directly and corely provides customs abnormal case data, usually a database officially released or authorized by the customs, and these data are highly authoritative and accurate. The supplementary extension channel refers to an auxiliary channel of the main channel, which is used as a supplement to the main data channel. These channels provide data in other fields closely related to the customs clearance process to enrich the analysis dimension and depth. Specifically, the main data channel establishes a data interface with the customs department and automatically downloads the latest customs abnormal case data regularly, including but not limited to violation records, delay reasons, inspection details, etc. The supplementary extension channel includes an international trade database, a logistics company, and a weather data platform. The supplementary extension channel is connected to the database of the International Trade Organization or a large data service provider to obtain information such as global trade trends and changes in trade policies. At the same time, cooperate with major logistics companies to obtain real-time status and delay reports during cargo transportation. Further, integrate meteorological data services to obtain weather conditions along ports and routes, such as severe weather warnings and natural disaster records.

[0043] After acquiring data from different channels, it is necessary to integrate them efficiently and ensure the consistency and comparability of the data. Remove duplicate, erroneous or irrelevant data entries to ensure the accuracy and completeness of the data. According to the timestamp of each case, all data are sorted and matched in chronological order to ensure that data from different sources can be analyzed in the same time frame. The timestamp refers to the time mark in the data record, which is used to determine the specific time point when the data occurs. In the data integration process, the timestamp is a key factor in ensuring that the data can be correctly aligned. Furthermore, the cleaned and aligned data are merged into a unified database to facilitate subsequent analysis and query.

[0044] The integrated data needs to be deeply classified and analyzed to extract the multi-source factors that affect customs clearance and form a multi-source impact data set. According to the nature, cause, impact range and other characteristics of abnormal cases, the cases are divided into different categories, such as policy changes, logistics delays, weather factors, etc. In-depth analysis is conducted on the cases in each category to identify the specific factors that cause the abnormalities, such as the specific content of policy adjustments, the specific links of logistics delays, the specific degree of weather impact, etc. The multi-source influencing factors obtained from the analysis are organized into a structured data set, including key information such as factor name, description, impact degree, and occurrence time, to complete the construction of the multi-source impact data set and provide data support for subsequent risk assessment and early warning.

[0045] Through the above steps, a multi-source impact data set based on abnormal import and export customs clearance cases was successfully constructed, providing strong data support for improving customs clearance efficiency and reducing risks.

[0046] Furthermore, the integrated import and export customs clearance exception cases are classified and analyzed to obtain the multi-source impact data set. Step S130 of this application also includes:

[0047] Step S131: extracting customs characteristics, international trade data characteristics, logistics characteristics, and weather characteristics from the abnormal import and export customs clearance cases respectively.

[0048] Step S132: clustering the customs features, international trade data features, logistics features, and weather features to identify the frequency of case features.

[0049] Step S133: configuring the classification priority of each case feature based on the frequency rate, determining the classification center based on the classification priority, performing feature aggregation, and obtaining a classification feature cluster.

[0050] Step S134: performing case feature analysis on each of the classification feature clusters respectively, determining the risk feature of each cluster using the classification center as a multi-source risk label, and obtaining the multi-source impact data set.

[0051] Specifically, four categories of key features are extracted from the integrated import and export customs clearance abnormal cases: customs features, international trade data features, logistics features, and weather features. The customs features refer to features directly related to customs procedures, regulations, operations, etc., such as the number of customs inspections, lists of prohibited import / export commodities, etc.; the international trade data features involve features of global trade environment, policies, market changes, etc., such as trade barriers, exchange rate fluctuations, and the impact of trade agreements. The logistics features are features related to logistics links such as cargo transportation, warehousing, and distribution, such as transportation time delays, cargo damage, and logistics route changes; and the weather features refer to natural environmental factors that affect transportation and customs clearance, such as bad weather, natural disasters, and seasonal climate changes. Natural language processing technology is used to extract keywords and phrases from text descriptions, such as extracting "number of inspections" from customs reports, and then using data analysis tools to directly extract quantitative data from international trade databases, logistics company systems, and weather data platforms, such as transportation time statistics and weather warning levels.

