Electrical complete set supply chain traceability method fusing blockchain technology

Abstract

This invention relates to the field of electrical complete equipment supply chain traceability and blockchain information processing technology, specifically a method for electrical complete equipment supply chain traceability integrating blockchain technology. The method includes: collecting multi-source process data and blockchain node status data; generating a unified data index through time alignment, field mapping, and unique identifier association; constructing an ideal benchmark graph based on object structure data and standard process data; generating a theoretical process anomaly simulation state and a theoretical residual graph set using anomaly operator library injection; mapping real-world data to generate a real-world residual graph and comparing its topological similarity with the theoretical residual graph set; controlling the process data to be stored on the blockchain based on similarity and dual thresholds, combined with blockchain node status, by directly uploading the data, rejecting its uploading, or adding anomaly tags before uploading, and adjusting the anomaly operator library or dual thresholds based on the completed handling results; and enabling off-chain multi-source process authenticity verification before uploading to the blockchain, preventing erroneous data from being reliably stored.

Description

Technical Field

[0001] This invention relates to the field of electrical complete equipment supply chain traceability and blockchain information processing technology, specifically to an electrical complete equipment supply chain traceability method that integrates blockchain technology. Background Technology

[0002] As key equipment in scenarios such as rail transit, substations, and industrial power distribution, electrical complete sets of equipment have many supply chain links, complex participating entities, and wide-ranging links. To ensure that the source of equipment is traceable, the manufacturing process is verifiable, and the delivery responsibility can be implemented, data of electrical complete sets of equipment is usually recorded in the stages of design issuance, material receipt, assembly, quality inspection, warehousing, transportation, and receipt. It is also possible to combine blockchain technology to store relevant process information to improve the transparency and tamper-proof capability of supply chain traceability. However, when storing electrical complete set supply chain processes on the blockchain, it is necessary to first determine the authenticity and consistency of off-chain multi-source process data. Off-chain data often comes from different sources such as business systems, logistics sensing nodes, and quality inspection systems. There are likely to be problems such as time deviation, inconsistent fields, missing records, or mutual conflicts between different sources. Existing methods usually process process data by rule comparison, single-item threshold judgment, or direct storage on the blockchain. This processing method tends to assume that the off-chain data itself is trustworthy, or to judge based on only local field anomalies. It is difficult to take into account the structural differences such as process path, time sequence relationship, spatial authorization, and quality inspection characteristics at the same time. This can easily lead to normal business fluctuations being misjudged as anomalies, or abnormal but artificially tampered data being directly written into the blockchain, which affects the accuracy and reliability of traceability results. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for tracing the supply chain of electrical complete sets of equipment by integrating blockchain technology. Specifically, the technical solution of this invention includes: Collect multi-source process data and blockchain node status data from the electrical complete equipment supply chain, perform time alignment, field mapping and unique identifier association based on a unified standard time base on the multi-source process data, and generate a unified data index; An ideal benchmark graph is constructed based on pre-configured object structure data and standard process data. The ideal benchmark graph is a directed graph representing the standard supply chain process. Based on the anomaly operators in the anomaly operator library, the ideal benchmark spectrum is parameterized and injected to generate multiple theoretical process anomaly simulation states. Then, the difference calculation is performed between each theoretical process anomaly simulation state and the ideal benchmark spectrum to generate a theoretical residual spectrum set. The multi-source process data corresponding to the unified data index is mapped to the nodes and edges of the ideal benchmark graph to generate the real mapping graph. The real mapping graph and the ideal benchmark graph are then differentially calculated to generate the real residual graph. A topological similarity comparison is performed between the actual residual map and the theoretical residual map set to determine the process anomaly category and similarity. Obtain preset dual thresholds, which include a release threshold and a blocking threshold, wherein the blocking threshold is strictly greater than the release threshold. Both are initialized by the system based on the confidence interval of the topological similarity distribution of historical compliant supply chain processes. Generate process status determination results based on the similarity and the preset dual thresholds. Based on the process status determination result and the blockchain node status data, control the process data to be stored corresponding to the unified data index to be uploaded to the blockchain through smart contract, refuse to upload to the blockchain, or upload to the blockchain after attaching a process abnormal label. The process status determination result is verified based on the actual compliance status of the target process as reported by manual or system audit, and the abnormal operator library or the dual threshold is adjusted based on the verification result.

[0004] Optionally, the multi-source process data includes: Structure list data, work order flow data, and quality inspection process data collected by the business system; Identification trajectory data, node timestamp data, and load characteristic data collected by logistics sensing nodes; Transaction log data and contract call log data collected by blockchain nodes; The blockchain node status data includes node online status, consensus completion status, and contract execution status.

[0005] Optionally, the construction of the ideal benchmark map includes: Based on the object structure data, determine the material hierarchy and process dependencies; Based on the aforementioned standard process data, standard processing routes, standard logistics routes, and standard quality inspection routes are determined; Organize the material nodes, process nodes, logistics nodes, and quality inspection nodes into a directed acyclic graph; The nodes and / or edges in the ideal baseline graph are associated with standard time attributes, standard spatial attributes, standard quality attributes, standard identity attributes, and standard resource consumption attributes.

[0006] Optionally, the anomaly operator library includes substitution operators, unauthorized processing operators, and report reuse operators; wherein each of the anomaly operators is encapsulated by a triggering rule, a target node or directed edge location label, and a mathematical perturbation function for modifying the target attribute value and graph topology; the parameterized injection includes: Modify the quality attributes and resource consumption attributes of the material node based on the aforementioned substitution operator; Based on the above-mentioned over-authorization processing operator, modify the spatial and temporal attributes of the process node or logistics node; Modify the quality inspection feature attributes of the quality inspection node based on the report reuse operator; And generate corresponding theoretical process anomaly simulation states based on the attribute changes and / or path changes after injection.

[0007] Optionally, the generation of the reality mapping map includes: Map the multi-source process data corresponding to the unified data index to the corresponding nodes and edges of the ideal benchmark graph; Missing fields are filled in according to preset data source rules, and consistency checks are performed on the filled in results; Perform priority adjudication on conflicting fields; The priority decision is executed in order of preset data trust levels. The data trust level is determined at least based on the data source, signature status and time integrity. Specifically, the data trust level is quantified by assigning preset weights to data acquisition terminals with different security levels, digital signature status with anti-tampering characteristics and continuous and unbroken timestamp sequences, and calculating a weighted evaluation total score to generate a real mapping map that is isomorphic to the ideal benchmark map.

[0008] Optionally, the generation of the theoretical residual map set and the actual residual map includes: Extract the baseline node attributes, baseline edge attributes, and baseline path sequences from the ideal baseline map; The simulation node attributes, simulation edge attributes, and simulation path sequences in each of the theoretical process abnormal simulation states are respectively compared with the corresponding baseline node attributes, baseline edge attributes, and baseline path sequences to generate a theoretical residual map. The real node attributes, real edge attributes, and real path sequences in the real mapping map are respectively compared with the corresponding reference node attributes, reference edge attributes, and reference path sequences to generate a real residual map. The differential calculation is used to characterize attribute offset, path offset and temporal offset. The path offset is the difference in node insertion, deletion and modification between the real or simulated path sequence and the baseline path sequence. The temporal offset is the difference or normalized difference between the corresponding node timestamp and the standard time attribute. Specifically, the difference calculation includes: for the numerical attributes of nodes and edges, using vector subtraction and extracting the difference as the attribute offset; for the topological structure of the graph path, using the topological symmetric difference operation between the node set and edge set of the baseline graph and each of the theoretical process abnormal simulation state or real mapping graphs, extracting the relatively inserted extra nodes and edges and the relatively missing deleted nodes and edges as the path offset.

