Method and system for sorting electronic archives and transferring single-set archives in four-property environment

Abstract

The invention relates to the technical field of electronic archive management, and discloses an electronic archive arrangement and single-set archive transfer method and system in a four-property environment, and the method comprises the steps: collecting electronic files in a project construction period; adding the electronic file into the approval process and carrying out electronic signature to generate a file stream containing signature stream data; pre-archiving and classifying the approved electronic files; performing four-property detection on the pre-archived file by adopting a detection strategy based on an adversarial detection model, wherein the four properties comprise authenticity, integrity, availability and security; and registering and warehousing the engineering construction period electronic file after the detection is passed, and archiving and transferring. According to the method, the approval efficiency of the engineering electronic file is improved, and the integration level of an electronic signature process and a project management system is enhanced; the coverage range of four-property detection is expanded to defect types such as time sequence abnormity and metadata logic errors which are difficult to identify by a traditional method, and the risks of missing detection and misjudgment can be effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic archive management, and in particular to a method and system for electronic archive organization and single-set archive transfer under a four-factor environment. Background Art

[0002] In today's digital age, the number of electronic files generated by engineering construction projects is huge and their importance is becoming increasingly prominent. Traditional engineering file management and file transfer methods mostly rely on manual operations, which are inefficient and prone to errors, and it is difficult to meet the efficiency, accuracy and security requirements of modern engineering projects for file management. With the development of electronic signature technology, although some engineering projects have begun to use electronic signatures for document signing, there are still deficiencies in the integration with project management systems. Interface compatibility issues between different platforms, inconsistent signature rules, and imperfect authority management have limited the application of electronic signatures in engineering file management and have been unable to give full play to their advantages. In the verification of engineering electronic files, the authenticity, integrity, availability and security (referred to as "four properties") of electronic files are crucial. However, existing detection methods often have problems such as incomplete detection and low accuracy. Some detection methods only detect one aspect of electronic files and cannot comprehensively evaluate the overall quality of the files; some detection technologies are inefficient when processing complex engineering file formats and large-scale file data, and it is difficult to meet the needs of actual projects. In addition, the existing file transfer methods are mostly dual-set, that is, paper files and electronic files coexist, which not only increases management costs, but also easily leads to inconsistent file versions. Therefore, there is an urgent need for a solution that can effectively integrate electronic signature technology, comprehensively detect the four properties of electronic files, and realize single-set file transfer, so as to improve the overall efficiency and security of engineering electronic file management. In view of the above problems, the existing technology needs to be improved. Summary of the invention

[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method and system for electronic archive organization and single-set archive transfer under a four-quality environment, so as to improve the degree of automation of electronic archive management and ensure the comprehensiveness and accuracy of four-quality detection.

[0004] In order to achieve the above object, the present invention provides the following technical solutions: The method of electronic archive organization and single-set archive transfer under a four-quality environment includes: collecting electronic files during the construction period; adding electronic files to the approval process and electronically signing and sealing them to generate a file flow containing signature flow data; pre-archiving and classifying the approved electronic files; using a detection strategy based on an adversarial detection model to perform four-quality detection on pre-archived files, the four qualities including authenticity, integrity, availability and security; registering and storing the electronic files during the construction period that pass the detection, and archiving and transferring them.

[0005] In the present invention, preferably, the adversarial detection model adopts an adversarial training strategy: creating virtual signature nodes to generate adversarial streaming data containing forged certificate chains and abnormal timestamps; constructing metadata conflict samples, including abnormal sample data such as out-of-bounds enumerated values, format misalignment, and duplicate fields; mixing the adversarial streaming data, abnormal sample data, and real data according to a preset ratio to generate a training set, and performing multiple adversarial trainings on the adversarial detection model using the training set until the misjudgment rate of the adversarial detection model for the input data is lower than the threshold.

[0006] In the present invention, preferably, the detection of the authenticity by the detection strategy specifically includes: extracting the timestamp sequence in the file stream, constructing a time series feature vector through an LSTM network; comparing the matching degree of the actual signature node with the time window of the approval process; calculating the cryptographic strength indicators of the certificate chain, including key length, hash collision probability, and signature timeliness; inputting the time series feature vector, time window matching degree, and cryptographic strength indicators into the adversarial detection model after adversarial training, and outputting a credibility score.

