A customs clearance compliance risk early warning method based on big data analysis

Abstract

This invention belongs to the field of risk warning technology, specifically a customs clearance compliance risk warning method based on big data analysis. It includes acquiring multi-dimensional attribute data of target commodities in customs declarations and associating them to form a commodity attribute structure set; acquiring historical commodity attribute structure sets; constructing a commodity attribute structure space; obtaining a stable set of commodity attribute structures through distribution aggregation identification; mapping the current commodity to this space to obtain its positional distribution status and determine its membership relationship; if it does not belong, calculating the degree of structural boundary offset with the stable structure set of the most recent time node; further decomposing to obtain a set of structural offset features; simultaneously, tracking the evolution of the stable structure set based on historical time series, generating an evolution path and judging consistency to obtain a credibility correction result; finally, combining the offset features and credibility correction result to perform a progressive risk judgment and output a warning level, thereby improving the accuracy and interpretability of customs clearance warnings.

Description

Technical Field

[0001] This invention relates to the field of risk warning technology, specifically a method for early warning of customs clearance compliance risks based on big data analysis. Background Technology

[0002] One big data analytics-based method for early warning of customs compliance risks involves customs and other regulatory departments integrating multi-source information such as enterprise declarations, logistics trajectories, payment records, and historical violation data to form a massive trade database. Based on this database, algorithms such as machine learning, correlation analysis, and anomaly detection automatically learn normal trade behavior patterns and dynamically identify hidden compliance issues such as price anomalies, incorrect commodity classification, and supply chain disruption risks. Once the system determines that the risk probability of a batch of goods or a particular enterprise exceeds a set threshold, it automatically triggers a tiered warning before or during customs clearance, prompting regulatory personnel to conduct precise inspections or take other appropriate measures.

[0003] In existing technologies, the first step is to integrate multi-source data such as enterprise declarations, logistics, and payments to build a unified database. Then, machine learning algorithms are used to learn from historical positive and negative cases to establish a dynamic risk model. Finally, at the moment of customs clearance, the declaration data is compared with the model baseline in multiple dimensions. Once potential violations such as abnormal prices or classification deviations are identified, an early warning is automatically triggered according to the risk level, assisting customs in achieving precise control.

[0004] The aforementioned solutions still have some problems in practical application. Existing technologies can only perform isolated threshold comparisons of customs declaration data or rely on preset static rule bases for anomaly matching, but they cannot perform structured modeling and spatial distribution analysis of the inherent correlation constraints between different attribute data. Therefore, it is difficult to dynamically depict the aggregation pattern of historical normal samples in the overall distribution space and their continuous evolution over time. This leads to the system being prone to misjudging the synergistic anomalies between attributes as random fluctuations when faced with complex false declarations and concealment behaviors. It also faces the problem of false or missed declarations accumulating due to the slow changes in normal trade patterns. It may even require frequent reliance on human experience for review, threshold adjustment, or repeated testing, resulting in a waste of regulatory resources and missed risk detection, which in turn reduces the accuracy and stability of customs clearance compliance risk warnings.

[0005] Therefore, this invention provides a method for early warning of customs clearance compliance risks based on big data analysis. Summary of the Invention

[0006] This application provides a method for early warning of customs clearance compliance risks based on big data analysis. It can issue warnings at the incipient stage of complex false declarations, avoid misjudging coordinated violations as random fluctuations, significantly improve the stability of risk warnings and early detection capabilities, and ensure the accurate allocation of regulatory resources and the efficient maintenance of customs clearance order.

[0007] To achieve the above objectives, this application adopts the following technical solution:

[0008] This application provides a method for early warning of customs clearance compliance risks based on big data analysis, the method including:

[0009] Obtain the multi-dimensional attribute data set of the target commodity during the customs declaration process, and combine and associate the attribute data according to the preset attribute association constraint rules to form a commodity attribute structure set;

[0010] Obtain a set of historical product attribute structures, analyze the set of historical product attribute structures to form a product attribute structure space, identify the distribution and clustering of the set of historical product attribute structures in the product attribute structure space, and construct a stable set of product attribute structures based on the identified clustering regions.

[0011] The current product attribute structure set is mapped to the product attribute structure space to obtain the position distribution state of the current product attribute structure set in the stable product attribute structure set. Based on the position distribution state, the membership relationship between the current product attribute structure set and the stable product attribute structure set is determined to obtain the determination result.

[0012] If the determination result is "not belonging", then the current product attribute structure set and the stable structure set of each product attribute arranged in chronological order that is closest to the current time are calculated to obtain the degree of structural boundary offset, and the degree of structural boundary offset is subjected to perturbation feature decomposition to obtain the structural offset feature set.

[0013] Based on historical time series, the structural evolution of the stable structure set of commodity attributes is tracked, a structural evolution path is generated, and the continuity of structural changes of the stable structure set of commodity attributes in the time series is judged based on the consistency of the evolution path between adjacent time nodes, and a credibility correction result is generated.

[0014] Based on the credibility correction results of the product attribute structure offset feature set and the product attribute stable structure set, a progressive risk assessment is performed, and the corresponding customs clearance compliance risk warning level is output.

[0015] In some possible implementations, obtaining the historical product attribute structure set and analyzing the historical product attribute structure set to construct a product attribute structure space includes:

[0016] Each set of product attribute structures in the historical compliant customs clearance sample data is taken as a sample point to obtain a set of sample points;

[0017] Calculate the difference value of attribute data between any two sample points in the sample point set in each attribute dimension, and sum the absolute values ​​of the difference values ​​in all attribute dimensions to obtain the structural distance between the two sample points. Then, obtain the structural distance between all pairs of sample points in the sample point set to form a structural distance set.

[0018] Based on the structural distance set, the relative position coordinates of each sample point in the sample point set in a preset dimension space are determined by a multidimensional scaling analysis algorithm, and the corresponding relationship of sample point coordinates is obtained.

[0019] Based on the coordinate correspondence of the sample points, interpolation is performed on the sparse regions of the coordinate results obtained from the multidimensional scaling analysis in the space of a preset number of dimensions to obtain the coordinate mapping function.

[0020] The product attribute structure space is formed by using the set of all coordinate points covered by the range of the coordinate mapping function in the preset dimension space.

[0021] In some possible implementations, the step of identifying the distribution clusters of the historical product attribute structure set in the product attribute structure space, and constructing a stable set of product attribute structures based on the identified clustered regions, includes:

[0022] Based on the coordinate mapping function, each set of commodity attribute structures corresponding to the historical compliant customs clearance sample data is mapped to coordinate points in the commodity attribute structure space to obtain a set of spatial points;

[0023] Calculate the spatial distance between any two spatial points in the set of spatial points, and divide spatial points with a spatial distance less than the preset distance threshold into the same candidate clustering region according to the preset distance threshold to obtain a set of candidate clustering regions;

[0024] Count the number of spatial points contained in each candidate cluster region, and determine the candidate cluster regions whose number of spatial points exceeds a preset threshold as valid cluster regions;

[0025] For each valid cluster region, extract the common attribute data range under each attribute dimension in the product attribute structure set corresponding to all spatial points contained in the valid cluster region, use the common attribute data range as the stable attribute constraint corresponding to the valid cluster region, and construct the product attribute stable structure set according to the stable attribute constraints corresponding to all valid cluster regions.

[0026] In some possible implementations, mapping the current product attribute structure set to the product attribute structure space to obtain the positional distribution state of the current product attribute structure set in the stable product attribute structure set includes:

[0027] Based on the coordinate mapping function, the set of product attribute structures of the current product to be evaluated is mapped to a current coordinate point in the product attribute structure space, thus obtaining the current coordinate point;

[0028] Calculate the spatial distance between the current coordinate point and each effective clustering region in the stable structure set of product attributes. Determine whether the current coordinate point falls within the coverage area of ​​any effective clustering region based on whether the spatial distance is less than a preset membership distance threshold, and obtain the positional distribution state of the current product attribute structure set in the stable structure set of product attributes.

[0029] In some possible implementations, based on the location distribution state, the membership relationship between the current set of product attribute structures and the stable set of product attribute structures is determined to obtain a determination result, including:

[0030] If the location distribution status indicates that the current coordinate point falls within the coverage area of ​​any effective clustering area, then it is determined that the current product attribute structure set belongs to the product attribute stable structure set, and the determination result is "belongs to".

