Method and system for automatic checking of ship entry and exit port declaration information based on artificial intelligence

Abstract

This application provides an automatic verification method and system for ship entry and exit declaration information based on artificial intelligence, belonging to the field of artificial intelligence technology. This application acquires the entry and exit declaration information of the target ship and a historical declaration verification dataset. Based on the historical data, it constructs a historical declaration sequence and a historical declaration deviation sequence. The historical declaration sequence is input into a trained prediction model to obtain a predicted declaration habit vector. A causal discovery algorithm is used to analyze the historical declaration deviation sequence, constructing a declaration deviation causal graph where nodes represent declaration item deviations and edges represent causal relationships. The difference between the current declaration and the predicted habit is calculated to obtain the declaration deviation vector, which is then input into the causal graph as observation evidence. Target nodes are determined through causal inference. A deviation threshold is calculated based on the historical deviation sequence. The current deviation value of the target node is compared with this threshold to determine the corresponding declaration information as abnormal declaration information, effectively identifying systematic and correlated declaration deviations in entry and exit declaration information.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to an automatic verification method and system for ship entry and exit declaration information based on artificial intelligence. Background Technology

[0002] The AI-based automatic verification method for ship entry and exit declarations is a key technology that utilizes machine learning, data mining, and other techniques to intelligently analyze massive amounts of declaration data, thereby improving maritime regulatory efficiency and water traffic safety. With the continuous development of smart ports and smart maritime systems, this method shows broad application prospects in achieving precise control over ship dynamics and combating illegal declarations.

[0003] In existing technologies, some AI-based automatic verification of ship entry and exit declaration information mainly relies on establishing statistical models of ship declaration data or employing unsupervised anomaly detection algorithms. For example, this involves comparing the current declaration data with the ship's historical declaration data distribution, or treating a single declaration as a whole vector to identify isolated points deviating from the normal data cluster, in order to determine whether anomalies exist.

[0004] However, existing technologies for automatically verifying ship entry and exit declaration information primarily focus on isolated anomalies in individual declarations. They lack the ability to effectively identify patterns where ship operators systematically conceal violations through subtle, coordinated deviations between multiple ship entry and exit declaration data over a long period. For example, when a ship consistently underreports a small amount of cargo while simultaneously reducing its declared draft, existing methods struggle to detect this logically interconnected combination of declaration deviations because all values ​​in a single declaration are within reasonable ranges. Therefore, existing technologies suffer from the technical problem of failing to effectively identify systematically correlated declaration deviations in port entry and exit declaration information. Summary of the Invention

[0005] The purpose of this application is to provide an artificial intelligence-based method, system, electronic device, and storage medium for automatically verifying ship entry and exit declaration information, in order to solve the problem that existing technologies cannot effectively identify systematic correlational declaration deviations in entry and exit declaration information.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides an artificial intelligence-based method for automatically verifying ship entry and exit port declaration information, comprising: Obtain the port entry and exit declaration information of the target vessel and a historical declaration and verification dataset including multiple historical declaration records and corresponding historical verification records; The declaration information of multiple target declaration items is extracted from the port entry and exit declaration information to construct the declaration data vector. Based on the multiple target declaration items in the historical declaration verification dataset, multiple data tuples including timestamps, historical declaration data vectors, and historical verification data vectors are determined. Based on the historical declaration data vector in each data tuple, a historical declaration sequence is constructed in time sequence, and the historical declaration sequence is input into the trained multivariate time series prediction model to obtain the predicted declaration habit vector. Based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation sequence is constructed in time sequence. Then, the conditional independence test and directed edge construction are performed on the multivariate deviation values ​​in the historical declaration deviation sequence through the causal discovery algorithm to determine the causal relationship between different declaration item deviations. A declaration deviation causal graph is constructed, where nodes represent declaration item deviations and edges represent the causal relationship between declaration item deviations. The declaration deviation vector is calculated based on the difference between the declared data vector and the predicted declaration habit vector. The declaration deviation vector is then used as observation evidence and input into the declaration deviation causal graph. The target node is determined in the declaration deviation causal graph through a causal inference algorithm. Based on the historical deviation values ​​associated with the declaration items corresponding to the target node in the historical declaration deviation sequence, the deviation threshold is calculated, and the target deviation value corresponding to the target node is determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

[0007] In one feasible implementation, the method further includes: The structural risk score is obtained by calculating the graph structure entropy of the cause-and-effect graph of the declaration deviation; Based on the structural risk score, the target threshold adjustment factor is determined in the pre-defined correspondence between the structural risk score and the threshold adjustment factor. Based on the historical deviation values ​​associated with the declaration items corresponding to the target node in the historical declaration deviation sequence, a deviation threshold is calculated. The target deviation value corresponding to the target node is then determined from the declaration deviation vector. If the target deviation value exceeds the deviation threshold, the declaration information corresponding to the target node is identified as abnormal declaration information, including: Based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor, the deviation threshold is calculated, and the target deviation value corresponding to the target node is determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

[0008] In one feasible implementation, a deviation threshold is calculated based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor. The target deviation value corresponding to the target node is then determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information, including: From the historical declaration deviation sequence, extract all historical declaration deviations associated with the declaration item corresponding to the target node to form a set of historical deviation values; Sort the absolute values ​​of all historical reported deviations in the historical deviation value set, and select the values ​​that are located at the preset quantiles after sorting as the first deviation threshold. The first deviation threshold is multiplied by the target threshold adjustment factor to generate the deviation threshold; The target deviation value corresponding to the target node is determined from the declaration deviation vector. If the absolute value of the target deviation value is greater than the deviation threshold, the declaration information associated with the declaration item deviation corresponding to the target node is determined to be abnormal declaration information.

[0009] In one feasible implementation, declaration information for multiple target declaration items is extracted from the port entry and exit declaration information to construct a declaration data vector. Based on these multiple target declaration items in the historical declaration verification dataset, multiple data tuples are determined, including timestamps, historical declaration data vectors, and historical verification data vectors, comprising: Extract multiple declaration information corresponding to multiple target declaration items from the port entry and exit declaration information, and construct a declaration data vector. The multiple target declaration items include multiple declaration items related to the physical loading status of the ship. From each historical declaration record and corresponding historical verification record in the historical declaration verification dataset, the declaration information corresponding to multiple target declaration items is extracted to form a historical declaration data vector and a historical verification data vector. The historical declaration data vector, the historical verification data vector, and the timestamp corresponding to the historical declaration record are combined according to the preset sorting order to obtain a data tuple corresponding to each historical declaration record in the historical declaration verification dataset.

[0010] In one feasible implementation, based on the historical declaration data vector in each data tuple, a historical declaration sequence is constructed in time sequence, and the historical declaration sequence is input into a trained multivariate time series prediction model to obtain a predicted declaration habit vector, including: Based on the historical declaration data vectors in all data tuples, arrange them in chronological order according to timestamps to construct a historical declaration sequence; A continuous sequence of a preset length is extracted from the end of the historical declaration sequence and used as the input sequence for the multivariate time series prediction model; The input sequence is fed into a trained multivariate time series prediction model to generate a predicted declaration habit vector. The multivariate time series prediction model is trained using a deep learning model based on a self-attention mechanism, based on training samples including historical declaration training sequences and corresponding label vectors.

[0011] In one feasible implementation, based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation sequence is constructed in chronological order. Then, a causal discovery algorithm is used to perform conditional independence tests and construct directed edges on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationships between different declaration item deviations, and a declaration deviation causal graph is constructed, including: By performing element-wise subtraction of the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation vector is obtained. The historical declaration deviation vectors corresponding to each data tuple are arranged in chronological order according to the timestamps to obtain a historical declaration deviation sequence. Each column of data in the historical declaration deviation sequence corresponds to a declaration item deviation. Construct an initial graph structure in which there exists an undirected edge between any two distinct deviations of the declaration items; Traverse the undirected edges in the initial graph structure, and use at least one declaration item deviation in the initial graph structure that is not connected to the current undirected edge as the condition set. Calculate the joint probability distribution of the two declaration item deviations connected by the current undirected edge under the given condition set, and compare the joint probability distribution with the product of the conditional probability distributions of the two declaration item deviations to obtain the independence statistic. Remove the undirected edge when the independence statistic meets the preset independence judgment condition to generate a skeleton graph. Based on the undirected edges in the skeleton graph, the direction of the undirected edges is determined by identifying and orienting the collision structure formed by the deviations of the three declaration items in the skeleton graph and the propagation rules to avoid the generation of directed loops, thereby constructing a causal graph of declaration deviations.