[0052] Next, the extracted customs features, international trade data features, logistics features, and weather features are clustered to identify the frequency of case features. Clustering is an unsupervised learning method used to group objects in a data set into multiple classes or clusters, so that objects in the same cluster have a high degree of similarity, while objects in different clusters have a low degree of similarity. The frequency rate refers to the frequency of a feature appearing in multiple cases, reflecting the importance and prevalence of the feature's impact on customs clearance anomalies. K-means, hierarchical clustering and other algorithms are used to cluster features, group them according to their similarities and differences, and count the number of occurrences of each feature cluster in the case set, and calculate the frequency rate to quantify the impact of the feature.

[0053] Furthermore, the classification priority of each case feature is configured based on the frequency, and the classification center is determined to perform feature aggregation to form a classification feature cluster. The classification priority refers to the classification order or weight assigned to each feature or feature cluster based on the frequency and importance of the feature. The classification center refers to the most representative feature or feature combination in a feature cluster, which is used to identify and represent the entire cluster. The process of feature aggregation is to combine similar or related features according to certain rules to form a higher-level feature representation. Optionally, a classification priority is assigned to each feature cluster based on the frequency and expert judgment to ensure that high-frequency and important features are given priority. The feature with the highest frequency or the most representative rate in each feature cluster is selected as the classification center. With the classification center as the core, other related features are aggregated to form a classification feature cluster with clear meaning and boundaries.

[0054] Finally, case feature analysis is performed on each classification feature cluster, and the classification center is used as the multi-source risk label to determine the risk characteristics of each cluster, thereby constructing a multi-source impact data set. The case feature analysis refers to an in-depth analysis of specific cases in the classification feature cluster to extract risk characteristics and details closely related to the classification center. Multi-source risk labels refer to labels used to identify and classify risk factors of different sources and types, which facilitates subsequent risk management and response measures. Specifically, the cases in each classification feature cluster are interpreted in detail to identify the specific risk factors and manifestations that lead to customs clearance anomalies. Based on the classification center, corresponding multi-source risk labels are assigned to each cluster, such as "policy change risk", "logistics delay risk", "weather disaster risk", etc. The risk characteristics and label information of all classification feature clusters are sorted and summarized to form a structured multi-source impact data set to provide data support for subsequent risk assessment, early warning and decision-making.

[0055] Furthermore, the customs features, international trade data features, logistics features, and weather features are clustered to identify the frequency of case features. Step S132 of the present application also includes:

[0056] Step S1321: Based on the alignment relationship, construct a feature distribution array, wherein the array is the case information in the horizontal direction and the case occurrence features in the vertical direction.

[0057] Step S1322: performing correlation analysis on each feature according to the feature distribution array to determine the correlation coefficient of each feature.

[0058] Step S1323: extracting case features based on the correlation coefficient, wherein the case features are one or more case features whose correlation coefficients reach a correlation threshold.

[0059] Step S1324: Calculate the risk probability based on the abnormal import and export customs clearance cases with the case features as the target, and determine the frequency rate of each case feature.

[0060] Exemplarily, the alignment relationship refers to ensuring that the feature data of each case is extracted and represented according to the same standards, formats and dimensions when processing multiple cases, so as to facilitate subsequent comparison and analysis. Therefore, a feature distribution array is constructed based on the alignment relationship between each case. The feature distribution array is a two-dimensional data structure used to show the correspondence between different cases and their respective features. The horizontal direction of this array structure represents different case information (i.e., each case occupies a row), and the vertical direction represents the various features that appear in the case (i.e., each feature occupies a column). Assuming that there are 100 cases of abnormal import and export customs clearance, and 5 types of features (customs, international trade, logistics, weather and others) are extracted for each case, then the feature distribution array will be a matrix of 100 rows and 5 columns, where each row represents a case and each column represents a type of feature.