[0031] Otherwise, it is determined that the current set of product attribute structures does not belong to the set of stable product attribute structures, and the determination result is "does not belong".

[0032] In some possible implementations, if the determination result is "not belonging", then the current product attribute structure set is compared with the stable product attribute structure set arranged in chronological order, and the one most recent to the current time is calculated to obtain the degree of structure boundary offset, including:

[0033] Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series to obtain the historical stable structure sequence;

[0034] Select the set of stable commodity attribute structures corresponding to the time node closest to the current time from the historical stable structure sequence to obtain the set of most recent stable structures;

[0035] Obtain the current coordinate point of the current product attribute structure set in the product attribute structure space, and calculate the shortest spatial distance from the current coordinate point to all spatial points contained in each effective cluster region of the nearest stable structure set to obtain the degree of structural boundary offset.

[0036] In some possible implementations, the perturbation feature decomposition of the structural boundary offset degree is performed to obtain a structural offset feature set, including:

[0037] From the set of most recently stable structures, extract the common attribute data range for each attribute dimension to obtain the common attribute data range corresponding to each attribute dimension;

[0038] From the current product attribute structure set, read the corresponding attribute data according to each attribute dimension to obtain the attribute data for each attribute dimension, which is used as the current attribute data set;

[0039] For each attribute dimension, the current attribute data is compared with the corresponding common attribute data range. If the current attribute data exceeds the common attribute data range, the excess amount is calculated as the offset sub-component of the current attribute dimension.

[0040] If the current attribute data does not exceed the range of the common attribute data, then the offset sub-component of the current attribute dimension is recorded as zero, and the offset sub-component corresponding to each attribute dimension is obtained.

[0041] Combine the offset sub-components corresponding to all attribute dimensions to obtain the set of structural offset features.

[0042] In some possible implementations, the step of tracking the structural evolution of the stable product attribute structure set based on historical time series, generating a structural evolution path, and judging the continuity of structural changes in the stable product attribute structure set over time series based on the consistency of evolution paths between adjacent time nodes, and generating a credibility correction result, includes:

[0043] Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series to obtain the historical stable structure sequence;

[0044] For each pair of adjacent time nodes in the historical stable structure sequence, the minimum spatial distance between the two commodity attribute stable structure sets is calculated. The minimum spatial distance is used as the evolution step size between adjacent time nodes, and the evolution steps are combined in chronological order to generate the structure evolution path.

[0045] Calculate the change between adjacent evolution steps in the structural evolution path, and determine the local consistency of the structural evolution path based on whether the change is less than a preset consistency threshold, and obtain the local consistency determination results.

[0046] The proportion of all local consistency judgment results that meet the consistency condition is counted. If the proportion is greater than the preset continuity threshold, the structural change is determined to be continuous in the time series, and a high-confidence correction result is generated. Otherwise, it is determined to be discontinuous, and a low-confidence correction result is generated.

[0047] In some possible implementations, the progressive risk assessment based on the credibility correction results of the product attribute structure offset feature set and the product attribute stable structure set, and the output of the corresponding customs clearance compliance risk warning level, includes:

[0048] The offset breadth is determined based on the number of non-zero offset sub-components in the structural offset feature set, and the offset depth is determined based on the value of each offset sub-component.

[0049] The offset breadth and offset depth are compared with a preset risk level threshold to obtain the first risk level;

[0050] The first risk level is adjusted based on the credibility adjustment result: if the credibility adjustment result is high credibility, the first risk level is maintained; if the credibility adjustment result is low credibility, the first risk level is increased by one level.

[0051] The revised risk level is output as the early warning level for customs clearance compliance risks.

[0052] The beneficial effects of this invention are as follows:

[0053] 1. The present invention discloses a customs clearance compliance risk early warning method based on big data analysis. First, it acquires multi-dimensional attribute data of target commodities in customs declarations and combines the attributes into a commodity attribute structure set reflecting the inherent dependency relationship according to preset association constraints. By spatially mapping and distribution clustering identification of a large number of historical structure sets, a stable structure set of commodity attributes representing normal trade patterns is constructed. Then, the current commodity is mapped to the same space for membership determination. The whole process is equivalent to establishing a pre-diagnosis and health management based on structural consistency for the customs clearance supervision system, thereby issuing an early warning at the incipient stage of complex false declaration behavior, avoiding misjudging coordinated violations as random fluctuations, significantly improving the stability of risk warning and early detection capability, and ensuring the accurate allocation of regulatory resources and efficient maintenance of customs clearance order.

[0054] 2. The customs clearance compliance risk early warning method based on big data analysis described in this invention further calculates the degree of boundary offset between the structure and the most recent stable structure, and obtains the precise offset sub-components in each attribute dimension through perturbation feature decomposition; simultaneously, it tracks the evolution path of the stable structure based on historical time series, generates a credibility correction result based on the consistency of the evolution path between adjacent nodes, and finally performs a progressive risk judgment and outputs an early warning level by combining the offset breadth, depth and credibility. This process, through the fine decomposition of structural offset features and dynamic calibration of evolution credibility, can adaptively track the slow changes in normal trade patterns, continuously maintain the effectiveness of the cluster boundary, and realize long-term pre-diagnosis and health management of customs clearance risk status. This multi-level progressive judgment method effectively suppresses the accumulation of false alarms and omissions caused by trade pattern drift, ensures the adaptability and reliability of the early warning level output, and provides a stable and interpretable risk basis for regulatory decisions. Attached Figure Description

[0055] The invention will now be further described with reference to the accompanying drawings.

[0056] Figure 1 This is a flowchart of a customs clearance compliance risk early warning method based on big data analysis according to the present invention; Detailed Implementation

[0057] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are used to distinguish different objects, not to limit a specific order.

[0058] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0059] Research has revealed that existing technologies can only perform isolated threshold comparisons of customs declaration data or rely on pre-set static rule bases for anomaly matching. They cannot perform structured modeling and spatial distribution analysis of the inherent correlations and constraints between different attribute data, making it difficult to dynamically depict the clustering patterns of historical normal samples in the overall distribution space and their continuous evolution over time. This leads to the system's tendency to misjudge complex false declarations and concealment as random fluctuations when faced with such behaviors. Furthermore, it faces the challenge of accumulating false or missed declarations due to the slow evolution of normal trade patterns, requiring frequent reliance on manual review, threshold adjustments, or repeated testing. This wastes regulatory resources and leads to missed risks, ultimately reducing the accuracy and stability of customs compliance risk warnings.

[0060] Example 1

[0061] To solve the above problems, such as Figure 1 As shown, this application provides a method for early warning of customs clearance compliance risks based on big data analysis, the method comprising:

[0062] Step 1: Obtain the multi-dimensional attribute data set of the target product during the customs declaration process, and combine and associate the attribute data according to the preset attribute association constraint rules to form a product attribute structure set;

[0063] In step one: target commodity multidimensional attribute data set: refers to a set of attribute data from multiple different dimensions collected for the same customs clearance commodity, such as commodity name, HS code, country of origin, declared price, quantity, weight, material, and purpose;

[0064] Preset attribute association constraint rules: These are pre-defined rules used to specify the combination methods and association logic between different attribute data. For example, they may stipulate that "HS code" and "product name" must appear in pairs, or that there is a logical verification relationship between "place of origin" and "declared price".

[0065] Combination association: refers to the operation of connecting and merging different attribute data in a master-slave, calculation or labeling manner according to the logical relationship of preset rules;

[0066] Product attribute structure set: refers to the structured data with hierarchical or related relationships formed by combining the attribute data according to the above association constraint rules.

[0067] For example: During customs declaration, the early warning platform receives an electronic customs declaration for imported goods and extracts multiple attribute data of the target commodity (such as commodity name, HS code, country of origin, declared unit price, declared quantity, packaging method, etc.) to form a multi-dimensional attribute data set. Subsequently, the early warning platform reads preset attribute association constraint rules, which stipulate that different attributes must appear in pairs or have a logical correspondence (for example, the first four digits of the HS code correspond to the material category of the commodity name; country of origin and declared unit price must be stored as a pair of associated data; declared quantity and packaging method must be stored as a pair of associated data). The early warning platform performs combination association according to these rules: First, it verifies the correspondence between the first four digits of the HS code and the material in the commodity name. After confirming a match, it combines the HS code and commodity name into an attribute pair as the basic layer of the structure; second, it combines the country of origin and declared unit price side-by-side to form a second attribute pair as the intermediate layer of the structure; finally, it combines the declared quantity and packaging method side-by-side to form a third attribute pair as the additional layer of the structure. The early warning platform merges the above attribute pairs in the order of "basic layer - intermediate layer - additional layer" and outputs a set containing multiple ordered attribute pairs. This set is the formed product attribute structure set.