[0007] In the present invention, preferably, the detection of the integrity by the detection strategy specifically includes: constructing a metadata topology graph to analyze the logical dependency relationship between fields; detecting abnormal connection edges and missing nodes through a graph neural network; calculating the semantic similarity between the file entity and the metadata description; dynamically generating adversarial texts containing randomly missing fields to verify the misjudgment rate of the adversarial detection model.

[0008] In the present invention, preferably, the detection of the usability by the detection strategy specifically includes: judging whether the format of the pre-archived electronic file meets the requirements of long-term storage; judging whether the metadata of the pre-archived electronic file is accessible and whether the metadata in the information package is readable.

[0009] In the present invention, preferably, the detection of the security by the detection strategy specifically includes: generating attack simulation data for data tampering attacks, signature forgery attacks, and structural damage attacks, mixing the attack simulation data and real data according to a preset ratio and putting them into the training set of the adversarial detection model to train the adversarial detection model until the misjudgment rate of the adversarial detection model for the input data is lower than the threshold; optimizing the generated adversarial detection model using gradient penalty, and using the optimized adversarial detection model to perform security detection on the electronic file.

[0010] In the present invention, preferably, it further includes dynamic threshold adjustment of the adversarial detection model: establishing a Bayesian probability distribution model of the four-property indicators based on historical detection data; calculating the KL divergence between the detection results of the current batch and the probability distribution in real time; when the divergence value exceeds the warning threshold, triggering the adversarial data generation module to update the training set; adjusting the weight coefficients of each detection dimension to form a closed-loop feedback and dynamic optimization of the adversarial detection model.

[0011] In the present invention, preferably, the archived and transferred electronic file package includes encrypted original electronic files, timestamped complete file streams, and verification matrices.

[0012] In the present invention, preferably, there is also provided an electronic file arrangement and single - set file transfer system in a four - property environment, including a receiving module, an approval module, an electronic signature module, an adversarial detection model, an electronic file storage module, and an archiving and transfer module. The receiving module is used to receive electronic files during the project construction period; the approval module initiates an approval process for the electronic files for hierarchical approval; the electronic signature module is communicatively connected to the Contract Lock platform to perform electronic signatures on the electronic files. Extract keywords of the electronic files that need to be signed, determine the signature positions based on the keywords, determine the signature rules according to the current review process, and merge and transmit the corresponding signature nodes, signature positions, signature files, and signature rules to the Contract Lock platform for signature; the adversarial detection model is based on a neural network structure, and determines its weights and parameters through pre - period adversarial training and post - period dynamic threshold adjustment, and is used to perform four - property detection on the electronic files; the electronic file storage module includes an ordinary storage unit and a stream file storage unit. The ordinary storage unit is used to store the original electronic files, and the stream file is used to store the file stream containing signature stream data; the archiving and transfer module is used to archive and transfer the electronic files that pass the four - property detection.

[0013] In the present invention, preferably, there is also provided a storage medium storing a computer program, and when the computer program is executed by a processor, the processor is caused to execute the steps of the method for electronic file arrangement and single - set file transfer in a four - property environment as described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: The method of the present invention discovers hidden logical errors by constructing a metadata topology graph and a graph neural network, and adopts a multi - dimensional joint evaluation mechanism. It analyzes time - series features through an LSTM network, and comprehensively judges the authenticity, integrity, availability, and security of files in combination with cryptographic strength indicators; at the same time, it improves the approval efficiency of engineering electronic files and enhances the integration degree of the electronic signature process and the project management system; the coverage of four - property detection is extended to defect types such as time - series anomalies and metadata logical errors that are difficult to identify by traditional methods; the integrity verification of archived files continuously optimizes the adversarial detection model by dynamically generating adversarial samples, which can effectively reduce the risks of missed detection and misjudgment; at the same time, the file transfer process uses encrypted storage and verification matrices to ensure the readability and anti - tampering of data during long - term storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1It is a schematic flow chart of the method for sorting electronic files and transferring single-set archives in the four-nature environment described in the present invention.