[0061] Next, statistical methods or machine learning algorithms are used to perform correlation analysis on each feature in the feature distribution array to determine the corresponding correlation coefficient. The purpose of this step is to determine which features have a strong correlation, as well as the direction and strength of this correlation, and then determine whether to classify them together according to their correlation coefficients that appear simultaneously in risk cases. Among them, the correlation coefficient is a statistical indicator used to quantify the strength and direction of the linear relationship between two variables. Common ones include the Pearson correlation coefficient and the Spearman rank correlation coefficient. For example, the Pearson correlation coefficient can be used to calculate the correlation between each feature. If a high positive correlation coefficient is found between the "number of customs inspections" and the "logistics delay time", it means that when the number of customs inspections increases, the logistics delay time tends to increase as well.

[0062] Then, based on the results of the correlation analysis, case features whose correlation coefficients reach the preset correlation threshold are extracted. These features are considered to be interrelated and have a significant impact on customs clearance anomalies. Assuming that the set correlation threshold is 0.7 (indicating strong correlation), after the correlation analysis, all feature pairs with correlation coefficients greater than 0.7 are screened out, such as "number of customs inspections" and "logistics delay time", etc. These feature pairs will be regarded as important case features. Among them, the correlation threshold is a preset numerical standard used to determine whether the correlation between two features is strong enough to warrant further attention and analysis. The value can be set based on actual conditions and is not limited here.

[0063] Finally, taking the extracted case features as the target, the risk probability is calculated based on the import and export customs clearance abnormality cases. The purpose of this step is to determine the frequency of each case feature in multiple cases, that is, the universality and importance of the feature's impact on customs clearance abnormalities. For each extracted case feature, count the number of times it appears in all cases and divide it by the total number of cases to get the frequency of the feature. For example, "customs inspection times" appears 80 times in 100 cases, then its frequency is 0.8, which can be used as an important indicator to evaluate the impact of the feature on customs clearance abnormalities. At the same time, the contribution of each feature to the risk of customs clearance abnormalities can be further quantified by combining the calculation method of risk probability (such as Bayesian formula, logistic regression, etc.).

[0064] By constructing feature distribution arrays, analyzing feature correlations, extracting key case features, and calculating their risk probability and frequency, we can effectively identify multi-source features that have a significant impact on import and export customs clearance anomalies and their degree of influence, thereby improving the comprehensiveness and accuracy of risk identification.

[0065] Furthermore, according to the risk features, screening and combining are performed to build a multi-feature recognition model. Step S200 of the present application also includes:

[0066] Step S210: Conduct in-depth mining of the risk features to determine the risk probability of each risk feature and identify its contribution.

[0067] Step S220: configuring the risk probability and identification contribution screening weights, performing multi-level screening, and determining screening features.

[0068] Step S230: reconstructing a combination based on the screened features, wherein the reconstructed combination is a previously unavailable feature combination relationship determined by nonlinear transformation and feature crossover.

[0069] Step S240: Evaluate the reconstruction combination to determine a risk feature combination.

[0070] Step S250: construct a training data set based on the screening features and the risk feature combination, perform multi-layer neural network training convergence, and obtain the multi-feature recognition model, which includes a multi-risk feature recognition layer, a weighting layer, and a fully connected layer.

[0071] Specifically, the collected risk features are first deeply mined. These features may include but are not limited to commodity categories, historical violation records, etc. The risk probability of each risk feature is determined through statistical analysis, machine learning algorithms (such as logistic regression, decision trees), etc., that is, the possibility of the feature causing customs clearance abnormalities when it exists alone. At the same time, the recognition contribution of each feature is evaluated, that is, the unique role and value of the feature in identifying abnormal cases. In order to streamline the model input and improve the recognition efficiency, the screening weights of risk probability and recognition contribution can be configured. These weights are adjusted based on expert experience, business logic, and preliminary model test results. Subsequently, multi-level screening is performed. The first layer may filter out obviously unimportant features based on the threshold of risk probability, and the second layer further considers the recognition contribution to ensure that the remaining features have both high risk predictiveness and can play a key role in the recognition process.