[0068] Among them: attribute pair: refers to a binary combination formed by placing two attribute data side by side according to a preset rule;

[0069] Base layer: refers to the attribute pairs at the beginning of the product attribute structure set, used to identify the core category of the product;

[0070] Intermediate layer: refers to the attribute pair layer located after the base layer, used to record origin and price information;

[0071] Additional layer: refers to the attribute pair layer located at the end, used to record quantity and packaging information.

[0072] It should be noted that by acquiring the original declaration data and combining and associating attributes according to preset rules, the originally scattered attribute data is transformed into a set of commodity attribute structures with a clear internal hierarchical order. This structured output provides standardized data units for subsequent mapping of commodities to the attribute structure space, enabling distance calculation and cluster identification between different commodities at a unified attribute pair level, thereby laying a data foundation for the quantitative assessment of customs clearance compliance risks.

[0073] Step 2: Obtain the historical product attribute structure set, analyze the historical product attribute structure set, and construct the product attribute structure space;

[0074] Furthermore: Each set of commodity attribute structures in the historical compliant customs clearance sample data is taken as a sample point to obtain a set of sample points;

[0075] Calculate the difference value of attribute data between any two sample points in the sample point set in each attribute dimension, and sum the absolute values ​​of the difference values ​​in all attribute dimensions to obtain the structural distance between the two sample points. Then, obtain the structural distance between all pairs of sample points in the sample point set to form a structural distance set.

[0076] Based on the structural distance set, the relative position coordinates of each sample point in the sample point set in a preset dimension space are determined by a multidimensional scaling analysis algorithm, and the corresponding relationship of sample point coordinates is obtained.

[0077] Based on the coordinate correspondence of the sample points, interpolation is performed on the sparse regions of the coordinate results obtained from the multidimensional scaling analysis in the space of a preset number of dimensions to obtain the coordinate mapping function.

[0078] The product attribute structure space is formed by using the set of all coordinate points covered by the range of the coordinate mapping function in the preset dimension space.

[0079] In step two: Historical commodity attribute structure set: refers to the total number of commodity attribute structure sets obtained from the declaration data of customs clearance procedures completed within a historical time period (such as the past year, the past three years, or the statistical window period preset by the system);

[0080] Historical compliant customs clearance sample data: refers to the set of declaration data and related structures of all goods that have been determined by customs to have no risk of violation and have successfully completed customs clearance in historical customs clearance records;

[0081] Sample point: refers to treating each set of product attribute structures as an independent data unit that can participate in calculations;

[0082] Sample point set: refers to the set consisting of all sample points;

[0083] Attribute dimension: refers to the independent measurement direction corresponding to each specific attribute in the product attribute structure set;

[0084] Structural distance: refers to the sum of the absolute values ​​of the differences in attribute data across all attribute dimensions between two sample points, used to quantify the overall degree of difference between two sets of product attribute structures;

[0085] Structural distance set: refers to the set of all pairwise structural distances between all sample points in the sample point set;

[0086] Multidimensional scaling analysis algorithm: refers to a statistical analysis method that maps the similarity or distance relationship between high-dimensional data points to a low-dimensional space while preserving the original distance relationship as much as possible;

[0087] Preset dimension space: refers to a pre-defined low-dimensional space used for dimensionality reduction visualization, whose number of dimensions (such as two or three dimensions) is determined before calculation;

[0088] The correspondence between sample point coordinates refers to the coordinate position assigned to each sample point in the low-dimensional space after multidimensional scaling analysis.

[0089] Sparse region: refers to a local area in a space of a predetermined number of dimensions where the distribution of coordinate points is relatively sparse and the density is significantly lower than the average level after multi-dimensional scaling analysis;

[0090] Interpolation completion: refers to the operation of estimating and supplementing missing coordinate points in a sparse region based on the data distribution of existing coordinate points around that region using mathematical methods.

[0091] Coordinate mapping function: refers to a mathematical function that can take any set of product attribute structures as input and output its unique coordinate position in a preset number of dimensions;

[0092] Range: refers to the set of all output values ​​(i.e., all possible coordinate points) of the coordinate mapping function;

[0093] Product attribute structure space: refers to the virtual geometric space consisting of all coordinate points covered by the range of the coordinate mapping function and the continuous region between these points.

[0094] For example: The early warning platform reads the set of commodity attribute structures corresponding to all goods that have been judged to be compliant by customs within a certain period of time from the historical compliant customs clearance sample database, and obtains multiple sample points, which together constitute the sample point set;

[0095] The early warning platform calculates the structural distance between any two sample points in the sample point set: for each pair of sample points, the difference in attribute data is calculated dimension by dimension, and the absolute values ​​of the differences in each dimension are added together to obtain the structural distance between the two sample points. The early warning platform repeats the above calculation for all pairwise combinations in the sample point set to form a structural distance set. Based on this structural distance set, the early warning platform calls a multidimensional scaling analysis algorithm (first, each historical sample point is regarded as a point in a high-dimensional space, and its original attribute dimension is the dimension of the high-dimensional space. Since the high-dimensional space cannot be directly visualized and subsequent geometric calculations can not be performed, it needs to be mapped to a low-dimensional space with a pre-set number of dimensions, such as two or three dimensions, through multidimensional scaling analysis. Second, the multidimensional scaling analysis algorithm takes the structural distance set as input and optimizes iteratively or decomposes features in the low-dimensional space). The algorithm finds a coordinate position for each sample point in the low-dimensional space such that the difference between the Euclidean distance (straight-line distance) between any two sample points in this low-dimensional space and the corresponding structural distance in the original high-dimensional space is minimized. Specifically, the algorithm continuously adjusts the coordinates of each sample point in the low-dimensional space until the sum of the squares of the distance errors between all point pairs, i.e., the differences between the low-dimensional space distance and the original structural distance, is minimized. Finally, the algorithm outputs a unique coordinate position for each sample point in a space of a preset number of dimensions. For example, it outputs horizontal and vertical coordinates for two-dimensional space and three coordinate values ​​for three-dimensional space, forming a coordinate correspondence between sample points. This maps each sample point to a space of a preset number of dimensions (e.g., three dimensions) while preserving the original structural distance relationship as much as possible, thereby assigning a low-dimensional space coordinate to each sample point and obtaining the coordinate correspondence between sample points.

[0096] When the early warning platform observes the distribution of coordinate points in low-dimensional space and finds that certain local areas have very few coordinate points and low density (i.e., sparse regions), in order to obtain a continuous product attribute structure space, the platform performs interpolation to complete these sparse regions. Specifically, for blank positions within a sparse region, the platform selects the coordinates of the two or more nearest known sample points around the blank position and uses linear interpolation to calculate new coordinate points according to a preset interpolation ratio (e.g., the midpoint, quarter-point, or third-quarter point of the line connecting the two points). These new coordinate points are then added to the sparse region as virtual coordinate points. For blank regions enclosed by multiple known points, planar interpolation or spatial grid interpolation can be used to calculate the position of the virtual coordinate points based on the weighted average of the coordinates of surrounding points. Through these interpolation completion operations, the point density in the sparse region is made essentially consistent with that in other regions. After interpolation completion, the early warning platform uses all known sample point coordinates and the added virtual coordinate points as known control points, and establishes a continuous functional relationship through mathematical interpolation or fitting. This function maps any input set of product attribute structures (i.e., the specific values ​​of each attribute dimension) to a unique coordinate position in a space with a preset number of dimensions. Specifically, for the input set of product attribute structures, the function first uses a coordinate mapping relationship determined by a multidimensional scaling analysis algorithm to find several known sample points that are closest to its attribute structure. Then, based on the coordinate values ​​of these known points and their structural distance weights from the input point, the spatial coordinates of the input point are calculated using a weighted average or similar method. The function constructed in this way is the coordinate mapping function.

