Customs declaration intelligent dispatching and collaborative processing system

Abstract

The present invention belongs to the field of customs declaration distribution, and particularly relates to a system for intelligent declaration dispatch and collaborative processing. This system uses a business declaration module, combined with a user's historical credit score and an initial review model, to conduct compliance audits on declaration information, marking the difficulty and risk level of approved information. An intelligent generation module generates compliant customs declarations based on pre-assigned tags and a compliance rule library, and encrypts and locks them. A collaborative dispatch module uses a distributed reinforcement dispatch model to accurately match declarations to reviewers based on information such as the difficulty of customs declaration processing, comprehensive risk assessment scores, and employee proficiency. The review feedback module monitors review progress and anomalies in real time, and implements dynamic closed-loop feedback through a declaration feedback adjustment chain. The present invention implements intelligent assessment, precise dispatch, and closed-loop collaborative management of customs declaration processing.

Description

Technical Field

[0001] The present invention belongs to the field of customs declaration distribution, and in particular relates to a customs declaration intelligent distribution and collaborative processing system. Background Art

[0002] In the context of the digital transformation of international trade, the dispatching and processing of customs declarations faces multiple technical bottlenecks; although the existing automation system has introduced a dispatching algorithm, the multimodal extraction model has not designed parameter weights for the customs declaration scenario, and is unable to identify implicit difficulty parameters such as the timeliness of special regulatory certificates and the risk of commodity classification disputes, resulting in a missing dimension of dispatch data; in the compliance review module, the employee skill map and the customs declaration risk level are not associated, the vertical transmission rules are not embedded in the dispatching priority rules, and the horizontal review ignores the historical processing efficiency of employees, resulting in insufficient matching accuracy of customs declaration processing difficulty, employee skills, and processing timeliness, and a high mismatch rate for high-difficulty orders; the transmission unit is only allocated based on load and review type, and a dynamic weight model including order urgency, skill matching, and order pool saturation is not constructed, and there is a lack of cross-node collaboration mechanism, resulting in a backlog of high-difficulty orders. For this reason, this application provides an intelligent dispatching and collaborative processing system for customs declarations. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention proposes an intelligent customs declaration dispatch and collaborative processing system. The system comprises a business declaration module, an intelligent generation module, a collaborative dispatch module, and an audit feedback module. The business declaration module combines the user's historical credit score and initial audit model to conduct compliance audits on declaration information, and labels the audited information with processing difficulty and risk level. The intelligent generation module generates compliant customs declarations based on pre-assigned tags and a compliance rule library and encrypts and locks them. The collaborative dispatch module utilizes a distributed reinforcement dispatch model to accurately match customs declarations to reviewers based on information such as customs declaration processing difficulty, comprehensive risk assessment scores, and employee proficiency. The audit feedback module monitors audit progress and anomalies in real time and implements dynamic closed-loop feedback through a declaration feedback adjustment chain. For declaration information anomalies, a direct traceability warning is issued to the declaring user. For dispatch delay anomalies, feedback is fed back to the intelligent pre-assignment node for dynamic adjustment and reallocation, ensuring audit efficiency and risk control. The present invention achieves intelligent assessment, precise dispatch, and closed-loop collaborative management of customs declaration processing.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] The customs declaration intelligent dispatching and collaborative processing system includes: business declaration module, intelligent generation module, collaborative dispatching module and review feedback module;

[0006] The business declaration module is used to obtain customs declaration business information and conduct an initial compliance review of the collected customs declaration business information in combination with the historical declaration credit score of the declaring user in combination with the preset initial review and assessment model. The module also marks the difficulty and risk level of customs declaration processing for those that pass the review, and obtains the declaration information sequence that passes the initial review;

[0007] The intelligent generation module generates compliant customs declarations based on the declaration information sequence that has passed the initial review, combined with the configured customs declaration generation model, and combined with the priority pre-assigned label markings and the preset customs declaration generation compliance rule library. The generated compliant customs declarations are encrypted and locked using an encryption algorithm;

[0008] The collaborative dispatch module is used to match and dispatch customs declarations based on the customs declarations on the chain, combined with the difficulty of customs declaration processing, comprehensive risk assessment scores, employee review proficiency and risk level, through the configured distributed enhanced dispatch model. It also monitors the review progress and abnormal results of compliant customs declarations in real time through the review feedback module, and feeds back the monitoring results to the corresponding user declaration end in real time through the preset declaration feedback adjustment chain.

[0009] Specifically, the declaration feedback adjustment chain includes the declaration end, the customs clearance cloud node and the review end; the customs clearance cloud node includes the risk assessment sub-node;

[0010] The declaration terminal is used to collect user declaration information and user historical declaration exception information and, in combination with the configured multimodal parsing algorithm, obtain the declaration text parsing space and the declaration image parsing space;

[0011] The risk assessment sub-node is used to obtain the corresponding risk score of the declared target and the corresponding assessment accuracy probability based on the declared text analysis space and the declared image analysis space combined with the preset risk level expert assessment. At the same time, through a comprehensive assessment algorithm combined with the user's historical declared abnormal information and the user's historical credit score, the corresponding user credit score and the corresponding assessment accuracy rate are obtained.

[0012] Specifically, the customs clearance cloud node also includes an initial review sub-node;

[0013] The initial review sub-node is used to obtain the comprehensive risk assessment score of the corresponding user through a weighted average algorithm based on the hazard score of the declared target and the corresponding assessment accuracy probability, the corresponding user credit score and the corresponding assessment accuracy rate. Based on the comprehensive risk assessment score, the declaration material information currently submitted by the user and the preset declaration information criteria for the declared target, it is used to judge whether the current materials are sufficient and compliant. If sufficient, the declaration material information currently submitted by the user is input into the priority assessment sub-node and the corresponding comprehensive risk assessment score and the hazard score of the declared target are marked; if insufficient or non-compliant, the supplementary or modified materials are directly rejected until the corresponding declaration material requirements under the corresponding comprehensive risk assessment score are met, and the supplementary materials are used to update the declaration text parsing space and the declaration image parsing space.

[0014] Specifically, the clearance cloud node also includes priority evaluation sub-nodes and intelligent pre-allocation sub-nodes;

[0015] The priority evaluation sub-node is used to obtain the declaration processing difficulty of the corresponding declaration target based on the complexity of the historical declaration process of the declaration target corresponding to the initial review compliance, the amount of declaration materials required, and the technical proficiency of the auditors assigned to the historical declaration review through the evaluation algorithm C1. At the same time, based on the declaration processing difficulty of the declaration target combined with the maximum effective declaration cycle length of the declaration target, the corresponding audit end load of the declaration target, and the auditor's skill proficiency, the evaluation algorithm C2 is used to obtain the declaration review overdue probability risk value of the declaration target. The declaration review overdue probability risk value is used as the declaration priority and marked on the corresponding declaration target;

[0016] Intelligent pre-allocation of sub-nodes is used to obtain the initial matching review sub-nodes and the predicted first review delay risk probability corresponding to the initial matching review sub-nodes based on the comprehensive risk assessment score corresponding to the declared target object, the declaration priority, the corresponding review sub-node load on the review end, and the reviewer's skill proficiency through matching algorithms combined with Bayesian algorithms.

