Supply chain whole-process data tracing and collaboration system based on block chain

Abstract

The invention, which relates to the technical field of supply chain digital management, discloses a supply chain full-process data tracing and collaboration system based on a block chain, comprising a supply chain collaboration management platform which is in communication connection with the following modules: a twinning edge simulation monitoring module which is used for integrating a digital twinning and edge computing technology and is used for monitoring the twinning edge simulation; a real-time behavior simulation model is constructed at each supply chain node, field operation behaviors are dynamically monitored, and abnormity is identified in real time through time sequence mode analysis. According to the method, digital twinning and edge calculation are integrated at key nodes of a supply chain, an entity behavior simulation model is constructed, actual and expected operation behaviors are compared in real time, potential fraud and abnormal behaviors are identified, abnormal data uplink is directly blocked at the edge side, abnormal identification and blocking are preposed to a data acquisition stage, and the safety of the supply chain is improved. False, repeated or malicious operation data is prevented from polluting a block chain data source fundamentally, and authenticity, timeliness and integrity of uplink data are ensured.

Description

Technical Field

[0001] This invention relates to the field of digital supply chain management technology, specifically to a blockchain-based supply chain end-to-end data traceability and collaboration system. Background Technology

[0002] With increasing market competition and rising consumer demands for product quality and source transparency, businesses face pressure from all sides and urgently need to optimize their supply chain management. Data-driven decision support systems have become the core of modern supply chain management, enabling real-time monitoring and analysis of data at each stage to improve efficiency and reduce costs. Meanwhile, collaboration among logistics, production, sales, and other stages helps achieve optimal resource allocation. All participants in the supply chain, such as suppliers, manufacturers, and distributors, need to ensure information sharing through transparent data flow in order to respond quickly to market changes and customer needs.

[0003] For example, the product supply chain traceability system and method based on industrial blockchain in Chinese patent publication number CN118520511A mainly involves the core technologies of user roles, full-process operation behavior, full-cycle supply chain data flow supervision, and full-process status and process traceability methods in the product supply chain.

[0004] Existing technologies rely on blockchain data comparison to identify tampering, but can only perform post-event verification after data is uploaded to the blockchain. They cannot identify abnormal behavior in real time during the data collection stage. In on-site processes such as logistics and transportation, there are fraudulent activities such as vehicles falsely reporting data by repeatedly weighing to obtain subsidies. Once such abnormal data is uploaded to the blockchain, it will pollute the entire traceability system and undermine the authenticity and timeliness of the data. Furthermore, even after blocking such abnormal data, due to the asynchronous updates and isolated management of data from multiple roles and stages in the supply chain, such as warehousing, transportation, and finance, it is difficult to dynamically build collaborative traceability paths based on real-time and reliable data. This results in insufficient transparency in the order fulfillment process, low efficiency of cross-stage collaboration, and an inability to support real-time and reliable supply chain decision-making and status tracking. Therefore, this paper proposes a blockchain-based supply chain end-to-end data traceability and collaboration system. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention is implemented through the following technical solution: a blockchain-based supply chain end-to-end data traceability and collaboration system, including a supply chain collaboration management platform, wherein the supply chain collaboration management platform has the following communication connections:

[0006] The twin edge simulation monitoring module is used to integrate digital twin and edge computing technologies to build real-time behavior simulation models at each supply chain node, dynamically monitor on-site operation behavior, and identify anomalies in real time through time-series pattern analysis to ensure the authenticity and reliability of the data source.

[0007] The smart contract interception and purification module, based on the results of twin edge simulation monitoring, automatically triggers smart contracts to intercept and clean abnormal on-site operation data in real time, obtains a clean blockchain dataset, prevents abnormal data from being uploaded to the chain, and ensures the integrity and timeliness of blockchain data.

[0008] The multi-agent collaboration engine is used to build multi-role intelligent agents covering warehousing, transportation and finance based on clean blockchain datasets. It combines the real-time status of each link in the supply chain to adjust the data synchronization strategy, realize efficient data synchronization across links, and improve order fulfillment transparency and collaborative response speed.

[0009] The dynamic path optimization module is used to continuously optimize the data traceability path and collaborative logic in the data synchronization strategy through reinforcement learning, adapt to the dynamic changes in the supply chain, support real-time and reliable supply chain status tracking and decision-making, and improve the efficiency of the entire chain operation.

[0010] The privacy and security collaborative traceability module is used to introduce a differential privacy mechanism in cross-link supply chain data sharing. It adds controllable noise to sensitive business fields, ensuring data availability and traceability consistency while protecting the business privacy of all participants.

[0011] Preferably, the twin edge simulation monitoring module includes a dynamic behavior simulation unit and a temporal anomaly identification unit;

[0012] The dynamic behavior simulation unit is used to combine digital twin and edge computing technologies to deploy entity behavior simulation models at edge nodes, simulate and compare actual on-site operation behavior data with expected operations in real time, identify behavior deviations and potential fraud, block malicious behaviors such as false weighing and repeated reporting from the source, and prevent pollution of blockchain data sources.

[0013] The time-series anomaly identification unit, based on time-series data analysis, monitors the temporal consistency of on-site operational behaviors in real time, automatically triggers an early warning mechanism, and realizes dynamic anomaly detection for vehicle entry and exit, weighing, loading and other links, thereby improving the real-time protection capability during the data acquisition stage.

[0014] Preferably, the dynamic behavior simulation unit performs the following steps:

[0015] Edge computing devices are deployed at each node of the supply chain. By combining physical parameters of the entity with business process models, a digital twin-based entity behavior simulation model is constructed. The entity behavior simulation model includes the real-time mapping relationship of vehicle positioning, weighing sensor data flow, and loading and unloading action sequence, so as to realize real-time, high-fidelity simulation mapping of on-site operation behavior and ensure the physical consistency between simulation and actual environment.

[0016] Based on the entity behavior simulation model, the actual on-site operation behavior data from the sensor cluster is received in real time, and the expected operation simulation is executed in parallel at the edge. The deviation between the actual behavior and the simulated behavior is calculated by the behavior trajectory comparison algorithm, realizing millisecond-level real-time comparison and deviation calculation, providing accurate quantitative basis for anomaly identification.

[0017] When the deviation exceeds the preset deviation threshold, it is determined to be an abnormal on-site operation behavior, an behavior deviation report is automatically generated, and the abnormal event is marked as a potential fraudulent behavior. At the same time, the transmission of the original data corresponding to the abnormal operation to the blockchain platform is blocked, realizing automatic identification of anomalies and real-time interception of data, preventing false data from polluting the blockchain from the source.