[0009] Optionally, the topological similarity comparison includes: Calculate the graph edit distance between the actual residual map and each of the theoretical residual maps, or encode the features of the actual residual map and each of the theoretical residual maps to generate residual feature vectors and calculate vector similarity; Candidate matching sequences are generated by sorting the graph edit distance from smallest to largest or the vector similarity from largest to smallest. The theoretical residual map that ranks first and meets the preset matching conditions is selected as the optimal matching result; wherein, the preset matching conditions include: the similarity of the first and second ranked positions is greater than the preset absolute similarity threshold, and the difference between the similarity of the first and second ranked positions is greater than the preset discrimination threshold. The theoretical residual map category corresponding to the optimal matching result is output as the process anomaly category.

[0010] Optionally, the generation of the process status determination result includes: When the similarity is greater than or equal to the blocking threshold, the process is determined to be in an abnormal state, and a rejection instruction for uploading to the chain is generated. When the similarity is less than or equal to the release threshold, the process is determined to be in a normal state, and a direct on-chain instruction is generated. When the similarity is greater than the allow threshold and less than the block threshold, the process is determined to be in a pending confirmation state, and an on-chain instruction with an additional process abnormal label is generated. The process anomaly label includes at least a process anomaly category identifier and an anomaly uncertainty identifier.

[0011] Optionally, adjusting the anomaly operator library or the dual threshold based on the process status determination result of completed processing includes: Extract the process status determination results of completed treatments and the corresponding actual residual maps; When the process status determination result is an abnormal process status, update the attribute perturbation parameters or rule templates of each abnormal operator in the abnormal operator library. When the process status determination result is a process pending confirmation status, the interval boundary of the dual threshold is updated; When the process status determination result is that the process is in a normal state, the abnormal operator library and the dual threshold remain unchanged.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This method constructs an ideal benchmark map representing the standard supply chain process and compares the topological similarity between the real residual map generated by mapping and differential analysis of multi-source process data and the theoretical residual map set generated based on anomaly operator injection. This overcomes the limitation of existing technologies that rely solely on local fields for judgment. This method can comprehensively evaluate structural differences such as attribute offset, path offset, and temporal offset. Combined with a preset dual threshold mechanism including release threshold and blocking threshold, it effectively distinguishes between normal supply chain business fluctuations and real process violations and anomalies, significantly improving the accuracy of real verification of multi-source process data. 2. This method, based on the process status determination results obtained from topological similarity comparison and combined with blockchain node status data, implements more refined differentiated evidence storage control for multi-source process data. This avoids the problem of forged compliant data being directly written into the blockchain due to reliance on only local verification, ensuring the data consistency and immutability of the on-chain ledger. At the same time, the system dynamically adjusts the parameter rules or dual threshold interval boundaries in the anomaly operator library based on the process status determination results of completed processing, enabling the traceability method to continuously adapt to changes in different projects and environments, and maintain a high standard of anti-tampering and verification capabilities in the long term. Attached Figure Description

[0013] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0015] like Figure 1 As shown, an electrical equipment supply chain traceability method integrating blockchain technology includes: Collect multi-source process data and blockchain node status data from the electrical complete equipment supply chain, perform time alignment, field mapping and unique identifier association on the multi-source process data based on a unified standard time base, and generate a unified data index; An ideal baseline graph is constructed based on pre-configured object structure data and standard process data. The ideal baseline graph is a directed graph representing the standard supply chain process. Based on the anomaly operators in the anomaly operator library, the ideal benchmark spectrum is parametrically injected to generate multiple theoretical process anomaly simulation states. Then, the anomaly simulation states of each theoretical process are differentially calculated with the ideal benchmark spectrum to generate a set of theoretical residual spectra. The multi-source process data corresponding to the unified data index is mapped to the nodes and edges of the ideal benchmark graph to generate the real mapping graph. The real mapping graph and the ideal benchmark graph are then differentially calculated to generate the real residual graph. A topological similarity comparison is performed between the actual residual map and the theoretical residual map set to determine the process anomaly category and similarity; Obtain preset dual thresholds, which include a release threshold and a blocking threshold, with the blocking threshold being strictly greater than the release threshold. Both are initialized by the system based on the confidence interval of the topological similarity distribution of historical compliant supply chain processes; generate process status determination results based on similarity and preset dual thresholds; Based on the process status determination results and blockchain node status data, control the process data to be stored and verified corresponding to the unified data index to be uploaded to the blockchain through smart contracts, refuse to upload to the blockchain, or add process abnormal tags to the blockchain. The results of the process status determination are verified based on the actual compliance status of the target process as reported by manual or system audits, and the abnormal operator library or dual thresholds are adjusted according to the verification results.