[0017] Specifically, the clearance cloud node also includes intelligent generation sub-nodes and distributed dispatch sub-nodes;

[0018] Intelligently generate a sub-node, which is used to obtain the customs declaration form for the corresponding declared object based on the declaration text parsing space and the declaration image parsing space of the initial review compliance, through the text generation algorithm combined with the customs declaration template corresponding to the declared object, and conduct customs declaration compliance review based on the generated customs declaration and the preset customs declaration compliance review rule library to obtain a compliant customs declaration;

[0019] The distributed dispatch sub-node is used to allocate the compliant customs declaration form to the corresponding review sub-node through a distributed reinforcement model based on the compliant customs declaration form combined with the corresponding initial matching review sub-node and the load information fed back in real time by the corresponding initial matching review sub-node.

[0020] Specifically, the audit end includes an audit sub-node, a monitoring sub-node, and a feedback sub-node;

[0021] A monitoring sub-node is used to monitor the audit progress of each audit sub-node in real time, monitor and identify the audit anomaly type of the declaration form corresponding to the declared object, and feed back the monitored information to the feedback sub-node;

[0022] The feedback sub-node is used to provide differential feedback based on the identified audit anomaly type. Specifically, if the audit anomaly type is incorrect or unqualified declaration information, it is determined to be an abnormal declaration information. Based on the abnormal declaration information, the preset traceability feature code in the distributed connection in the declaration feedback adjustment chain is triggered, and the corresponding abnormality of the declared target object is directly fed back to the corresponding declaration end through the distributed connection to issue an abnormality warning;

[0023] If the audit exception type is that the difference between the second review delay risk probability and the first review delay risk probability of the corresponding audit sub-node monitored by the monitoring sub-node is greater than the preset risk limit threshold, it is determined to be a declaration assignment exception, and the declaration assignment exception triggers the preset traceability feature code in the distributed connection in the declaration feedback adjustment chain, and the corresponding declaration assignment exception is fed back to the intelligent pre-assignment sub-node for allocation warning. According to the load information monitored by the declaration end and the difference between the second review delay risk probability and the first review delay risk probability, the audit sub-node information initially predicted and assigned by the intelligent generation sub-node is adjusted, and the adjusted audit sub-node is used to re-audit the customs declaration business information of the declared target object with the declaration assignment exception until the difference between the second review delay risk probability and the first review delay risk probability is less than or equal to the preset risk limit threshold.

[0024] Specifically, the steps for constructing distributed connections in the feedback regulation chain include:

[0025] Based on the declaration interaction information between all declaration terminals and the customs clearance cloud node, a declaration channel connection set is established; based on the dispatch interaction information between the customs clearance cloud node and each audit sub-node under the audit terminal, a dispatch channel connection set is established;

[0026] A first feedback channel connection set is established based on the feedback interaction information between the review sub-node corresponding to each declared object under the review end and the declaration end corresponding to the declared object. At the same time, a second feedback channel connection set is established based on the feedback interaction information between the review end and each sub-node under the customs clearance cloud node;

[0027] Based on the declaration end, customs clearance cloud node and corresponding sub-nodes and the review end and corresponding sub-nodes combined with the declaration channel connection set, the dispatch channel connection set, the first feedback channel connection set and the second feedback channel connection set, a declaration feedback adjustment chain is constructed through the blockchain algorithm.

[0028] Specifically, the steps for constructing the distributed connection in the feedback regulation chain also include:

[0029] Based on the comprehensive risk assessment score of the corresponding declared object at each reporting end of the declaration feedback adjustment chain and the hazard score of the declared object, combined with the preset encryption rules, the declared objects of different hazard levels are differentially encrypted on the chain;

[0030] Based on the declaration form ID, initial review timestamp, risk score, user credit score, number of declared objects and destination corresponding to each declared object, a random number generator and hash algorithm are combined to form a hash calibration sequence for each declared object;

[0031] Based on the hash calibration sequence of each declared target object and the number N of corresponding connection relationships of each declared target object, a random anchor point segmentation algorithm is used to obtain 2N hash calibration sequence segments.

[0032] Specifically, the steps for constructing the distributed connection in the feedback regulation chain also include:

[0033] Assign 2N hash calibration sequences to both ends of the N connection relationships corresponding to each declared target object, and obtain the packet loss probability of the child nodes corresponding to the two adjacent connection relationships of each declared target object and multiply it by the comprehensive risk assessment score of the declared target object to obtain the product of all child nodes corresponding to the declared target object;

[0034] Based on the product of all child nodes corresponding to the declared target object, a relationship connection site sequence of a length corresponding to the product size is randomly extracted from one end of the connection relationship connected to the corresponding child node through a random algorithm, and the extracted relationship connection site sequence is inserted into the hash calibration sequence segment configured at the corresponding end to form an identification feature code generation sequence;

[0035] Based on the identification feature code generation sequence corresponding to the adjacent ends of the two adjacent connection relationships of each declared target object, the identification feature codes corresponding to the two adjacent ends are generated using the SHA3-256 algorithm. The generated identification feature codes of the two adjacent ends are combined with the smart contract in the blockchain and stored on the chain in the child nodes corresponding to the two ends, thus obtaining a declaration feedback adjustment chain with traceability identification;

[0036] Specifically, the declaration feedback adjustment chain with traceability identification is also used to: establish a verification mechanism between each sub-node based on the identification feature code. When any sub-node receives data, it verifies the integrity and authenticity of the data transmission process by comparing the identification feature codes at both ends with the corresponding values ​​stored in the smart contract. If the verification fails, the abnormal feedback process is triggered;

[0037] The abnormality feedback process includes: generating a sequence of relational connection sites in a sequence based on the identification feature code, tracing the specific connection relationship where the data transmission abnormality occurs; locating the specific position of the abnormal data in the transmission path in combination with the hash calibration sequence segment of the declared target object; and feeding back the abnormal information and tracing results to the corresponding declaration end and customs clearance cloud node through the first feedback channel connection and the second feedback channel connection.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] In response to the deficiencies of the existing technology, the present invention constructs an intelligent audit system through multi-module collaboration and blockchain technology, significantly improving the efficiency and security of customs declarations. Among them, the business declaration module uses a multimodal analysis algorithm to deeply extract declaration information, and combines credit scoring to achieve initial compliance review and risk grading. The intelligent generation module generates encrypted customs declarations based on the compliance rule base and pre-assigned tags to ensure the legitimacy and security of the data. The collaborative dispatch module uses a distributed reinforcement model to comprehensively process multi-dimensional indicators such as difficulty, risk assessment, and skill proficiency to achieve the optimal match between customs declarations and audit resources. The declaration feedback adjustment chain builds a distributed connection based on the blockchain, and through hash calibration, feature code verification and exception tracing mechanisms, a full-process data traceability system is formed to ensure transmission integrity and authenticity. This application constructs a closed-loop management system through dynamic load balancing, intelligent path optimization and exception handling mechanisms to achieve intelligence, precision and efficiency in customs declaration business, and provide reliable customs clearance guarantees for cross-border trade. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a module diagram of the customs declaration intelligent dispatching and collaborative processing system according to embodiment 1 of the present invention;

[0041] Figure 2 This is a diagram of the feedback regulation chain architecture for Example 1 of the present invention. DETAILED DESCRIPTION

[0042] Example 1

[0043] See also Figure 1 The present invention provides an embodiment: a customs declaration intelligent dispatching and collaborative processing system, comprising: a business declaration module, an intelligent generation module, a collaborative dispatching module and an audit feedback module;

[0044] The business declaration module is used to obtain customs declaration business information and conduct an initial compliance review of the collected customs declaration business information in combination with the historical declaration credit score of the declaring user in combination with the preset initial review and assessment model. The module also marks the difficulty and risk level of customs declaration processing for those that pass the review, and obtains the declaration information sequence that passes the initial review;

[0045] The intelligent generation module generates compliant customs declarations based on the declaration information sequence that has passed the initial review, combined with the configured customs declaration generation model, and combined with the priority pre-assigned label markings and the preset customs declaration generation compliance rule library. The generated compliant customs declarations are encrypted and locked using an encryption algorithm;

[0046] The collaborative dispatch module is used to match and dispatch customs declarations based on the customs declarations on the chain, combined with the difficulty of customs declaration processing, comprehensive risk assessment scores, employee review proficiency and risk level, through the configured distributed enhanced dispatch model. It also monitors the review progress and abnormal results of compliant customs declarations in real time through the review feedback module, and feeds back the monitoring results to the corresponding user declaration end in real time through the preset declaration feedback adjustment chain.