[0097] The set of all coordinate points covered by the range of the coordinate mapping function in the preset dimension space constitutes the product attribute structure space.

[0098] Among them: virtual coordinate points: refer to coordinate points that are artificially generated by interpolation completion methods and do not correspond to the actual product attribute structure set. They are used to fill sparse areas to ensure spatial continuity.

[0099] It should be noted that by transforming the historical compliant product attribute structure set into a sample point set, calculating pairwise structural distances, and using a multidimensional scaling analysis algorithm to map it to a low-dimensional space, and then interpolating and completing sparse regions, a continuous and complete product attribute structure space is finally constructed. This space provides a geometric reference framework for subsequently mapping the current product into the space for cluster identification and membership determination, enabling the structured data output in step one to be quantitatively compared in a unified space. This ensures a complete logical chain from data preprocessing to spatial modeling and reduces the probability of misjudgment due to spatial discontinuity.

[0100] Step 3: In the product attribute structure space, perform distribution clustering identification on the historical product attribute structure set, and construct a stable product attribute structure set based on the identified clustering regions;

[0101] Furthermore: Based on the coordinate mapping function, each set of commodity attribute structures corresponding to the historical compliant customs clearance sample data is mapped to coordinate points in the commodity attribute structure space to obtain a set of spatial points;

[0102] Calculate the spatial distance between any two spatial points in the set of spatial points, and divide spatial points with a spatial distance less than the preset distance threshold into the same candidate clustering region according to the preset distance threshold to obtain a set of candidate clustering regions;

[0103] Count the number of spatial points contained in each candidate cluster region, and determine the candidate cluster regions whose number of spatial points exceeds a preset threshold as valid cluster regions;

[0104] For each valid cluster region, extract the common attribute data range under each attribute dimension in the product attribute structure set corresponding to all spatial points contained in the valid cluster region, use the common attribute data range as the stable attribute constraint corresponding to the valid cluster region, and construct the product attribute stable structure set according to the stable attribute constraints corresponding to all valid cluster regions.

[0105] In step three: Distribution clustering identification: refers to the process of finding the areas where coordinate points are clustered in the product attribute structure space by calculating the distance and density between coordinate points;

[0106] Clustered area: refers to a localized, densely populated area in the product attribute structure space, consisting of multiple coordinate points that are close to each other;

[0107] Spatial point set: refers to the set of all coordinate points obtained after transforming each commodity attribute structure set corresponding to historical compliant customs clearance sample data through a coordinate mapping function;

[0108] Spatial distance: refers to the Euclidean distance (i.e., straight-line distance) between any two coordinate points in a set of spatial points, used to measure how close or far two points are in the product attribute structure space;

[0109] Preset distance threshold: refers to a pre-set value used to determine whether two spatial points belong to the same candidate clustering region; if the spatial distance between two points is less than the threshold, they are considered to belong to the same region;

[0110] Candidate clustering region: refers to a region initially defined based on a distance threshold, consisting of spatial points that meet the distance conditions, and has not yet undergone quantity screening;

[0111] Candidate clustering region set: refers to the set consisting of all candidate clustering regions;

[0112] Preset quantity threshold: refers to a pre-set integer used to determine whether a candidate clustering region is a valid clustering region; if the number of spatial points contained in the region exceeds this threshold, it is considered valid.

[0113] Effective clustering area: refers to a clustering area with a sufficient number of spatial points that has been confirmed after quantitative screening, representing the distribution area of ​​typical attribute combinations in historical compliant samples;

[0114] The range of common attribute data refers to the interval between the minimum and maximum values ​​of each attribute dimension within the original product attribute structure set corresponding to all spatial points within a valid aggregation area. For categorical attributes, the range of common attribute data is all category values ​​that have appeared within that area.

[0115] Stable attribute constraints: refer to a set of rules consisting of the range of common attribute data, used to describe the restrictions on attribute values ​​in an effective clustering region;

[0116] Stable structure set of product attributes: refers to the set constructed based on the stable attribute constraints corresponding to all valid clustered regions, used to represent the typical stable distribution pattern of compliant products in the attribute space in history.

[0117] For example: The early warning platform, based on a coordinate mapping function, maps each set of product attribute structures corresponding to historical compliant customs clearance sample data to coordinate points in the product attribute structure space, obtaining a set of spatial points. The early warning platform calculates the spatial distance between any two spatial points in the set and, according to a preset distance threshold, assigns spatial points with a spatial distance less than the threshold to the same candidate clustering region, thus obtaining a set of candidate clustering regions. The early warning platform counts the number of spatial points contained in each candidate clustering region and determines the candidate clustering regions with a number of spatial points exceeding a preset threshold as valid clustering regions, while regions with insufficient numbers are considered noise regions and discarded. For each valid clustering region, the early warning platform extracts the original set of product attribute structures corresponding to all spatial points within the region and counts the range of common attribute data under each attribute dimension (e.g., for numerical attributes, the range consisting of the minimum and maximum values; for categorical attributes, all occurrences of category values). The early warning platform uses the range of common attribute data corresponding to each valid clustering region as the stable attribute constraint for that region, and the set of stable attribute constraints for all valid clustering regions as the set of stable product attribute structures.

[0118] Where: Euclidean distance: refers to the straight-line distance between two points in space, and the calculation formula is the square root of the sum of the squares of the differences between the coordinates;

[0119] Noise region: refers to the region in the candidate cluster where the number of spatial points does not exceed the preset threshold. These regions are considered to be statistically insignificant and do not represent a typical compliance pattern.

[0120] It should be noted that: by clustering and identifying historical compliant sample points in the constructed commodity attribute structure space, using spatial distance and quantity thresholds to filter out statistically significant effective clustering areas, and extracting the common data range of each attribute dimension in each area as stable attribute constraints, a stable set of commodity attributes is finally formed.

[0121] Step 4: Map the current product attribute structure set to the product attribute structure space to obtain the position distribution state of the current product attribute structure set in the stable product attribute structure set, and based on the position distribution state, determine the membership relationship between the current product attribute structure set and the stable product attribute structure set to obtain the determination result;

[0122] Furthermore: Based on the coordinate mapping function, the set of product attribute structures of the current product to be evaluated is mapped to a current coordinate point in the product attribute structure space to obtain the current coordinate point;

[0123] Calculate the spatial distance between the current coordinate point and each effective clustering region in the stable structure set of product attributes. Determine whether the current coordinate point falls within the coverage of any effective clustering region based on whether the spatial distance is less than a preset membership distance threshold, and obtain the positional distribution state of the current product attribute structure set in the stable structure set of product attributes.

[0124] Furthermore: if the location distribution status indicates that the current coordinate point falls within the coverage area of ​​any effective clustering area, then it is determined that the current product attribute structure set belongs to the product attribute stable structure set, and the determination result is "belongs to".

[0125] Otherwise, it is determined that the current set of product attribute structures does not belong to the set of stable product attribute structures, and the determination result is "does not belong".

[0126] In step four: Current coordinate point: refers to the unique coordinate position obtained in the product attribute structure space after inputting the product attribute structure set of the current product to be evaluated into the coordinate mapping function;

[0127] Preset membership distance threshold: refers to a pre-set value used to determine whether the current coordinate point is sufficiently close to a certain effective clustering area in terms of spatial distance; if the shortest spatial distance from the current coordinate point to a certain effective clustering area is less than the threshold, then the current coordinate point is considered to fall within the coverage area of ​​that area.

[0128] Coverage range: refers to the area surrounded by a spherical neighborhood with a radius equal to the membership distance threshold, based on all spatial points within the effective cluster area; if the current coordinate point falls into this area, it means that it is spatially adjacent to the effective cluster area, and its attribute structure characteristics are similar to the typical distribution pattern of historical compliant samples.

[0129] Location distribution status: refers to the spatial relationship between the current coordinate point and each effective cluster area, specifically whether it falls within the coverage area of ​​a certain effective cluster area, and (optionally) the distance between it and the nearest effective cluster area.