[0016] This embodiment provides an electrical complete set supply chain traceability mechanism that integrates blockchain technology. Specifically, it takes the delivery process of a complete set of high-voltage switchgear equipment used in a subway traction substation as the main scenario. The equipment consists of components such as cabinets, circuit breakers, busbars, relay protection units, and insulation parts, involving multiple stages such as design issuance, incoming materials, sub-assembly, final assembly, quality inspection, warehousing, transportation, arrival receipt, and pre-installation re-inspection. This mechanism does not directly assume that off-chain data has inherent credibility. Instead, it first constructs a standardized ideal benchmark map, and then matches the results of theoretical anomaly simulation and real process playback through dual differential analysis to determine whether relevant data can be written to the blockchain. Specifically, during the data acquisition phase, the system collects process data and node status data from the Enterprise Resource Planning (ERP) system, Manufacturing Execution System (MES), Logistics Sensing Nodes, and Blockchain Nodes, respectively. For ease of explanation, assume that a high-voltage switchgear with the serial number GZ-KG-021 corresponds to three types of original records: the first type is a work order record, recording the busbar assembly completion time as 10:02; the second type is a logistics record, recording that the cabinet passed through the factory transfer door at 10:05; and the third type is a quality inspection record, recording the withstand voltage test completion time as 10:14. Since the clocks of different systems may have deviations, for example, the MES clock is 2 minutes ahead of the standard time base, the logistics sensing node clock is 2 minutes ahead of the standard time base, and the quality inspection workstation clock is 1 minute behind the standard time base, time alignment is performed first, and the three types of records are uniformly converted to the same standard time base, for example, becoming 10:00, 10:03, and 10:15 respectively after conversion. Perform field mapping, for example, map the process code OP23 in the manufacturing execution system to the unified field assembly node, map the station L5 in the logistics sensing node data to the in-plant transfer edge, and map the test order number Q89 in the quality inspection to the quality inspection node; through unique identifier association, merge the cabinet serial number, work order number, purchase batch number and RFID tag number into the unified primary key device instance ID=GZ-KG-021; after the above processing, a unified data index can be generated, under which the multi-source records of the device in each stage can be attached; Furthermore, the system constructs an ideal baseline map based on pre-configured object structure data and standard process data; here, the object structure data can be understood as the engineering bill of materials and process dependencies of the switch cabinet, while the standard process data corresponds to the manufacturing and logistics standard operating procedures approved and executed by the enterprise. For ease of understanding, this graph can be simplified into five sequential nodes: N1 incoming material confirmation, N2 busbar assembly, N3 final assembly, N4 pressure testing, and N5 shipment. These nodes are connected sequentially by four directed edges. Each node and edge has standard attributes. For example, the standard completion time window for N2 is 09:50 to 10:10, N4 requires execution at the authorized quality inspection station Q-Station-3, and the corresponding shipping location for N5 must be Dock-2. The resulting ideal baseline graph is a directed graph representing how a standard supply chain should occur. Based on this, the system does not directly use real data for simple threshold comparisons. Instead, it first calls the anomaly operator library to parameterize the ideal baseline graph, generating multiple theoretical process anomaly simulation states. For simulation deduction, three types of anomalies can be set: one is substitute material, which changes the quality attribute of the busbar copper material from the standard value Q=100 to Q=72, and at the same time changes the resource consumption attribute under the resource consumption dimension from C=100 to C=63; another is unauthorized processing, which means that the assembly that should have been completed at the authorized station A is changed to the unauthorized station X, and the completion time is 3 hours longer than the standard path; the third is report reuse, which means that the quality inspection node uses the historical report fingerprint of other equipment, only making local modifications to the sampling time field. For each type of simulation state, the difference is then calculated with the ideal baseline graph to form the corresponding theoretical residual graph. For example, the residual generated by the unauthorized processing simulation is not only a 180-minute time delay, but also includes two structural differences: the appearance of an additional stop node X in the path and the mismatch of authorized space attributes; the system maps the real multi-source process data corresponding to the unified data index back to the ideal benchmark map to generate a real mapping map; For example, in reality, the following situation occurs with equipment GZ-KG-021: the completion time of N2 busbar assembly is recorded as 13:05, the logistics track shows that it stopped at the outsourcing point P7 between 11:20 and 12:40, and the fingerprint of the quality inspection report has a higher similarity than the report of another equipment GZ-KG-018 three days ago than the preset similarity threshold; the system fills these data into nodes and edges that are isomorphic to the ideal baseline map, that is, the number of nodes and the connection relationship are kept aligned, but the node attributes and path details are allowed to carry real differences; then the real mapping map is subtracted from the ideal baseline map to obtain the real residual map; the real residual map is used to characterize the degree of deviation, the position of deviation, and the structural differences of the actual process relative to the standard process; Next, the system performs a topological similarity comparison between the actual residual map and the theoretical residual map set. For ease of explanation, it is assumed that the system pre-stores three theoretical residual templates R1, R2, and R3, corresponding to substitute materials, unauthorized processing, and report reuse, respectively. After matching the same actual residual map, the similarity results are 0.41, 0.86, and 0.79, respectively. The similarity here can be derived from the graph edit distance conversion value or from the encoded vector similarity. Since 0.86 is the highest and corresponds to the unauthorized processing category, the system first marks the current abnormal category as suspected unauthorized processing. During the status determination phase, the system employs a dual-threshold mechanism. For illustration, assume the allow threshold is 0.35 and the block threshold is 0.80. If the similarity is not higher than 0.35, the actual deviation is considered closer to normal supply chain business fluctuations, such as temporary queuing or authorized internal workstation changes, and can be directly uploaded to the blockchain. If the similarity is not lower than 0.80, the actual deviation is considered to have met the preset conditions for matching the topological features of a certain type of abnormal template, and should be rejected from being uploaded to the blockchain. If it is between the two, it indicates that there are abnormal signs but not enough to completely block the process. In this case, it is allowed to be uploaded to the blockchain after attaching an abnormal label for subsequent review. For the example above with a similarity of 0.86, the system outputs the abnormal status of the process. During the on-chain control phase, the system does not only consider abnormal results but also uses blockchain node status data to make contract decisions. For example, if all consensus nodes are online and the target contract is executing normally, data that is directly uploaded to the chain is submitted immediately. Data that is rejected from being uploaded to the chain is not written to the business's official chain but instead the blocking event is written to the regulatory audit chain or local isolation log. For data uploaded to the chain with additional tags, an anomaly category identifier and an uncertainty measurement identifier are added to the on-chain data structure. For example, for batches suspected of unauthorized processing but not yet meeting the blocking conditions, an anomaly tag = spatial unauthorized processing / uncertainty can be added to the on-chain field. During the closed-loop handling phase, the system adjusts the anomaly operator library or dual thresholds based on the results of completed manual review, on-site audit, or accountability. If it is subsequently confirmed that this was indeed an illegal outsourcing project, the weight of the spatial dwell parameter in this type of anomaly operator is increased. If it is found that this was just a temporary transfer within the park and there was no violation, the blocking threshold of the relevant category is appropriately increased or its similarity mapping sensitivity is decreased. This forms a closed loop of generating theoretical anomalies - detecting actual deviations - on-chain control - post-event calibration. As a supplementary anomaly handling mechanism, the system executes processing logic when the following boundary conditions occur: If a device only collects partial data, such as missing logistics tracks, the system still generates an incomplete reality mapping map, but additionally marks the evidence as sparse in the difference results; If multiple theoretical residual templates have the highest similarity, the template with smaller structural modifications is selected first, or multiple candidate categories are output for verification; If the blockchain node is in an abnormal state, such as incomplete consensus or contract execution failure, even if the process is normal, the data to be stored will be temporarily stored in a buffer queue, and the on-chain event will be triggered again after the node recovers, without directly losing business records; In this subway project, the logistics trajectory of a high-voltage switchgear showed that it had deviated from the standard factory path and stopped at an unauthorized location before leaving the factory. At the same time, the quality inspection report almost overlapped with that of another piece of equipment. After system comparison, it was found that this deviation was closest to the theoretical residual of unauthorized processing, with a similarity of 0.86, which was higher than the blocking threshold of 0.80. Therefore, it triggered a rejection of uploading to the chain and wrote a blocking event for batch GZ-KG-021 with suspected unauthorized processing in the audit chain. After subsequent verification, it was confirmed that the equipment was indeed partially assembled by an unregistered outsourcing point. Therefore, the system updated the spatial perturbation template of the corresponding abnormal operator. The purpose of this step is to complete the structured verification of the authenticity of off-chain multi-source processes before writing to the blockchain, so as to distinguish between normal business fluctuations and real violations, and avoid the problem of erroneous data being reliably stored due to relying solely on the blockchain's tamper-proof feature. In this embodiment, the multi-source process data includes: Structure list data, work order flow data, and quality inspection process data collected by the business system; Identification trajectory data, node timestamp data, and load characteristic data collected by logistics sensing nodes; Transaction log data and contract call log data collected by blockchain nodes; The blockchain node status data includes node online status, consensus completion status, and contract execution status.

[0017] This embodiment provides a refined collection mechanism for multi-source process data. Specifically, if the aforementioned solution only collects manufacturing-side work order data, although it can reconstruct part of the production path, when materials are replaced midway or processing occurs outside the authorized path, the data from a single source has a high risk of being tampered with by humans. Therefore, it is necessary to introduce the joint collection of business system data, logistics perception data, and blockchain node operation data. Specifically, in the same high-voltage switchgear scenario, the structural list data is used to characterize the composition relationship of the equipment, such as the cabinet assembly corresponding to circuit breaker module A, busbar module B, and protection module C; the work order flow data is used to record the workstations, operators, completion times, and rework information of each module; the quality inspection process data records the results such as withstand voltage value, insulation resistance, and temperature rise test curve; at the same time, the logistics sensing nodes collect RFID or QR code identification trajectories, sampling timestamps of each station, and load characteristic data; the load characteristics are not limited to weight, but can also be volume, vibration level, or cabinet tilt angle; for example, if the standard factory load of a certain equipment should be 950 kg, but the sensing node collects a load of 905 kg at the transfer point after leaving the factory, then this difference can be used as one of the potential evidences of anomalies; Furthermore, blockchain nodes also collect transaction logs and contract call logs; transaction logs can record whether a batch of devices has applied for on-chain access, transaction hashes, and submission times; contract call logs can record which business role, when, which on-chain interface was called, and whether the return status was successful; node status data includes at least three dimensions: node online status indicates whether the node can participate in the network, consensus completion status indicates whether a certain write has been confirmed by the network, and contract execution status indicates whether the business logic is running correctly; in this way, the system can distinguish between two different problems: abnormalities in the process itself and abnormalities in the on-chain facilities. As a supplementary processing mechanism, if a logistics sensing node goes offline, the system marks the corresponding time slice of that site as missing, and does not directly identify it as an anomaly; if the business system data and the on-chain transaction log are in reverse chronological order, for example, the business side shows a submission at 15:00 while the on-chain log shows completion at 14:58, the system will prioritize retaining the original timestamp and marking the clock deviation, and hand it over to the subsequent time alignment module for processing; if the contract call log is missing, but the transaction log exists, the record can be marked as a semi-transparent transaction, still allowing participation in auditing but with a reduced trust level; During a shipment of a subway traction substation project, the system read from the manufacturing execution system that the final assembly was completed, from the logistics node that the equipment had left the factory and passed through the L2 transit station, and from the blockchain node that the notarization application had been submitted but the contract execution failed. Since the business process and the logistics process are basically the same, and the only problem on the chain is the contract failure, the system will not misjudge it as a supply chain violation, but will classify it as a system anomaly branch that needs to be retried on the chain. The purpose of this step is to enhance the integrity of the evidence chain through mutual verification between heterogeneous data sources, thereby enabling full-chain observation of materials, processes, logistics, and on-chain behavior. In this embodiment, the construction of the ideal reference map includes: Determine the material hierarchy and process dependencies based on the object structure data; Standard processing routes, standard logistics routes, and standard quality inspection routes are determined based on standard process data; Organize the material nodes, process nodes, logistics nodes, and quality inspection nodes into a directed acyclic graph; Among them, the nodes and / or edges in the ideal benchmark graph are associated with standard time attributes, standard spatial attributes, standard quality attributes, standard identity attributes, and standard resource consumption attributes.