[0016] Figure 2 It is a schematic flow chart of the adversarial training strategy of the method for sorting electronic files and transferring single-set archives in the four-nature environment described in the present invention.

[0017] Figure 3 It is a schematic flow chart of the detection process for authenticity in the method for sorting electronic files and transferring single-set archives in the four-nature environment described in the present invention.

[0018] Figure 4 It is a schematic flow chart of the dynamic threshold adjustment of the adversarial detection model in the method for sorting electronic files and transferring single-set archives in the four-nature environment described in the present invention.

[0019] Figure 5 It is a structural block diagram of the system for sorting electronic files and transferring single-set archives in the four-nature environment described in the present invention. Detailed implementation manners

[0020] Next, the technical solutions in the present application will be clearly and completely described in conjunction with the accompanying drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Usually, the components of the present application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application to be protected, but only represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application. It should be noted that: similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, the terms "first", "second", etc. are only used for distinguishing descriptions, and cannot be understood as indicating or implying relative importance.

[0021] In the prior art, the number of electronic documents generated in engineering construction projects is huge and their importance is increasing day by day. Traditional engineering document management relies on manual operations, resulting in low efficiency and easy errors. Although electronic signature technology has been partially applied, there are problems such as insufficient platform integration, poor interface compatibility, and imperfect permission management. Existing document verification methods mostly adopt single-dimensional detection technologies, which are difficult to meet the comprehensive evaluation requirements of authenticity, integrity, availability, and security at the same time, especially with low efficiency when dealing with complex formats and large-scale data.

[0022] Please also refer to Figures 1 to 4, a preferred embodiment of the present invention provides a method for organizing electronic files and transferring single - set archives in a four - property environment, including collecting electronic documents during the project construction period; adding the electronic documents to the approval process and performing electronic signature, generating a document stream containing signature stream data; pre - archiving and classifying the approved electronic documents; performing four - property detection on the pre - archived documents using a detection strategy based on an adversarial detection model; registering and storing the electronic documents during the project construction period after passing the detection, and conducting archiving and transfer.

[0023] Among them, the electronic documents during the project construction period refer to structured or unstructured data generated during the entire life cycle of the project, including the submission of five - party materials before construction, the start - up order, as well as quality management, progress management, daily control, safety management, and inspection management documents during construction, all data such as the completion acceptance of construction defects and the defect liability period. Specifically, they can be collected through Internet of Things sensors, BIM systems, and project management software, forming an electronic document set including design drawings, construction logs, and acceptance reports. The electronic documents can be uploaded in their entirety or in batches according to the progress. The file data of the same project is integrated into one electronic document. When the electronic document is complete, it enters the approval process. After approval, it is electronically signed according to requirements or directly approved and transferred. When electronic signature is required during approval, the electronic signature process is realized by integrating a third - party certification platform. The RSA encryption algorithm and a timestamp server can be used to generate non - tamperable signature stream data, forming a complete evidence chain including the signer's identity, time node, and approval status. After approval, the electronic documents are pre - archived. The pre - archiving classification is completed by setting project - stage tags and file - type tags. Four - property detection is performed on the pre - archived electronic documents, that is, authenticity, integrity, availability, and security detection. An adversarial detection model is used to perform intelligent detection. The adversarial detection model is constructed based on a deep neural network. In the early stage, iterative training is carried out with training data containing forged features so that the adversarial detection model can identify four - property defects such as abnormal signatures, tampering traces, abnormal certificate chains, timestamp conflicts, and missing metadata in electronic documents. Based on the results of the adversarial detection model, four - property detection can be quickly completed, while ensuring the confidential transfer of electronic documents, realizing high - efficiency electronic file archiving, and the entire processing process is safe and highly credible.