[0072] After screening out key features, in order to explore the potential relationship between features, the screened features can be reconstructed and combined. This process includes nonlinear transformations (such as logarithmic transformations, exponential transformations) to adjust feature distribution, and feature crossovers (such as multiplying two or more features or performing other operations) to create feature combination relationships that did not exist before. These new combination features can often capture more complex risk patterns and improve the recognition ability of the model. A comprehensive evaluation is conducted on the reconstructed combinations obtained through nonlinear transformations and feature crossovers, including evaluating their contribution to model performance, whether new noise or redundant information is introduced, and whether good interpretability is maintained. The risk feature combination is finally determined through a series of evaluation indicators (such as accuracy, recall, F1 score) and cross-validation techniques.

[0073] Finally, a training data set is constructed based on the selected features and the determined risk feature combination. The data set is used to train a multi-layer neural network model, which includes a multi-risk feature recognition layer (for processing different types of risk features), a weighting layer (automatically adjusting weights according to feature importance), and a fully connected layer (for synthesizing information and outputting the final recognition result). During training, the training process is continuously iterated through optimization algorithms such as gradient descent until the model converges, that is, the performance of the model on the training set reaches a stable state. For example, suppose that during the screening process, it is found that the risk probability of the two features "high-value goods" and "frequent changes in import and export locations" is high when they exist alone, but their recognition contribution is not outstanding. However, when these two features are cross-combined, it is found that almost all cases under this combination are abnormal customs clearance, so the combined feature is given an extremely high weight. Therefore, during the model training process, this combined feature becomes one of the key factors for identifying anomalies, significantly improving the recognition accuracy and efficiency of the model.

[0074] Furthermore, the risk features are further mined to determine the risk probability and contribution of each risk feature. Step S210 of the present application further includes:

[0075] Based on abnormal cases of import and export customs clearance, the risk impact probability of each risk feature and the contribution of risk identification results are analyzed respectively to determine the risk probability and identification contribution of each risk feature.

[0076] The expression of risk probability is: , Features Conditional probability under clearance risk, The probability of customs clearance risk occurring, Features The prior probability of occurrence of is the i-th risk feature; the expression for identifying contribution is: , where D is the original data set including customs clearance risk, is the entropy of the original data set, Risk characteristics The subset with value v, For subset The entropy of Contribution to the identification of the i-th risk feature.

[0077] Furthermore, based on the detailed data of abnormal import and export customs clearance cases, in-depth analysis is conducted on each risk feature, which may cover multiple aspects such as commodity attributes and external environment, such as commodity category, import and export location, logistics method, weather conditions, etc. In order to quantify the impact of each risk feature on customs clearance risk, the concept of risk probability is introduced. Given a risk characteristic Under the existing conditions, customs clearance risks The conditional probability of occurrence. The calculation of this probability usually relies on historical data and statistical methods. Specifically, the expression of risk probability is: , Features Conditional probability under clearance risk, The probability of customs clearance risk occurring, Features The prior probability of occurrence of is the ith risk feature. The prior probability of occurrence refers to the probability of an event occurring without any conditional restrictions.

[0078] In addition to the risk probability, it is also necessary to evaluate the contribution of each risk feature in identifying clearance anomalies, which is usually measured by the concept of entropy in information theory. Reflects the characteristics The ability to reduce the uncertainty of a data set (i.e., reduce entropy), that is, the ability to reduce the overall confusion or uncertainty of a data set. Specifically, the expression for identifying contribution is: , where D is the original data set including customs clearance risk, is the entropy of the original data set, Risk characteristics The subset with value v, For subset The entropy of Contribution to the identification of the i-th risk feature.

[0079] Through the above steps, not only the risk probability of each risk feature is determined, but also their contribution in identifying customs clearance anomalies is evaluated. This information provides key data support and theoretical basis for subsequent model construction.

[0080] Furthermore, based on the graph model, after performing a full-process risk fusion analysis on the risk probability of each node in the goods circulation path and obtaining the customs clearance identification risk, step S500 of the present application further includes:

[0081] Step S510: Based on the graph model, analysis is performed according to the import and export order information to determine the flow constraint information of each node.