[0018] Preferably, the timing anomaly identification unit performs the following steps:

[0019] It receives actual on-site operational behavior data, extracts multi-dimensional time-series features including vehicle entry and exit time series, continuous weighing value series, and loading operation intervals, and constructs high-precision time-series feature vectors in real time to provide a standardized input basis for abnormal pattern recognition, ensuring the accuracy and consistency of data analysis.

[0020] By applying time series pattern recognition algorithms, the multi-dimensional time series features are analyzed in real time to detect whether there are time series patterns that violate business logic, including repeated weighing in a short period of time, reversed loading and weighing order, and abnormal vehicle dwell time. Violations are dynamically identified and located, effectively preventing data fraud and process abnormalities, and improving the compliance and credibility of on-site operations in the supply chain.

[0021] When an abnormal timing pattern is detected, a real-time warning signal is automatically triggered, and the warning event, associated timestamp, and operation node identifier are sent to the smart contract interception and purification module to initiate the data interception process. This enables the immediate reporting of abnormal events and process blocking, preventing contaminated data from being uploaded to the blockchain and ensuring the cleanliness and timeliness of the blockchain data source.

[0022] Preferably, the smart contract interception and purification module includes a contract triggering interception unit and a data cleaning and archiving unit;

[0023] The contract triggering interception unit is used to automatically execute the smart contract, mark the abnormal data, and block the abnormal data from being written to the blockchain when abnormal on-site operation behavior is detected, so as to achieve real-time purification of data before it is put on the chain and avoid contamination of the post-event traceability system.

[0024] The data cleaning and archiving unit is used to clean, mark and archive the intercepted abnormal data, record the abnormality type, time and operator information, form an abnormal data log, and integrate it to obtain a clean blockchain dataset, which supports subsequent auditing and behavior tracing and improves system traceability.

[0025] Preferably, the contract triggering interception unit performs the following steps:

[0026] The system monitors abnormal event signals from the twin edge simulation monitoring module. When an abnormal warning is received, it automatically calls the interception smart contract pre-installed on the blockchain platform to achieve an immediate response to abnormal signals, ensuring that warning information is not lost or delayed, and improving real-time interception efficiency.

[0027] The interception smart contract is used to parse the content of abnormal events, determine the level of abnormality according to a predefined set of rules, and mark the data records to be uploaded to the chain generated by the current operation, assigning them a pending review status. Based on standardized rules, the abnormality level is accurately determined, providing a clear basis for subsequent classification and processing, and enhancing the targeting and reliability of the interception.

[0028] The interception logic in the smart contract is executed to prevent abnormal data records marked as pending audit from being written into a new block of the blockchain. At the same time, an interception log containing an abnormal summary, interception time, and triggering contract address is generated and stored in an off-chain database. This effectively blocks abnormal data from being uploaded to the chain and prevents pollution of the main chain data. In addition, a structured log is generated to ensure that the entire interception operation is traceable and auditable.

[0029] Preferably, the data cleaning and archiving unit performs the following steps:

[0030] Read the interception logs generated by the interception unit triggered by the contract from the off-chain database, extract the corresponding original abnormal data records, ensure the rapid location and full restoration of abnormal data, and provide complete input for subsequent cleaning;

[0031] Based on the anomaly type, the corresponding data cleaning rules are invoked to repair, remove, or label the original abnormal data, generate a clean data version, and associate it with the anomaly metadata (including anomaly type, occurrence time, related operator ID, edge node ID) to achieve accurate repair and removal of data anomalies and ensure the business compliance of clean data.

[0032] The clean data version, associated abnormal metadata, and cleaning process records are jointly archived in the off-chain audit database to form a structured abnormal data log. This log can be used for subsequent audit tracing. At the same time, the clean data version is synchronized to the blockchain platform and updated into a valid clean blockchain dataset that can be uploaded to the chain, thus building a fully traceable audit chain and ensuring the consistency between on-chain data and off-chain archives.

[0033] Preferably, the multi-agent collaborative engine performs the following steps:

[0034] Based on the clean blockchain dataset output by the smart contract interception and purification module, corresponding smart agents are instantiated for the warehouse management, transportation scheduling, and financial settlement processes. Each smart agent encapsulates the business logic and data access interface of its process, realizing the independent encapsulation and decoupling of the business logic of each process, and improving the modularity and maintainability of the system.

[0035] Each intelligent agent monitors the status update events related to its own link on the blockchain in real time, and dynamically adjusts the priority and frequency of data synchronization through negotiation protocols based on the overall performance progress of the supply chain. This enables near real-time synchronization of order status, inventory changes, transportation trajectory, and payment information across links, significantly reducing cross-link data synchronization delays and ensuring dynamic coordination and efficient response in the order fulfillment process.

[0036] By communicating and sharing states among intelligent agents, a global order fulfillment view is constructed, and key collaborative events are recorded on the blockchain to ensure the auditability and transparency of operations at each stage, forming a complete and trustworthy global traceability chain and providing all participants with tamper-proof records of collaborative operations.

[0037] Preferably, the dynamic path optimization module performs the following steps:

[0038] Using historical collaborative data and real-time status of the entire supply chain as input, a reinforcement learning environment is constructed. The state space includes data synchronization delay, resource load, and order urgency at each stage, while the action space includes the selection of data traceability paths and the adjustment of synchronization strategy parameters. This enables comprehensive and dynamic perception and quantitative modeling of the supply chain's operational status, providing precise input for intelligent decision-making.

[0039] The agent explores and performs actions in the reinforcement learning environment, evaluates the effectiveness of the actions through a reward function (rewarding efficient synchronization and low-latency tracing), and continuously iterates and optimizes its policy network to learn the optimal strategy that adapts to dynamic supply chain scenarios, including optimal data tracing and collaborative paths. During the training process, it automatically learns collaborative paths and parameter combinations that can significantly shorten data synchronization time and improve tracing efficiency.

[0040] The optimal strategy obtained from training is deployed to the multi-agent collaborative engine to guide each agent to dynamically adjust the data flow and processing logic in complex scenarios. This enables real-time and accurate tracking of the supply chain status and efficient collaborative decision support, ensuring that the data synchronization latency across the entire supply chain is consistently below 2 seconds and that resource load remains balanced, thereby improving the real-time performance and reliability of overall collaborative decision-making.