[0024] Traditional methods rely on manual verification of the authenticity of signatures or simply verify the digest value to judge the authenticity of signatures. In this application, the adversarial detection model is used to automatically identify forged certificates and abnormal timestamps, making the verification more accurate and efficient. Existing technologies mostly use fixed rules to detect the authenticity and integrity of files. This method discovers hidden logical errors by constructing a metadata topology graph and a graph neural network. Traditional file verification systems independently process four - property indicators. This solution adopts a multi - dimensional joint evaluation mechanism, analyzes time - series features through an LSTM network, and comprehensively judges the authenticity, integrity, availability, and security of files in combination with cryptographic strength indicators.

[0025] Through the above technical solutions, the approval efficiency of engineering electronic documents is improved, and the integration degree of the electronic signature process and the project management system is enhanced. The coverage of the four-nature detection is extended to defect types that are difficult to identify by traditional methods such as time-series anomalies and metadata logical errors. The integrity verification of the archived documents continuously optimizes the adversarial detection model by dynamically generating adversarial samples, which can effectively reduce the risks of missed detection and misjudgment. At the same time, the encrypted storage and verification matrix are adopted in the file transfer process to ensure the readability and anti-tampering property of the data during long-term preservation.

[0026] In this embodiment, the adversarial detection model adopts an adversarial training strategy, including creating virtual signature nodes to generate adversarial flow data containing forged certificate chains and abnormal timestamps, constructing metadata conflict samples, and mixing the adversarial flow data and real data in a preset ratio to generate a training set for multiple adversarial trainings until the false positive rate is lower than the threshold.

[0027] Among them, the virtual signature node refers to a false process node generated by simulating unauthorized signature operations in the approval process, which is generated through preset illegal user identity parameters and forged permission identifiers and is used to simulate unauthorized file tampering scenarios. The forged certificate chain refers to a digital certificate sequence issued by a non-certifying authority or with a key length less than the standard, which is generated by modifying the certificate issuing authority field and shortening the RSA key length to less than 512 bits and is used to test the adversarial detection model's ability to identify illegal authentication credentials. The abnormal timestamp refers to a time mark that conflicts with the logical time window of the approval process. For example, a final approval timestamp is generated before the approval process is started, or the timestamps of each node form a reverse sequence, which is used to detect timing logic anomalies. The metadata conflict sample refers to an abnormal data instance that violates the engineering file metadata specification, which is constructed by randomly deleting required fields, filling integer fields with string content, and repeatedly generating metadata records with the same field name but different values. The generated metadata is used as adversarial sample data to train the adversarial detection model to identify data structure anomalies.

[0028] Specifically, the implementation process of the adversarial training strategy is divided into two stages: adversarial sample data construction and iterative optimization of the adversarial detection model. In the stage of adversarial sample data construction, adversarial file streams are generated by virtual signature nodes and forged certificate chains to construct abnormal timestamps. Meanwhile, metadata conflict samples with missing fields, incorrect types, and duplicate definitions are constructed. These adversarial samples cover typical abnormal forms of electronic files in terms of certificate legitimacy verification, approval time sequence logic, and data structure integrity. In the stage of adversarial detection model training, the constructed adversarial sample data and real engineering files are mixed in a preset ratio and then input into the detection model. Through multiple training iterations, the model establishes an association mapping between abnormal features and error types. During the training process, the false positive rate index of the adversarial detection model for the mixed data set is continuously monitored. When the adversarial detection model can stably identify adversarial samples and the false positive rate for real data is lower than the preset threshold, the training is terminated, so as to ensure that the adversarial detection model has the ability to distinguish real business data from potential attack samples.

[0029] Traditional engineering file detection methods usually use fixed rules to separately detect the authenticity, integrity, availability, and security of electronic documents, with low detection efficiency, poor accuracy, or rely on historical real data for intelligent model training, but the generalization ability of the intelligent model is insufficient when facing new attack means or abnormal data. This method actively generates adversarial samples containing multi-dimensional attack features and adopts a training strategy with a dynamic mixing ratio, enabling the adversarial detection model to comprehensively cover abnormal patterns that may occur during the process of engineering file transfer, accurately identify the authenticity, integrity, availability, and security of electronic documents, and improve the recognition accuracy of unknown error types.