[0018] This embodiment provides a mechanism for constructing an ideal benchmark map. Specifically, if the aforementioned scheme directly uses historical actual data as the baseline, it is easy to introduce hidden violation features in historical samples, which may lead to misjudging abnormal patterns as compliant patterns. Therefore, this embodiment does not use historical averages, but constructs an ideal benchmark map based on engineering design and standard operating procedures. Specifically, in this high-voltage switchgear scenario, the object structure data provides the material hierarchy, such as the primary main circuit component G1, secondary protection component G2, and mechanical interlock component G3 under the cabinet node G0; G1 is further subdivided into circuit breaker unit M1, busbar unit M2, and insulation support component M3; the process dependency relationship stipulates that M2 must first complete cutting and bending, then enter surface treatment, and then enter assembly; the standard process data further specifies the standard processing path, standard logistics path, and standard quality inspection path; for example, the circuit breaker unit enters the cleaning station P1, the assembly station P2, the commissioning station P3 from the incoming material inspection area I1, then enters the withstand voltage quality inspection area Q1, and finally enters the waiting-to-ship area S1; In terms of graph organization, material nodes, process nodes, logistics nodes, and quality inspection nodes are uniformly organized into a directed acyclic graph. For ease of simulation verification, the following simplified relationship can be constructed: Material node M2 ​​forms a semi-finished product through process node P2, and then is transferred to P3 through logistics edge E23, entering quality inspection node Q1. The direction of the edge in this graph represents the legal flow direction. It is not allowed to return from Q1 to P2 in reverse unless there is a rework branch definition in the graph. Each node and edge is also associated with standard attributes: standard time attributes are used to describe planned working hours and allowed windows, standard spatial attributes are used to describe authorized workstations and standard paths, standard quality attributes are used to describe inspection thresholds and process quality indicators, standard identity attributes are used to describe the executable entity, such as which type of work group or which equipment is allowed to operate, and standard resource consumption attributes are used to describe reasonable resource usage or energy consumption ranges. To illustrate the attribute organization method, assume that the standard time window for busbar assembly node P2 is 30 minutes, the standard space is assembly line A-2, the standard identity is certified assembly worker group Z3, the standard quality is the assembly torque range of 45 to 50 Nm, and the standard resource consumption is the electrical energy consumed in this process of 60 kWh. If the actual data shows that assembly line B-7 is completed and takes 120 minutes, even if the final output can pass some tests, this deviation will be explicitly reflected in the subsequent differential calculation. As a supplementary processing mechanism, if the object structure data itself has multiple versions in parallel, such as a subway project-specific version and a general version of the same model, the system first selects the corresponding baseline map according to the contract configuration code; if the standard process allows legal branches, such as export packaging and domestic sales packaging being different, then the branch nodes are explicitly set in the ideal baseline map, rather than treating them as anomalies during subsequent inspection; if some attributes cannot be quantified, such as manual visual confirmation, the system can use discrete level coding, such as three levels A, B, and C, and participate in the difference; In subway projects, the withstand voltage test of the project-specific switchgear must be completed in laboratory Q1, while ordinary civilian switchgear can be completed in laboratory Q2. In this embodiment, after constructing the baseline map according to the project configuration, if GZ-KG-021 is recorded as having completed the withstand voltage test in Q2, a spatial attribute offset will be immediately formed, and it will not be mistakenly absorbed as a normal variant. The purpose of this step is to establish a standard process reference system that is independent of historical noise, computable and traceable, so as to achieve a unified coordinate basis for subsequent theoretical simulation and reality mapping. In this embodiment, the anomaly operator library includes substitution operators, unauthorized processing operators, and report reuse operators; each anomaly operator is encapsulated by a triggering rule, a target node or directed edge location label, and a mathematical perturbation function for modifying target attribute values ​​and graph topology; parameterized injection includes: Modify the quality and resource consumption attributes of material nodes based on substitution operators; Modify the spatial and temporal attributes of process nodes or logistics nodes based on the overstepping processing operator; Modify the quality inspection feature attributes of the quality inspection node based on the report reuse operator; And generate corresponding theoretical process anomaly simulation states based on the attribute changes and / or path changes after injection.

[0019] This embodiment provides an anomaly operator library and parameterized injection mechanism; specifically, an ideal baseline graph alone is insufficient to identify complex violations, because real-world anomalies are often not just simple time delays or missing fields, but rather combinations of deviations with industry characteristics; therefore, this embodiment pre-abstracts typical risk patterns into injectable anomaly operators. Specifically, the substitution operator is used to simulate the replacement of low-specification materials or the replacement of key components with non-original parts. Taking the busbar unit as an example, ideally its quality attribute vector can be simplified to a combination of attributes such as conductivity 100, impedance stability 100, and surface coating consistency 100, while the resource consumption attribute in the resource consumption dimension is 100. After applying the substitution operator, its quality attribute combination can be changed to 88, 74, and 93, and the resource consumption attribute in the resource consumption dimension can be changed to 67. The simulation state generated in this way does not necessarily change the processing path, but it will deviate from the standard in both the quality and resource consumption dimensions, forming a theoretical residual with industry-specific characteristics. The unauthorized processing operator is used to simulate production at unauthorized workstations, illegal subcontracting at unregistered sites, or cross-regional unregistered processing. For example, process P3 should originally be completed in factory A, with a standard time of 40 minutes. After applying this operator, the spatial attribute is replaced with the external point P7, the time attribute increment is set to +180 minutes, and a round-trip path outside the factory is inserted on the logistics edge. The resulting simulation state has both changes in node attributes and changes in path structure. The report reuse operator is used to simulate the cloning of quality inspection reports or the misappropriation of test results. For example, ideally, the test characteristics of quality inspection node Q1 include a withstand voltage waveform curve and a set of timestamp sequences. After applying this operator, the curve is not generated arbitrarily, but a feature template is extracted from a historical compliance report, and only non-core fields such as report time and batch number are fine-tuned. The theoretical anomaly formed in this way is closer to the real data tampering pattern, because in reality, the anomaly injection process is usually based on the cloning of features from historical compliance reports, rather than randomly generating new data. Furthermore, the system can configure an injection intensity range for each type of operator; for example, the time increment of the over-authorization processing operator can be set to 60 to 240 minutes, the spatial offset can be limited to within 30 kilometers outside the authorized area, and the feature similarity of the report reuse operator can be set to above 0.95; the system generates multiple simulation states in batches according to these ranges, so that the same anomaly category covers different strengths; As a supplementary processing mechanism, if a certain type of anomaly is theoretically impossible in the current project, such as when the contract explicitly prohibits outsourcing and there are no outsourcing nodes in the actual organizational structure, the system can temporarily disable the corresponding operator to prevent meaningless matching. If a new type of deviation not covered by the operator library occurs in reality, such as a legal workstation using uncalibrated temporary experimental equipment, the deviation will be presented as failing to reach the similarity threshold with any preset template in subsequent matching and will be sent to the learning entry of the new rules instead of being forcibly classified. In the subway project, to address the risk of circuit breaker modules originating from unauthorized channels, the system applies a substitution operator to the circuit breaker nodes, generating a set of theoretical simulation states where resource consumption is below a preset lower limit, but appearance and weight remain within preset tolerances, and test curves show numerical deviations less than a preset alarm threshold. Simultaneously, to address the risk of unauthorized busbar welding outside the factory, an overstepping processing operator is applied, generating a set of theoretical simulation states where the path detours to external cooperative points and there are abnormal stops before final assembly. Subsequent real-world residuals can be more accurately identified as long as they closely resemble a certain type of structure. The purpose of this mechanism is to transform violation patterns from industry experience into computable and comparable anomaly templates, thereby achieving a transition from generalization bias detection to categorical anomaly identification. In this embodiment, the generation of the reality mapping map includes: Map the multi-source process data corresponding to the unified data index to the corresponding nodes and edges of the ideal benchmark graph; Missing fields are filled in according to preset data source rules, and consistency checks are performed on the filled in results; Perform priority adjudication on conflicting fields; The priority adjudication is executed in order of preset data trust levels. The data trust level is determined at least based on the data source, signature status and time integrity. Specifically, preset weights are assigned to data acquisition terminals with different security levels, digital signature status with anti-tampering characteristics and continuous and unbroken timestamp sequences, and the weighted evaluation total score is calculated to quantify the data trust level in order to generate a real mapping map that is isomorphic to the ideal benchmark map.