[0012] In one feasible implementation, a declaration deviation vector is calculated based on the difference between the declared data vector and the predicted declaration habit vector. This vector is then used as observational evidence and input into a declaration deviation causal graph. A causal inference algorithm is then used to determine the target node within the declaration deviation causal graph, including: A vector operation is performed to subtract the declared data vector from the predicted declaration habit vector element by element to generate a declaration deviation vector; Deviation values ​​whose absolute value in the reported deviation vector is greater than the preset anomaly judgment threshold are identified as abnormal deviation values. In the cause-effect graph of the reported deviation, all ancestor nodes that can reach the node corresponding to the abnormal deviation value through directed edge paths are identified, forming a set of candidate cause nodes; The contribution of each candidate cause node is obtained by calculating the number of times each candidate cause node in the candidate cause node set is the ancestor node of the node corresponding to the abnormal deviation value in the cause-effect graph of the reported deviation. The candidate cause node with the highest contribution is identified as the target node.

[0013] Secondly, this application provides an artificial intelligence-based automatic verification system for ship entry and exit declaration information, including: The acquisition module is used to acquire the port entry and exit declaration information of the target vessel and the historical declaration and verification dataset, which includes multiple historical declaration records and corresponding historical verification records. The determination module is used to extract the declaration information of multiple target declaration items from the port entry and exit declaration information to construct the declaration data vector, and to determine multiple data tuples including timestamps, historical declaration data vectors and historical verification data vectors based on the multiple target declaration items in the historical declaration verification dataset; The prediction module is used to construct a historical declaration sequence in time order based on the historical declaration data vector in each data tuple, and input the historical declaration sequence into the trained multivariate time series prediction model to obtain the predicted declaration habit vector. The module is used to construct a historical declaration deviation sequence in chronological order based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple. It then uses a causal discovery algorithm to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationship between different declaration item deviations and construct a declaration deviation causal graph, where nodes represent declaration item deviations and edges represent the causal relationship between declaration item deviations. The determination module is also used to calculate the declaration deviation vector based on the difference between the declaration data vector and the predicted declaration habit vector, and input the declaration deviation vector as observation evidence into the declaration deviation causal graph, and determine the target node in the declaration deviation causal graph through the causal reasoning algorithm; The determination module is also used to calculate the deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, and to determine the target deviation value corresponding to the target node from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute computer programs to implement the steps of the artificial intelligence-based automatic verification method for ship entry and exit declaration information as described in the first aspect above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the artificial intelligence-based automatic verification method for ship entry and exit declaration information as described in the first aspect above.

[0016] The AI-based automatic verification method for ship entry and exit declaration information provided in this application constructs and utilizes historical declaration deviation sequences to discover the causal relationships between deviations in different declaration items, thereby establishing a deviation causal graph for each ship that characterizes its systematic declaration deviation pattern. When verifying current declaration information, the method first identifies abnormal behaviors that deviate from its historical declaration habits using a predictive model, and then uses the deviation causal graph to attribute these abnormal behaviors causally, tracing and locating the root cause nodes leading to the declaration anomalies. This method can penetrate the superficial compliance facade of single declaration data and identify combined declaration deviation patterns with inherent logical connections, consisting of minor coordinated deviations between multiple declaration items, thus effectively solving the technical problem that existing technologies cannot identify systematic and correlated declaration deviations of ships.

[0017] Furthermore, by calculating the graph structure entropy of the causal graph of declaration deviations, a structural risk score is obtained to quantify the complexity of systematic declaration deviation patterns for vessels. This structural risk score is used to dynamically adjust the deviation threshold used to judge anomalies; as the deviation pattern becomes more complex and difficult to predict, a stricter judgment standard is automatically applied. This allows the application not only to identify systematic declaration deviations but also to adaptively adjust the sensitivity of supervision based on the predictability risk of the deviation pattern itself. This enables earlier and more accurate risk warnings when facing vessels with chaotic declaration behavior and varied fraudulent methods, further improving the accuracy of identifying systemically related declaration deviations. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an artificial intelligence-based automatic verification method for ship entry and exit declaration information provided in this application embodiment; Figure 2 A flowchart illustrating a method for determining abnormal declaration information provided in an embodiment of this application; Figure 3 A schematic diagram of a cause-effect graph of a declaration deviation provided in an embodiment of this application; Figure 4 A schematic diagram of the structure of an automatic verification system for ship entry and exit declaration information based on artificial intelligence, provided for an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0020] Addressing the limitation of existing technologies, which can only identify isolated declaration anomalies and cannot uncover long-term, systematically correlated deviations, this application proposes a design that deconstructs ship declaration behavior into two dimensions: declaration habits and declaration deviations. A predictive model learns the ship's declaration habits to identify current behavioral anomalies, and causal discovery techniques are then used to construct an internal logical map of historical deviations. Finally, by tracing and locating the root causes of the identified anomalies, accurate verification of systematically correlated deviations is achieved.

[0021] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To address the problems of existing technologies, embodiments of this application provide a method, apparatus, equipment, computer storage medium, and computer program product for automatically verifying ship entry and exit declaration information based on artificial intelligence. The method for automatically verifying ship entry and exit declaration information based on artificial intelligence provided in this application embodiment will be described first below.

[0023] Figure 1 This illustration shows a flowchart of an artificial intelligence-based automatic verification method for ship entry and exit port declaration information according to an embodiment of this application. Figure 1 As shown, the method includes steps S110 to S150.

[0024] S110: Obtain the port entry and exit declaration information of the target vessel and the historical declaration and verification dataset, which includes multiple historical declaration records and corresponding historical verification records.

[0025] Port entry / exit declaration information refers to the electronic data submitted by the target vessel for its current voyage, related to its current voyage status. This can include numerical declarations related to the vessel's physical loading status, such as draft, cargo tonnage, ballast water volume, and fuel oil inventory, as well as declarations related to vessel operation and safety, such as the vessel's unique identifier, number of crew members, estimated arrival time, navigation status, and cargo type. The historical declaration verification dataset is a collection of data established for the target vessel, consisting of multiple historical declaration records and corresponding historical verification records. Historical declaration records refer to the declaration data submitted by the target vessel during a previously completed voyage, while historical verification records refer to the actual values ​​corresponding to the various values ​​in the historical declaration records, obtained by regulators through methods such as draft surveys, sensor measurements, or on-site inspections.

[0026] The port entry and exit declaration information of the target vessel can be obtained in real time through the relevant regulatory system. Then, using the unique vessel identification code extracted from the port entry and exit declaration information, a data retrieval request is initiated to the historical database to retrieve all historical data associated with this target vessel. The database has paired and stored the port entry and exit declaration data of each vessel for each historical voyage with the confirmed port entry and exit verification data. All paired and stored historical declaration records and historical verification records associated with this target vessel are retrieved, and the collection of these paired records is used as the historical declaration and verification dataset.

[0027] For example, taking the entry of vessel A into port as an example, the port entry and exit declaration information obtained is a set of electronic records, such as (vessel draft: 8.52 meters, cargo tonnage: 25,130 tons, ballast water: 4,980 tons, fuel oil: 855.5 tons, crew: 22, navigation status: anchored, cargo type: iron ore). Meanwhile, the obtained historical declaration verification dataset contains paired data from multiple past voyages of the vessel. For example, data from one historical voyage shows that its historical declaration record is (vessel draft: 8.25 meters, cargo tonnage: 24,300 tons, ballast water: 6,150 tons, fuel oil: 910.2 tons, crew: 22, navigation status: underway, cargo type: coal), while the corresponding historical verification record is (vessel draft: 8.31 meters, cargo tonnage: 24,750 tons, ballast water: 6,150 tons, fuel oil: 912.0 tons, crew: 22). The first voyage's declaration record shows the following: (ship's draft: 8.61 meters, cargo tonnage: 25,600 tons, ballast water: 4,450 tons, fuel oil: 890.0 tons, crew: 21 people, voyage status: berthed, cargo type: iron ore). The corresponding historical verification record shows the same information: (ship's draft: 8.61 meters, cargo tonnage: 25,600 tons, ballast water: 4,450 tons, fuel oil: 935.8 tons, crew: 22 people, voyage status: berthed, cargo type: iron ore).