[0041] Preferably, the privacy and security collaborative tracing module performs the following steps:

[0042] Before data sharing across different stages, sensitive business fields in the data are identified, including specific transaction amounts, supplier costs, and inventory details of specific products. Sensitive field identification and classification can dynamically adjust privacy protection strategies, enabling precise isolation and classification management of key information.

[0043] By applying a differential privacy mechanism, Laplace noise or Gaussian noise that meets a preset privacy budget is added to the sensitive business fields to generate perturbed data that satisfies ε-differential privacy. The noise injection can resist external inference and correlation attacks, ensuring data availability while effectively preventing privacy leakage risks. The perturbed data is used for cross-agent collaborative tracing and global analysis, ensuring that the logical consistency of the supply chain traceability chain and the availability of key business insights are maintained while protecting the business privacy of all participants. The perturbed data supports cross-chain tracing and analysis tasks, ensuring the consistency of business logic and the validity of statistical conclusions, achieving a balance between privacy and business.

[0044] This invention provides a blockchain-based end-to-end supply chain data traceability and collaboration system. It offers the following advantages:

[0045] (i) This blockchain-based supply chain end-to-end data traceability and collaboration system integrates digital twins and edge computing at key nodes in the supply chain to build entity behavior simulation models, compare actual and expected operational behaviors in real time, identify potential fraud and abnormal behavior using multi-dimensional time series analysis, and directly block abnormal data from being uploaded to the chain at the edge. This brings the identification and blocking of abnormalities to the data collection stage, fundamentally preventing false, duplicate or malicious operational data from polluting the blockchain data source and ensuring the authenticity, timeliness and integrity of the data uploaded to the chain.

[0046] (II) This blockchain-based supply chain end-to-end data traceability and collaboration system constructs a multi-role intelligent agent collaboration engine based on a clean blockchain dataset. Each intelligent agent achieves near real-time data synchronization across links through an event-driven architecture and smart contract rules. By dynamically adjusting the data synchronization priority and push frequency, it achieves efficient flow and consistency of order status, inventory changes, transportation trajectory and payment information, controls data delay at each link to the second level, significantly improves the transparency of the order fulfillment process and the speed of end-to-end collaborative response, and supports real-time and reliable supply chain status tracking and decision-making.

[0047] (III) This blockchain-based supply chain end-to-end data traceability and collaboration system, by introducing reinforcement learning technology for dynamic path optimization, can continuously learn and optimize data traceability paths and synchronization strategy parameters based on multi-dimensional states such as real-time synchronization delay, resource load, and order urgency. The trained strategy can guide the agents in each link to dynamically adjust the data processing logic in complex business scenarios, achieve resource load balancing and traceability efficiency improvement, enable the system to adapt to dynamic changes in the supply chain, and enhance overall operational efficiency and decision-making intelligence.

[0048] (iv) This blockchain-based supply chain end-to-end data traceability and collaboration system integrates a differential privacy mechanism in cross-link data sharing. By automatically identifying sensitive business fields and injecting noise in accordance with the privacy budget, it effectively protects the sensitive business information of all participants while ensuring data availability and traceability continuity. The disturbed data supports secure collaborative analysis and global traceability. All privacy operations are recorded in audit logs, realizing compliance, controllability and auditability of the data sharing process, and balancing the contradiction between business collaboration needs and commercial privacy protection. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the workflow of a blockchain-based supply chain end-to-end data traceability and collaboration system according to the present invention.

[0050] Figure 2 This is a data flow diagram of a blockchain-based supply chain end-to-end data traceability and collaboration system according to the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a blockchain-based supply chain end-to-end data traceability and collaboration system, including a supply chain collaboration management platform, which has the following communication modules:

[0053] The twin edge simulation monitoring module is used to integrate digital twin and edge computing technologies to build real-time behavior simulation models at each supply chain node, dynamically monitor on-site operation behavior, and identify anomalies in real time through time-series pattern analysis to ensure the authenticity and reliability of the data source. The twin edge simulation monitoring module includes a dynamic behavior simulation unit and a time-series anomaly identification unit.

[0054] The dynamic behavior simulation unit combines digital twin and edge computing technologies to deploy entity behavior simulation models at edge nodes. It performs real-time simulations and comparisons of actual on-site operational behavior data with expected operations, identifying behavioral deviations and potential fraud. This prevents malicious behaviors such as false weighing and duplicate reporting from the source, thus preventing contamination of the blockchain data source. Edge computing devices are deployed at each supply chain node, combining entity physical parameters and business process models to construct entity behavior simulation models based on digital twins. These models include real-time mapping relationships between vehicle positioning, weighing sensor data streams, and loading / unloading action sequences, achieving real-time, high-fidelity simulation mapping of on-site operational behavior and ensuring accurate correlation between simulation and actual operation. The physical consistency of the environment is achieved by using an entity behavior simulation model to receive real-time data on actual on-site operation behavior from a sensor cluster. The expected operation simulation is executed in parallel at the edge. The deviation between the actual behavior and the simulated behavior is calculated through a behavior trajectory comparison algorithm, achieving millisecond-level real-time comparison and deviation calculation. This provides a precise quantitative basis for anomaly identification. When the deviation exceeds a preset deviation threshold, it is determined to be an abnormal on-site operation behavior, and a behavior deviation report is automatically generated. The abnormal event is marked as a potential fraudulent behavior, and the transmission of the original data corresponding to the abnormal operation to the blockchain platform is blocked. This achieves automatic anomaly identification and real-time data interception, preventing false data from polluting the blockchain from the source.