[0020] This embodiment provides a reality mapping map generation mechanism; specifically, the aforementioned solution will face the following technical defects when deployed in practice: the off-chain data of the same device is not always complete and consistent; for example, the manufacturing execution system shows that it has been completed, but the logistics system has not recorded the transfer, and the quality inspection system still has missing fields; if the field completion and conflict resolution are not completed first, the subsequent difference results will introduce systematic matching errors caused by the missing records from multiple sources; Specifically, the system first projects data from various sources onto the nodes and edges corresponding to the ideal baseline graph based on a unified data index. For example, if the N3 assembly node requires three fields: completion time, workstation, and operator signature, but only the first two are collected in reality, the operator signature is missing, the system completes the data according to preset data source rules. For instance, if the manufacturing execution system workstation terminal login information matches the assembly time window, the identity of the currently logged-in personnel can be used as a candidate completion value. If the candidate completion value matches the access control record or workstation camera attendance record, it is completed as a formal field; otherwise, it is only used as a low-confidence placeholder and not considered strong evidence. For conflicting fields, the system performs priority adjudication. For example, for the arrival time of the same transit edge E34, the logistics sensing node records it as 10:12, while the manual scanning system records it as 10:18. If the logistics sensing node data has a device signature and a complete time series, while the manual scanning record has no signature and has a breakpoint, then 10:12 is preferentially retained as the actual mapping value, while 10:18 is stored in the bypass evidence set. The trust level can be formed by combining the data source, signature status, and time integrity. For example, records with device signatures and continuous time chains can be set as the highest level among the three levels, while records manually entered without signatures can be set as the lowest level. To perform micro-level extrapolation, we can assume that a node has three source values: V1 from the RFID logistics sensing node, with a confidence score of 0.9; V2 from manual input, with a confidence score of 0.5; and V3 from video analysis results, with a confidence score of 0.7. If the three conflict, V1, corresponding to 0.9, will be used first. If V1 is missing and the difference between V2 and V3 is less than or equal to the allowable error (e.g., the time difference does not exceed 2 minutes), then the weighted average of the two or the higher confidence source value can be used. If the difference is greater than the allowable error, then this field is marked as conflict unresolved and an uncertain marker is retained in the real-world mapping map. After completion and adjudication, the system generates a real-world mapping graph that is isomorphic to the ideal baseline graph. In this embodiment, isomorphism mainly refers to the one-to-one correspondence between node types and main connection frames. For example, if the ideal graph has four main nodes: incoming materials, assembly, quality inspection, and outbound, the real-world mapping graph also retains four main nodes, but the attribute values ​​and specific edge details may carry completion values, conflict states, or abnormal markers. As a supplementary processing mechanism, if a key field cannot be completed and has no reliable alternative source, such as the original quality inspection waveform file being lost, the system marks the node as missing key evidence and increases the uncertainty weight in subsequent similarity calculations. If two conflicting fields have the same level of credibility, the system does not forcibly select one of them, but generates two parallel reality candidate branches to participate in subsequent matching. When the parallel reality candidate branches participate in matching, the system follows a conservative evaluation strategy, prioritizing the branch that results in a higher similarity in the overall matching, i.e., is more inclined to the abnormal state of the process, as the final reality mapping map, thereby avoiding the concealment of potential violations due to random selection. If the completion result fails the consistency check, such as the access control record showing that the operator is not present, the completion value is rolled back and marked for manual verification. In the subway project, the final assembly node of GZ-KG-021 lacked an operator identity, but the person logged into the manufacturing execution system terminal was Zhang. The access control showed that Zhang entered the assembly area during that time period, and the camera check-in also corresponded to this. Therefore, the system completed the completion of this field. Another example is that there is a 6-minute conflict between the outbound time and the logistics access control. Since the access control data is signed and continuous, the system ultimately uses the access control time as the actual mapping value. The purpose of this step is to first form a unified and clearly defined real-world process expression in a multi-source, heterogeneous, and uneven data environment, so as to achieve the stability of subsequent differential and comparison. In this embodiment, the generation of the theoretical residual map set and the actual residual map includes: Extract the baseline node attributes, baseline edge attributes, and baseline path sequences from the ideal baseline map; The simulation node attributes, simulation edge attributes, and simulation path sequences in each theoretical process abnormal simulation state are respectively compared with the corresponding baseline node attributes, baseline edge attributes, and baseline path sequences to generate a theoretical residual map. The real node attributes, real edge attributes, and real path sequences in the real mapping graph are respectively compared with the corresponding reference node attributes, reference edge attributes, and reference path sequences to generate a real residual graph. Among them, differential calculation is used to characterize attribute offset, path offset and time offset. Path offset is the difference between the node insertion, deletion and modification of the real or simulated path sequence and the baseline path sequence. Time offset is the difference or normalized difference between the corresponding node timestamp and the standard time attribute. Specifically, the difference calculation includes: for the numerical attributes of nodes and edges, vector subtraction is used to extract the difference as the attribute offset; for the topological structure of the graph path, the topological symmetric difference operation of the node set and edge set of the benchmark graph and each theoretical process abnormal simulation state or real mapping graph is used to extract the relatively inserted extra nodes and edges as well as the relatively missing deleted nodes and edges, as the path offset.