[0028] S120: Extract the declaration information of multiple target declaration items from the port entry and exit declaration information to construct a declaration data vector, and determine multiple data tuples including timestamps, historical declaration data vectors and historical verification data vectors based on the multiple target declaration items in the historical declaration verification dataset.

[0029] A target declaration item refers to a pre-defined set of multiple port entry and exit declaration items selected from the vessel's port entry and exit declaration information. These can include vessel draft, cargo tonnage, ballast water volume, fuel oil inventory, and crew number. A declaration data vector is a numerical vector formed by arranging the values ​​corresponding to the target declaration item in the target vessel's current port entry and exit declaration information in a pre-defined order. A data tuple is a data unit obtained by structuring the data from each historical voyage in the historical declaration and verification dataset according to a unified format. Each data tuple contains a timestamp, a historical declaration data vector, and a historical verification data vector. The historical declaration data vector and historical verification data vector are numerical vectors formed by extracting the values ​​corresponding to the target declaration item from historical declaration records and historical verification records, respectively, and arranging them in the same order as the declaration data vector.

[0030] First, a target set of declaration items is pre-determined, covering key declaration items that may have systematic correlation biases. Then, based on this target set of declaration items, corresponding declaration values ​​are extracted from the arrival and departure declaration information obtained in step S110 and arranged in a preset order to construct the declaration data vector required for this verification. Simultaneously, each paired record in the historical declaration verification dataset obtained in step S110 is traversed. Again based on this target set of declaration items, corresponding values ​​are extracted from historical declaration records and historical verification records to construct historical declaration data vectors and historical verification data vectors. The completion time associated with the historical voyage is used as a timestamp and combined with these two vectors, thereby transforming the unstructured historical data into a set composed of multiple structured data tuples.

[0031] S130: Based on the historical declaration data vector in each data tuple, construct a historical declaration sequence in time sequence, and input the historical declaration sequence into the trained multivariate time series prediction model to obtain the predicted declaration habit vector.

[0032] A historical declaration sequence refers to a multivariate time series data set formed by arranging the historical declaration data vectors within multiple data tuples according to their timestamp order. A multivariate time series prediction model is a pre-trained deep learning model that can be trained using multiple training samples, including an input sequence extracted from the historical declaration sequence and the next historical declaration data vector that is temporally adjacent to the input sequence, as corresponding labels. A predicted declaration habit vector is a prediction vector output by the model after the historical declaration sequence of the target vessel is input into the trained multivariate time series prediction model. This vector represents the most likely combination of declaration data to occur in this voyage based on the vessel's declaration behavior pattern.

[0033] First, based on the timestamps in each data tuple, all historical declaration data vectors are sorted in ascending order to construct a historical declaration sequence reflecting the temporal evolution of the target vessel's declaration behavior. Next, a continuous sequence of a predetermined length is extracted from the end of this historical declaration sequence; for example, the N most recent historical declaration data vectors are extracted to form an input sequence. This input sequence is fed into a trained multivariate time series prediction model based on the Transformer architecture. This model uses its internal self-attention mechanism to process the input sequence to capture the nonlinear correlations and long-term dependencies between various declaration values. After receiving this input sequence, the model performs a forward computation and finally outputs a numerical vector with the same dimension as the declaration data vectors. This vector is the predicted declaration habit vector. For example, if a historical declaration data vector is represented as Vt=[At, Bt, Ct, ...], then the input sequence is [Vt-N+1, ..., Vt-1, Vt], and the predicted declaration habit vector output by the model is V't+1=[A't+1, B't+1, C't+1, ...].

[0034] S140: Based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, construct a historical declaration deviation sequence in chronological order, and use a causal discovery algorithm to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationship between different declaration item deviations, and construct a declaration deviation causal graph, where nodes represent declaration item deviations and edges represent the causal relationship between declaration item deviations.

[0035] A historical declaration deviation sequence refers to a multivariate time series data formed by arranging multiple data tuples according to their timestamps, and then considering the differences between the historical declaration data vectors and the historical verification data vectors. Each column of this sequence represents the historical deviation of a target declaration item. Causal discovery algorithms are a class of algorithms that can extract causal relationships, rather than just correlations, between variables from observed data, such as constraint-based probabilistic computation (PC) algorithms. A declaration deviation causal graph is a directed acyclic graph constructed for a specific target vessel. Nodes in this graph represent deviations of a declaration item, and directed edges indicate a direct causal impact of one declaration item's deviation on another. The entire graph structurally represents the long-term, systematic, and correlated declaration deviation patterns of the vessel.

[0036] First, a deviation vector is obtained by subtracting the historical verification data vector from the historical declaration data vector within each generated data tuple element-by-element. This yields a deviation vector quantifying the various declaration deviations for that voyage. For example, if the declared cargo tonnage for a voyage is 24,300 tons and the verified actual tonnage is 24,750 tons, the corresponding element in the vector representing the cargo tonnage deviation is +450 tons. Next, based on the timestamp of each data tuple, all calculated deviation vectors are sorted in ascending order to construct a historical declaration deviation sequence. Finally, this historical declaration deviation sequence is input into a constraint-based causal discovery algorithm. This algorithm first constructs an initial fully connected graph where any two declaration deviations are connected. Then, it gradually removes connections without direct causal relationships by performing a series of conditional independence tests, forming a skeleton graph. Finally, it determines the causal direction of the remaining connections in the skeleton graph by identifying collision structures and other directional judgment rules, ultimately outputting a customized declaration deviation causal graph for the target vessel.

[0037] S150: Based on the difference between the declared data vector and the predicted declared habit vector, the declared deviation vector is calculated, and the declared deviation vector is used as observation evidence to input into the declared deviation causal graph. The target node is determined in the declared deviation causal graph through the causal reasoning algorithm.

[0038] The declaration deviation vector is the difference vector obtained by subtracting the constructed declaration data vector from the predicted declaration habit vector element by element. Each element of this vector quantifies the degree to which the current declaration behavior deviates from its historical habits on the corresponding declaration item. Causal inference algorithms are a class of algorithms that can infer the state of other variables or find the cause based on the observed states of some variables in a known causal structure graph. The target node is the node in the declaration deviation causal graph that, as determined by the causal inference algorithm, best explains the root cause of the abnormal components in the currently observed declaration deviation vector.

[0039] First, a declaration deviation vector is generated by subtracting the declared data vector of this voyage from the predicted declaration habit vector generated by the model element-wise. For example, if the cargo tonnage in the declared data vector is 25,130 tons, while the corresponding cargo tonnage in the predicted habit vector is 25,200 tons, then the element corresponding to the cargo tonnage deviation in the declaration deviation vector is -70 tons. This declaration deviation vector is then used as a set of observational evidence and input into a declaration deviation causal graph constructed for the ship. Next, a causal inference algorithm is executed. This algorithm first identifies deviations in the declaration deviation vector whose absolute value is greater than a preset anomaly threshold as one or more anomalous deviation values. Subsequently, for each anomalous deviation value, all ancestor nodes that can reach the corresponding node of the anomalous deviation value through directed paths are traced back in the causal graph, and all traced ancestor nodes form a candidate cause node set. Finally, the total number of times each node in the candidate cause node set appears as an ancestor node is calculated, and the candidate cause node with the highest total number of occurrences is determined as the final target node.

[0040] S160: Calculate the deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, and determine the target deviation value corresponding to the target node from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, determine the declaration information corresponding to the target node as abnormal declaration information.

[0041] The deviation threshold refers to the numerical limit used to determine whether the deviation of the current declaration behavior is abnormal. The target deviation value refers to the deviation component corresponding to the target node extracted from the declaration deviation vector. This value quantifies the degree to which the current declaration deviates from its own habits in terms of the root cause. Abnormal declaration information refers to the specific declaration items and their values ​​in the current port entry / exit declaration information corresponding to the target node that are ultimately determined to be abnormal by this method.