[0082] Step S520: configure necessary nodes and optimized nodes in the graph model according to the flow constraint information, wherein the necessary nodes are circulation path nodes with high constraint strength that cannot be adjusted, and the optimized nodes are circulation path nodes with low constraint strength that can be optimized and adjusted.

[0083] Step S530: Based on the risk probability of each node, taking the necessary nodes as constraints, optimizing the path with the minimum risk probability according to the graph model, and determining a low-risk optimization path.

[0084] Step S540: Based on the low-risk optimal path as the starting point, the optimization is performed in the direction of path shortening to construct a set of multiple intermediate paths.

[0085] Step S550: configuring the risk probability and path shortening weight, performing path evaluation based on the weight, screening the multiple intermediate path sets, integrating the screened paths with the low-risk optimal path, generating an optimized recommended path for feedback.

[0086] Specifically, based on the constructed graph model (the model uses nodes to represent various links in logistics, such as warehouses, ports, customs, etc., and edges represent the flow relationship between goods in these links), the potential risks of each node in the goods circulation path are analyzed in a full-process risk fusion analysis. Historical data, real-time information, and external factors (such as policy changes, weather conditions, etc.) are comprehensively considered to quantify the risk probability of each node. After completing this analysis, special attention is paid to risk identification in the customs clearance link, because customs clearance efficiency and compliance directly affect the smoothness of the entire logistics chain.

[0087] Next, we use the specific information of import and export orders (such as the type, quantity, destination, etc. of goods) to deeply analyze the constraints of this information on the logistics path. These constraints come from laws and regulations, transportation restrictions, customer needs and other aspects. Through analysis, we can clarify which nodes (i.e. links on the logistics path) are necessary nodes (such as certain types of goods must pass certain customs inspections, these nodes have strong constraints and cannot be adjusted), and which are optimization nodes (such as goods can be flexibly allocated between multiple warehouses, these nodes have low constraints and have room for optimization). Then, based on the risk probability of each node obtained from the analysis and the necessary nodes as unshakable constraints, we use the graph model to optimize the path with the minimum risk probability. This optimization process is similar to finding the shortest path from the starting point to the end point in a graph, but the "shortest" here means the lowest risk probability. Through algorithmic calculation, a low-risk optimization path can be determined, which minimizes the overall risk while ensuring compliance and security.

[0088] Furthermore, starting from the low-risk optimal path, we further explore how to shorten the path length and improve logistics efficiency by adjusting the optimization nodes while ensuring that the risk is controllable. This step constructs a set of multiple potential optimization paths, namely, a set of multiple intermediate paths, each of which attempts to find a new balance between risk and efficiency.

[0089] In order to select the optimal solution from multiple intermediate paths, the weights of risk probability and path shortening can be configured. These weights reflect the importance that enterprises attach to different optimization goals. Furthermore, based on these weights, a path evaluation mechanism can be designed to score each path. By comparing the scores, the paths that meet the risk control requirements and can effectively shorten the path length are selected. Finally, the selected paths are integrated and analyzed with the original low-risk optimization path to generate the final optimized recommended path. This path not only takes into account risk minimization, but also takes into account the improvement of logistics efficiency. According to this recommended path, the logistics strategy is adjusted to realize the intelligent and efficient flow of goods. At the same time, this optimization result will be promptly fed back to the relevant system or decision makers to continuously optimize the logistics network.

[0090] The above steps realize intelligent path planning from risk minimization to efficiency optimization by comprehensively analyzing the risk probability and flow constraints in the goods circulation path, and finally generate an optimized recommended path that is both safe and efficient, providing scientific and intelligent logistics decision-making support for the circulation of goods.

[0091] Through the technical solutions of the above embodiments, a method for identifying risks in customs clearance of imported and exported goods provided by the present application solves the technical problems of low efficiency, poor accuracy and weak adaptability in the identification of risks in customs clearance of imported and exported goods existing in the prior art. By constructing a multi-feature identification model based on a multi-source influencing data set and combining it with a graph model for full-process risk fusion analysis, intelligent identification and assessment of customs clearance risks of imported and exported goods are realized, the accuracy and efficiency of risk identification are effectively improved, the adaptability and flexibility of freight circulation are enhanced, and the technical effects of improving customs clearance efficiency, reducing operating costs and ensuring trade security are achieved.