[0055] The specific work involves deploying edge devices with edge computing capabilities, i.e., industrial IoT gateways, at key nodes in the supply chain (warehouse loading and unloading areas, logistics transfer stations, and factory shipping outlets). These devices must be equipped with at least a 4-core CPU, 8GB of RAM, and at least 128GB of solid-state storage, supporting industrial protocol access such as ModbusTCP and OPCUA. Based on actual physical parameters and business process rules, a simulation model of entity behavior will be constructed, including a vehicle positioning model, a weighing sensor data flow model, and a loading and unloading action sequence model. The vehicle positioning model integrates a GPS / BeiDou positioning module, with a positioning accuracy requirement of ≤2 meters and a sampling frequency of at least [missing information]. The data flow model for the weighing sensor is set at 1Hz, based on the weighing characteristics of the weighbridge, with a range of 0-100 tons, a graduation value of 50kg, and a data reporting interval of 5 seconds. The loading and unloading action sequence model needs to define the theoretical time window and sequential logic of each action node (forklift positioning, pallet clamping, and cargo placement) according to the standard operating procedure (SOP). Each parameter needs to be calibrated on-site and approved by process documents to ensure that the simulation model matches the actual operating environment. The edge device continuously receives the actual operation data uploaded by the sensor cluster, including real-time vehicle coordinates, instantaneous weighing values, loading and unloading action trigger signals, etc., and synchronously runs the expected operation simulation thread on the edge side. The theoretical behavioral trajectory is calculated in parallel at millisecond intervals (20ms recommended). A multidimensional Euclidean distance metric is used to compare the behavioral trajectory and calculate the comprehensive deviation between the actual sequence and the simulated sequence in each dimension. The deviation thresholds are set according to business fault tolerance requirements: the vehicle positioning trajectory deviation threshold is set to 10 meters, the weighing data fluctuation threshold is set to ±1.5% of the range, and the loading and unloading action timing deviation threshold is set to ±20% of the standard duration. All threshold parameters are determined through statistical analysis of historical normal operation data and can be dynamically adjusted through the management interface. When the deviation in any dimension exceeds the preset deviation threshold, it is immediately judged as an abnormal on-site operation. Edge devices automatically generate structured behavior deviation reports, recording the anomaly type, deviation value, occurrence timestamp, associated device number, and operator ID. These reports are then uploaded to the supply chain collaborative management platform via an encrypted link. Simultaneously, the anomaly is marked as a potential fraudulent activity, and the edge device immediately initiates a data interception protocol to block the transmission of the original data corresponding to the batch of abnormal operations to the blockchain platform. The interception mechanism employs a whitelist filtering and content discarding strategy to ensure that abnormal data does not enter the subsequent data cleaning and on-chain process. All interception operations generate audit logs, recording the interception time, data packet characteristics, and triggering rules for subsequent traceability and verification.

[0056] The time-series anomaly detection unit, based on time-series data analysis, monitors the temporal consistency of on-site operational behaviors in real time, automatically triggering an early warning mechanism. It achieves dynamic anomaly detection for vehicle entry / exit, weighing, and loading, enhancing real-time protection capabilities during the data acquisition phase. It receives actual on-site operational behavior data, extracts multi-dimensional time-series features including vehicle entry / exit time sequences, continuous weighing value sequences, and loading operation intervals, and constructs high-precision time-series feature vectors in real time. This provides a standardized input basis for anomaly pattern recognition, ensuring the accuracy and consistency of data analysis. Time-series pattern recognition algorithms are applied to analyze the multi-dimensional time-series features. Real-time analysis is performed to detect any time-series patterns that violate business logic, including repeated weighing within a short period of time, reversed loading and weighing order, and abnormal vehicle dwell time. Violations are dynamically identified and located to effectively prevent data fraud and process anomalies, improve the compliance and credibility of supply chain operations, and automatically trigger real-time warning signals when abnormal time-series patterns are identified. The warning event, associated timestamp, and operation node identifier are sent to the smart contract interception and purification module to initiate the data interception process, realize the immediate reporting of abnormal events and process blocking, prevent contaminated data from being uploaded to the blockchain, and ensure the cleanliness and timeliness of the blockchain data source.

[0057] The specific work involves: using edge computing devices deployed at key nodes in the supply chain to receive real-time operational data from sensor clusters. The data interface uses Modbus TCP or OPCUA industrial protocols, collecting vehicle GPS / BeiDou positioning coordinates at a frequency of at least 1Hz. Continuous weight values ​​from weighbridge sensors are obtained via RS485 bus, with a range of 0-100 tons, a division value of 50kg, and a reporting interval of 5 seconds. Loading and unloading signals are triggered by PLC digital output nodes, recording timestamps for key actions such as forklift positioning, pallet picking, and cargo placement. The edge devices have a built-in time-series feature extraction engine to perform sliding window (window length 3) processing on the raw operational data. The system is segmented (0 seconds, 5-second step) to extract vehicle entry / exit time series, continuous weighing data series, and loading operation interval series (calculating the time difference between adjacent actions). All time series features are normalized and labeled with data source node number, operator ID, and equipment identifier to form a standardized time series feature vector. Based on the extracted multidimensional time series features, a lightweight time series pattern recognition algorithm (based on sequence alignment using Dynamic Time Warping (DTW)) is applied for real-time analysis. A normal time series pattern template is preset for the supply chain business logic: vehicle entry / exit time must meet the scheduled deviation of ±5 minutes, weighing data fluctuation should not exceed ±0.5% of the range within 5 consecutive seconds, and the loading action sequence must follow forklift positioning → pallet clamping → cargo loading. The sequence of object placement → vehicle departure should be strictly followed, with the interval between adjacent actions within the 20-40 second range specified in the standard operating procedure. The system then calculates the similarity score between the actual sequence and the template sequence (threshold set at 0.85) to detect any timing patterns that violate business logic. These include anomalies such as the same vehicle being weighed repeatedly within 60 seconds with a weight difference of less than 50 kg, loading actions being triggered before the weighing completion signal, and vehicles remaining in the loading / unloading area for more than the preset limit of 15 minutes. When an abnormal timing pattern is detected, a three-level warning mechanism is automatically triggered, including a primary warning, a secondary warning, and a primary warning. A primary warning records minor anomalies with a deviation score below the threshold of 0.85, while a secondary warning indicates a reversed or repeated timing logic. For medium-risk events such as operation, advanced warnings are issued for high-risk behaviors such as vehicles staying for too long or repeated abnormalities. The warning signal includes anomaly type code, deviation value, occurrence timestamp, associated node identifier and operator ID, encapsulated as a JSON message, and pushed in real time to the smart contract interception and purification module of the blockchain platform through a TLS encrypted channel. At the same time, the edge device generates a warning log locally, records the feature vector, matching template and confidence score of the triggering algorithm, and initiates the data interception process. The original sensor data in the abnormal time sequence window is marked as pending review and its transmission to the blockchain data pool is suspended. All warning events and interception operations are synchronously written to the off-chain audit database to form a complete abnormal time sequence tracing chain.