[0047] For further explanation, please refer to Figure 2 In this embodiment, the declaration feedback adjustment chain includes the declaration end, the customs clearance cloud node, and the audit end; the customs clearance cloud node includes the risk assessment sub-node, the initial audit sub-node, the priority assessment sub-node, the intelligent pre-allocation sub-node, the intelligent generation sub-node, and the distributed dispatch sub-node; the audit end includes the audit sub-node, the monitoring sub-node, and the feedback sub-node; A1 to An are n declaration ends, and B1 to Bq are m audit sub-nodes under the audit end;

[0048] The declaration terminal is used to collect user declaration information and user historical declaration exception information and, in combination with the configured multimodal parsing algorithm, obtain the declaration text parsing space and the declaration image parsing space;

[0049] It should be further explained that the multimodal parsing algorithm in this embodiment is preferably a multimodal algorithm constructed by combining Transformer-based OCR technology and Faster R-CNN image detection algorithm; a specific implementation process is as follows:

[0050] The collected declaration information of the declared objects is preprocessed using text cleaning and image enhancement algorithms. A pre-trained language model is then used to extract text semantic features and entity relationships. Faster R-CNN is used to locate the form / certificate area in the image and combined with OCR to extract text content. The Cross-Attention mechanism is then used to achieve semantic alignment of text and image features. A gated fusion strategy is used to integrate multimodal information. Finally, a structured verification is performed on the extracted key information based on the customs declaration domain knowledge base (such as the regulatory document rule base). A declaration text parsing space containing an entity relationship graph and a declaration image parsing space with key area features annotated are generated. At the same time, the abnormal pattern library is used to identify declaration risk characteristics. The regulatory document rule library and the abnormal pattern library are constructed by those skilled in the art through a graph database based on historical declaration process information and declaration rule information, and will not be described in detail here. It should be noted that the gated fusion strategy in this embodiment is based on the GRU algorithm combined with a multi-head attention network to perform multimodal fusion of extracted text information and image information. Key information includes but is not limited to quantity, value, and place of origin; text semantic features and entity relationships include but are not limited to product names, HS codes, etc.; the abnormal pattern library includes but is not limited to abnormal certificate validity and product classification disputes.

[0051] It should be further explained that the reporting end in this embodiment also includes a risk warning sub-module. The risk warning sub-module is used to predict future data transmission risks through the LSTM algorithm based on the historical data of transmission anomalies in the reporting feedback adjustment chain, combined with the comprehensive risk assessment score and danger score of the reported target object. When the predicted risk value exceeds the set threshold, a warning message is sent to the user, and the user is prompted to supplement or correct the reported information to reduce the risk.

[0052] The risk assessment sub-node is used to obtain the risk score of the corresponding declared object and the corresponding assessment accuracy probability based on the declared text analysis space and the declared image analysis space combined with the preset risk level expert assessment. At the same time, a comprehensive assessment algorithm is used to combine the user's historical declared abnormal information and the user's historical credit score to obtain the corresponding user credit score and the corresponding assessment accuracy rate;

[0053] It should be further explained that one implementation of the risk level expert assessment in this embodiment is as follows:

[0054] First, CNN is used to identify and extract features of hazardous materials logos and packaging features in the declared image parsing space. NLP technology is used to extract semantic features such as product names, ingredients, and uses from the declared text parsing space. Secondly, a hazard assessment index system is constructed based on expert experience, and the weight of each index is determined using AHP. Thirdly, the features extracted from the image and text are mapped to the corresponding evaluation indicators, and the eigenvalues ​​are converted into a fuzzy evaluation matrix using fuzzy membership functions. Fourthly, the hazard score of the declared target is obtained through fuzzy evaluation matrix operations, and the probability of accurate evaluation corresponding to the score is calculated based on the comparison of model training and historical evaluation data, thus achieving a scientific hazard level assessment of the declared target. The hazard assessment index system includes but is not limited to dimensions such as toxicity, flammability, and radioactivity.

[0055] The initial review sub-node is used to obtain the comprehensive risk assessment score of the corresponding user through a weighted average algorithm based on the hazard score of the declared target and the corresponding assessment accuracy probability, the corresponding user credit score and the corresponding assessment accuracy rate. Based on the comprehensive risk assessment score, the declaration material information currently submitted by the user and the preset declaration information criteria for the declared target, it is used to judge whether the current materials are sufficient and compliant. If sufficient, the declaration material information currently submitted by the user is input into the priority assessment sub-node and the corresponding comprehensive risk assessment score and the hazard score of the declared target are marked; if insufficient or non-compliant, the supplementary or modified materials are directly rejected until the corresponding declaration material requirements under the corresponding comprehensive risk assessment score are met, and the supplementary materials are used to update the declaration text parsing space and the declaration image parsing space.

[0056] It should be further explained that in this embodiment, one implementation method of the comprehensive risk assessment score of the corresponding user is:

[0057] Step 1: Based on the reported target's hazard score and the corresponding probability of assessment accuracy, the user's credit score and the corresponding assessment accuracy, determine the weight of each indicator through the analytic hierarchy process, and calculate the comprehensive risk assessment score using the weighted average algorithm; the weight of each indicator includes but is not limited to the weight of the hazard score, the weight of the probability of assessment accuracy, the weight of the credit score, and the weight of the assessment accuracy;

[0058] Step 2: Based on the comprehensive risk assessment score and the application materials currently submitted by the user, the rules engine algorithm is used to match the preset application information criteria for the application target, verify the completeness and compliance of the materials, and obtain a material sufficiency judgment result. The application information criteria for the application target include but are not limited to different material requirements corresponding to different risk levels, such as the requirement for high-risk products to provide an additional dangerous goods test report. The completeness of the materials includes but is not limited to whether required fields are missing. Compliance includes but is not limited to consistency verification of the HS code and product description.

[0059] Step 3: If the materials are insufficient or non-compliant, the rejection notice generation algorithm will trigger a supplementary modification process until the materials meet the declaration requirements under the corresponding comprehensive risk assessment score. Based on the supplemented materials, the declaration text parsing space and the declaration image parsing space are updated through the incremental learning algorithm to ensure that the parsing model is iteratively optimized as the materials are improved; the rejection notice generation algorithm will prioritize the pre-trained text generation algorithm; the rejection notice generation algorithm will include a list of missing materials and correction instructions; the updated declaration text parsing space will include the semantic vectors of the newly added product ingredient descriptions; the updated declaration image parsing space will include but is not limited to the supplementation of the dangerous goods packaging image feature library;

[0060] Step 4: If the materials are sufficient and compliant, the comprehensive risk assessment score and the hazard score of the declared target object are embedded into a data structure through a data labeling algorithm and input into the priority assessment sub-node. The data structure is specifically: the risk_score and danger_level fields of the JSON object. The data labeling algorithm is preferably a clustering labeling algorithm constructed by K-Means, which clusters targets with the same hazard score and embeds the labels at the same time.