[0092] Embodiment 2 is based on the same inventive concept as a method for identifying risks in customs clearance of goods import and export in the above embodiment. Figure 2 As shown, the present application provides a device for identifying risks in customs clearance of goods for import and export, the device comprising:

[0093] The case analysis module 11 is used to analyze abnormal import and export customs clearance cases and determine a multi-source impact data set, which includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type.

[0094] The identification model building module 12 is used to screen and combine the risk features to build a multi-feature identification model.

[0095] The risk identification module 13 is used to deploy the multi-feature identification model to the monitoring and identification platform, obtain the order analysis features output by the goods import and export analysis module, identify the order analysis features through the multi-feature identification model, and obtain risk identification results.

[0096] The graph model construction module 14 is used to construct a graph model and fit the risk identification result into the graph model, wherein the graph nodes represent key points in the cargo circulation network, the graph edges represent the circulation paths of the cargo between the graph nodes, and the weights of the graph edges correspond to the risk identification results and represent the risk value.

[0097] The risk fusion analysis module 15 is used to perform a full-process risk fusion analysis on the risk probability of each node in the goods circulation path based on the graph model to obtain customs clearance identification risks.

[0098] Furthermore, the case analysis module 11 is also used to perform the following steps:

[0099] Configure case collection channels, including a main data channel and a supplementary extension channel. The main data channel includes customs abnormal case data, and the supplementary extension channel includes an international trade database, a logistics company, and a weather data platform.

[0100] The case data obtained by the main data channel and the supplementary extension channel are integrated, and the cases are aligned based on timestamps.

[0101] The integrated import and export customs clearance exception cases are classified and analyzed to obtain the multi-source impact data set.

[0102] Furthermore, the case analysis module 11 is also used to perform the following steps:

[0103] The customs characteristics, international trade data characteristics, logistics characteristics and weather characteristics of the abnormal import and export clearance cases are extracted respectively.

[0104] The customs characteristics, international trade data characteristics, logistics characteristics, and weather characteristics are clustered to identify the frequency of case characteristics.

[0105] The classification priority of each case feature is configured based on the frequency rate, the classification center is determined based on the classification priority, and feature aggregation is performed to obtain a classification feature cluster.

[0106] Case feature analysis is performed on each of the classification feature clusters respectively, and the risk feature of each cluster is determined using the classification center as a multi-source risk label to obtain the multi-source impact data set.

[0107] Furthermore, the case analysis module 11 is also used to perform the following steps:

[0108] Based on the alignment relationship, a feature distribution array is constructed, wherein the horizontal direction of the array is the case information and the vertical direction of the array is the case appearance features.

[0109] The correlation analysis of each feature is performed according to the feature distribution array to determine the correlation coefficient of each feature.

[0110] Based on the correlation coefficient, case features are extracted, where the case features are one or more case features whose correlation coefficients reach a correlation threshold.

[0111] The risk probability calculation is performed based on the abnormal import and export customs clearance cases with the case characteristics as the target, and the frequency rate of each case characteristic is determined.

[0112] Furthermore, the recognition model building module 12 is also used to perform the following steps:

[0113] Conduct in-depth research on the risk characteristics, determine the risk probability of each risk characteristic, and identify its contribution.

[0114] The risk probability and the screening weights of the identification contribution are configured, multi-level screening is performed, and the screening characteristics are determined.

[0115] A combination is reconstructed based on the screened features, wherein the reconstructed combination is a previously unknown feature combination relationship determined by nonlinear transformation and feature crossover.

[0116] The reconstruction combinations are evaluated to determine risk characteristic combinations.

[0117] A training data set is constructed based on the screening features and the risk features, and multi-layer neural network training convergence is performed to obtain the multi-feature recognition model, which includes a multi-risk feature recognition layer, a weighting layer, and a fully connected layer.