[0058] The smart contract interception and purification module, based on the results of twin edge simulation monitoring, automatically triggers smart contracts to intercept and clean abnormal on-site operation behavior data in real time, obtains a clean blockchain dataset, prevents abnormal data from being uploaded to the chain, and ensures the integrity and timeliness of blockchain data. The smart contract interception and purification module includes a contract triggering interception unit and a data cleaning and archiving unit.

[0059] The contract-triggered interception unit is used to automatically execute smart contracts when abnormal on-site operations are detected, mark abnormal data, and block the writing of abnormal data to the blockchain. This achieves real-time purification of data before it is uploaded to the chain, preventing contamination of the post-event traceability system. It listens for abnormal event signals from the twin edge simulation monitoring module. When an abnormal warning is received, it automatically calls the interception smart contract pre-installed on the blockchain platform to achieve an immediate response to the abnormal signal, ensuring that the warning information is not lost or delayed, and improving the efficiency of real-time interception. It uses the interception smart contract to parse the content of abnormal events, judges the abnormal level according to the predefined rule set, and marks the data records to be uploaded to the chain generated by the current operation, assigning them a pending review status. It achieves accurate judgment of the abnormal level according to standardized rules, providing a clear basis for subsequent classification and processing, enhancing the targeting and reliability of interception. It executes the interception logic in the smart contract to prevent abnormal data records marked as pending review from being written to new blocks of the blockchain. At the same time, it generates an interception log containing an abnormal summary, interception time, and triggering contract address, which is stored in an off-chain database, effectively blocking abnormal data from being uploaded to the chain and preventing contamination of the main chain data. It also generates a structured log to ensure that the entire interception operation is traceable and auditable.

[0060] The specific tasks are as follows: Using an intercept smart contract with a specific address deployed on the blockchain platform, the system continuously monitors abnormal event signals from the twin edge simulation monitoring module. These signals are pushed in real-time via an encrypted communication channel in a structured data format, containing key fields such as anomaly type code, deviation value, timestamp, node identifier, and operator ID. When a blockchain node receives a valid abnormal event signal, it automatically verifies its digital signature and format compliance, triggering a call to the pre-built intercept smart contract. During the contract call, the abnormal event signal is passed as a transaction parameter, and the trigger time, source edge device number, and event sequence number are recorded to ensure traceability for each call. This process is based on the blockchain consensus mechanism, with all participating nodes synchronously verifying the authenticity of the event and call permissions to prevent malicious or repeated triggering. The intercept smart contract uses a multi-layered rule set to parse the abnormal event content and determine the anomaly level. The rule set is predefined according to business security policies, classifying anomalies into three levels: primary, intermediate, and advanced. Primary anomalies correspond to a deviation score between 0.80 and 0.85 with no timing logic errors; intermediate anomalies include reversed timing sequences or the same operation occurring within 60 seconds. Repeated occurrences; advanced anomalies encompass vehicle dwell times exceeding 15 minutes or more than 3 consecutive instances of abnormal behavior. After parsing the anomaly event, the contract automatically assigns an anomaly level based on the rule matching results and marks the associated data records to be uploaded to the blockchain (indexed by transaction hash) as pending review. The status identifier is written to the block temporary storage area in the form of metadata, including the anomaly level code, marking time, review timeout threshold (default set to 300 seconds), and the responsible person's role, ensuring that data cannot enter the formal on-chain process before review. When the smart contract executes the interception logic, it checks the status identifier of the data records to be uploaded to the blockchain; if it is pending review... If the record is not found, the transaction confirmation process of writing the record to the new block is immediately stopped, and the record is isolated to the off-chain buffer pool. At the same time, the contract automatically generates a structured interception log, which includes an anomaly summary, the precise timestamp of the interception, the triggering contract address, the associated block height, and the operator ID. This log is synchronously written to a dedicated table in the off-chain audit database through a secure interface. The database uses a time-series structure for storage and supports multi-dimensional retrieval by time range, node identifier, and anomaly level. All interception operations record the operation serial number and establish a two-way index with the original anomaly warning event to ensure traceability of the entire chain from warning to interception.

[0061] The data cleaning and archiving unit is used to clean, mark, and archive intercepted abnormal data, record the abnormality type, time, and operator information, form an abnormal data log, and integrate it to obtain a clean blockchain dataset. This supports subsequent auditing and behavior tracing, improves system traceability, reads the interception logs generated by the contract-triggered interception unit from the off-chain database, extracts the corresponding original abnormal data records, ensures rapid location and full restoration of abnormal data, provides complete input for subsequent cleaning, calls the corresponding data cleaning rules according to the abnormality type, repairs, removes, or marks the original abnormal data, generates a clean data version, and associates it with abnormal metadata (including abnormality type, occurrence time, related operator ID, edge node ID) to achieve accurate repair and removal of data abnormalities, ensure the business compliance of clean data, and archive the clean data version, associated abnormal metadata, and cleaning process records to the off-chain audit database to form a structured abnormal data log. This log can be used for subsequent auditing and tracing. At the same time, the clean data version is synchronized to the blockchain platform, updated to a valid clean blockchain dataset that can be uploaded to the chain, builds a fully traceable audit chain, and ensures the consistency between on-chain data and off-chain archives.

[0062] The specific tasks are as follows: After the interception operation is triggered, the interception log generated by the smart contract is read from the off-chain audit database. This log is stored in a time-series structure database and supports efficient retrieval by timestamp, triggering contract address, and associated block height. During the reading process, based on the original transaction hash field recorded in the interception log, the original abnormal data record stored in the off-chain buffer pool is accurately located. Each record contains a complete original sensor data packet, and the data format follows the industrial communication protocol specification, covering vehicle positioning coordinates, weighing values, and loading / unloading action timestamp sequences. The extraction process ensures data integrity and compares it with the log data. Initial association is performed on abnormal metadata (including abnormality level code, edge node number, and operator identifier); based on the abnormality type code recorded in the interception log, a pre-set rule-based cleaning strategy library is invoked. This strategy library, defined according to business security procedures, sets differentiated processing procedures for different abnormality levels: for primary abnormalities, a data repair algorithm is executed, using linear interpolation to correct the numerical values ​​of abnormal fluctuation points within the compliance time window; for intermediate abnormalities, logical verification and sequence rearrangement are performed, logically reconstructing the timestamps of disordered actions based on the order of forklift positioning, pallet picking, and goods placement defined in the standard operating procedures; for high-level abnormalities... For level-one anomalies, a data removal operation is performed, marking all associated sensor data within that period as invalid and isolating it. During the cleaning process, the rule number, processing time, and operation result of each operation are recorded, generating a structured intermediate record containing the clean data version and its corresponding anomaly metadata (anomaly type, occurrence time, related operator ID, edge node ID). After cleaning, the generated clean data version, associated anomaly metadata, and detailed cleaning process records (including the rule number used, processing parameters, and operator audit trajectory) are written together into a dedicated archive table in the off-chain audit database. This table adopts a partitioned storage design, establishes a composite index by date and anomaly level, and supports millisecond-level multi-dimensional conditional retrieval to ensure the efficiency and completeness of subsequent audit tracing. At the same time, the clean data version is synchronized to the blockchain platform through a secure data channel, the clean data is encapsulated into a structured transaction that conforms to the on-chain data format standard, and the verified data on-chain smart contract is called to submit the transaction to the blockchain network. After verification by the consensus node, the data is written to a new block, and the global state is updated to an audited, traceable, and valid clean blockchain dataset. The entire process ensures the consistency of clean data on-chain and off-chain, providing a real and reliable data foundation for the supply chain collaboration engine.