[0042] First, from the constructed historical declaration deviation sequence, all historical deviation values ​​associated with the declaration item corresponding to the target node are extracted, forming a historical deviation value set. Then, statistical analysis is performed on this set to calculate the deviation threshold; for example, the 95th percentile of the absolute values ​​of all deviation values ​​in the set can be used as the deviation threshold. Simultaneously, the target deviation value corresponding to the target node is extracted from the declaration deviation vector generated in step S150. Finally, the absolute value of the target deviation value is compared with the calculated deviation threshold. If the absolute value of the target deviation value is greater than the deviation threshold, the current declaration information corresponding to the target node is determined to be abnormal, and a corresponding verification conclusion is generated. This leads to further verification of the target vessel's port entry and exit information or control measures for the target vessel's port entry and exit.

[0043] In one feasible implementation, the method further includes: The structural risk score is obtained by calculating the graph structure entropy of the cause-and-effect graph of the reporting deviation.

[0044] Graph structure entropy is an indicator used to quantify the complexity and uncertainty of a network or graph structure. A higher value generally indicates a more complex and disordered graph structure. Structural risk score is used to quantify the degree of disorder and unpredictability of a particular vessel's historical reporting deviation patterns; a higher score indicates a higher risk of systematic deviation patterns for the vessel.

[0045] First, the graph entropy calculation method from information theory is applied to calculate the topological structure of the causal graph of the reported deviations constructed for the target vessel, in order to obtain an initial graph structure entropy value. This calculation process considers structural features such as the number of nodes, the number of edges, the degree distribution of nodes, and the connectivity of the graph. Finally, the calculated graph structure entropy value can be converted into a structural risk score between 0 and 100 points through a preset mapping function, such as a normalization function. For example, for vessel A, if the causal graph structure of the reported deviations constructed for it is relatively complex, the structural risk score obtained after calculating its graph structure entropy is 80 points, indicating that the vessel's systematic deviation pattern is relatively chaotic and unpredictable.

[0046] Based on the structural risk score, the target threshold adjustment factor is determined in the pre-defined correspondence between the structural risk score and the threshold adjustment factor.

[0047] The threshold adjustment factor is a multiplier factor used to dynamically adjust the leniency or strictness of the deviation threshold. The correspondence between structural risk scores and threshold adjustment factors refers to a pre-established table defining which threshold adjustment factor should be used for structural risk scores in different intervals. For example, the correspondence between structural risk scores and threshold adjustment factors is shown in Table 1.

[0048] Table 1: Comparison Table of Structural Risk Scores and Threshold Adjustment Factors

[0049] The relationship between structural risk scores and threshold adjustment factors is inversely proportional; that is, a higher structural risk score corresponds to a smaller threshold adjustment factor, reflecting stricter regulation of high-risk patterns. The calculated structural risk score is input into this relationship to determine the target threshold adjustment factor required for this verification. For vessel A in the aforementioned example, its structural risk score is 85.7 points, which, according to Table 1, falls within the range of (75.0, 100). Therefore, the determined target threshold adjustment factor is 0.82, indicating that a stricter judgment standard will be applied to this vessel.

[0050] Step S160 calculates the deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, and determines the target deviation value corresponding to the target node from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information, including: Based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor, the deviation threshold is calculated, and the target deviation value corresponding to the target node is determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

[0051] Figure 2 A flowchart illustrating a method for determining abnormal reporting information according to an embodiment of this application is shown.

[0052] In one feasible implementation, a deviation threshold is calculated based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor. The target deviation value corresponding to the target node is then determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information. Figure 2 As shown, the method includes: S210: Extract all historical declaration deviations associated with the declaration items corresponding to the target node from the historical declaration deviation sequence to form a set of historical deviation values.

[0053] The historical deviation value set refers to the set extracted from the historical declaration deviation sequence for a specific declaration item, containing the deviation values ​​for that item across all historical voyages of the vessel. The target node represents a specific declaration item. For example, if the root cause of the current declaration anomaly is determined to be cargo tonnage deviation, then the target node is cargo tonnage deviation. Then, in the historical declaration deviation sequence constructed in step S140, the specific data column corresponding to that declaration item is located. Finally, all values ​​in that column are extracted, and these values ​​form a one-dimensional historical deviation value set; for example, the extracted historical deviation value set might be [D1, D2, D3, ...].

[0054] S220: Sort the absolute values ​​of all historical reported deviations in the historical deviation value set, and select the values ​​that are located at the preset quantiles after sorting as the first deviation threshold.

[0055] A preset percentile is a percentage value between 0 and 100 used to select a value at a specific position in a set of sorted data. For example, the 95th percentile means that the value is greater than or equal to 95% of the values ​​in the dataset.

[0056] First, the absolute value of each deviation value in the historical deviation value set is calculated, ignoring the directionality of the deviation and focusing only on its magnitude. Next, all calculated absolute values ​​are sorted in ascending order. Finally, based on a preset quantile, the value at the corresponding position in the sorted absolute value sequence is found and determined as the first deviation threshold. For example, for the historical deviation value set of cargo tonnage deviations, after sorting the absolute values, the value at the 95th percentile is selected. Assuming this value is calculated to be 450 tons, the first deviation threshold is determined to be 450 tons.

[0057] S230: Multiply the first deviation threshold with the target threshold adjustment factor to generate the deviation threshold.

[0058] The first deviation threshold and the target threshold adjustment factor are combined using the formula shown in Equation (1) to generate the structural risk-adjusted deviation threshold. For example, the deviation threshold can be calculated using the following formula: (1) in, It is the final generated deviation threshold. It is the first deviation threshold. This is the target threshold adjustment factor. If the first deviation threshold is Te and the target threshold adjustment factor is 0.82, then the generated deviation threshold is... .

[0059] S240: Determine the target deviation value corresponding to the target node from the declaration deviation vector, and if the absolute value of the target deviation value is greater than the deviation threshold, determine that the declaration information associated with the declaration item deviation corresponding to the target node is abnormal declaration information.

[0060] First, the deviation component corresponding to the target node is extracted from the declared deviation vector; this component is the target deviation value. For example, if the target node is cargo tonnage deviation, the target deviation value extracted from the current declared deviation vector is -480 tons. Then, the absolute value of this target deviation value is compared with the deviation threshold generated in the previous step. Finally, if the absolute value of the target deviation value is greater than the deviation threshold, the current declaration information corresponding to the target node is determined to be abnormal, and a corresponding verification conclusion is generated. In this example, because the absolute value of the target deviation value |-480| is greater than the calculated deviation threshold... Ultimately, the declared cargo tonnage was determined to be abnormal.

[0061] After determining that the declaration information corresponding to the target node is abnormal, a verification result is generated, including specific abnormal items, deviation values, and risk levels. This verification result can serve as the basis for triggering subsequent regulatory actions. For example, it can automatically send warnings to on-site law enforcement personnel, suggesting that specific declaration items of the vessel be subject to focused verification, or, when certain high-risk conditions are met, directly link with the port scheduling system to temporarily control the vessel's entry and exit plans.

[0062] This embodiment calculates the graph structure entropy of the causal graph of declaration deviations to obtain a structural risk score that quantifies the complexity of systematic declaration deviation patterns for ships. The structural risk score is used to dynamically adjust the deviation threshold used to judge anomalies; as the deviation pattern becomes more complex and difficult to predict, stricter judgment criteria are automatically applied. This allows the application not only to identify systematic declaration deviations but also to adaptively adjust the sensitivity of supervision based on the predictability risk of the deviation pattern itself. This enables earlier and more accurate risk warnings when facing ships with chaotic declaration behavior and varied fraudulent methods, further improving the accuracy of identifying systemically related declaration deviations.

[0063] In one feasible implementation, step S120 extracts declaration information for multiple target declaration items from the port entry and exit declaration information to construct a declaration data vector, and determines multiple data tuples, including timestamps, historical declaration data vectors, and historical verification data vectors, based on the multiple target declaration items in the historical declaration verification dataset. Extract multiple declaration information corresponding to multiple target declaration items from the port entry and exit declaration information, and construct a declaration data vector. The multiple target declaration items include multiple declaration items related to the physical loading status of the ship.