[0030] Through the above technical solutions, this application effectively solves the problems of low efficiency, poor accuracy, and insufficient generalization ability of traditional detection methods. By actively constructing training data with attack features, the detection model can accurately identify malicious files containing forged certificate chains, abnormal timestamps, and metadata conflicts, and control the false positive rate of the four-nature detection of engineering files within an acceptable range. The adoption of the mixed training strategy not only retains the processing ability of the adversarial detection model for normal business data but also enhances its defense performance against new attack means, achieving a double improvement in the robustness and accuracy of the detection model.

[0031] In this embodiment, specific implementation steps for detecting the authenticity of electronic archives based on the four-nature environment are proposed: extracting the timestamp sequence in the file stream and constructing a time series feature vector through an LSTM network; comparing the matching degree between the actual signature node and the time window of the approval process; calculating the cryptographic strength indicators of the certificate chain, including key length, hash collision probability, and signature timeliness; inputting the time series feature vector, time window matching degree, and cryptographic strength indicators into the adversarial detection model after adversarial training, and outputting a credibility score.

[0032] Among them, the timestamp sequence refers to a set of time-stamped data generated at each operation node in the approval process of electronic documents. It stores timestamp fields including creation time, modification time, and approval time in JSON format to reflect the dynamic timing characteristics of the document transfer process. The LSTM network refers to the long short-term memory neural network. In this embodiment, a bidirectional network structure with 64 hidden units is adopted. The timestamp sequence segments are intercepted through a sliding window to capture the abnormal fluctuations in the time intervals in the approval process. The time window matching degree is the degree of compliance between the actual signature operation time and the corresponding node time range in the preset approval process. The matching degree score is calculated using the absolute value of the time difference between the timestamp and the preset time range of the process node, which is used to detect illegal operations or process skipping behaviors. The cryptographic strength indicators of the certificate chain include three dimensions: the key length can be set with parameters above 2048 bits in the RSA algorithm, the hash collision probability can be evaluated by calculating the theoretical collision value of the SHA-256 algorithm, and the signature timeliness is determined by verifying the overlapping interval between the validity period of the digital certificate and the file generation time, forming a multi-dimensional security evaluation system.

[0033] Specifically, in the approval process of electronic documents for engineering projects, the timestamp data generated at each operation node is extracted as a continuous time series. This series is subjected to time series modeling through the LSTM network to capture the time distribution characteristics of different approval stages, such as the time interval distribution pattern in the normal process. When abnormal approval operations occur, such as the time interval between operations of the same user at different nodes being too short or showing a time reversal, the time series feature vector output by the LSTM will exhibit significant fluctuations. At the same time, the preset approval process time window defines the reasonable range of operation times for each node. By calculating the deviation degree between the actual signature time and the window boundary, operation behaviors that violate the process timing constraints can be identified. In terms of cryptographic verification, a comprehensive evaluation is made on whether the key length of the certificate chain meets the security standard, whether the hash algorithm has collision resistance, and whether the digital signature is within the validity period, avoiding security verification loopholes in a single dimension. After the above multi-dimensional features are normalized, they are input into the adversarial detection model, and the neural network model learns the non-linear associations between the features, and finally outputs a comprehensive credibility score. When the score is lower than the preset threshold, the manual review process is automatically triggered.

[0034] Traditional authenticity detection methods usually only adopt static timestamp verification and key length verification. For example, existing technologies may only check the legality of a single timestamp and cannot capture the dynamic associations of multiple timestamps in the approval process; in certificate verification, they may only focus on whether the key length meets the standard, ignoring the dimensions of hash algorithm strength and signature timeliness. This solution realizes the dynamic monitoring of the approval process through time series feature modeling and constructs a composite security evaluation mechanism through multi-dimensional cryptographic indicators, solving the problem of single detection dimension in traditional methods.

[0035] Through the above technical solutions, this application realizes the dynamic timing monitoring of the electronic document approval process, effectively identifies abnormal fluctuations in the timestamp sequence, establishes a timing correlation verification mechanism between the signature nodes and the approval process to prevent irregular operations such as process skipping or time inversion, and constructs a multi-dimensional cryptographic strength evaluation system to avoid verification vulnerabilities caused by defects in a single security dimension.