[0061] It should be further explained that in this embodiment, a specific implementation method for verifying the adequacy of the application materials is as follows: based on the comprehensive risk assessment score and the application materials information currently submitted by the user, a matching algorithm is used to match the preset application information criteria for the application target, verify the completeness and compliance of the materials, and obtain the material adequacy judgment result. The application information criteria for the application target include but are not limited to different material requirements corresponding to different risk levels, such as the requirement for high-risk products to provide additional dangerous goods test reports;

[0062] The priority evaluation sub-node is used to assign the technical proficiency of the reviewers to the historical declaration process complexity, the amount of required declaration materials, and the historical declaration review of the declaration target corresponding to the initial review compliance, and obtain the declaration processing difficulty of the corresponding declaration target through evaluation algorithm C1. At the same time, based on the declaration processing difficulty of the declaration target, combined with the maximum effective declaration cycle length of the declaration target, the corresponding review end load of the declaration target, and the skill proficiency of the reviewers, the evaluation algorithm C2 is used to obtain the probability risk value of the declaration review overdue of the declaration target, and use the probability risk value of the declaration review overdue as the declaration priority and mark it on the corresponding declaration target; the evaluation algorithm C1 is preferably a hierarchical analysis method or a multi-factor weighted regression algorithm. The specific process is as follows: based on the historical declaration process complexity, the amount of required materials, and the technical proficiency of the historical reviewers of the declaration target that has initial review compliance, the declaration processing difficulty is obtained through indicator quantification, hierarchical analysis method to determine weights and weighted calculation, and an assessment of the time consumption and risk of the declaration target processing is achieved. The evaluation algorithm C2 is preferably a queuing theory model or a survival analysis algorithm; the indicator quantification includes but is not limited to the conversion of process links, material missing rate, and personnel skills into numerical values; the weights include but are not limited to process complexity, material quantity, and personnel proficiency;

[0063] Intelligent pre-allocation of sub-nodes is used to obtain the initial matching review sub-nodes and the predicted first review delay risk probability corresponding to the initial matching review sub-nodes based on the comprehensive risk assessment score of the declared target object, the declaration priority, the corresponding review sub-node load on the review end, and the reviewer's skill proficiency through a matching algorithm combined with a Bayesian algorithm;

[0064] It should be further explained that one way to implement the first review delay risk probability in this embodiment is as follows:

[0065] Based on the comprehensive risk assessment score and declaration priority of the declared object, a cost matrix is ​​constructed using a multi-objective optimization matching algorithm combined with the real-time load data and skill proficiency of each sub-node on the review side to obtain an initial matching review sub-node. Simultaneously, a Bayesian algorithm is used to probabilistically update the delay records of the sub-node's historical processing of similar customs declarations, combining historical delay rates and current load, to obtain a predicted first-inspection delay risk probability. The multi-objective optimization matching algorithm is preferably the Hungarian algorithm; real-time load data includes, but is not limited to, the current number of queued tasks and average processing time.

[0066] An intelligent generation subnode is used to obtain a customs declaration form for the corresponding declared object based on the declaration text parsing space and the declaration image parsing space that have been initially reviewed for compliance, using a text generation algorithm in combination with a customs declaration form template corresponding to the declared object. A customs declaration compliance review is performed based on the generated customs declaration form in combination with a preset customs declaration compliance review rule base to obtain a compliant customs declaration form. Furthermore, the customs declaration compliance review rule base in this embodiment is constructed by those skilled in the art based on existing customs declaration standards in combination with a distributed database, and will not be described in detail here. The text generation algorithm is preferably a pre-trained Bert model.

[0067] The distributed dispatch subnode is used to allocate the compliant customs declaration to the corresponding review subnode based on the compliant customs declaration, the corresponding initial matching review subnode, and the load information fed back in real time by the corresponding initial matching review subnode, through a distributed reinforcement model. It should be further explained that the distributed reinforcement model in this embodiment is constructed by a distributed framework and the SAC algorithm.

[0068] It should be noted that the trigger information corresponding to the distributed reinforcement model in this embodiment includes:

[0069] The detailed implementation process of the in-depth triggering information of the basic attributes of the customs declaration is as follows: when the processing difficulty of the customs declaration is marked as high complexity, and the criteria for determining this level include the number of materials requiring special qualification review exceeding three, and the average processing time of similar declarations in history exceeding 50% of the industry standard, the distributed reinforcement model is triggered to start the emergency assessment process, giving priority to matching the review sub-nodes with special qualification review experience;

[0070] If the danger level of the customs declaration is judged to be high-risk, and the system identifies that the declared goods belong to the highly toxic or explosive categories specified in the list of hazardous chemicals, or contain radioactive material identification, the model will immediately trigger a special dispatch channel for high-risk customs declarations to ensure that the customs declaration is handled by an audit team with dangerous goods audit qualifications and zero errors in the past three months; when the comprehensive risk assessment score of the customs declaration is in an extremely high range, and the user credit score in the score composition is lower than the basic threshold and the danger score of the declared target exceeds the preset danger score threshold, the distributed reinforcement model will forcibly intervene and allocate resources for a second review of the customs declaration, giving priority to senior auditors with skill proficiency above proficient and an audit accuracy rate above 95%.

[0071] The resource status of the audit end dynamically triggers information. The specific implementation process is as follows: when the real-time task backlog of a certain audit sub-node exceeds 1.5 times of its regular processing capacity, and the proportion of high-priority customs declarations in the current task queue exceeds 40%, and the node is in full-load working state for 2 consecutive hours, and the system detects that its processing efficiency has dropped by more than 20%, the distributed reinforcement model is triggered to reallocate subsequent customs declarations, and new tasks are preferentially assigned to audit sub-nodes with a current load rate of less than 30% and rich experience in processing customs declarations of the same type; if the skill proficiency of a certain audit sub-node in a specific field (such as medical device customs declaration audit) is lower than the system average, and the customs declaration to be assigned is The single audit involves complex classification problems in this field. At the same time, there are idle personnel with advanced skill proficiency ratings in other audit sub-nodes. The distributed reinforcement model automatically triggers the cross-node allocation mechanism to allocate the customs declaration form to a more suitable audit sub-node and simultaneously updates the node load status; when the load difference between multiple audit sub-nodes exceeds the balance threshold set by the system, for example, the difference in task volume between the highest-load node and the lowest-load node exceeds 50%, and the duration reaches 1 hour, and there are auditors who can undertake the task at the low-load node, the distributed reinforcement model starts the load balancing algorithm and re-dispatches the unassigned customs declaration forms in transit to ensure that the load of each node is balanced.

[0072] The pre-allocation stage accurately triggers information. The specific implementation process is as follows: if the first review delay risk probability predicted by the intelligent pre-allocation sub-node exceeds the benchmark threshold, and the delay caused by the mismatch of auditor skills accounts for more than 60% of the risk structure, and the comprehensive risk assessment score of the customs declaration form is in the medium or high risk range, the distributed reinforcement model will immediately terminate the original allocation plan and re-select audit resources with a higher matching degree for allocation based on the real-time skill map and load situation of the audit sub-node. When the load status of the pre-assigned initial review sub-node changes significantly due to a sudden surge in tasks within 5 minutes after the allocation instruction is issued, for example, the load rate rises from 40% to 70%, and the average remaining time of the node's current processing tasks exceeds 1.2 times the expected processing time requirement for the new customs declaration, the distributed reinforcement model automatically triggers the reallocation process and transfers the customs declaration to a backup review sub-node with stable load and corresponding processing capacity; if the review error rate of the review sub-node involved in the pre-assignment plan for the same type of customs declaration exceeds the system tolerance limit in the past week, and the comprehensive risk assessment score of the current customs declaration to be allocated is higher than the average level of this type, the distributed reinforcement model will force the replacement of the review sub-node, and give priority to nodes with high historical review accuracy and fast processing speed for allocation.