[0064] First, a target declaration item set is pre-determined, consisting of multiple declaration items. For example, the target declaration item set could be defined as (ship draft, cargo tonnage, ballast water volume, number of crew members). Then, based on this target declaration item set, the corresponding declaration values ​​for each declaration item in the acquired port entry and exit declaration information are extracted. Finally, these extracted declaration values ​​are combined into a numerical vector according to a pre-defined order in the target declaration item set. This vector is the declaration data vector required for this verification. For example, if the order of the target declaration item set is [ship draft, cargo tonnage, ballast water volume, number of crew members], and the corresponding values ​​in the current port entry and exit declaration information are 8.52 meters, 25130 tons, 4980 tons, and 22 people, then the constructed declaration data vector would be [8.52, 25130, 4980, 22].

[0065] From each historical declaration record and corresponding historical verification record in the historical declaration verification dataset, the declaration information corresponding to multiple target declaration items is extracted to form a historical declaration data vector and a historical verification data vector.

[0066] First, iterate through each paired record in the acquired historical declaration verification dataset. Then, for each paired record, based on the same target declaration item set determined in the previous step, extract the values ​​corresponding to each declaration item in the target declaration item set from both the historical declaration records and historical verification records within that paired record. Finally, arrange the values ​​extracted from the historical declaration records according to the preset order of the target declaration item set to form the historical declaration data vector for this voyage; simultaneously, arrange the values ​​extracted from the historical verification records in the same order to form the historical verification data vector for this voyage. For example, for a historical voyage, if its historical declaration record is (ship draft: 8.25 meters, cargo tonnage: 24,300 tons, ballast water volume: 6,150 tons, number of crew members: 22), then the historical declaration data vector generated for it is [8.25, 24,300, 6,150, 22]; if its corresponding historical verification record is (ship draft: 8.31 meters, cargo tonnage: 24,750 tons, ballast water volume: 6,150 tons, number of crew members: 22), then the historical verification data vector generated for it is [8.31, 24,750, 6,150, 22].

[0067] The historical declaration data vector, the historical verification data vector, and the timestamp corresponding to the historical declaration record are combined according to the preset sorting order to obtain a data tuple corresponding to each historical declaration record in the historical declaration verification dataset.

[0068] First, for each paired record in the historical declaration and verification dataset, the voyage completion time associated with that historical declaration record is extracted as a timestamp. Then, this timestamp is combined with the historical declaration data vector and historical verification data vector generated for that voyage in the previous step. Finally, these three data entities are encapsulated into a structured data unit, which is a data tuple, according to a preset order, such as [timestamp, historical declaration data vector, historical verification data vector]. For example, for the historical voyage in the aforementioned example, if its completion time is October 15, 2024, the final data tuple generated for that voyage would be ('2024-10-15', [8.25, 24300, 6150, 22], [8.31, 24750, 6150, 22]). By performing this operation on all historical records, the entire historical dataset is ultimately transformed into a collection of multiple such structured data tuples.

[0069] By traversing the historical declaration and verification dataset, a standardized data tuple containing a timestamp, historical declaration data, and historical verification data is generated for each historical voyage. For example, if the timestamp of a historical voyage is T1, its declaration data vector is Vd1, and its verification data vector is Vi1, then the generated data tuple is (T1, Vd1, Vi1); for another voyage with a timestamp of T2, (T2, Vd2, Vi2) is generated. Finally, a set consisting of multiple data tuples such as (T1, Vd1, Vi1), (T2, Vd2, Vi2), ... is obtained.

[0070] In one feasible implementation, step S130 constructs a historical declaration sequence according to the historical declaration data vector in each data tuple, and inputs the historical declaration sequence into a trained multivariate time series prediction model to obtain a predicted declaration habit vector, including: Based on the historical declaration data vectors in all data tuples, arrange them in chronological order according to timestamps to construct a historical declaration sequence.

[0071] First, all data tuples are sorted in ascending order using the timestamp in each tuple as the sorting key. Then, the historical declaration data vectors from each sorted data tuple are extracted sequentially and arranged in the same chronological order, thus constructing a multivariate time series data reflecting the temporal evolution of the target vessel's declaration activities. This sequence is the historical declaration sequence. For example, if the historical declaration verification dataset contains records of 20 historical voyages, this step will generate a historical declaration sequence [Vd1, Vd2, ..., Vd20] composed of 20 historical declaration data vectors, sorted by time.

[0072] A continuous sequence of a preset length is extracted from the end of the historical reporting sequence and used as the input sequence for the multivariate time series prediction model.

[0073] First, a preset input sequence length N is set. This parameter N represents the number of historical flights that the model needs to reference when making predictions. For example, N can be set to 5. Then, a continuous vector sequence of length N is extracted from the end of the constructed historical declaration sequence, that is, the most recent part in time. This extracted sequence is the input sequence required for this prediction. For example, for the aforementioned historical declaration sequence [Vd1, ..., Vd20] containing 20 vectors, if N is set to 5, the extracted input sequence will be [Vd16, Vd17, Vd18, Vd19, Vd20].

[0074] The input sequence is fed into a trained multivariate time series prediction model to generate a predicted declaration habit vector. The multivariate time series prediction model is trained using a deep learning model based on a self-attention mechanism, based on training samples including historical declaration training sequences and corresponding label vectors.

[0075] First, the generated input sequence is fed into a pre-trained multivariate time series prediction model, such as one based on the Transformer architecture. This model has learned the ship's declaration habits by learning from multiple training samples, where each training sample consists of a historical declaration training sequence and a corresponding label vector. After receiving the input sequence, the model uses its internal self-attention mechanism to process the input sequence to capture the nonlinear correlations and long-term dependencies between the declaration values, and performs a forward computation, finally outputting a numerical vector with the same dimension as the declaration data vector. This vector is the predicted declaration habit vector. For example, if the input sequence is [[8.25, 24300, 6150, 22], ..., [8.61, 25600, 4450, 21]], then the predicted declaration habit vector output by the model might be [8.55, 25200, 4850, 22].

[0076] As an embodiment of this application, the training process of training a multivariate time series prediction model can be as follows: First, a training sample set is obtained. The training sample set is generated by applying a sliding window method to the historical declaration sequence. Each training sample includes a historical declaration training sequence consisting of multiple consecutive historical declaration data vectors, and a next historical declaration data vector that is temporally adjacent to the historical declaration training sequence as the corresponding label vector.

[0077] Then, for each training sample in the training sample set, the following steps are performed: input the historical declaration training sequence into a preset deep learning model based on a self-attention mechanism to obtain a prediction vector; determine a loss function value based on the corresponding label vector and prediction vector; if the loss function value does not meet a preset training stopping condition, adjust the model parameters of the deep learning model based on the loss function value to obtain an updated deep learning model, and return to input the historical declaration training sequence into a preset deep learning model based on a self-attention mechanism to obtain a prediction vector, until the training stopping condition is met, thereby obtaining a trained multivariate time series prediction model.

[0078] In one feasible implementation, step S140 constructs a historical declaration deviation sequence in chronological order based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple. Then, using a causal discovery algorithm, it performs conditional independence tests and constructs directed edges on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationships between different declaration item deviations, and constructs a declaration deviation causal graph, including: By performing element-wise subtraction of the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation vector is obtained. The historical declaration deviation vectors corresponding to each data tuple are then arranged in chronological order according to the timestamps to obtain a historical declaration deviation sequence. Each column of data in the historical declaration deviation sequence corresponds to a declaration item deviation.

[0079] The historical declaration deviation vector is the difference vector obtained by subtracting the historical verification data vector and the historical declaration data vector of a single historical voyage element by element. Each element of the vector quantifies the degree and direction of deviation of the voyage on the corresponding declaration item.

[0080] First, all data tuples are traversed. For each data tuple, a vector operation is performed by subtracting the historical verification data vector and the historical declaration data vector element by element to obtain a historical declaration deviation vector that quantifies the deviations of various declarations for that voyage. For example, if the historical declaration data vector for a certain voyage is [8.25, 24300, 6150, 22] and the historical verification data vector is [8.31, 24750, 6150, 22], then the generated historical declaration deviation vector is [+0.06, +450, 0, 0]. Finally, based on the timestamp of each data tuple, all calculated historical declaration deviation vectors are sorted in ascending order to construct a historical declaration deviation sequence. Each column of this sequence represents the historical deviation of a target declaration item.