[0130] For example: After the early warning platform completes the construction of the product attribute structure set for the current product and the construction of the stable product attribute structure set, it will enter the membership determination stage;

[0131] First, the early warning platform obtains the set of product attribute structures for the product to be evaluated (for example, the base layer of this set contains HS code and product name, the intermediate layer contains country of origin and declared unit price, and the additional layer contains declared quantity and packaging method; specific attribute values ​​have been combined and associated according to preset rules). The early warning platform uses this set of product attribute structures as input and calls a pre-built coordinate mapping function. This function calculates the unique coordinates of the current product in the product attribute structure space based on the structural distance relationships between historical sample points and the interpolation completion results, thus obtaining the current coordinate point (for example, the coordinates of this point in three-dimensional space are three numerical values).

[0132] Secondly, the early warning platform acquires a stable set of product attributes, which contains multiple effective clustering regions. Each effective clustering region corresponds to a set of stable attribute constraints and has a clear spatial location range in the product attribute structure space (i.e., the coordinate distribution range of all historical sample points within that region). The platform calculates the spatial distance from the current coordinate point to each effective clustering region. Specifically, for each effective clustering region, it finds the point within that region with the closest Euclidean distance to the current coordinate point and uses this minimum Euclidean distance as the spatial distance from the current coordinate point to that effective clustering region. The early warning platform then reads a preset membership distance threshold (this threshold is a pre-set value based on the dispersion of historical compliant samples, for example, determined by multiplying the standard deviation of the distance between each point within the effective clustering region and the region center by a coefficient). For each effective clustering region, the early warning platform determines whether the spatial distance from the current coordinate point to that region is less than the preset membership distance threshold.

[0133] If at least one valid clustering region exists such that the spatial distance from the current coordinate point to that region is less than the membership distance threshold, the early warning platform determines that the current coordinate point falls within the coverage area of ​​that valid clustering region, meaning the current product attribute structure set's positional distribution status within the product attribute stable structure set is "falling within the coverage area." Conversely, if the spatial distance between all valid clustering regions is greater than or equal to the membership distance threshold, it is determined that "not falling within any coverage area."

[0134] When the location distribution status is "falls within the coverage area", it is determined that the current product attribute structure set belongs to the stable product attribute structure set, and the output judgment result is "belongs to". This means that the attribute combination of the current product is consistent with the typical stable distribution pattern of compliant products in history, and it is initially determined that there is no significant deviation risk;

[0135] When the location distribution status is "not within any coverage area", it is determined that the current product attribute structure set does not belong to the stable product attribute structure set, and the output judgment result is "does not belong". This means that the attribute combination of the current product deviates from the typical distribution pattern of historical compliant products, which may be abnormal or have a high risk, and the subsequent progressive risk assessment process needs to be initiated.

[0136] Wherein: Euclidean distance: refers to the straight-line distance between two points in three-dimensional space, and the calculation formula is the square root of the sum of the squares of the differences between the coordinates; in this step, it is used to calculate the spatial distance between the current coordinate point and each spatial point in the effective aggregation area;

[0137] Spatial distance (in the context of membership determination): refers to the shortest Euclidean distance from the current coordinate point to a certain effective clustering region, that is, the distance value that is closest to the current coordinate point among all spatial points in that region;

[0138] The preset membership distance threshold refers to a pre-defined numerical radius used to define the coverage area of ​​the effective clustering region. This threshold can be a fixed empirical value or can be adaptively set according to the internal dispersion of different effective clustering regions (for example, taking twice the average distance from each point in the region to the center of the region). This embodiment uses a globally unified threshold to simplify implementation.

[0139] Coverage range: refers to the union of spherical neighborhoods formed by all spatial points within the effective aggregation area as sphere centers and the membership distance threshold as radius; the current coordinate point falling into this union is considered to fall within the coverage range.

[0140] It should be noted that by mapping the current product to the product attribute structure space and calculating its spatial distance from the historical compliant cluster area, and using a preset membership distance threshold for coverage judgment, a quantitative determination of whether the current product belongs to a historical stable distribution pattern is achieved.

[0141] Step 5: If the determination result is "not belonging", then calculate the degree of structural boundary offset by comparing the current product attribute structure set with the one most recent in the stable structure set of each product attribute arranged in chronological order.

[0142] Further: Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series, and obtain the historical stable structure sequence;

[0143] Select the set of stable commodity attribute structures corresponding to the time node closest to the current time from the historical stable structure sequence to obtain the set of most recent stable structures;

[0144] Obtain the current coordinate point of the current product attribute structure set in the product attribute structure space, and calculate the shortest spatial distance from the current coordinate point to all spatial points contained in each effective cluster region of the nearest stable structure set to obtain the degree of structural boundary offset.

[0145] In step five: Historical stable structure sequence: refers to the sequence of stable structures of commodity attributes constructed at each historical time point, arranged in chronological order; for example, with each month or quarter as the time point, each point corresponds to a stable structure set constructed based on historical compliance samples up to that point;

[0146] The most recent stable structure set refers to the set of stable commodity attribute structures selected from the historical stable structure sequence that is closest in time to the current declaration time.

[0147] Structural boundary offset: This refers to a value used to quantify the degree to which the current product attribute structure set deviates from the boundary of the most recent stable structure set; specifically, it is the shortest spatial distance from the current coordinate point to all spatial points within each effective cluster region of the most recent stable structure set; the larger this value, the higher the degree to which the current product deviates from the historical stable pattern.

[0148] For example, when the judgment result is "not belonging," the early warning platform will pre-store multiple historical stable structure sets of commodity attributes at fixed time intervals (e.g., every quarter or every six months). Each set is constructed as follows: based on historical compliant customs clearance sample data up to that moment, compliant samples before that moment are mapped into the commodity attribute structure space. By calculating the Euclidean distance between spatial points, using preset distance and quantity thresholds for cluster identification, effective clustering areas are obtained. The common attribute data range of each attribute dimension within each effective clustering area is extracted, thus forming the stable structure set of commodity attributes at that moment. These sets are arranged chronologically from earliest to latest, collectively forming a historical stable structure sequence. For example, the sequence may include stable structure sets from the end of the previous quarter, stable structure sets from the end of the previous year, etc.

[0149] Secondly, the early warning platform selects the set of stable commodity attribute structures corresponding to the most recent time point from this historical stable structure sequence, denoted as the most recent stable structure set. This set represents the typical stable pattern of commodity attribute distribution within the most recent verified historical compliance period.

[0150] Then, for each valid cluster region in the set of most recently stable structures, the platform calculates the Euclidean distance from the current coordinate point to all spatial points within that region (i.e., the coordinates of historical sample points contained in that valid cluster region), and takes the minimum of these distances as the spatial distance from the current coordinate point to that region. Among all valid cluster regions, the platform then takes the minimum of these spatial distances, which is the shortest spatial distance from the current coordinate point to the entire set of most recently stable structures. This shortest spatial distance is defined as the degree of structural boundary offset.

[0151] For example, if the minimum spatial distance from the current coordinate point to a valid clustering region in the most recent stable structure set is a certain value d, then the degree of structural boundary offset is d. The larger this value is, the greater the difference between the current product's attribute combination and the most recent historical stable pattern; conversely, if this value is very small and only slightly greater than the membership distance threshold, it indicates that although the current product does not fall into any coverage area, it is very close to the boundary of the historical stable pattern and is in a critical state.

[0152] Where: Shortest spatial distance: refers to the minimum Euclidean distance from the current coordinate point to all spatial points contained in all valid cluster regions in the nearest stable structure set. The calculation method is as follows: for each valid cluster region, traverse all spatial points in the region, calculate the Euclidean distance between the current coordinate point and each spatial point, and take the minimum distance in the region; then take the minimum value among the minimum distances of all valid cluster regions.

[0153] Time point: refers to the cutoff point of historical data used when constructing a stable set of product attributes, such as the end of the first quarter of 2024, the end of the second quarter of 2024, etc. The interval between adjacent time points can be preset (such as monthly, quarterly or annual).

[0154] It should be noted that when a product is determined to "not belong" to the current product attribute stable structure set, it only indicates that it deviates from the latest stable model currently in use, but it is not enough to completely determine the risk level. By introducing a historical stable structure sequence and selecting the stable structure set most recent to the current time to calculate the boundary offset, the deviation of the current product from the most recent historical compliance model can be quantified. This offset provides a basic numerical value for the perturbation feature decomposition and progressive risk assessment in subsequent steps, enabling risk warnings to distinguish between minor deviations and severe anomalies.