[0073] It should be noted that the implementation logic of the abnormal depth trigger mechanism of the audit process is: when the difference between the second review delay risk probability and the first review delay risk probability monitored by the monitoring sub-node exceeds the preset delay risk judgment threshold, and the difference is caused by the mismatch of the audit sub-node skills and excessive load, and the comprehensive risk assessment score of the corresponding customs declaration form is in the high-risk range, the distributed reinforcement model will activate the emergency intervention mechanism; specifically, it includes: reallocating the customs declaration form to the audit sub-node with higher skill matching; dynamically evaluating and optimizing the load of the original assigned node through the minimum distance load optimization algorithm to reduce the probability of recurrence of similar abnormalities.

[0074] If, during the review process, the same review sub-node has two consecutive declaration allocation exceptions due to insufficient skills, and the comprehensive risk assessment scores of the abnormal customs declarations are higher than the system average, the distributed reinforcement model will trigger a skill training warning for the review sub-node and strictly restrict the subsequent customs declaration allocation of the node, giving priority to low-complexity tasks until its skill level passes the assessment and meets the standards; when a customs declaration is reallocated due to a declaration allocation exception during the review process, and the same exception occurs again (such as the delay risk is too high due to skill mismatch), and the customs declaration has been reallocated more than twice, the model will mark the customs declaration as a difficult case and automatically submit it to the expert review team for processing, and conduct an in-depth review and optimization of the entire allocation process.

[0075] The review sub-node is used to review the assigned customs declaration form;

[0076] It should be further explained that the audit end in this embodiment also includes an audit sub-node indication diagram, and the specific construction process is as follows:

[0077] By collecting the real-time load data, skill proficiency scores, processing frequency and successful processing frequency of customs declarations of different hazard levels of each audit sub-node (historical data is statistically analyzed by hazard level), and integrating auxiliary information such as the node's geographical location and the efficiency of the team to which it belongs, after data standardization preprocessing, the audit sub-nodes are used as graph vertices, and the correlation between the nodes in collaborative processing of customs declarations is used as the edge. A topological structure is constructed using a graph convolutional network or a force-directed graph algorithm, in which the node size maps the load, the color gradient represents the skill proficiency, the thickness of the edge is positively correlated with the frequency of joint processing, and the edge weight is associated with the probability of successful processing. Finally, a dynamically updated audit sub-node indicator map is generated; the node load balancing status, skill matching degree and hazard level processing capacity distribution are intuitively presented; real-time load data includes but is not limited to the task backlog and processing time; skill proficiency scores are obtained based on professional qualifications, historical audit accuracy and other parameters combined with a comprehensive fuzzy algorithm;

[0078] The monitoring sub-node is used to monitor the audit progress of each audit sub-node in real time, monitor and identify the audit anomaly type of the corresponding declaration form of the declared object, and feed back the monitoring information to the feedback sub-node;

[0079] The feedback sub-node is used to provide differential feedback based on the identified audit anomaly type. Specifically, if the audit anomaly type is incorrect or unqualified declaration information, it is determined to be an abnormal declaration information. Based on the abnormal declaration information, the preset traceability feature code in the distributed connection in the declaration feedback adjustment chain is triggered, and the corresponding abnormality of the declared target object is directly fed back to the corresponding declaration end through the distributed connection to issue an abnormality warning;

[0080] If the audit exception type is that the difference between the second review delay risk probability and the first review delay risk probability of the corresponding audit sub-node monitored by the monitoring sub-node is greater than the preset risk limit threshold, it is determined to be a declaration assignment exception, and the declaration assignment exception triggers the preset traceability feature code in the distributed connection in the declaration feedback adjustment chain, and the corresponding declaration assignment exception is fed back to the intelligent pre-assignment sub-node for allocation warning. According to the load information monitored by the declaration end and the difference between the second review delay risk probability and the first review delay risk probability, the audit sub-node information initially predicted and assigned by the intelligent generation sub-node is adjusted, and the adjusted audit sub-node is used to re-audit the customs declaration business information of the declared target object with the declaration assignment exception until the difference between the second review delay risk probability and the first review delay risk probability is less than or equal to the preset risk limit threshold.

[0081] The intelligent dispatching and collaborative processing system for customs declarations in this application builds a full-link intelligent audit system through multi-module collaboration and blockchain technology, effectively solving the technical bottlenecks of the existing system. The business declaration module uses the Transformer-based multimodal parsing algorithm combined with OCR and Faster R-CNN to deeply analyze the declared text and image information, extract key entities and risk characteristics such as commodity names and HS codes, fill in the missing dimensions of the dispatched data, and provide accurate basic data for subsequent audits. The risk assessment and initial review sub-nodes use CNN, AHP algorithms and rule engines to quantify the risk score and credit rating, dynamically verify the adequacy of materials, ensure the compliance of declarations and mark the risk level, and realize hierarchical management and control of the audit process. The priority assessment and intelligent pre-allocation sub-nodes use the hierarchical analysis method, queuing theory model and Bayesian algorithm to comprehensively process difficulty, overdue risk, load status and skill proficiency, build a multi-dimensional matching mechanism, improve the matching accuracy of customs declarations and audit resources, and reduce the mismatch rate of high-difficulty orders. The distributed reinforcement model combines real-time load, skill maps, and multi-dimensional trigger information to dynamically adjust dispatch strategies. Through cross-node deployment and load balancing algorithms, it achieves intelligent scheduling and resource optimization for audit tasks. The declaration feedback adjustment chain builds distributed connections based on blockchain. Through hash calibration, feature code generation, and smart contract evidence storage, it verifies the integrity of data transmission and manages traceability. It also utilizes the LSTM algorithm to predict transmission risks, forming a closed "initial review-feedback-review" loop. The audit sub-node indicator diagram uses graph algorithms to visualize load and skill distribution, assisting in resource scheduling decisions. By integrating multiple technologies, this system significantly improves the efficiency, accuracy, and security of customs declaration processing, providing intelligent and reliable customs clearance solutions for cross-border trade.

[0082] It should be further explained that the steps of constructing the distributed connection in the declaration feedback adjustment chain in this embodiment include:

[0083] Based on the declaration interaction information between all declaration terminals and the customs clearance cloud node, a declaration channel connection set is established; based on the dispatch interaction information between the customs clearance cloud node and each audit sub-node under the audit terminal, a dispatch channel connection set is established;

[0084] A first feedback channel connection set is established based on the feedback interaction information between the review sub-node corresponding to each declared object under the review end and the declaration end corresponding to the declared object. At the same time, a second feedback channel connection set is established based on the feedback interaction information between the review end and each sub-node under the customs clearance cloud node;

[0085] Based on the declaration end, the customs clearance cloud node and its corresponding sub-nodes, the review end and its corresponding sub-nodes, combined with the declaration channel connection set, the dispatch channel connection set, the first feedback channel connection set and the second feedback channel connection set, a declaration feedback adjustment chain is constructed through the blockchain algorithm;

[0086] Based on the comprehensive risk assessment score of the corresponding declared object at each reporting end of the declaration feedback adjustment chain and the hazard score of the declared object, combined with the preset encryption rules, the declared objects of different hazard levels are differentially encrypted on the chain;

[0087] Based on the declaration form ID, initial review timestamp, risk score, user credit score, number of declared objects and destination corresponding to each declared object, a random number generator and hash algorithm are combined to form a hash calibration sequence for each declared object;

[0088] Based on the hash calibration sequence of each declared target object and the number N of corresponding connection relationships of each declared target object, a random anchor point segmentation algorithm is used to obtain 2N hash calibration sequence segments.