[0081] Construct an initial graph structure in which there exists an undirected edge between any two distinct deviations of the declaration items.

[0082] The initial graph structure refers to a fully connected undirected graph used to represent the potential correlations between all declaration item deviations. First, based on the determined set of target declaration items, a graph node is created for each declaration item deviation. Then, an undirected edge is added between any two distinct nodes to indicate a possible direct causal relationship between the two declaration item deviations. Finally, an initial graph structure is generated, which is a fully connected undirected graph. For example, if the set of target declaration items is (ship draft, cargo tonnage, ballast water volume, number of crew members), the initial graph structure contains four nodes, each representing a deviation from one of these four declaration items, and any two nodes are connected by an undirected edge, forming a tetrahedral structure.

[0083] Traverse the undirected edges in the initial graph structure, and use at least one declaration item deviation in the initial graph structure that is not connected to the current undirected edge as the condition set. Calculate the joint probability distribution of the two declaration item deviations connected by the current undirected edge under the given condition set, and compare the joint probability distribution with the product of the conditional probability distributions of the two declaration item deviations to obtain the independence statistic. Remove the undirected edge when the independence statistic meets the preset independence judgment condition, and generate a skeleton graph.

[0084] A condition set is a set of one or more background variables introduced to determine whether a direct relationship exists between two target variables when conducting a conditional independence test. When testing the relationship between deviations in any two declaration items, such as cargo tonnage deviation and fuel inventory deviation, other declaration item deviations, such as ship draft deviation, are selected as the condition set. By controlling or specifying the variables in the condition set, it is possible to observe whether the relationship between the target variables still exists, thus eliminating indirect influences transmitted by the condition set variables. An independence statistic is a numerical value used to quantify the degree to which two or more variables deviate from the statistical independence assumption. It transforms the abstract concept of independence into a measurable indicator. An ideal independence statistic will approach a specific theoretical value, which is usually zero, if and only if the variables are truly independent. For example, the KL divergence, mutual information, or the p-value in the chi-square test can all be used as independence statistics.

[0085] A skeleton graph is an undirected graph that remains after removing all edges deemed to lack direct causal relationships from an initial fully connected graph structure through a conditional independence test. The skeleton graph represents the skeletal network of direct relationships between all reported deviations, but the directions of these relationships are not yet determined.

[0086] First, the algorithm begins with the simplest test: the zeroth-order conditional independence test. It iterates through every undirected edge in the initial graph, such as the edge connecting the cargo tonnage deviation (variable X) and the ship draft deviation (variable Y). At this point, the condition set is empty. The algorithm then calculates the joint probability distribution of X and Y. and multiply it by their respective marginal probability distributions. The comparison yields a zero-order independence statistic. If this statistic satisfies the independence criteria—specifically, if its value is less than a preset independence threshold—it means that there is no direct correlation between X and Y, and the edge will be removed directly.

[0087] Secondly, for all edges that survive the zero-order test, the algorithm proceeds to the first-order conditional independence test. It iterates through these edges again, such as the edge connecting X and Y. But this time, it selects another node in the graph as the condition set, for example, the ballast water deviation, i.e., the variable Z. The algorithm then calculates the joint probability distribution of X and Y given Z. and the product of them with their respective conditional probabilities. By comparing these values, we obtain a first-order independence statistic. If this statistic satisfies the condition, it means that the relationship between X and Y is entirely transmitted through the intermediary Z, and there is no direct connection between them. Therefore, the edge connecting X and Y will be removed.

[0088] This process continues, gradually increasing the number of variables in the condition set to perform higher-order conditional independence tests. This iterative process continues until no more edges can be removed from the graph. Throughout this process, the probability distribution at each step is calculated using methods such as frequency counting or density estimation, based on actual data from historical declaration deviation sequences. Finally, the undirected graph remaining after these layers of filtering constitutes the skeleton graph.

[0089] Based on the undirected edges in the skeleton graph, the direction of the undirected edges is determined by identifying and orienting the collision structure formed by the deviations of the three declaration items in the skeleton graph and the propagation rules to avoid the generation of directed loops, thereby constructing a causal graph of declaration deviations.

[0090] A collision structure, also known as a V-structure, refers to a basic causal pattern in a graph consisting of three nodes, such as A→B←C. In this structure, two cause nodes, A and C, both point to a result node, B, and there are no direct edges connecting A and C. A key statistical characteristic of a collision structure is that A and C are independent, but become correlated once their common result, B, is observed or controlled. Propagation rules are a set of logically deterministic inference rules used to determine the direction of remaining undirected edges in a graph given some existing directed edges. The core objective of these rules is to ensure that the resulting directed graph is globally consistent with all conditional independence relationships learned from the data, and that no directed cycles exist in the graph, as causal relationships cannot form loops. For example, a well-known propagation rule is that if a structure A→B—C exists in the graph, to avoid creating a new collision structure, B—C must be directed to B→C.

[0091] First, the collision structure is identified and oriented. The algorithm systematically traverses all local structures in the skeleton graph consisting of three claim item deviation nodes, in the form A—B—C, where A and C are not directly connected by edges. For each such structure, the algorithm queries the results recorded in the aforementioned conditional independence test step. If the query result shows that nodes A and C are statistically independent without any other variable as a condition, but become statistically correlated when the intermediate node B is considered, then this local structure satisfies the definition of a collision structure. In this case, the two undirected edges connecting A and B, and C and B, are both determined to be directed edges pointing to the intermediate node B, forming the structure A→B←C. For example, if the skeleton graph contains a structure "cargo tonnage deviation"—"ship draft deviation"—"ballast water deviation," and the test result shows that "cargo tonnage deviation" and "ballast water deviation" are independent on their own, but become correlated given "ship draft deviation," then the two edges are oriented as cargo tonnage deviation → ship draft deviation ← ballast water deviation.

[0092] Next, propagation rules are applied to perform directional diffusion. After orienting all collision structures, some undirected edges may still remain in the graph. At this point, a set of pre-defined propagation rules are iteratively applied, using the already determined directed edges in the graph as anchor points to infer the direction of the remaining undirected edges. For example, one rule is that if there is a structure A→B—C in the graph, and there is no edge between A and C, in order to avoid creating a new collision structure A→B←C that does not match the data, the undirected edge B—C must be oriented to B→C. Another rule is that if orienting an edge would lead to a directed loop in the graph, such as A→B→C→A, then this direction is not allowed. By repeatedly applying these logical rules until all edges in the graph are assigned a definite direction, and the entire graph satisfies the characteristics of directed acyclicity and consistency with the conditional independence relationship of the data, the final graph is the causal graph of the reported deviation.

[0093] Figure 3 A schematic diagram of the structure of a cause-effect graph of reporting deviation provided in one embodiment of this application is shown.

[0094] In one feasible implementation, a cause-effect graph of reported deviations is as follows: Figure 3As shown, each ellipse represents a node of a declaration item deviation, and the directed arrow segments between nodes represent edges between nodes, i.e., direct causal relationships. The direction of the arrows indicates the direction of transmission of causal influence. For example, the arrow pointing from the cargo tonnage deviation node to the ship draft deviation node indicates that the cargo tonnage deviation is a direct cause of the ship draft deviation. At the same time, the diagram also reveals more complex causal patterns, such as crew number deviation and voyage duration deviation jointly leading to fuel inventory deviation, which in turn further affects the ship draft deviation. The entire diagram structurally reveals the interwoven systematic correlation patterns among multiple declaration item deviations present in the ship's historical declaration activities.

[0095] In one feasible implementation, step S150 calculates the declaration deviation vector based on the difference between the declaration data vector and the predicted declaration habit vector, and inputs the declaration deviation vector as observation evidence into the declaration deviation causal graph. The target node is then determined in the declaration deviation causal graph using a causal inference algorithm, including: A vector operation is performed to subtract the declared data vector from the predicted declared habit vector element by element to generate a declared deviation vector.