[0063] The multi-agent collaboration engine is used to build multi-role intelligent agents covering warehousing, transportation and finance based on clean blockchain datasets. It combines the real-time status of each link in the supply chain to adjust the data synchronization strategy, realize efficient data synchronization across links, and improve order fulfillment transparency and collaborative response speed.

[0064] The dynamic path optimization module is used to continuously optimize the data traceability path and collaborative logic in the data synchronization strategy through reinforcement learning, adapt to the dynamic changes in the supply chain, support real-time and reliable supply chain status tracking and decision-making, and improve the efficiency of the entire chain operation.

[0065] The privacy and security collaborative traceability module is used to introduce a differential privacy mechanism in cross-link supply chain data sharing. It adds controllable noise to sensitive business fields, ensuring data availability and traceability consistency while protecting the business privacy of all participants.

[0066] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the multi-agent collaborative engine executes the following steps: based on the clean blockchain dataset output by the smart contract interception and purification module, corresponding intelligent agents are instantiated for warehouse management, transportation scheduling, and financial settlement. Each intelligent agent encapsulates the business logic and data access interface of its segment, realizing independent encapsulation and decoupling of the business logic of each segment, improving the modularity and maintainability of the system. Each intelligent agent listens to the status update events related to its segment on the blockchain in real time, and dynamically adjusts the priority and frequency of data synchronization through negotiation protocol according to the overall performance progress of the supply chain, realizing near real-time synchronization of order status, inventory changes, transportation trajectory, and payment information across segments, significantly reducing cross-segment data synchronization delay, ensuring dynamic coordination and efficient response in the order performance process, constructing a global order performance view through message transmission and status sharing between intelligent agents, and recording key collaborative events on the blockchain, ensuring the auditability and process transparency of operations in each segment, forming a complete and reliable global traceability chain, and providing all participants with tamper-proof collaborative operation records;

[0067] The specific work involves the following: In implementation, based on the clean blockchain dataset, corresponding intelligent agents are instantiated for three core aspects: warehouse management, transportation scheduling, and financial settlement. Each intelligent agent encapsulates the business logic and data access interface for its respective aspect. Specifically, the warehouse management intelligent agent handles inventory counting, inbound / outbound verification, and warehouse location status tracking; the transportation scheduling intelligent agent manages vehicle trajectories, delivery timeliness, and route optimization; and the financial settlement intelligent agent handles payment verification, invoice matching, and account reconciliation. Each intelligent agent registers with the blockchain network during initialization, binding itself to the on-chain data model with a unique identifier, enabling it to accurately parse and respond to transaction events within its own business scope. The intelligent agents listen in real-time to state update events related to their own aspect in new blocks through WebSocket or event subscription interfaces provided by the blockchain nodes, ensuring that business logic remains synchronized with on-chain data. Each intelligent agent dynamically adjusts the priority and frequency of data synchronization based on the overall supply chain fulfillment progress through a consensus-based negotiation protocol. For example, when an order enters the outbound stage, the warehouse management intelligent agent prioritizes the synchronization of inventory change events. The system is optimized to push status updates at the highest level, with the frequency adjusted to once per second. Once the vehicle is loaded, the transportation dispatching agent immediately writes the loading status and estimated departure time to the blockchain, triggering the financial settlement agent to initiate pre-settlement verification. Through an event-driven architecture and pre-defined collaborative rules in smart contracts, near real-time synchronization of order status, inventory changes, transportation trajectory, and payment information is achieved between the warehouse, transportation, and financial stages. Data latency at each stage is controlled within 2 seconds, ensuring the continuity and consistency of the supply chain status throughout the entire chain. A global order fulfillment view is constructed through point-to-point message passing and status sharing between agents. For each key collaborative event, including completion of outbound shipment, in-transit status update, and settlement statement generation, the initiating agent encapsulates it into a standard event message, digitally signs it, and writes it to the blockchain as a transaction, recording it as an immutable collaborative log. The global order fulfillment view aggregates the status of each stage, generating a panoramic timeline on the chain from order creation to delivery completion. This allows any participant to trace the entire fulfillment process according to their permissions. All collaborative operations of agents are notarized through the blockchain, ensuring the auditability and transparency of each stage's operations.

[0068] The dynamic path optimization module executes the following steps: Using historical collaborative data and real-time status of the entire supply chain as input, a reinforcement learning environment is constructed. The state space includes data synchronization latency, resource load, and order urgency at each stage; the action space includes the selection of data traceability paths and the adjustment of synchronization strategy parameters. This enables comprehensive, dynamic perception and quantitative modeling of the supply chain's operational status, providing precise input for intelligent decision-making. Agents explore and execute actions in the reinforcement learning environment, evaluating the effectiveness of these actions through a reward function (rewarding efficient synchronization and low-latency traceability). The module continuously iterates and optimizes its strategy network to learn the optimal strategy for adapting to dynamic supply chain scenarios, including optimal data traceability and collaborative paths. During training, it automatically learns collaborative paths and parameter combinations that significantly shorten data synchronization time and improve traceability efficiency. The trained optimal strategy is deployed to the multi-agent collaborative engine, guiding each agent to dynamically adjust data flow and processing logic in complex scenarios. This achieves real-time, accurate tracking of the supply chain status and efficient collaborative decision support, ensuring that the data synchronization latency across the entire supply chain remains consistently below 2 seconds and that resource load remains balanced, thereby improving the real-time performance and reliability of overall collaborative decision-making.