[0089] Assign 2N hash calibration sequences to both ends of the N connection relationships corresponding to each declared target object, and obtain the packet loss probability of the child nodes corresponding to the two adjacent connection relationships of each declared target object and multiply it by the comprehensive risk assessment score of the declared target object to obtain the product of all child nodes corresponding to the declared target object;

[0090] Based on the product of all child nodes corresponding to the declared target object, a relationship connection site sequence of a length corresponding to the product size is randomly extracted from one end of the connection relationship connected to the corresponding child node through a random algorithm, and the extracted relationship connection site sequence is inserted into the hash calibration sequence segment configured at the corresponding end to form an identification feature code generation sequence;

[0091] Based on the identification feature code generation sequence corresponding to the adjacent ends of the two adjacent connection relationships of each declared target object, the identification feature codes corresponding to the two adjacent ends are generated using the SHA3-256 algorithm. The generated identification feature codes of the two adjacent ends are combined with the smart contract in the blockchain and stored on the chain in the child nodes corresponding to the two ends, thus obtaining a declaration feedback adjustment chain with traceability identification;

[0092] This embodiment realizes the safe, reliable and efficient flow of data in the whole process of customs declaration business by constructing a distributed connection of declaration feedback adjustment chain; based on interactive information such as declaration, dispatch, feedback, etc., a multi-channel connection set is constructed using blockchain algorithm to ensure the integrity and non-tamperability of data in each link, forming a reliable business collaboration network; by combining comprehensive risk assessment scores and hazard levels to implement differential on-chain encryption, the security of data with different sensitivity levels is guaranteed; hash calibration sequence is generated and segmented to give each declared target a unique and traceable identification, thereby enhancing data recognition; hash segments are combined with connection relationships, and the product calculation of packet loss probability and risk assessment score is introduced to make the reliability assessment of data transmission path more in line with actual network conditions; by randomly extracting relationship connection site sequences and generating identification feature codes, combined with smart contract on-chain storage, a strict inter-node data verification mechanism is established. Once a data transmission anomaly occurs, the problem connection and abnormal position can be accurately traced based on the feature code and hash calibration sequence. This technical system not only ensures the authenticity and integrity of data during transmission, but also significantly improves the risk prevention and control capabilities and exception handling efficiency of the customs declaration system through dynamic assessment and intelligent traceability, providing solid guarantees for efficient customs clearance of cross-border trade.

[0093] It should be further explained that one implementation method of the declaration feedback adjustment chain with traceability identification in this embodiment is:

[0094] A structured concatenation is performed based on each declared object's declaration form ID, initial review timestamp, risk score, user credit score, quantity of objects, and destination. A 128-bit random salt value is generated using a random number generator and appended to the end of the concatenated string. This salt is then encrypted using the SHA-256 hash algorithm to generate a 256-bit hash value, ultimately forming a unique hash sequence to ensure the uniqueness and collision resistance of each declared object's identification. For example, the structured concatenation is: "declaration form ID_timestamp_risk score_credit score_quantity_destination";

[0095] Based on the number N of connection relationships corresponding to each declared target object, the hash calibration sequence is segmented using a random anchor point segmentation algorithm: first, the number of segmentation anchor points is determined to be N-1. A uniformly distributed random number generator is used to generate N-1 non-repeating random anchor points within the hash sequence index range [1,255]. The hash sequence is segmented into N segments after being arranged in ascending order. Each segment is then reverse-joined and parity bits are added to ultimately obtain 2N hash calibration sequence segments, achieving structured segmentation and redundancy check of the hash sequence. The random anchor point segmentation algorithm is preferably constructed by combining the Fisher-Yates shuffle algorithm with the Mersenne Twister random number generator.

[0096] 2N hash calibration sequence segments are respectively assigned to both ends of the N connection relationships corresponding to the declared target object. Then, the packet loss probability of the adjacent sub-nodes of each connection relationship is obtained and multiplied by the comprehensive risk assessment score of the declared target object to obtain the product value of each sub-node. The product values ​​of all sub-nodes are then normalized to obtain the product matrix of all sub-nodes corresponding to the declared target object. It should be further explained that the packet loss probability in this embodiment is obtained by combining real-time network monitoring data with a communication algorithm and is in the range of [0, 1]. It should be noted that a specific implementation process of a packet loss probability in this embodiment includes:

[0097] Based on a combination of active detection and passive monitoring, the number of packets sent, received, and lost between adjacent sub-nodes in each connection relationship is collected in real time. The packet loss rate within the window is calculated using a sliding window algorithm, and the data is smoothed using an exponential moving average algorithm to eliminate the impact of sudden fluctuations. For cross-node transmission scenarios, distributed monitoring agents are used to collect statistical information on the network interfaces of each node and synchronized to the blockchain node through a consensus mechanism. Finally, the ratio of the number of lost packets to the total number of packets sent is normalized to the interval [0,1] to obtain the packet loss probability value between adjacent sub-nodes. Active detection is preferably ICMP ping or TCP / UDP heartbeat packets; passive monitoring is preferably a method of capturing network traffic and analyzing ACK responses; the window size of the sliding window algorithm is preferably 5-10 minutes; the distributed monitoring agent is preferably an Exporter algorithm based on Prometheus.

[0098] Based on the subnode product matrix, a roulette wheel random algorithm is used to extract a sequence of connection sites of corresponding length from the end of each connection relationship connected to the subnode: first, the subnode product is converted into a selection probability (e.g., the larger the product, the higher the probability), then a site in the connection relationship is randomly selected based on the probability distribution, and the integer part of the length equal to the subnode product × 10 is extracted. Finally, the extracted site sequence is inserted into the specified position of the hash calibration sequence segment configured on the corresponding end to form an identification feature code generation sequence, realizing the fusion of network connection characteristics and hash identification. In this embodiment, each site corresponds to a MAC address fragment or API interface identifier in the network protocol;

[0099] For each connection relationship of the declared target object, the identification feature code generation sequence of the adjacent two ends is taken, and a 256-bit identification feature code is generated respectively using the SHA3-256 algorithm. The generated feature code is then bound to the parameters of the evidence function in the blockchain smart contract. The feature code is stored on the chain in the corresponding child nodes connected at both ends through the consensus mechanism, and finally a declaration feedback adjustment chain with traceability identification is formed, realizing the unique identification of the data transmission path and blockchain evidence. The parameter binding of the evidence function includes but is not limited to metadata such as the declaration form ID, timestamp, and connection relationship ID;

[0100] It should be further explained that the intelligent pre-allocation sub-node in this embodiment is also used to: upon receiving abnormal feedback information, re-evaluate the transmission reliability of each connection relationship based on the product of all sub-nodes corresponding to the declared target object; if the product of the sub-nodes of a connection relationship is lower than a preset threshold, automatically activate the backup connection relationship and reallocate the customs declaration transmission path through the dispatch channel connection set;

[0101] It should be further explained that the declaration feedback adjustment chain with traceability identification in this embodiment is also used to establish a verification mechanism between each child node based on the identification feature code. When any child node receives data, it verifies the integrity and authenticity of the data transmission process by comparing the identification feature codes at both ends with the corresponding values ​​stored in the smart contract. If the verification fails, the abnormal feedback process is triggered. A specific implementation process includes:

[0102] A two-way verification mechanism is established between each child node based on the identification feature code. When any child node receives data, it automatically retrieves the identification feature codes stored at both ends, recalculates them using the SHA3-256 algorithm, and performs a triple comparison with the hash value stored in the smart contract. Specifically, the first comparison is with the sender's feature code, the second comparison is with the receiver's feature code, and the third comparison is with the full-link feature code chain. If any comparison result is inconsistent, a multi-level abnormality feedback process is immediately triggered: first, the data transmission status is marked as suspected tampering, second, the timestamp of the current transmission link and the node interaction log are locked, and finally, the abnormality signal is broadcast to the blockchain network through the distributed consensus mechanism.