[0021] This embodiment provides a residual map generation mechanism. Specifically, if the actual mapping map is directly compared with the theoretical simulation state as a whole, it is easily interfered with by differences in absolute scale and differences in individual project parameters. Therefore, this embodiment first performs differences relative to the same ideal benchmark map, and uniformly converts the comparison objects into data forms that represent the degree of deviation from the standard. Specifically, the system first extracts the baseline node attributes, baseline edge attributes, and baseline path sequence from the ideal baseline map. For ease of deduction, it is assumed that the baseline path sequence is N1 incoming material, N2 assembly, N3 quality inspection, and N4 outbound, with standard completion times of 08:00, 10:00, 11:00, and 12:00, respectively. A certain unauthorized processing theoretical simulation state is injected by inserting an unauthorized node X between N2 and N3, and N3 and N4 are correspondingly delayed to 13:30 and 14:10. Then, its path offset can be represented as the insertion of X between N2 and N3, and the timing offset can be represented as N1 and N2 offsets of 0, N3 delayed by 150 minutes, and N4 delayed by 130 minutes. If the quality attribute of a node also changes, the difference in that attribute is recorded in the node residual. Similarly, the system performs differential analysis on the real-world mapping map; for example, the actual real-world path is obtained by inserting an unauthorized outsourcing stop point P7 between N2 and N3; the real-world times are 08:02, 10:05, 11:40, 13:20, and 14:05 respectively; compared with the baseline, the path offset is also manifested by inserting an extra node between N2 and N3, and the time offset is manifested by N3 and N4 being significantly lagging behind; if the report fingerprint of the quality inspection node has a higher overlap with the historical report features than a preset threshold, then the node will also have a quality inspection feature offset; To achieve a simplified representation, the residual of each node can be compressed into a triplet representation containing attribute offset, path offset, and temporal offset. For example, the actual residual of node N3 can be represented as a quality inspection fingerprint similarity anomaly of 0.82, one preceding inserted node, and a delay of 140 minutes. After organizing the residuals of multiple nodes along the graph structure, a real residual map is formed. The theoretical residual map set also adopts the same representation method, so the two are comparable. Furthermore, the timing offset can be either an absolute value or a normalized value; for example, the standard interval between final assembly and quality inspection is 60 minutes, but the actual interval is 180 minutes, so the absolute offset is +120 minutes and the normalized offset is +2.0. When performing the above normalization calculation, in order to avoid the error of division by zero due to the standard time interval or standard attribute value being zero, the system introduces a preset minimal positive constant into the denominator for smoothing. For example, when the standard time for a check-in node at a certain instant is defined as 0 minutes, the normalization denominator takes this minimum positive constant to ensure the stability of the calculation logic in any extreme parallel process; if different projects have different durations, using a normalized value with a smoothing term is more conducive to cross-project comparison; path offset can be described by node insertion, deletion, and replacement; for example, skipping a necessary quality inspection station is deletion, passing through an unauthorized station is insertion, and replacing an authorized station with an unauthorized station is replacement; As a supplementary processing mechanism, if there are rework branches in the real-world graph, but there is already a valid rework template in the ideal graph, then during the difference process, it will not be directly identified as an abnormal insertion, but will first be mapped to a valid rework path; if the timestamp is missing and the absolute offset cannot be calculated, then the interval offset or missing test mark can be used as a substitute; if the attribute units of a node are different, such as cost being amount and quality being score, then the system will first normalize it and then write it into the residual to avoid a single large numerical field dominating the result; In the subway project, the actual path of GZ-KG-021 showed a situation where it went to the outsourcing point P7 after assembly and then returned to the factory for quality inspection, and the quality inspection time was 145 minutes later than the standard. The actual residual map generated by the system was highly consistent with the residual map of the over-authorization processing theory in terms of path insertion and time lag. If the quality inspection characteristics were found to be duplicated with historical reports, the attribute offset of the report reuse would be superimposed to provide a basis for subsequent multi-template competitive matching. The purpose of this step is to convert the raw process data into a deviation signal expression under a unified semantics, thereby enabling a structured comparison between different anomaly patterns. In this embodiment, topological similarity comparison includes: Calculate the graph edit distance between the actual residual map and each theoretical residual map, or encode the features of the actual residual map and each theoretical residual map to generate residual feature vectors and calculate vector similarity; Candidate matching sequences are generated by sorting the graph edit distance from smallest to largest or the vector similarity from largest to smallest. The theoretical residual map that ranks first and meets the preset matching conditions is selected as the optimal matching result. The preset matching conditions include: the similarity of the first and second ranked items is greater than the preset absolute similarity threshold, and the difference between the similarity of the first and second ranked items is greater than the preset discrimination threshold. Among them, the theoretical residual map category corresponding to the optimal matching result is output as the process anomaly category.

[0022] This embodiment provides a topological similarity comparison mechanism; specifically, residual maps alone are insufficient to output anomaly categories, because different anomalies may exhibit similarities in a single attribute, such as both potentially causing time delays; therefore, this embodiment introduces graph-level comparison to determine anomaly types based on the overall structure of the deviation rather than single-field exceedances. Specifically, the first method is to calculate the graph editing distance. Editing operations can be simplified to node insertion, node deletion, node replacement, and attribute modification, each with a cost. For example, the cost of inserting an unauthorized logistics node is set to 3, the cost of replacing an authorized workstation with a spatial overreach workstation is set to 2, and the cost of simply delaying the time by 60 minutes is set to 1. Assuming that the actual residual graph and the overreach processing theoretical residual graph only require one node replacement and one attribute modification, the total cost is 3; while the relationship with the report reuse theoretical residual graph requires two attribute modifications and one path insertion, the total cost is 5; and the relationship with the alternative material theoretical residual graph requires three node attribute rewrites, the total cost is 6. Therefore, the graph with the smallest editing distance corresponds to the overreach processing. The second approach is to first perform feature encoding. For example, each residual map is encoded into a feature vector of length 6. The six components of the vector can represent the strength of quality offset, the strength of resource consumption offset, the degree of spatial overweighting, the number of path insertions, the degree of temporal lag, and the degree of report duplication, respectively. If the component values ​​of a real residual vector are 0.10, 0.05, 0.90, 0.80, 0.85, and 0.20, and the component values ​​of the three theoretical vectors are 0.80, 0.70, 0.10, 0.05, 0.15, and 0.10 for A, 0.05, 0.10, 0.92, 0.78, 0.88, and 0.15 for B, and 0.02, 0.05, 0.15, 0.10, 0.20, and 0.95 for C, the system performs matching based on the cosine similarity calculation rule, that is, it calculates the similarity score in the form of the quotient of the inner product of the two vectors and the product of their magnitudes. To prevent a few theoretical simulation states from exhibiting zero vectors after feature encoding, which could lead to a division-by-zero anomaly due to a zero product of the modulus and length, the system uniformly adds a preset, minimal positive constant to the denominator for safety bias. Furthermore, considering the business sensitivity of different dimensions, the system can apply a pre-defined weight matrix to each component, for example, assigning higher weights to spatial overreach and report duplication, thereby effectively preventing normal fluctuations in secondary dimensions from interfering with the identification of the main violation features. Through the above calculations, the weighted cosine similarity between the real vector and B is the highest, and it can be considered the closest to the overreach processing category. After sorting, the system generates a candidate matching sequence. If similarity sorting is used, for example, B equals 0.91, C equals 0.63, and A equals 0.28. The system selects B, which ranks first, as the candidate optimal result, but it also needs to meet preset matching conditions, such as a similarity of at least 0.75 and a difference of at least 0.10 from the second-ranked result, to avoid ambiguity in the geometric consistency measure due to multi-class confusion. Only when these conditions are met simultaneously will the category be output as the final anomaly category. As a supplementary processing mechanism, if the first matching result does not meet the preset matching conditions, such as the highest similarity being only 0.52, the system will not force classification and can output "not matched existing template" and send it for manual review; if the difference between the first and second results is less than the preset discrimination threshold, such as 0.78 and 0.76, it is difficult to distinguish them by automation alone, so a composite suspected category can be output, such as "overstepping processing / report reuse pending confirmation"; if the graph edit distance method and the vector similarity method have inconsistent conclusions, the system can process them according to the preset strategy, such as giving priority to the graph edit distance result which is more sensitive to paths, or performing a weighted fusion of the two. In the subway project, the actual residual of GZ-KG-021, after being encoded, has a similarity of 0.91 with the theoretical vector of unauthorized processing, 0.63 with report reuse, and 0.28 with substitute materials. The difference meets the preset discrimination condition, so the system directly outputs unauthorized processing as a process anomaly category. If another piece of equipment subsequently shows obvious outsourcing delays and duplicate reports, with its two highest similarities being 0.81 and 0.79 respectively, the system marks it as a composite anomaly pending confirmation, rather than directly classifying it into a single category. The purpose of this step is to map real-world deviations to interpretable anomaly patterns through structural-level matching, thereby achieving accuracy and auditability of anomaly category outputs. In this embodiment, the generation of the process status determination result includes: When the similarity is greater than or equal to the blocking threshold, it is determined to be an abnormal process state, and a rejection instruction for uploading to the chain is generated. When the similarity is less than or equal to the release threshold, the process is determined to be in a normal state, and a direct on-chain instruction is generated. When the similarity is greater than the allow threshold and less than the block threshold, the process is determined to be in a pending confirmation state, and an on-chain instruction with an additional process exception label is generated. The process exception label includes at least a process exception category identifier and an exception uncertainty identifier.