[0118] Furthermore, the recognition model building module 12 is also used to perform the following steps:

[0119] Based on abnormal cases of import and export customs clearance, the risk impact probability of each risk feature and the contribution of risk identification results are analyzed respectively to determine the risk probability and identification contribution of each risk feature.

[0120] The expression of risk probability is: , Features Conditional probability under clearance risk, The probability of customs clearance risk occurring, Features The prior probability of occurrence of is the i-th risk feature; the expression for identifying contribution is: , where D is the original data set including customs clearance risk, is the entropy of the original data set, Risk characteristics The subset with value v, For subset The entropy of Contribution to the identification of the i-th risk feature.

[0121] Furthermore, the risk fusion analysis module 15 is further configured to perform the following steps:

[0122] Based on the graph model, analysis is performed based on the import and export order information to determine the flow constraint information of each node.

[0123] The necessary nodes and optimized nodes in the graph model are configured according to the circulation constraint information. The necessary nodes are circulation path nodes with high constraint strength that cannot be adjusted, and the optimized nodes are circulation path nodes with low constraint strength that can be optimized and adjusted.

[0124] Based on the risk probability of each node, taking the necessary nodes as constraints, optimizing the path with the minimum risk probability is performed according to the graph model to determine a low-risk optimizing path.

[0125] Based on the low-risk optimization path as the starting point, optimization is performed in the direction of path shortening to construct a set of multiple intermediate paths.

[0126] The risk probability and path shortening weight are configured, path evaluation is performed based on the weight, and the multiple intermediate path sets are screened, the screened paths are integrated with the low-risk optimal path, and an optimized recommended path is generated for feedback.

[0127] Through the above-mentioned detailed description of a method for identifying risks in customs clearance of goods import and export, this specification allows those skilled in the art to clearly understand a device for identifying risks in customs clearance of goods import and export in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0128] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying risks in customs clearance of goods for import and export, characterized in that: The method for identifying risks in customs clearance of goods for import and export includes: Analyze abnormal import and export customs clearance cases to determine a multi-source impact data set, which includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type; Screen and combine the risk features to build a multi-feature recognition model; Deploy the multi-feature recognition model to the monitoring and recognition platform, obtain the order analysis features output by the goods import and export analysis module, identify the order analysis features through the multi-feature recognition model, and obtain risk identification results; Constructing a graph model and fitting the risk identification result into the graph model, wherein the graph nodes represent key points in the goods circulation network, the graph edges represent the circulation paths of the goods between the graph nodes, and the weights of the graph edges correspond to the risk identification result and represent the risk value; Based on the graph model, the risk probability of each node in the cargo circulation path is analyzed through the whole process of risk fusion to obtain the customs clearance identification risk; Among them, screening and combining are performed according to the risk characteristics to build a multi-feature recognition model, including: Conduct in-depth research on the risk characteristics, determine the risk probability of each risk characteristic, and identify its contribution; Configure the risk probability and identification contribution screening weights, perform multi-level screening, and determine screening features; Reconstructing a combination based on the screening features, wherein the reconstructed combination is a previously unknown feature combination relationship determined by nonlinear transformation and feature crossover; evaluating the reconstruction combination to determine a risk characteristic combination; Constructing a training data set based on the screening features and the risk features, performing multi-layer neural network training convergence, and obtaining the multi-feature recognition model, wherein the multi-feature recognition model includes a multi-risk feature recognition layer, a weighting layer, and a fully connected layer; Conduct in-depth research on the risk characteristics to determine the risk probability and contribution of each risk characteristic, including: Based on the abnormal import and export customs clearance cases, the risk impact probability of each risk feature and the contribution of risk identification results are analyzed to determine the risk probability and identification contribution of each risk feature; The expression of risk probability is: , Features Conditional probability under clearance risk, The probability of customs clearance risk occurring, Features The prior probability of occurrence of is the i-th risk feature; the expression for identifying contribution is: , where D is the original data set including customs clearance risk, is the entropy of the original data set, Risk characteristics The subset with value v, For subset The entropy of Contribution to the identification of the i-th risk feature.