[0103] The abnormality feedback process includes: tracing the specific connection relationship where the data transmission abnormality occurred based on the relationship connection site sequence in the identification feature code generation sequence; locating the specific position of the abnormal data in the transmission path by combining the hash calibration sequence segment of the declared target object; and feeding back the abnormality information and tracing results to the corresponding declaration end and customs clearance cloud node through the first feedback channel connection and the second feedback channel connection. A specific implementation process is as follows:

[0104] By parsing the relationship connection site sequence in the identification feature code generation sequence, combined with the transmission path hash chain stored in the blockchain, a two-way tracing algorithm is used to locate the specific connection segment where the anomaly occurred; at the same time, the hash calibration sequence segment of the declared target object is retrieved, and the data stream of the abnormal period is matched through the timestamp index to generate a four-dimensional anomaly report containing the abnormal node ID, transmission time, data segment hash, and risk level. The report is synchronously pushed to the declaration end risk control module and the customs clearance cloud node dispatch center in the form of encrypted data packets through the first feedback channel connection set and the second feedback channel connection set; the two-way tracing algorithm is specifically forward from the declaration end to the review end, and reverse from the abnormal node to the source node; the hash calibration sequence segment includes but is not limited to the declaration form ID encrypted segment, the risk score hash value, and the quantity destination mixed code;

[0105] Furthermore, in this embodiment, when the connection corresponding to the application target object fails, a specific implementation method is:

[0106] When the ratio of a certain connection relationship is lower than the industry standard threshold, the following three methods are selected through the simulation algorithm to select a backup connection;

[0107] First, triggering a preset standby connection relationship candidate pool; the preset standby connection relationship candidate pool in this embodiment is constructed by those skilled in the art based on historical connection data combined with a graph algorithm, and will not be further described here;

[0108] Second, the Dijkstra algorithm is used in combination with constraints to calculate the backup path. In this embodiment, the constraints prioritize links with a packet loss rate of less than 5% and a transmission delay of less than 200ms.

[0109] Third, the distributed reinforcement model dynamically adjusts the weight of the dispatch path based on the historical abnormal frequency, current load rate, and the skill matching of the auditors. When redistributing, the system automatically switches the customs declaration transmission path to the optimal backup path and sends a timestamped path change instruction to the relevant audit sub-nodes through the dispatch channel connection set.

[0110] This embodiment realizes the trusted traceability and secure transmission of the entire process of customs declaration data by constructing a declaration feedback adjustment chain with traceability identification; generates a unique hash calibration sequence based on declaration form ID, timestamp and other information, combines random salt value and SHA-256 algorithm to ensure the uniqueness and collision resistance of the identification, laying the foundation for data traceability. The random anchor point splitting algorithm structures the hash sequence and adds check bits, achieving both redundant verification and enhancing data tamper resistance. The product of packet loss probability and comprehensive risk assessment score is introduced to integrate network status and business risk into transmission path reliability assessment, enhancing the scientific nature of path selection. The roulette wheel algorithm extracts relational connection points and embeds them into hash segments, integrating network features and identifiers, providing a multi-dimensional basis for subsequent signature verification. The SHA3-256-generated identifier signature is combined with smart contract on-chain storage to establish a two-way verification mechanism, ensuring data transmission integrity through triple comparison. When the exception feedback process is triggered, the two-way tracing algorithm and hash calibration segments are used to locate the anomaly, generating a four-dimensional report containing node ID, timestamp, and other information, enabling accurate anomaly tracing and rapid response. When a connection fails, the path weight is dynamically adjusted using a backup connection candidate pool, the Dijkstra algorithm, and a distributed reinforcement model to ensure transmission continuity. This system, through the integration of multiple technologies, forms a closed loop from identifier generation, transmission verification, to exception handling, significantly improving the security, reliability, and exception handling efficiency of the customs declaration system.

[0111] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned specific embodiments are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are protected by the present invention.

Claims

1. Customs declaration intelligent dispatching and collaborative processing system, characterized by: include: Business declaration module, intelligent generation module, collaborative dispatch module and review feedback module; The business declaration module is used to obtain customs declaration business information and conduct an initial compliance review of the collected customs declaration business information in combination with the historical declaration credit score of the declaring user in combination with a preset initial review and evaluation model, and mark the customs declaration processing difficulty and risk level of those that pass the review, thereby obtaining a sequence of declaration information that has passed the initial review; The intelligent generation module generates a compliant customs declaration based on the declaration information sequence that has passed the initial review, combined with the configured customs declaration generation model, and combined with the priority pre-assigned label mark and the preset customs declaration generation compliance rule library, and encrypts and locks the generated compliant customs declaration using an encryption algorithm; The collaborative dispatch module is used to match and dispatch customs declarations based on the customs declarations uploaded to the blockchain, combined with the customs declaration processing difficulty, comprehensive risk assessment score, employee review proficiency, and risk level, using a configured distributed reinforcement dispatch model. The module also monitors the review progress and abnormal results of compliant customs declarations in real time through the review feedback module, and feeds back the monitoring results to the corresponding user declaration terminal in real time through a preset declaration feedback adjustment chain. The declaration feedback adjustment chain includes the declaration terminal, the customs clearance cloud node, and the review terminal. The declaration feedback adjustment chain is constructed based on the declaration interaction information between all the declaration terminals and the customs clearance cloud node, and the dispatch interaction information between the customs clearance cloud node and each review sub-node under the review terminal, combined with the blockchain algorithm. The audit end includes a feedback sub-node; the feedback sub-node is used to provide differential feedback based on the identified audit anomaly type. Specifically, if the audit anomaly type is incorrect or unqualified declaration information, it is determined to be an abnormal declaration information. Based on the abnormal declaration information, a preset traceability feature code in the distributed connection in the declaration feedback adjustment chain is triggered, and the abnormality corresponding to the declared target object is directly fed back to the corresponding declaration end through the distributed connection to issue an abnormality warning; If the audit exception type is that the difference between the second review delay risk probability and the first review delay risk probability of the corresponding audit sub-node monitored by the monitoring sub-node is greater than the preset risk limit threshold, it is determined to be a declaration assignment exception, and the preset traceability feature code in the distributed connection in the declaration feedback adjustment chain is triggered according to the declaration assignment exception, and the corresponding declaration assignment exception is fed back to the intelligent pre-assignment sub-node for allocation warning, and according to the load information monitored by the declaration end and the difference between the second review delay risk probability and the first review delay risk probability, the audit sub-node information initially predicted and assigned by the intelligent generation sub-node is adjusted, and the adjusted audit sub-node is used to re-audit the customs declaration business information of the declared target object with the declaration assignment exception, until the difference between the second review delay risk probability and the first review delay risk probability is less than or equal to the preset risk limit threshold.