[0096] First, a declaration deviation vector is generated by subtracting the declaration data vector and the predicted declaration habit vector, which are identical in dimension and order, element by element. Each element of this vector quantifies the degree to which the current declaration behavior deviates from its historical habits on the corresponding declaration item. For example, if the declaration data vector is [8.52, 25130, 4980, 22] and the predicted declaration habit vector is [8.55, 25200, 4850, 22], then the generated declaration deviation vector is [-0.03, -70, +130, 0].

[0097] Deviation values ​​whose absolute value in the reported deviation vector is greater than the preset anomaly judgment threshold are identified as abnormal deviation values.

[0098] Anomaly detection thresholds are numerical thresholds used to initially screen out noteworthy anomalous components in the declared deviation vector. First, an independent anomaly detection threshold is set for each declaration item. This threshold can be based on expert experience or statistical analysis of a large amount of historical data from similar vessels. Then, each deviation value in the declared deviation vector is iterated over, and its absolute value is calculated. Finally, the absolute value of each deviation value is compared with the anomaly detection threshold set for its corresponding declaration item. If the absolute value is greater than the threshold, the deviation value is determined as an anomalous deviation value. For example, for the aforementioned generated declared deviation vector [-0.03, -70, +130, 0], if the anomaly detection thresholds set for cargo tonnage deviation and ballast water deviation are 50 tons and 100 tons respectively, then the absolute values ​​of -70 and +130 are both greater than their respective thresholds; therefore, -70 and +130 are determined as anomalous deviation values.

[0099] In the cause-effect graph of the reported deviation, all ancestor nodes that can reach the node corresponding to the abnormal deviation value through directed edge paths are identified, forming a set of candidate cause nodes.

[0100] An ancestor node is any node in a directed graph that can be reached from the target node through one or more consecutive directed edge paths. The candidate cause node set is the set of all ancestor nodes identified as nodes corresponding to abnormal deviation values ​​in the reported deviation cause-effect graph.

[0101] First, for each identified abnormal deviation value, its corresponding node is located in the reported deviation cause-effect graph. Then, starting from that node, a reverse traversal is performed along all directed edges pointing to it, adding all traversed nodes to a set. This reverse traversal process continues until a starting node with no incoming edges is reached. Finally, the ancestor nodes traced back to all abnormal deviation values ​​are merged, and duplicates are removed, thus forming a candidate cause node set. For example, if cargo tonnage deviation and ballast water volume deviation are identified as abnormal deviation values, and the cause-effect graph contains paths like voyage duration deviation → cargo tonnage deviation and cargo tonnage deviation → ballast water volume deviation, then tracing the ancestor of cargo tonnage deviation yields (voyage duration deviation), and tracing the ancestor of ballast water volume deviation yields (cargo tonnage deviation, voyage duration deviation). The merged candidate cause node set is (cargo tonnage deviation, voyage duration deviation).

[0102] The contribution of each candidate cause node is obtained by calculating the number of times each candidate cause node in the candidate cause node set is the ancestor node of the node corresponding to the abnormal deviation value in the cause-effect graph of the reported deviation.

[0103] Contribution is used to measure the likelihood that a candidate cause node is the root cause. First, a counter is initialized to record the contribution of each candidate cause node. Then, all anomalous deviation values ​​are iterated over. For each anomalous deviation value, all its ancestor nodes are traced back in the reported deviation cause-effect graph. Whenever an ancestor node is traced, its count in the counter is incremented by one. After iterating over all anomalous deviation values, the value recorded in the counter for each candidate cause node is its contribution. For example, for the aforementioned set of candidate cause nodes (cargo tonnage deviation, voyage duration deviation) and anomalous deviation values ​​(-70, +130), since cargo tonnage deviation is the ancestor of ballast water deviation, its contribution is 1; while voyage duration deviation is the ancestor of both cargo tonnage deviation and ballast water deviation, its contribution is 2.

[0104] The candidate cause node with the highest contribution is identified as the target node. First, the contribution values ​​of all nodes in the calculated set of candidate cause nodes are compared. Then, the candidate cause node with the highest contribution value is selected. Finally, this candidate cause node with the highest contribution value is identified as the target node for this verification. If multiple nodes have the same highest contribution value, one can be selected according to a preset priority rule, such as selecting the first identified node as the target node, or selecting all of them as target nodes. For example, in the previous example, since the contribution value of the voyage duration deviation is 2, which is greater than the contribution value of the cargo tonnage deviation (1), the node corresponding to the voyage duration deviation is identified as the target node.

[0105] This embodiment constructs and utilizes historical declaration deviation sequences to discover the causal relationships between deviations in different declaration items, thereby establishing a deviation causal graph for each vessel that characterizes its systematic declaration deviation patterns. When verifying current declaration information, anomalies deviating from its historical declaration habits are first identified using a predictive model. Then, the deviation causal graph is used to attribute these anomalies causally, tracing and locating the root cause nodes leading to the declaration anomalies. This method can penetrate the facade of compliance in individual declaration data, identifying combined declaration deviation patterns with inherent logical connections, composed of minor, coordinated deviations between multiple declaration items. This effectively solves the technical problem that existing technologies cannot identify systematically correlated declaration deviations in vessels.

[0106] Figure 4 This application provides a schematic diagram of a specific implementation of an artificial intelligence-based automatic verification system for ship entry and exit declaration information, referring to... Figure 4 The system may include: The acquisition module 410 is used to acquire the port entry and exit declaration information of the target vessel and the historical declaration and verification dataset, which includes multiple historical declaration records and corresponding historical verification records. The module 420 is used to extract the declaration information of multiple target declaration items from the port entry and exit declaration information to construct the declaration data vector, and to determine multiple data tuples including timestamps, historical declaration data vectors and historical verification data vectors based on the multiple target declaration items in the historical declaration verification dataset; The prediction module 430 is used to construct a historical declaration sequence in time order based on the historical declaration data vector in each data tuple, and input the historical declaration sequence into the trained multivariate time series prediction model to obtain the predicted declaration habit vector. The construction module 440 is used to construct a historical declaration deviation sequence in time order based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple. The module then uses a causal discovery algorithm to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence, determine the causal relationship between different declaration item deviations, and construct a declaration deviation causal graph. Nodes represent declaration item deviations, and edges represent the causal relationship between declaration item deviations. The determination module 420 is also used to calculate the declaration deviation vector based on the difference between the declaration data vector and the predicted declaration habit vector, and input the declaration deviation vector as observation evidence into the declaration deviation causal graph, and determine the target node in the declaration deviation causal graph through the causal reasoning algorithm; The determination module 420 is also used to calculate the deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, and to determine the target deviation value corresponding to the target node from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

[0107] The AI-based automatic verification system for ship entry and exit declaration information in this application is used to implement the aforementioned AI-based automatic verification method for ship entry and exit declaration information. Therefore, the specific implementation of the AI-based automatic verification system for ship entry and exit declaration information can be found in the embodiment section of the AI-based automatic verification method for ship entry and exit declaration information above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0108] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application is shown.

[0109] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0110] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0111] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0112] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0113] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the artificial intelligence-based automatic verification methods for ship entry and exit declaration information in the above embodiments.

[0114] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0115] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0116] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0117] The electronic device can execute the AI-based automatic verification method for ship entry and exit declaration information in the embodiments of this application, thereby realizing the AI-based automatic verification method for ship entry and exit declaration information described in conjunction with the accompanying drawings.

[0118] Furthermore, in conjunction with the AI-based automatic verification method for ship entry and exit declaration information in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the AI-based automatic verification methods for ship entry and exit declaration information in the above embodiments.