[0069] The specific work involves: constructing a reinforcement learning environment that supports continuous decision-making based on historical collaborative data and real-time status across the entire supply chain. The state space of this environment consists of three quantifiable metrics: real-time data synchronization latency at each stage, measured in milliseconds; resource load calculated comprehensively based on the CPU utilization, memory usage, and network bandwidth utilization of edge nodes and servers; and dynamic assessment of order urgency based on preset priority tags and promised delivery time differences, categorized into low, medium, and high levels, corresponding to values ​​1, 2, and 3 respectively. The action space includes two types of operations: data traceability path selection, i.e., choosing one from multiple predefined paths; and synchronization... The strategy parameters can be adjusted, specifically including the status push frequency (which can be set to 0.5Hz, 1Hz, or 2Hz) and the data batch size (ranging from 128KB to 1MB). Initial environment parameters must be calibrated using historical data and dynamically calibrated after deployment to match the dynamic changes in actual business scenarios. The agent undergoes iterative training in this reinforcement learning environment using a temporal difference learning framework. In each training round, the agent observes the current state and outputs an action based on the policy network. The environment then provides an immediate reward. The reward function is designed as follows: the efficient synchronization reward is calculated based on the reduction ratio of data synchronization latency; for every 10% reduction in latency, the reward value increases by 0. 0.1; The low-latency traceability reward is negatively penalized based on the deviation between the end-to-end traceability completion time and the target time. The reward value decreases linearly when the deviation exceeds 2 seconds. A resource load balancing penalty is also introduced; if the load of any node continuously exceeds 85% for more than 5 minutes, the corresponding reward is deducted. During training, the agent uses a proximal policy optimization algorithm to update the policy network parameters. The learning rate is set to 0.0003, and the discount factor is fixed at 0.99. Each training cycle contains 10,000 interaction steps. After at least 50 iterations, the policy network gradually converges to a stable state, capable of outputting the optimal data traceability path adapted to dynamic supply chain scenarios. Synchronous parameter combination; The optimal strategy after training is exported through a lightweight model and deployed to the strategy execution module of the multi-agent collaborative engine. This module is embedded in each agent in the form of a microservice, receives real-time status input from the reinforcement learning environment, and outputs action instructions within a 2-second decision cycle. Each agent dynamically adjusts its internal data flow and processing logic according to the strategy instructions, so as to achieve the collaborative goal of controlling the synchronization delay of the entire link status within 2 seconds and the resource load balancing deviation not exceeding 15%. All strategy execution records are stored on the blockchain to support subsequent effect evaluation and strategy iteration, and continuously improve the real-time tracking accuracy of the supply chain status and the efficiency of collaborative decision-making.

[0070] The privacy and security collaborative traceability module performs the following steps: Before cross-link data sharing, it identifies sensitive business fields in the data, including specific transaction amounts, supplier costs, and specific product inventory details. Sensitive field identification and classification can dynamically adjust privacy protection strategies to achieve precise isolation and classification management of key information. It applies a differential privacy mechanism to add Laplace noise or Gaussian noise that meets the preset privacy budget to sensitive business fields, generating perturbed data that meets ε-differential privacy. Noise injection can resist external inference and correlation attacks, ensuring data availability while effectively preventing privacy leakage risks. The perturbed data is used for cross-agent collaborative traceability and global analysis to ensure the logical coherence of the supply chain traceability chain and the availability of key business insights while protecting the business privacy of all participants. The perturbed data supports cross-chain traceability and analysis tasks, ensuring the coherence of business logic and the effectiveness of statistical conclusions, achieving a balance between privacy and business.

[0071] The specific work involves: automatically identifying and classifying sensitive fields in the data structure to be shared across different stages; scanning key fields in the data package based on a predefined business metadata model, including highly commercially sensitive information such as transaction amount, supplier cost, and specific product inventory details; identifying the transaction amount field with accuracy to currency units and two decimal places; associating the supplier cost field with material codes and purchase batches; and differentiating product SKUs, real-time inventory quantities, and safety stock thresholds in the inventory details field. The identification process uses a combination of regular expression matching and semantic tagging to ensure coverage of both structured and unstructured data. After identification, a privacy level label is attached to each sensitive field, with the level set according to the business segment to which the data belongs, the agreements between participating parties, and legal and regulatory requirements, forming a sensitive dataset to be processed; and applying differential privacy budget parameters to the identified sensitive fields. To mitigate noise injection, for continuous data such as transaction amounts, a Laplace noise mechanism is employed. The noise scale parameter is calculated based on the field's global sensitivity and a preset ε value (set within the range of 0.1 to 1.0, dynamically adjusted according to data sharing frequency and business scenarios) to ensure that the generated data meets ε-differential privacy requirements. For count-type data such as inventory details, a Gaussian noise mechanism is prioritized. Its noise standard deviation is calculated based on the data distribution range and privacy budget allocation model. The noise injection process is completed before the data leaves the local edge node. The system records the amount of noise added to each field, the actual ε consumption value, and the timestamp, generating a privacy protection audit log. After perturbation, the data maintains its original data structure and key relationships, ensuring logical consistency in subsequent collaborative traceability. The perturbed data is securely transmitted to the supply chain collaboration platform and applied to cross-agent collaborative traceability and global analysis tasks. The data is transmitted via TLS. 1.3 Encrypted channel transmission: After the receiver verifies the data integrity signature, the data is loaded into the collaborative analysis process. In the global analysis, perturbed data is used to perform tasks such as inventory turnover trend analysis, supplier performance evaluation, and transportation cost correlation analysis. All aggregation calculations take into account the statistical errors introduced by noise and calibrate the output results through confidence intervals. In the cross-agent traceability scenario, each agent collaboratively constructs an order fulfillment status view based on the perturbed data. The continuity of the traceability chain is maintained through the fuzzy matching algorithm of the associated fields. At the same time, the original sensitive data is only accessible locally to the data owner. Any cross-link sharing and calculation is based on the dataset after differential privacy protection, supporting full-chain traceability and collaborative decision-making of the supply chain while ensuring business privacy.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A blockchain-based supply chain end-to-end data traceability and collaboration system, comprising a supply chain collaborative management platform, characterized in that, The supply chain collaborative management platform has the following communication modules: The twin edge simulation monitoring module is used to integrate digital twin and edge computing technologies to build real-time behavior simulation models at each supply chain node, dynamically monitor on-site operational behavior, and identify anomalies in real time through time-series pattern analysis. The smart contract interception and purification module automatically triggers smart contracts to intercept and clean abnormal on-site operation data in real time based on the results of twin edge simulation monitoring, thereby obtaining a clean blockchain dataset. A multi-agent collaboration engine is used to build multi-role intelligent agents based on clean blockchain datasets and adjust data synchronization strategies in combination with the real-time status of each link in the supply chain. The dynamic path optimization module is used to continuously optimize the data tracing path and collaborative logic in the data synchronization strategy through reinforcement learning. The privacy and security collaborative traceability module is used to introduce a differential privacy mechanism in cross-link supply chain data sharing, adding controllable noise to sensitive business fields.