2. The intelligent dispatching and collaborative processing system for customs declarations according to claim 1, characterized in that: The customs clearance cloud node includes a risk assessment sub-node; The declaration terminal is used to collect user declaration information and user historical declaration abnormality information and obtain declaration text analysis space and declaration image analysis space in combination with the configured multimodal analysis algorithm; The risk assessment sub-node is used to obtain the corresponding declared target object hazard score and the corresponding assessment accuracy probability based on the declared text analysis space and the declared image analysis space combined with the preset hazard level expert assessment, and at the same time obtain the corresponding user credit score and the corresponding assessment accuracy rate through a comprehensive assessment algorithm combined with the user's historical declared abnormal information and the user's historical credit score.

3. The intelligent dispatching and collaborative processing system for customs declarations according to claim 2, characterized in that: The customs clearance cloud node also includes an initial review sub-node; The initial review sub-node is used to obtain the corresponding user's comprehensive risk assessment score through a weighted average algorithm based on the declared target object's hazard score and the corresponding assessment accuracy probability, the corresponding user's credit score and the corresponding assessment accuracy rate, and to determine whether the current materials are sufficient and compliant based on the comprehensive risk assessment score, the user's currently submitted declaration materials information, and the preset declaration information criteria for the declared target object. If sufficient, the user's currently submitted declaration materials information is input into the priority assessment sub-node and marked with the corresponding comprehensive risk assessment score and declared target object hazard score; If it is insufficient or non-compliant, the supplementary or modified materials will be directly rejected until the corresponding application material requirements under the corresponding comprehensive risk assessment score are met, and the supplementary materials will be used to update the application text parsing space and application image parsing space.

4. The intelligent dispatching and collaborative processing system for customs declarations according to claim 3, characterized in that: The customs clearance cloud node also includes a priority evaluation sub-node and an intelligent pre-allocation sub-node; the audit terminal also includes an audit sub-node and a monitoring sub-node; The priority assessment sub-node is used to obtain the declaration processing difficulty of the corresponding declaration target object through the evaluation algorithm C1 based on the complexity of the historical declaration process of the declaration target object corresponding to the initial review compliance, the amount of required declaration materials, and the technical proficiency of the auditors assigned to the historical declaration review. At the same time, based on the declaration processing difficulty of the declaration target object, combined with the maximum effective declaration cycle length of the declaration target object, the corresponding audit end load of the declaration target object, and the skill proficiency of the auditors, the evaluation algorithm C2 is used to obtain the declaration review overdue probability risk value of the declaration target object, and the declaration review overdue probability risk value is used as the declaration priority and marked on the corresponding declaration target object; The intelligent pre-allocation sub-node is used to obtain the initial matching review sub-node and the predicted first review delay risk probability corresponding to the initial matching review sub-node through a matching algorithm combined with a Bayesian algorithm based on the comprehensive risk assessment score corresponding to the declared target object, the declaration priority, the load of the review sub-node corresponding to the review end, and the skill proficiency of the reviewer.

5. The intelligent dispatching and collaborative processing system for customs declarations according to claim 4, characterized in that: The clearance cloud node also includes intelligent generation sub-nodes and distributed dispatch sub-nodes; The intelligent generation sub-node is used to obtain the customs declaration form corresponding to the declared object based on the declaration text parsing space and the declaration image parsing space that have been initially reviewed for compliance, through a text generation algorithm combined with the customs declaration form template corresponding to the declared object, and perform a customs declaration compliance review based on the generated customs declaration form combined with a preset customs declaration compliance review rule library to obtain a compliant customs declaration form; The distributed dispatch sub-node is used to allocate the compliant customs declaration form to the corresponding review sub-node through a distributed reinforcement model based on the compliant customs declaration form combined with the corresponding initial matching review sub-node and the load information fed back in real time by the corresponding initial matching review sub-node.

6. The intelligent dispatching and collaborative processing system for customs declarations according to claim 5, characterized in that: The review sub-node is used to review the assigned customs declaration form; The monitoring sub-node is used to monitor the audit progress corresponding to each audit sub-node in real time and monitor and identify the audit abnormality type of the corresponding declaration form of the corresponding declaration target object, and feed back the monitored information to the feedback sub-node.

7. The intelligent dispatching and collaborative processing system for customs declarations according to claim 6, characterized in that: The steps of constructing the distributed connection in the feedback regulation chain include: Based on the declaration interaction information between all the declaration terminals and the customs clearance cloud node, a declaration channel connection set is established; based on the dispatch interaction information between the customs clearance cloud node and each audit sub-node under the audit terminal, a dispatch channel connection set is established; A first feedback channel connection set is established based on feedback interaction information between the review sub-node corresponding to each declared object under the review end and the declaration end corresponding to the declared object, and a second feedback channel connection set is established based on feedback interaction information between the review end and each sub-node under the customs clearance cloud node; Based on the declaration end, customs clearance cloud node and corresponding sub-nodes and the review end and corresponding sub-nodes combined with the declaration channel connection set, the dispatch channel connection set, the first feedback channel connection set and the second feedback channel connection set, a declaration feedback adjustment chain is constructed through the blockchain algorithm.

8. The intelligent dispatching and collaborative processing system for customs declarations according to claim 7, characterized in that: The step of constructing the distributed connection in the reporting feedback regulation chain also includes: Based on the comprehensive risk assessment score of the corresponding declared object at each reporting end of the declaration feedback adjustment chain and the hazard score of the declared object, combined with the preset encryption rules, the declared objects of different hazard levels are differentially encrypted on the chain; Based on the declaration form ID, initial review timestamp, risk score, user credit score, number of declared objects and destination corresponding to each declared object, a random number generator and hash algorithm are combined to form a hash calibration sequence for each declared object; Based on the hash calibration sequence of each declared target object and the number N of corresponding connection relationships of each declared target object, a random anchor point segmentation algorithm is used to obtain 2N hash calibration sequence segments.

9. The intelligent dispatching and collaborative processing system for customs declarations according to claim 8, characterized in that: The step of constructing the distributed connection in the reporting feedback regulation chain also includes: Assign 2N hash calibration sequences to both ends of the N connection relationships corresponding to each of the declared targets, and obtain the packet loss probability of the child nodes corresponding to the two adjacent connection relationships of each declared target and multiply it by the comprehensive risk assessment score of the declared target to obtain the product of all child nodes corresponding to the declared target; Based on the product of all child nodes corresponding to the declared target object, a relationship connection site sequence of a length corresponding to the product size is randomly extracted from one end of the connection relationship connected to the corresponding child node through a random algorithm, and the extracted relationship connection site sequence is inserted into the hash calibration sequence segment configured at the corresponding end to form an identification feature code generation sequence; Based on the identification feature code generation sequence corresponding to the adjacent ends of the two adjacent connection relationships of each of the declared targets, the identification feature codes corresponding to the two adjacent ends are generated through the SHA3-256 algorithm, and the generated identification feature codes of the two adjacent ends are combined with the smart contract in the blockchain and stored on the chain in the sub-nodes corresponding to the two ends, so as to obtain a declaration feedback adjustment chain with a traceability identification.

10. The customs declaration intelligent dispatching and collaborative processing system according to claim 9, characterized in that: The declaration feedback adjustment chain with traceability identification is also used to: establish a verification mechanism between each child node based on the identification feature code. When any child node receives data, it verifies the integrity and authenticity of the data transmission process by comparing the identification feature codes at both ends with the corresponding values ​​stored in the smart contract. If the verification fails, the abnormal feedback process is triggered; The abnormality feedback process includes: tracing the specific connection relationship where the data transmission abnormality occurs based on the relationship connection site sequence in the identification feature code generation sequence; locating the specific position of the abnormal data in the transmission path in combination with the hash calibration sequence segment of the declared target object; and feeding back the abnormal information and tracing results to the corresponding declaration end and customs clearance cloud node through the first feedback channel connection and the second feedback channel connection.

Images