[0119] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0120] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0121] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0122] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0123] The foregoing has provided a detailed description of the artificial intelligence-based automatic verification method, system, electronic device, and storage medium for ship entry and exit declaration information provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for automatically verifying ship entry and exit port declaration information based on artificial intelligence, characterized in that, include: Obtain the port entry and exit declaration information of the target vessel and a historical declaration and verification dataset including multiple historical declaration records and corresponding historical verification records; From the port entry and exit declaration information, the declaration information of multiple target declaration items is extracted to construct a declaration data vector, and based on the multiple target declaration items in the historical declaration verification dataset, multiple data tuples including timestamps, historical declaration data vectors, and historical verification data vectors are determined; Based on the historical declaration data vector in each data tuple, a historical declaration sequence is constructed in chronological order, and the historical declaration sequence is input into a trained multivariate time series prediction model to obtain a predicted declaration habit vector. Based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation sequence is constructed in chronological order. Then, a causal discovery algorithm is used to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationship between different declaration item deviations. A declaration deviation causal graph is constructed, where nodes represent declaration item deviations and edges represent the causal relationship between declaration item deviations. Based on the difference between the declared data vector and the predicted declaration habit vector, a declaration deviation vector is calculated, and the declaration deviation vector is used as observation evidence to input into the declaration deviation causal graph. The target node is determined in the declaration deviation causal graph through a causal inference algorithm. Based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, a deviation threshold is calculated, and the target deviation value corresponding to the target node is determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

2. The method according to claim 1, characterized in that, The method further includes: The structural risk score is obtained by calculating the graph structure entropy of the reported deviation cause-effect graph; Based on the structural risk score, the target threshold adjustment factor is determined in the preset correspondence between structural risk scores and threshold adjustment factors; The step of calculating a deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, determining the target deviation value corresponding to the target node from the declaration deviation vector, and determining the declaration information corresponding to the target node as abnormal declaration information if the target deviation value is greater than the deviation threshold includes: Based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor, a deviation threshold is calculated, and the target deviation value corresponding to the target node is determined from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

3. The method according to claim 2, characterized in that, The step of calculating a deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence and the target threshold adjustment factor, and determining the target deviation value corresponding to the target node from the declaration deviation vector, and determining the declaration information corresponding to the target node as abnormal declaration information if the target deviation value is greater than the deviation threshold, includes: Extract all historical declaration deviations associated with the declaration item corresponding to the target node from the historical declaration deviation sequence to form a set of historical deviation values; The absolute values ​​of all historical reported deviations in the historical deviation value set are sorted, and the values ​​that are located at a preset quantile after sorting are selected as the first deviation threshold. The first deviation threshold is multiplied by the target threshold adjustment factor to generate the deviation threshold; The target deviation value corresponding to the target node is determined from the declaration deviation vector, and if the absolute value of the target deviation value is greater than the deviation threshold, the declaration information associated with the declaration item deviation corresponding to the target node is determined to be abnormal declaration information.

4. The method according to claim 1, characterized in that, The process involves extracting multiple target declaration items from the port entry / exit declaration information to construct a declaration data vector, and then determining multiple data tuples, including timestamps, historical declaration data vectors, and historical verification data vectors, from the historical declaration verification dataset based on these multiple target declaration items. Extract multiple declaration information corresponding to the multiple target declaration items from the port entry and exit declaration information, and construct the declaration data vector. The multiple target declaration items include multiple declaration items related to the physical loading status of the ship. From each historical declaration record and the corresponding historical verification record in the historical declaration verification dataset, the declaration information corresponding to the multiple target declaration items is extracted to form the historical declaration data vector and the historical verification data vector; The historical declaration data vector, the historical verification data vector, and the timestamp corresponding to the historical declaration record are combined according to a preset sorting order to obtain a data tuple corresponding to each historical declaration record in the historical declaration verification dataset.

5. The method according to claim 1, characterized in that, The step of constructing a historical declaration sequence in chronological order based on the historical declaration data vector in each data tuple, and inputting the historical declaration sequence into a trained multivariate time series prediction model to obtain a predicted declaration habit vector includes: Based on the historical declaration data vectors in all the data tuples, arrange them in chronological order according to the timestamps to construct the historical declaration sequence; A continuous sequence of a preset length is extracted from the end of the historical declaration sequence and used as the input sequence for the multivariate time series prediction model; The input sequence is fed into the trained multivariate time series prediction model to generate the predicted declaration habit vector. The multivariate time series prediction model is trained using a deep learning model based on a self-attention mechanism, based on training samples including historical declaration training sequences and corresponding label vectors.

6. The method according to claim 1, characterized in that, The process involves constructing a historical declaration deviation sequence in chronological order based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, and then using a causal discovery algorithm to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence to determine the causal relationships between different declaration item deviations, and constructing a declaration deviation causal graph, including: By performing element-wise subtraction on the historical declaration data vector and the historical verification data vector in each data tuple, a historical declaration deviation vector is obtained. The historical declaration deviation vectors corresponding to each data tuple are arranged in chronological order according to the timestamps to obtain the historical declaration deviation sequence. Each column of data in the historical declaration deviation sequence corresponds to a declaration item deviation. Construct an initial graph structure in which there exists an undirected edge between any two different declaration item deviations; Traverse the undirected edges in the initial graph structure, and use at least one declaration item deviation in the initial graph structure that is not connected to the current undirected edge as a condition set. Calculate the joint probability distribution of the two declaration item deviations connected by the current undirected edge under the given condition set, and compare the joint probability distribution with the product of the conditional probability distributions of the two declaration item deviations to obtain an independence statistic. When the independence statistic satisfies a preset independence judgment condition, remove the undirected edge to generate a skeleton graph. Based on the undirected edges in the skeleton graph, the direction of the undirected edges is determined by identifying and orienting the collision structure formed by the three declaration item deviations in the skeleton graph and the propagation rules to avoid generating directed loops, thereby constructing the declaration deviation causal graph.

7. The method according to claim 6, characterized in that, The step of calculating a declaration deviation vector based on the difference between the declared data vector and the predicted declaration habit vector, and inputting the declaration deviation vector as observational evidence into the declaration deviation causal graph, and determining the target node in the declaration deviation causal graph through a causal inference algorithm, includes: The vector of declarations is subtracted element-by-element from the vector of predicted declaration habits to generate the vector of declaration deviations. Deviation values ​​whose absolute value in the reported deviation vector is greater than a preset anomaly determination threshold are identified as abnormal deviation values. In the reported deviation cause-effect graph, all ancestor nodes that can reach the node corresponding to the abnormal deviation value through directed edge paths are identified, forming a candidate cause node set; The contribution of each candidate cause node is obtained by calculating the number of times each candidate cause node in the candidate cause node set is an ancestor node of the node corresponding to the abnormal deviation value in the cause-effect graph of the reported deviation. The candidate cause node with the largest contribution is determined as the target node.

8. An automatic verification system for ship entry and exit declaration information based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire the port entry and exit declaration information of the target vessel and the historical declaration and verification dataset, which includes multiple historical declaration records and corresponding historical verification records. The determination module is used to extract the declaration information of multiple target declaration items from the port entry and exit declaration information to construct a declaration data vector, and to determine multiple data tuples including timestamps, historical declaration data vectors and historical verification data vectors in the historical declaration verification dataset based on the multiple target declaration items; The prediction module is used to construct a historical declaration sequence in time order based on the historical declaration data vector in each data tuple, and input the historical declaration sequence into a trained multivariate time series prediction model to obtain a predicted declaration habit vector. The construction module is used to construct a historical declaration deviation sequence in time order based on the difference between the historical declaration data vector and the historical verification data vector in each data tuple, and to perform conditional independence tests and directed edge construction on the multivariate deviation values ​​in the historical declaration deviation sequence through a causal discovery algorithm to determine the causal relationship between different declaration item deviations and construct a declaration deviation causal graph, where nodes represent declaration item deviations and edges represent the causal relationship between declaration item deviations. The determination module is also used to calculate the declaration deviation vector based on the difference between the declaration data vector and the predicted declaration habit vector, and input the declaration deviation vector as observation evidence into the declaration deviation causal graph, and determine the target node in the declaration deviation causal graph through a causal reasoning algorithm; The determination module is further configured to calculate a deviation threshold based on the historical deviation value associated with the declaration item corresponding to the target node in the historical declaration deviation sequence, and determine the target deviation value corresponding to the target node from the declaration deviation vector. If the target deviation value is greater than the deviation threshold, the declaration information corresponding to the target node is determined to be abnormal declaration information.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for automatically verifying ship entry and exit declaration information based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the automatic verification method for ship entry and exit declaration information based on artificial intelligence as described in any one of claims 1 to 7.

Images