[0155] Step 6: Perform perturbation feature decomposition on the degree of structural boundary offset to obtain a set of structural offset features;

[0156] Furthermore: from the set of most recently stable structures, extract the common attribute data range for each attribute dimension to obtain the common attribute data range corresponding to each attribute dimension;

[0157] From the current product attribute structure set, read the corresponding attribute data according to each attribute dimension to obtain the attribute data for each attribute dimension, which is used as the current attribute data set;

[0158] For each attribute dimension, the current attribute data is compared with the corresponding common attribute data range. If the current attribute data exceeds the common attribute data range, the excess amount is calculated as the offset sub-component of the current attribute dimension.

[0159] If the current attribute data does not exceed the range of the common attribute data, then the offset sub-component of the current attribute dimension is recorded as zero, and the offset sub-component corresponding to each attribute dimension is obtained.

[0160] Combine the offset sub-components corresponding to all attribute dimensions to obtain the set of structural offset features.

[0161] In step six: Offset sub-component: refers to the amount by which the current product's attribute data exceeds the range of common attribute data for each attribute dimension; if the current data is within the common range, the offset sub-component is zero; if the current data is greater than the maximum value of the common range, the excess is the current value minus the maximum value; if the current data is less than the minimum value of the common range, the excess is the minimum value minus the current value (taking a positive value).

[0162] Structural offset feature set: refers to the set (or vector) composed of offset sub-components of all attribute dimensions, used to quantitatively describe the specific deviation of the current product from the nearest stable structural set in each attribute dimension; this set reflects the detailed deviation pattern of the current product's attribute combination.

[0163] For example: First, the early warning platform extracts the common attribute data range for each attribute dimension from the most recently stable structure set. For instance, a valid cluster region in the most recently stable structure set contains the following common attribute data ranges: in the basic layer, the first four digits of the HS code are fixed to a certain value, and the product name is fixed to a certain type of material; in the intermediate layer, the place of origin is several allowed categories, and the declared unit price is between the minimum and maximum values; in the supplementary layer, the declared quantity is fixed to a certain value, and the packaging method is several allowed types. The early warning platform lists these ranges one by one according to the attribute dimension, as the common attribute data range corresponding to each dimension.

[0164] Secondly, the early warning platform reads the corresponding attribute data according to the same attribute dimensions based on the current set of product attribute structures. For example, the specific values ​​of the first four digits of the current product's HS code, product name, country of origin, declared unit price, declared quantity, and packaging method constitute the current attribute data set;

[0165] If the current attribute data exceeds the common range (for example, the current declared unit price is higher than the maximum value or lower than the minimum value in the common range; or the current country of origin is not in the list of allowed categories), then the excess amount is calculated as the offset sub-component of that dimension. For numerical attributes, the excess amount is the absolute value of the difference between the current value and the nearest boundary value; for categorical attributes, if the current value is not in the common category set, the excess amount is recorded as 1 (or assigned a value according to the preset category difference weight).

[0166] If the current attribute data does not exceed the common range (i.e., the value is within the range, or the category is in the allowed set), then the offset sub-component of this dimension is recorded as zero;

[0167] The early warning platform repeats the above comparison and calculation for all attribute dimensions to obtain the offset sub-components corresponding to each dimension.

[0168] Finally, the early warning platform combines the offset sub-components corresponding to all attribute dimensions according to a predetermined attribute order, forming a set or vector with the same number of attribute dimensions. This result is the structural offset feature set. For example, the structural offset feature set can be represented as (base layer offset, intermediate layer origin offset, intermediate layer unit price offset, additional layer quantity offset, additional layer packaging method offset, etc.), where non-zero components indicate which specific attributes of the current product deviate from the historical stable pattern, and zero components indicate that the attribute conforms to historical patterns.

[0169] Wherein: Excess Amount: For numerical attributes, if the current value is higher than the maximum value in the common range, the excess amount = current value - maximum value; if the current value is lower than the minimum value in the common range, the excess amount = minimum value - current value. For categorical attributes, if the current value is not within the common category set, the excess amount is 1 (or a specific value is calculated based on the preset category distance function); if it is within the set, the excess amount is 0.

[0170] It should be noted that: through perturbation feature decomposition, the overall structural boundary offset degree (scalar) obtained in step five is further decomposed into offset sub-components (vectors) on each attribute dimension, thereby obtaining a set of structural offset features. This set can reveal the specific reasons why the current commodity deviates from the historical stable pattern (e.g., abnormal price, abnormal place of origin, or abnormal quantity), providing a more detailed basis for subsequent credibility correction and progressive risk assessment, making the early warning results interpretable.

[0171] Step 7: The structure evolution of the stable structure set of commodity attributes is tracked based on historical time series, generating a structure evolution path. The continuity of the structural changes of the stable structure set of commodity attributes in the time series is judged based on the consistency of the evolution path between adjacent time nodes, and a credibility correction result is generated.

[0172] Further: Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series, and obtain the historical stable structure sequence;

[0173] For each pair of adjacent time nodes in the historical stable structure sequence, the minimum spatial distance between the two commodity attribute stable structure sets is calculated. The minimum spatial distance is used as the evolution step size between adjacent time nodes, and the evolution steps are combined in chronological order to generate the structure evolution path.

[0174] Calculate the change between adjacent evolution steps in the structural evolution path, and determine the local consistency of the structural evolution path based on whether the change is less than a preset consistency threshold, and obtain the local consistency determination results.

[0175] The proportion of all local consistency judgment results that meet the consistency condition is counted. If the proportion is greater than the preset continuity threshold, the structural change is determined to be continuous in the time series, and a high-confidence correction result is generated. Otherwise, it is determined to be discontinuous, and a low-confidence correction result is generated.

[0176] In step seven: Evolution step size: refers to the minimum spatial distance between the stable structure sets of commodity attributes corresponding to two adjacent time nodes in the historical stable structure sequence, which is used to quantify the overall change amplitude of the stable structure sets at two adjacent time points;

[0177] Structural evolution path: refers to the sequence formed by arranging the evolution steps between all adjacent time nodes in chronological order, reflecting the trajectory of the stable structural set of commodity attributes changing over time;

[0178] The change between adjacent evolutionary steps in the evolutionary path: refers to the absolute value of the difference between the (i+1)th evolutionary step and the ith evolutionary step in the structural evolutionary path, used to measure the degree of change in evolutionary speed;

[0179] Preset consistency threshold: refers to a pre-set value used to determine whether two adjacent evolutionary steps are locally consistent; if the change in adjacent evolutionary steps is less than the threshold, the local area is considered to be continuous, otherwise it is considered to have occurred.

[0180] Local consistency determination result: refers to the binary result (satisfying consistency or not satisfying consistency) obtained by comparing the change amount with the consistency threshold for each pair of adjacent evolution steps.

[0181] Continuity threshold: refers to a pre-set proportion used to determine whether the structural evolution over the entire time series is generally continuous; if the proportion of the number of local determination results that meet the consistency condition to the total number of local determination results exceeds this threshold, it is determined to be generally continuous.

[0182] High-credibility correction results: These are results generated when the structural evolution path is generally continuous over time, indicating that the changes in the set of historical stable structures are gradual and smooth, thus providing a positive correction to the credibility of subsequent risk assessments.

[0183] Low credibility correction result: refers to the result generated when the structural evolution path is generally discontinuous in the time series, indicating that there are sudden changes or abnormal fluctuations in the set of historical stable structures, and thus the credibility of subsequent risk assessment is negatively corrected.

[0184] For example: First, the early warning platform obtains the constructed historical stable structure sequence. This sequence contains multiple stable structure sets of commodity attributes arranged in chronological order, such as the stable structure set at the end of the previous quarter, the stable structure set at the end of the previous year, etc. Each stable structure set has been constructed according to the method in step three and has a clear geometric spatial location (i.e., the coordinate distribution range of each effective cluster region).