2. The method for identifying risks in customs clearance of goods import and export as claimed in claim 1, characterized in that: The above analysis is based on the abnormal import and export customs clearance cases to determine the multi-source impact data set, including: Configure case collection channels, including a main data channel and a supplementary extension channel. The main data channel includes customs abnormal case data, and the supplementary extension channel includes an international trade database, a logistics company, and a weather data platform; Integrate the case data obtained by the main data channel and the supplementary extension channel, and align the cases based on timestamps; The integrated import and export customs clearance exception cases are classified and analyzed to obtain the multi-source impact data set.

3. The method for identifying risks in customs clearance of goods import and export as claimed in claim 2, characterized in that: The integrated import and export customs clearance exception cases are classified and analyzed to obtain the multi-source impact data set, including: Extract customs characteristics, international trade data characteristics, logistics characteristics, and weather characteristics from the abnormal import and export customs clearance cases respectively; Clustering the customs characteristics, international trade data characteristics, logistics characteristics, and weather characteristics to identify the frequency of case characteristics; Based on the frequency rate, configure the classification priority of each case feature, determine the classification center based on the classification priority, perform feature aggregation, and obtain a classification feature cluster; Case feature analysis is performed on each of the classification feature clusters respectively, and the risk feature of each cluster is determined using the classification center as a multi-source risk label to obtain the multi-source impact data set.

4. The method for identifying risks in customs clearance of goods import and export as claimed in claim 3, characterized in that: Cluster the customs features, international trade data features, logistics features, and weather features to identify the frequency of case features, including: Based on the alignment relationship, construct a feature distribution array, wherein the array is case information in the horizontal direction and case occurrence features in the vertical direction; Performing correlation analysis on each feature according to the feature distribution array to determine the correlation coefficient of each feature; Based on the correlation coefficient, extract case features, wherein the case features are one or more case features whose correlation coefficients reach a correlation threshold; The risk probability calculation is performed based on the abnormal import and export customs clearance cases with the case characteristics as the target, and the frequency rate of each case characteristic is determined.

5. The method for identifying risks in customs clearance of goods import and export as claimed in claim 1, characterized in that: Based on the graph model, the risk probability of each node in the goods circulation path is analyzed through the whole process of risk fusion. After obtaining the customs clearance identification risk, it also includes: Based on the graph model, the import and export order information is analyzed to determine the flow constraint information of each node; According to the circulation constraint information, necessary nodes and optimized nodes in the graph model are configured, wherein the necessary nodes are circulation path nodes with large constraint strength that cannot be adjusted, and the optimized nodes are circulation path nodes with small constraint strength that can be optimized and adjusted; Based on the risk probability of each node, taking the necessary nodes as constraints, searching for the path with the minimum risk probability according to the graph model, and determining a low-risk optimization path; Based on the low-risk optimization path as the starting point, optimization is performed in the direction of path shortening to construct a set of multiple intermediate paths; The risk probability and path shortening weight are configured, path evaluation is performed based on the weight, and the multiple intermediate path sets are screened, the screened paths are integrated with the low-risk optimal path, and an optimized recommended path is generated for feedback.

6. A risk identification device for goods import and export clearance, characterized in that: The device is used to implement a method for identifying risks in customs clearance of goods import and export as described in any one of claims 1 to 5, comprising: A case analysis module is used to analyze abnormal import and export customs clearance cases and determine a multi-source impact data set, which includes multi-dimensional risk characteristics of cargo type, import and export policies, and risk type; An identification model building module is used to screen and combine risk features to build a multi-feature identification model; A risk identification module is used to deploy the multi-feature identification model to the monitoring and identification platform, obtain the order analysis features output by the goods import and export analysis module, identify the order analysis features through the multi-feature identification model, and obtain risk identification results; A graph model construction module, used to construct a graph model and fit the risk identification result into the graph model, wherein the graph nodes represent key points in the goods circulation network, the graph edges represent the circulation paths of the goods between the graph nodes, and the weights of the graph edges correspond to the risk identification result and represent the risk value; The risk fusion analysis module is used to perform a full-process risk fusion analysis of the risk probability of each node in the cargo circulation path based on the graph model to obtain customs clearance identification risks.

Images