[0036] In this embodiment, the detection steps of the detection strategy for integrity are proposed, specifically including: constructing a metadata topology graph to analyze the logical dependency relationship between fields; detecting abnormal connection edges and missing nodes through a graph neural network; calculating the semantic similarity between the file entity and the metadata description; and dynamically generating adversarial texts containing randomly missing fields to verify the false positive rate of the adversarial detection model.

[0037] Among them, the metadata topology graph refers to a network structure diagram that visually presents the logical association relationship between electronic document metadata fields. For example, it is implemented by constructing an entity-attribute-relationship model using graph database technology. This feature can structurally represent the nested relationship of multi-level metadata in complex engineering documents. Among them, the graph neural network refers to a deep learning model for feature learning based on graph-structured data. For example, it is implemented by using a graph attention network or a graph convolutional network. This feature effectively identifies abnormal connection patterns by automatically extracting the node features and edge weights of the metadata topology graph. Among them, the semantic similarity calculation refers to an index that quantifies the consistency degree between the file content and the metadata description. For example, it is implemented by calculating the cosine similarity after extracting text vectors using a pre-trained language model. This feature realizes the cross-verification mechanism between the content layer and the description layer. Dynamically generating adversarial texts means automatically constructing test samples with specific defects through algorithms. For example, it is implemented by combining Markov chain random sampling with field masking technology. This feature continuously verifies the robustness of the model by actively injecting abnormal data.

[0038] Specifically, when processing engineering documents with complex metadata structures, first construct a topology graph based on the attributes and constraint conditions of the metadata fields, taking each metadata field as a node and the logical dependency between fields as an edge. Subsequently, traverse the topology graph through a graph neural network to identify abnormal connection edges that violate business rules and missing nodes with undefined data types. At the same time, use natural language processing technology to perform semantic parsing on the text content of the document and calculate the similarity threshold between it and the metadata description. To further verify the reliability of the detection model, the system automatically generates test texts containing randomly missing fields, and dynamically adjusts the sensitivity parameters of the detection strategy by observing the recognition accuracy of the model for abnormal samples.

[0039] Compared with the prior art, traditional integrity detection mainly relies on a preset rule engine for format verification and cannot effectively handle metadata fields with dynamic association relationships. When detecting metadata topology anomalies, conventional methods usually adopt a static rule library matching method, which is difficult to adapt to the changing field association patterns in engineering documents. In contrast, this solution automatically learns the potential association rules of metadata fields through a graph neural network and continuously optimizes the model by combining dynamically generated adversarial samples, significantly improving the anomaly detection coverage rate in complex scenarios.

[0040] Through the above technical solution, this application solves the technical problems of inaccurate recognition of abnormal connections and lack of semantic consistency verification in the integrity detection of engineering documents. By combining the metadata topology graph and the graph neural network, in-depth analysis of complex field association relationships is achieved; a two-way verification mechanism for content and description is established through semantic similarity calculation; the dynamic adversarial text generation technology ensures that the detection model can continuously adapt to new data anomaly patterns and form a self-optimizing detection closed-loop. In a specific embodiment, when using this method and the traditional method to detect the same completed engineering electronic file, the detection comparison results are shown in Table 1 below. This method can construct a metadata topology graph with 5304 nodes and 7821 edges in a short time, detect 12 abnormal associations (such as illegal mappings between concrete grades and seismic grades), and improve the recognition rate of field deletion attacks by the adversarial detection model from 70% to 93%.

[0041] Table 1.

[0042] In this embodiment, detection steps for the usability of the detection strategy are proposed, which specifically include: determining whether the format of the pre-archived electronic file meets the requirements for long-term storage, determining whether the metadata of the pre-archived electronic file is accessible, and whether the metadata in the information package is readable.