[0185] Secondly, the early warning platform calculates the minimum spatial distance between two stable product attribute structure sets for each pair of adjacent time nodes in the historical stable structure sequence (e.g., from time node T1 to T2, from T2 to T3, ..., Tn to Tn+1). Specifically, for two adjacent stable structure sets A and B, the platform obtains the coordinates of all spatial points within all valid clustering regions contained in each set. It calculates the Euclidean distance from each spatial point in set A to each spatial point in set B, and takes the minimum of all point-to-point distances as the minimum spatial distance between the two sets. This minimum spatial distance is defined as the evolutionary step size from time node A to time node B. The early warning platform arranges the evolutionary step sizes corresponding to all adjacent time nodes in chronological order, forming a sequence (e.g., step size sequence S1, S2, S3, ..., Sn), which represents the structural evolution path.

[0186] Then, the early warning platform calculates the change in adjacent evolutionary step lengths along the structural evolution path, i.e., it calculates |S2−S1|, |S3−S2|, |S4−S3|, etc. For each change, the early warning platform reads a preset consistency threshold (this threshold is a value set based on historical experience, for example, 10% of the average of all evolutionary step lengths) and determines whether the change is less than the consistency threshold. If it is less, the change in the local evolutionary step length is determined to be continuous, and the local consistency determination result is recorded as "satisfied"; if it is not less, it is determined as "unsatisfied", thus obtaining a set of local consistency determination results.

[0187] Finally, the early warning platform calculates the proportion of all local consistency judgment results that meet the consistency criteria. Specifically, let the total number of local judgment results be N (i.e., the number of evolutionary steps minus 1), and the number of those meeting the consistency criteria be M, then the proportion = M / N. The early warning platform reads a preset continuity threshold (e.g., 0.7, i.e., 70%). If this proportion is greater than the continuity threshold, the structural evolution path is determined to be generally continuous in the time series, indicating that the historical stable structure set shows a smooth and gradual trend over time, without any drastic abrupt changes, and a high-confidence correction result is generated accordingly. Conversely, if the proportion is less than or equal to the continuity threshold, the structural evolution is determined to be discontinuous, indicating that the historical stable pattern has abrupt changes or abnormal fluctuations, and a low-confidence correction result is generated accordingly.

[0188] Wherein: Consistency threshold: a pre-set positive number, set according to the actual data distribution (e.g., 0.5 times the standard deviation of the evolution step sequence), the change is considered to be local consistency if it is less than this threshold;

[0189] Continuity threshold: A proportional value between 0 and 1, such as 0.7 or 0.8, which can be adjusted according to the required stringency.

[0190] It should be noted that by tracking the evolution and analyzing the consistency of historical stable structural sequences, it is possible to determine whether the pattern of changes in the stable structure of commodity attributes over time is stable. When the historical structural evolution is continuous, it indicates that the attribute distribution pattern of compliant commodities has good temporal stability, and the risk assessment results obtained in steps four to six are highly reliable. When the evolution is discontinuous, it indicates that the historical pattern has undergone abrupt changes (e.g., due to policy adjustments or changes in the trade environment), and the assessment results for the current commodity may be biased, requiring a reduction in reliability. This reliability correction provides a weighting basis for subsequent progressive risk assessments, improving the adaptability of the early warning system.

[0191] Step 8: Based on the credibility correction results of the product attribute structure offset feature set and the product attribute stable structure set, perform progressive risk assessment and output the corresponding customs clearance compliance risk warning level;

[0192] Furthermore: the offset breadth is determined based on the number of non-zero offset sub-components in the structural offset feature set, and the offset depth is determined based on the value of each offset sub-component;

[0193] The offset breadth and offset depth are compared with a preset risk level threshold to obtain the first risk level;

[0194] The first risk level is adjusted based on the credibility adjustment result: if the credibility adjustment result is high credibility, the first risk level is maintained; if the credibility adjustment result is low credibility, the first risk level is increased by one level.

[0195] The revised risk level is output as the early warning level for customs clearance compliance risks.

[0196] In step eight: Offset breadth: refers to the number of non-zero offset sub-components in the structural offset feature set, used to quantify the number of attribute dimensions that the current product deviates from the historical stable pattern; for example, if the current product has offsets in the two dimensions of declared unit price and country of origin, then the offset breadth is 2;

[0197] Offset depth: refers to the degree of deviation calculated based on the values ​​of each offset sub-component, such as the average, maximum, or weighted sum of all non-zero offset sub-components; used to quantify the severity of deviation of the current product in a specific attribute;

[0198] Preset risk level threshold: refers to the pre-defined numerical range or judgment rule used to map the offset breadth and offset depth to the risk level; for example, three levels can be set: low risk, medium risk, and high risk, corresponding to different combinations of breadth and depth.

[0199] First risk level: refers to the risk level initially determined based on the breadth and depth of the offset, which has not yet been revised for credibility.

[0200] Upgrade by one level: This means raising the risk level to a higher level, such as from medium risk to high risk, or from low risk to medium risk; if it is already the highest risk level, it will remain unchanged.

[0201] Customs clearance compliance risk warning level: refers to the final output level used to characterize the current level of customs clearance compliance risk of the goods, such as low risk, medium risk, and high risk.

[0202] For example: First, the early warning platform counts the number of non-zero offset sub-components from the structural offset feature set, using this as the offset breadth. For instance, if the structural offset feature set contains five attribute dimensions, and the offset sub-components for the declared unit price and country of origin are non-zero, while the rest are zero, then the offset breadth is 2. Simultaneously, the early warning platform calculates the offset depth based on the values ​​of each offset sub-component: a simple method is to take the average of all non-zero offset sub-components, or the maximum value among them. For example, if the declared unit price exceeds the public range by $50, and the country of origin offset sub-component is 1 (category exceeded), then the maximum offset depth is taken as 50, or a weighted sum (e.g., 50 × weight + 1 × weight). This embodiment uses the maximum value as the offset depth.

[0203] Secondly, the early warning platform reads preset risk level thresholds. For example, the preset rules are: if the offset breadth = 0, it is low risk; if the offset breadth ≥ 1 and the offset depth < 10, it is low risk; if the offset breadth ≥ 1 and 10 ≤ offset depth < 100, it is medium risk; if the offset breadth ≥ 1 and the offset depth ≥ 100, it is high risk. Based on the current offset breadth of 2 and offset depth of 50, the first risk level is determined to be medium risk.

[0204] Then, the early warning platform obtains the credibility correction result. If the credibility correction result is high credibility (indicating that the historical stable structure evolution is continuous and the judgment result is highly reliable), the first risk level remains unchanged and is still medium risk; if the credibility correction result is low credibility (indicating that the historical pattern has changed abruptly and the current judgment may be biased), the first risk level is upgraded by one level, that is, from medium risk to high risk (if the first risk level is already the highest, it remains unchanged).

[0205] Finally, the early warning platform outputs the revised risk level as the early warning level for customs clearance compliance risks. For example, it outputs "high risk" and can include an explanation: the offset dimensions are the declared unit price and country of origin, and the credibility is revised to low credibility (historical patterns have recently fluctuated).

[0206] Among them: non-zero offset sub-components: refer to the components in the structural offset feature set whose values ​​are not zero. Each non-zero component indicates that there is a deviation in the corresponding attribute dimension;

[0207] Offset depth calculation method: In this embodiment, the maximum value of all non-zero offset sub-components is used as the offset depth. Other methods such as weighted sum or square root can also be used according to business needs.

[0208] Preset risk level thresholds: These can be set based on historical violation case statistics or expert experience, with different threshold ranges corresponding to different risk levels;

[0209] Upgrade by one level: Risk levels are usually arranged as low, medium and high. Upgrading by one level means moving to a higher risk level; low → medium, medium → high, and if it is already high, it remains unchanged.

[0210] It should be noted that by using both the breadth and depth of deviation as dimensions for judgment, the extent and severity of attribute deviations can be distinguished, avoiding misjudgments caused by a single indicator (for example, the risk nature of a large deviation in a single dimension is different from that of a small deviation in multiple dimensions). Simultaneously, the first risk level is dynamically adjusted based on the credibility correction results, so that the system automatically increases its warning sensitivity when historical evolution patterns become unstable, thereby enhancing the robustness and adaptability of risk warnings. The final output of the customs clearance compliance risk warning level provides a quantitative basis for subsequent actions (such as release, inspection, and key control).