2. The blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 1, characterized in that: The twin edge simulation monitoring module includes a dynamic behavior simulation unit and a temporal anomaly identification unit; The dynamic behavior simulation unit is used to combine digital twin and edge computing technologies to deploy entity behavior simulation models at edge nodes, simulate and compare actual on-site operation behavior data with expected operations in real time, and identify behavior deviations and potential fraud. The timing anomaly identification unit, based on time series data analysis, monitors the timing consistency of on-site operational behaviors in real time and automatically triggers an early warning mechanism.

3. The blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 2, characterized in that: The dynamic behavior simulation unit performs the following steps: Edge computing devices are deployed at each node of the supply chain. By combining physical parameters of the entity with business process models, a digital twin-based entity behavior simulation model is constructed. The entity behavior simulation model includes the real-time mapping relationship of vehicle positioning, weighing sensor data flow, and loading and unloading action sequence. Based on the entity behavior simulation model, real-time data of actual field operation behavior from sensor cluster is received, and the expected operation simulation is executed in parallel at the edge. The deviation between actual behavior and simulated behavior is calculated by behavior trajectory comparison algorithm. When the deviation exceeds a preset deviation threshold, it is determined to be an abnormal on-site operation, an automatic behavior deviation report is generated, and the abnormal event is marked as a potential fraudulent behavior. At the same time, the transmission of the original data corresponding to the abnormal operation to the blockchain platform is blocked.

4. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 2, characterized in that: The timing anomaly identification unit performs the following steps: Receive actual on-site operational data and extract multi-dimensional time-series features, including vehicle entry and exit time series, continuous weighing value series, and loading operation intervals. The time series pattern recognition algorithm is applied to analyze the multi-dimensional time series features in real time to detect whether there are time series patterns that violate business logic, including repeated weighing in a short period of time, reversed loading and weighing order, and abnormal vehicle dwell time. When an abnormal timing pattern is detected, a real-time warning signal is automatically triggered, and the warning event, associated timestamp, and operation node identifier are sent to the smart contract interception and purification module to initiate the data interception process.

5. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 2, characterized in that: The smart contract interception and purification module includes a contract triggering interception unit and a data cleaning and archiving unit; The contract triggering interception unit is used to automatically execute the smart contract, mark the abnormal data, and block the abnormal data from being written to the blockchain when abnormal on-site operation behavior is detected. The data cleaning and archiving unit is used to clean, mark and archive the intercepted abnormal data, record the abnormality type, time and operator information, form an abnormal data log, and integrate them to obtain a clean blockchain dataset.

6. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 5, characterized in that: The contract trigger interception unit performs the following steps: Listen for abnormal event signals from the twin edge simulation monitoring module. When an abnormal warning is received, automatically invoke the interception smart contract pre-installed on the blockchain platform. The intercepted smart contract is used to parse the content of abnormal events, the abnormality level is determined according to the predefined rule set, and the data records to be uploaded to the chain generated by the current operation are marked and assigned a pending review status identifier. The interception logic in the smart contract is executed to prevent abnormal data records marked as pending review from being written into a new block of the blockchain. At the same time, an interception log containing an abnormal summary, interception time, and triggering contract address is generated and stored in an off-chain database.

7. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 5, characterized in that: The data cleaning and archiving unit performs the following steps: Read the interception logs generated by the contract-triggered interception unit from the off-chain database and extract the corresponding original abnormal data records; Based on the anomaly type, the corresponding data cleaning rules are invoked to repair, remove, or label the original abnormal data, generate a clean data version, and associate it with the abnormal metadata. The clean data version, associated abnormal metadata, and cleaning process records are archived together in the off-chain audit database to form a structured abnormal data log. At the same time, the clean data version is synchronized to the blockchain platform and updated into a valid clean blockchain dataset that can be uploaded to the chain.

8. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 5, characterized in that: The multi-agent collaborative engine performs the following steps: Based on the clean blockchain dataset output by the smart contract interception and purification module, corresponding smart agents are instantiated for the warehouse management, transportation scheduling and financial settlement processes, and each smart agent encapsulates the business logic and data access interface of its process. Each intelligent agent monitors the status update events related to its own link on the blockchain in real time, and dynamically adjusts the priority and frequency of data synchronization through negotiation protocols based on the overall performance progress of the supply chain. By using message passing and state sharing among intelligent agents, a global order fulfillment view is constructed, and key collaborative events are recorded on the blockchain.

9. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 8, characterized in that: The dynamic path optimization module performs the following steps: Using historical collaborative data and real-time status of the entire supply chain as input, a reinforcement learning environment is constructed. The state space includes data synchronization delay, resource load, and order urgency at each stage, while the action space includes the selection of data traceability paths and the adjustment of synchronization strategy parameters. The agent explores and performs actions in the reinforcement learning environment, evaluates the effectiveness of the actions through a reward function, and continuously iterates and optimizes its policy network to learn the optimal strategy that adapts to dynamic supply chain scenarios, including optimal data traceability and collaborative paths. The optimal strategy obtained from training is deployed to the multi-agent collaborative engine to guide each agent to dynamically adjust the data flow and processing logic in complex scenarios.

10. A blockchain-based supply chain end-to-end data traceability and collaboration system according to claim 9, characterized in that: The privacy and security collaborative tracing module performs the following steps: Before sharing data across different stages, identify sensitive business fields in the data, including specific transaction amounts, supplier costs, and inventory details of specific products; By applying a differential privacy mechanism, Laplace noise or Gaussian noise that meets a preset privacy budget is added to the sensitive business fields to generate perturbed data that satisfies ε-differential privacy, and the perturbed data is used for cross-agent collaborative tracing and global analysis.

Images