[0023] This embodiment provides a dual-threshold process status determination mechanism. Specifically, if only a single threshold is used, two types of problems will be encountered in real industrial scenarios: if the threshold is set too low, a large number of normal fluctuations will be easily misjudged; if the threshold is set too high, some real violations will be missed. Therefore, this embodiment uses the threshold transition range between the release threshold and the blocking threshold to handle uncertain samples. Specifically, the system pre-sets a blocking threshold higher than the release threshold; for ease of deduction, it is assumed that the release threshold is 0.35 and the blocking threshold is 0.80; when the similarity between a certain actual residual and the optimal theoretical residual is not higher than 0.35, it is considered that the deviation is insufficient to constitute a known abnormal pattern, the process is output as normal, and a direct on-chain instruction is generated; for example, if there is only a 20-minute logistics delay, but the path and quality inspection are normal, the similarity may be 0.21, and it can be directly on-chain. When the similarity is not less than 0.80, it is considered to be highly overlapping with a known abnormal template, the abnormal state of the process is output, and a rejection instruction for uploading to the chain is generated; for example, the similarity of GZ-KG-021 in the above example is 0.86, which falls into the blocking interval, and its formal traceability data should be rejected from being uploaded to the chain to avoid polluting the trusted records on the chain. When the similarity is between 0.35 and 0.80, the system does not simply allow or block the process, but outputs a status indicating that the process is pending confirmation. At this time, an on-chain instruction with an additional anomaly label is generated, so that the chain retains the fact that the batch had anomaly signs, while clarifying that it still needs to be verified. The anomaly label includes at least an anomaly category identifier and an anomaly uncertainty identifier. For example, it can be written as an anomaly category identifier corresponding to suspected report reuse and an anomaly uncertainty identifier corresponding to medium confidence. Furthermore, attaching a label to the blockchain does not equate to confirming a violation; it serves two purposes in the project: first, downstream installation, maintenance, or monitoring systems can detect the risk in this batch and should prioritize a review; second, if subsequent manual review confirms that there are no errors, the cancellation statement can be added to the blockchain as a supplementary transaction to form a complete audit chain. As a supplementary processing mechanism, if the similarity is exactly equal to the threshold boundary, it will be executed according to the preset rules. For example, if it is equal to the release threshold, it will be classified as normal, and if it is equal to the blocking threshold, it will be classified as abnormal. If the same device repeatedly triggers different states in different time windows, the system can update the state according to the principle of prioritizing the latest evidence or prioritizing the highest risk. If the blockchain node is temporarily unavailable when the attached tag is uploaded to the chain, the tag and the original evidence summary will be retained in the local trusted cache first, and then written after the network is restored to prevent the loss of the event to be confirmed. In the subway project, a batch of auxiliary relays had an extra storage stop on the logistics route compared to the standard route. However, the stop occurred within the registered park, and the quality inspection and resource consumption were normal, resulting in a final similarity of 0.47. The system determined this batch to be in a process confirmation pending state and added a label to the on-chain record indicating a slight spatial path anomaly and a category bias towards unauthorized processing with moderate uncertainty. If subsequent verification confirmed that it was just a temporary storage location adjustment within the park, another anomaly removal explanation could be added to the on-chain record. The purpose of this step is to handle samples with insignificant features that fall between normal and abnormal by setting a risk buffer, thereby achieving robustness and prudence in on-chain evidence storage. In this embodiment, adjusting the anomaly operator library or dual thresholds based on the process status determination results of completed processing includes: Extract the process status determination results of completed treatments and the corresponding actual residual maps; When the process status determination result is an abnormal process status, update the attribute perturbation parameters or rule templates of each abnormal operator in the abnormal operator library. When the process status determination result is that the process is pending confirmation, update the interval boundary of the dual thresholds; When the process status determination result is that the process is in a normal state, the abnormal operator library and the dual threshold remain unchanged.

[0024] This embodiment provides a closed-loop self-adjustment mechanism. Specifically, if the aforementioned scheme uses fixed templates and fixed thresholds for a long time, it will face two problems: first, the abnormal patterns are dynamically evolving, and the old templates may gradually become ineffective; second, the normal fluctuation range will also change under different projects, different seasons, and even different logistics environments. Therefore, this embodiment updates the abnormal operator library and dual thresholds based on the results of the completed handling. Specifically, the system first extracts the judgment results that have completed the manual review, on-site investigation, or responsibility determination process, and simultaneously extracts the corresponding actual residual map. If the final conclusion is that the process is in an abnormal state, it means that the actual residual does indeed represent a type of real violation. At this time, it is necessary to update the attribute disturbance parameters or rule templates in the abnormal operator library. For example, if multiple outsourcing events are characterized by inserting outsourcing point P7, delaying the final assembly to quality inspection by 120 to 180 minutes, and having an abnormal return to the factory before quality inspection, the typical time disturbance range of the unauthorized processing operator can be converged to a more realistic range, and the abnormal return to the factory can be added to the rule template. If the final conclusion is that the process is pending confirmation, it indicates that the original dual thresholds are insufficient for handling ambiguity in this type of sample, and the interval boundaries need to be updated. For example, if the system continuously finds multiple batches of samples with similarity between 0.45 and 0.52, and after verification, they are all normal transfers within the park, then the similarity mapping for this type of pattern can be lowered, or the sensitive interval width between the release threshold and the blocking threshold can be appropriately reduced. Conversely, if samples with a similarity of around 0.70 are repeatedly confirmed as genuine anomalies, then the blocking threshold can be lowered from 0.80 to 0.75 to improve the timeliness of interception. If the final conclusion is that the process is in a normal state, it means that the existing template and threshold do not need to be modified for this sample. The system keeps the abnormal operator library and dual threshold unchanged to avoid model drift caused by individual normal samples. To facilitate micro-level analysis, we can assume there are 10 events to be confirmed within a month. Of these, 8 have a similarity between 0.38 and 0.43 and are ultimately confirmed as normal, while 2 have a similarity between 0.72 and 0.76 and are ultimately confirmed as abnormal. Based on this, the system can fine-tune the release threshold from 0.35 to 0.40 and the blocking threshold from 0.80 to 0.75. For example, if 3 confirmed report reuse events all show a report fingerprint repetition rate higher than 0.97, but the sampling time is only changed by 1 to 3 minutes, the time perturbation template of the report reuse operator can be narrowed accordingly, thereby enhancing the targeting of identification. As a supplementary processing mechanism, if the number of samples that have been processed is insufficient, for example, if a new anomaly occurs only once, the system can temporarily store it as a candidate rule without immediately modifying the formal operator library; if different personnel have conflicting conclusions on the same event, the final archived conclusion shall prevail, and the disputed label shall be retained; if adjusting the threshold leads to a large-scale reclassification risk of recent normal samples, the system can first run the new threshold in the background parallel verification mode, compare the old and new judgment results in parallel, and then officially switch after passing the test. During the execution of the subway project, the system intercepted four high-voltage switchgear units with outsourced manufacturing issues. Review revealed that these four units all exhibited a common pattern in their residual performance: after assembly, the equipment left the factory, quality inspection was delayed by approximately 150 minutes before return, and the outsourced workstations were concentrated at P7 and P9. Based on this, the system updated the spatial template and timing disturbance parameters of the unauthorized processing operator. Meanwhile, six other batches of pending events were verified to be compliant transfers within the park. Therefore, the system appropriately relaxed the release boundaries for spatial path changes with deviations less than the preset tolerance threshold, reducing subsequent false alarms. The purpose of this mechanism is to enable the system to absorb the effective experience of verified cases during long-term operation, thereby achieving continuous calibration of the anomaly template and judgment threshold.