[0211] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for customs compliance risk early warning based on big data analysis, characterized in that, The method includes: Obtain the multi-dimensional attribute data set of the target commodity during the customs declaration process, and combine and associate the attribute data according to the preset attribute association constraint rules to form a commodity attribute structure set; Obtain a set of historical product attribute structures, analyze the set of historical product attribute structures to form a product attribute structure space, identify the distribution and clustering of the set of historical product attribute structures in the product attribute structure space, and construct a stable set of product attribute structures based on the identified clustering regions. The current product attribute structure set is mapped to the product attribute structure space to obtain the position distribution state of the current product attribute structure set in the stable product attribute structure set. Based on the position distribution state, the membership relationship between the current product attribute structure set and the stable product attribute structure set is determined to obtain the determination result. If the determination result is "not belonging", then the current product attribute structure set and the stable structure set of each product attribute arranged in chronological order that is closest to the current time are calculated to obtain the degree of structural boundary offset, and the degree of structural boundary offset is subjected to perturbation feature decomposition to obtain the structural offset feature set. Based on historical time series, the structural evolution of the stable structure set of commodity attributes is tracked, a structural evolution path is generated, and the continuity of structural changes of the stable structure set of commodity attributes in the time series is judged based on the consistency of the evolution path between adjacent time nodes, and a credibility correction result is generated. Based on the credibility correction results of the product attribute structure offset feature set and the product attribute stable structure set, a progressive risk assessment is performed, and the corresponding customs clearance compliance risk warning level is output.

2. The customs compliance risk early warning method based on big data analysis according to claim 1, characterized in that, The process of obtaining a set of historical product attribute structures, analyzing the set of historical product attribute structures to form a product attribute structure space includes: Each set of product attribute structures in the historical compliant customs clearance sample data is taken as a sample point to obtain a set of sample points; Calculate the difference value of attribute data between any two sample points in the sample point set in each attribute dimension, and sum the absolute values ​​of the difference values ​​in all attribute dimensions to obtain the structural distance between the two sample points. Then, obtain the structural distance between all pairs of sample points in the sample point set to form a structural distance set. Based on the structural distance set, the relative position coordinates of each sample point in the sample point set in a preset dimension space are determined by a multidimensional scaling analysis algorithm, and the corresponding relationship of sample point coordinates is obtained. Based on the coordinate correspondence of the sample points, interpolation is performed on the sparse regions of the coordinate results obtained from the multidimensional scaling analysis in the space of a preset number of dimensions to obtain the coordinate mapping function. The product attribute structure space is formed by using the set of all coordinate points covered by the range of the coordinate mapping function in the preset dimension space.

3. The customs compliance risk early warning method based on big data analysis according to claim 2, characterized in that, The step of identifying the distribution and clustering of the historical product attribute structure set in the product attribute structure space, and constructing a stable set of product attribute structures based on the identified clustering regions, includes: Based on the coordinate mapping function, each set of commodity attribute structures corresponding to the historical compliant customs clearance sample data is mapped to coordinate points in the commodity attribute structure space to obtain a set of spatial points; Calculate the spatial distance between any two spatial points in the set of spatial points, and divide spatial points with a spatial distance less than the preset distance threshold into the same candidate clustering region according to the preset distance threshold to obtain a set of candidate clustering regions; Count the number of spatial points contained in each candidate cluster region, and determine the candidate cluster regions whose number of spatial points exceeds a preset threshold as valid cluster regions; For each valid cluster region, extract the common attribute data range under each attribute dimension in the product attribute structure set corresponding to all spatial points contained in the valid cluster region, use the common attribute data range as the stable attribute constraint corresponding to the valid cluster region, and construct the product attribute stable structure set according to the stable attribute constraints corresponding to all valid cluster regions.

4. The customs compliance risk early warning method based on big data analysis according to claim 3, characterized in that, The step of mapping the current product attribute structure set to the product attribute structure space to obtain the positional distribution state of the current product attribute structure set in the stable product attribute structure set includes: Based on the coordinate mapping function, the set of product attribute structures of the current product to be evaluated is mapped to a current coordinate point in the product attribute structure space, thus obtaining the current coordinate point; Calculate the spatial distance between the current coordinate point and each effective clustering region in the stable structure set of product attributes. Determine whether the current coordinate point falls within the coverage area of ​​any effective clustering region based on whether the spatial distance is less than a preset membership distance threshold, and obtain the positional distribution state of the current product attribute structure set in the stable structure set of product attributes.

5. The customs compliance risk early warning method based on big data analysis according to claim 4, characterized in that, Based on the location distribution state, the membership relationship between the current product attribute structure set and the stable product attribute structure set is determined, and the determination result is obtained, including: If the location distribution status indicates that the current coordinate point falls within the coverage area of ​​any effective clustering area, then it is determined that the current product attribute structure set belongs to the product attribute stable structure set, and the determination result is "belongs to". Otherwise, it is determined that the current set of product attribute structures does not belong to the set of stable product attribute structures, and the determination result is "does not belong".

6. The method for early warning of customs clearance compliance risks based on big data analysis according to claim 5, characterized in that, If the determination result is "not belonging", then the current product attribute structure set is compared with the stable structure set of each product attribute arranged in chronological order, and the one most recent to the current time is calculated to obtain the degree of structure boundary offset, including: Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series to obtain the historical stable structure sequence; Select the set of stable commodity attribute structures corresponding to the time node closest to the current time from the historical stable structure sequence to obtain the set of most recent stable structures; Obtain the current coordinate point of the current product attribute structure set in the product attribute structure space, and calculate the shortest spatial distance from the current coordinate point to all spatial points contained in each effective cluster region of the nearest stable structure set to obtain the degree of structural boundary offset.

7. A method for early warning of customs clearance compliance risks based on big data analysis according to claim 6, characterized in that, The perturbation feature decomposition of the structural boundary offset degree to obtain a structural offset feature set includes: From the set of most recently stable structures, extract the common attribute data range for each attribute dimension to obtain the common attribute data range corresponding to each attribute dimension; From the current product attribute structure set, read the corresponding attribute data according to each attribute dimension to obtain the attribute data for each attribute dimension, which is used as the current attribute data set; For each attribute dimension, the current attribute data is compared with the corresponding common attribute data range. If the current attribute data exceeds the common attribute data range, the excess amount is calculated as the offset sub-component of the current attribute dimension. If the current attribute data does not exceed the range of the common attribute data, then the offset sub-component of the current attribute dimension is recorded as zero, and the offset sub-component corresponding to each attribute dimension is obtained. Combine the offset sub-components corresponding to all attribute dimensions to obtain the set of structural offset features.

8. The method for early warning of customs clearance compliance risks based on big data analysis according to claim 7, characterized in that, The process involves tracing the structural evolution of the stable product attribute structure set based on historical time series data, generating a structural evolution path, and determining the continuity of structural changes in the stable product attribute structure set over time series based on the consistency of evolution paths between adjacent time nodes, thereby generating a credibility correction result, including: Obtain the set of stable structures of product attributes corresponding to each time node in the historical time series to obtain the historical stable structure sequence; For each pair of adjacent time nodes in the historical stable structure sequence, the minimum spatial distance between the two commodity attribute stable structure sets is calculated. The minimum spatial distance is used as the evolution step size between adjacent time nodes, and the evolution steps are combined in chronological order to generate the structure evolution path. Calculate the change between adjacent evolution steps in the structural evolution path, and determine the local consistency of the structural evolution path based on whether the change is less than a preset consistency threshold, and obtain the local consistency determination results. The proportion of all local consistency judgment results that meet the consistency condition is counted. If the proportion is greater than the preset continuity threshold, the structural change is determined to be continuous in the time series, and a high-confidence correction result is generated. Otherwise, it is determined to be discontinuous, and a low-confidence correction result is generated.

9. A method for early warning of customs clearance compliance risks based on big data analysis according to claim 8, characterized in that, The progressive risk assessment based on the credibility correction results of the commodity attribute structure offset feature set and the commodity attribute stable structure set outputs the corresponding customs clearance compliance risk warning level, including: The offset breadth is determined based on the number of non-zero offset sub-components in the structural offset feature set, and the offset depth is determined based on the value of each offset sub-component. The offset breadth and offset depth are compared with a preset risk level threshold to obtain the first risk level; The first risk level is adjusted based on the credibility adjustment result: if the credibility adjustment result is high credibility, the first risk level is maintained; if the credibility adjustment result is low credibility, the first risk level is increased by one level. The revised risk level is output as the early warning level for customs clearance compliance risks.

Images