[0025] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology, characterized in that, include: Collect multi-source process data and blockchain node status data from the electrical complete equipment supply chain, perform time alignment, field mapping and unique identifier association based on a unified standard time base on the multi-source process data, and generate a unified data index; An ideal benchmark graph is constructed based on pre-configured object structure data and standard process data. The ideal benchmark graph is a directed graph representing the standard supply chain process. Based on the anomaly operators in the anomaly operator library, the ideal benchmark spectrum is parameterized and injected to generate multiple theoretical process anomaly simulation states. Then, the difference calculation is performed between each theoretical process anomaly simulation state and the ideal benchmark spectrum to generate a theoretical residual spectrum set. The multi-source process data corresponding to the unified data index is mapped to the nodes and edges of the ideal benchmark graph to generate the real mapping graph. The real mapping graph and the ideal benchmark graph are then differentially calculated to generate the real residual graph. A topological similarity comparison is performed between the actual residual map and the theoretical residual map set to determine the process anomaly category and similarity. Obtain preset dual thresholds, which include a release threshold and a blocking threshold, wherein the blocking threshold is strictly greater than the release threshold. Both are initialized by the system based on the confidence interval of the topological similarity distribution of historical compliant supply chain processes. Generate process status determination results based on the similarity and the preset dual thresholds. Based on the process status determination result and the blockchain node status data, control the process data to be stored corresponding to the unified data index to be uploaded to the blockchain through smart contract, refuse to upload to the blockchain, or upload to the blockchain after attaching a process abnormal label. The process status determination result is verified based on the actual compliance status of the target process as reported by manual or system audit, and the abnormal operator library or the dual threshold is adjusted based on the verification result.

2. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The multi-source process data includes: Structure list data, work order flow data, and quality inspection process data collected by the business system; Identification trajectory data, node timestamp data, and load characteristic data collected by logistics sensing nodes; Transaction log data and contract call log data collected by blockchain nodes; The blockchain node status data includes node online status, consensus completion status, and contract execution status.

3. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The construction of the ideal baseline map includes: Based on the object structure data, determine the material hierarchy and process dependencies; Based on the aforementioned standard process data, standard processing routes, standard logistics routes, and standard quality inspection routes are determined; Organize the material nodes, process nodes, logistics nodes, and quality inspection nodes into a directed acyclic graph; The nodes and / or edges in the ideal baseline graph are associated with standard time attributes, standard spatial attributes, standard quality attributes, standard identity attributes, and standard resource consumption attributes.

4. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The anomaly operator library includes substitution operators, unauthorized processing operators, and report reuse operators; each of the anomaly operators is encapsulated by a triggering rule, a target node or directed edge location label, and a mathematical perturbation function for modifying target attribute values ​​and graph topology; the parameterized injection includes: Modify the quality attributes and resource consumption attributes of the material node based on the aforementioned substitution operator; Based on the above-mentioned over-authorization processing operator, modify the spatial and temporal attributes of the process node or logistics node; Modify the quality inspection feature attributes of the quality inspection node based on the report reuse operator; And generate corresponding theoretical process anomaly simulation states based on the attribute changes and / or path changes after injection.

5. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The generation of the reality mapping map includes: Map the multi-source process data corresponding to the unified data index to the corresponding nodes and edges of the ideal benchmark graph; Missing fields are filled in according to preset data source rules, and consistency checks are performed on the filled in results; Perform priority adjudication on conflicting fields; The priority decision is executed in order of preset data trust levels. The data trust level is determined at least based on the data source, signature status and time integrity. Specifically, the data trust level is quantified by assigning preset weights to data acquisition terminals with different security levels, digital signature status with anti-tampering characteristics and continuous and unbroken timestamp sequences, and calculating a weighted evaluation total score to generate a real mapping map that is isomorphic to the ideal benchmark map.

6. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The generation of the theoretical residual map set and the actual residual map includes: Extract the baseline node attributes, baseline edge attributes, and baseline path sequences from the ideal baseline map; The simulation node attributes, simulation edge attributes, and simulation path sequences in each of the theoretical process abnormal simulation states are respectively compared with the corresponding baseline node attributes, baseline edge attributes, and baseline path sequences to generate a theoretical residual map. The real node attributes, real edge attributes, and real path sequences in the real mapping map are respectively compared with the corresponding reference node attributes, reference edge attributes, and reference path sequences to generate a real residual map. The differential calculation is used to characterize attribute offset, path offset and temporal offset. The path offset is the difference in node insertion, deletion and modification between the real or simulated path sequence and the baseline path sequence. The temporal offset is the difference or normalized difference between the corresponding node timestamp and the standard time attribute. Specifically, the difference calculation includes: for the numerical attributes of nodes and edges, using vector subtraction and extracting the difference as the attribute offset; for the topological structure of the graph path, using the topological symmetric difference operation between the node set and edge set of the baseline graph and each of the theoretical process abnormal simulation state or real mapping graphs, extracting the relatively inserted extra nodes and edges and the relatively missing deleted nodes and edges as the path offset.

7. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The topological similarity comparison includes: Calculate the graph edit distance between the actual residual map and each of the theoretical residual maps, or encode the features of the actual residual map and each of the theoretical residual maps to generate residual feature vectors and calculate vector similarity; Candidate matching sequences are generated by sorting the graph edit distance from smallest to largest or the vector similarity from largest to smallest. The theoretical residual map that ranks first and meets the preset matching conditions is selected as the optimal matching result; wherein, the preset matching conditions include: the similarity of the first and second ranked positions is greater than the preset absolute similarity threshold, and the difference between the similarity of the first and second ranked positions is greater than the preset discrimination threshold. The theoretical residual map category corresponding to the optimal matching result is output as the process anomaly category.

8. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, The generation of the process status determination result includes: When the similarity is greater than or equal to the blocking threshold, the process is determined to be in an abnormal state, and a rejection instruction for uploading to the chain is generated. When the similarity is less than or equal to the release threshold, the process is determined to be in a normal state, and a direct on-chain instruction is generated. When the similarity is greater than the allow threshold and less than the block threshold, the process is determined to be in a pending confirmation state, and an on-chain instruction with an additional process abnormal label is generated. The process anomaly label includes at least a process anomaly category identifier and an anomaly uncertainty identifier.

9. The method for tracing the supply chain of electrical complete sets of equipment integrating blockchain technology as described in claim 1, characterized in that, Adjusting the anomaly operator library or the dual threshold based on the process status determination results of completed processing includes: Extract the process status determination results of completed treatments and the corresponding actual residual maps; When the process status determination result is an abnormal process status, update the attribute perturbation parameters or rule templates of each abnormal operator in the abnormal operator library. When the process status determination result is a process pending confirmation status, the interval boundary of the dual threshold is updated; When the process status determination result is that the process is in a normal state, the abnormal operator library and the dual threshold remain unchanged.

Images