[0043] Among them, the requirements for long-term storage refer to that the file format has backward compatibility and cross-platform parsing capabilities, which are implemented by adopting the long-term preservation format standard defined by the International Organization for Standardization in this embodiment. For example, converting the electronic file into the PDF / A format to ensure that the file can still be correctly parsed by standard software decades later. Metadata accessibility means that the system can obtain metadata through API interfaces or database query methods. For example, the OAuth2.0 authorization protocol is adopted to implement access control, ensuring that authorized users can retrieve metadata fields in real time. The readability of metadata in the information package means that the metadata maintains a physical storage association with the file entity it describes. The binding storage of metadata and the file entity is achieved through the XML encapsulation specification, and the data readability is verified by checking the integrity of the encapsulation structure.

[0044] Specifically, in the pre-archiving stage of electronic files, an availability guarantee framework is constructed through a dual verification mechanism. First, a technical screening of the file format is carried out, and only open format standard files are retained to eliminate the risk of unreadable files caused by accidental deletion at the carrier level. At the same time, hierarchical detection of metadata is implemented. At the logical level, it is verified whether the permission configuration of the access interface conforms to the security policy, and at the physical level, the structural integrity of the metadata encapsulation package is detected. For example, when it is detected that there is a broken link between the metadata storage path and the file entity, an automatic repair mechanism can be triggered to re-establish the index association. This dual verification from format to metadata can discover potential availability problems caused by packaging errors or improper permission configurations.

[0045] Compared with the prior art, traditional availability detection usually only verifies file format compatibility and does not establish a dual detection mechanism for metadata access permissions and physical storage. The existing methods cannot detect the problem of restricted metadata access caused by incorrect permission configurations, nor can they identify the broken association between the metadata in the encapsulation package and the file entity. This solution can actively discover authorization policy vulnerabilities by adding steps for verifying metadata accessibility, and can timely repair abnormal storage structures through detecting the readability of metadata in the information package.

[0046] Through the above technical solutions, this application effectively solves the problem of insufficient long-term readability guarantee of electronic files and avoids the risk of future unresolvability caused by using private formats. At the same time, it overcomes the defect of the mismatch between metadata access permissions and physical storage status, ensuring that the archived electronic files can be fully accessed within the authorized scope, and the metadata and file entities always maintain an accurate correspondence.

[0047] In this embodiment, a detection step for security is proposed, which specifically includes: generating attack simulation data for data tampering attacks, signature forgery attacks, and structure destruction attacks, mixing the attack simulation data and real data into the training set of the adversarial detection model for training according to a preset ratio until the misjudgment rate of the adversarial detection model for the input data is lower than the threshold; optimizing the generated adversarial detection model using gradient penalty, and using the optimized adversarial detection model to detect the security of electronic files.

[0048] Among them, data tampering attack refers to the act of destroying data integrity by modifying the content or metadata of a file. For example, attack samples are generated by means of random bit flipping, injecting malicious code, or tampering with hash values, which are used to train the model to identify unauthorized modification behaviors. Signature forgery attack refers to the act of counterfeiting a legitimate electronic signature or certificate chain. For example, adversarial samples are constructed by means of generating a forged certificate chain, timestamp offset, or reverse engineering of the signature algorithm, which are used to enhance the model's ability to identify forged signatures. Structure destruction attack refers to the act of making a file invalid by deleting key fields or disrupting the logical structure of the file. For example, training samples are generated by means of randomly deleting metadata fields, scrambling the file encoding order, or inserting illegal characters, which are used to improve the model's recognition accuracy for abnormal structures. Gradient penalty optimization refers to imposing constraint conditions on the gradient of parameter updates during the model training process. For example, it is achieved by using a regularization term to limit the gradient amplitude or forcing a smooth transition in the parameter space, which is used to prevent the model from overfitting to specific attack patterns during adversarial training.

[0049] Specifically, the attack simulation data generation module constructs three types of attack samples: data tampering, signature forgery, and structure destruction, covering the main attack types that electronic files may suffer. The attack samples and real data are mixed in a preset ratio and then input into the adversarial detection model for iterative training. During the training process, the false positive rate of the model for normal files is continuously monitored. When the false positive rate is lower than the preset threshold, the training is terminated to ensure that the model controls the false alarm risk while maintaining a high detection rate. The model after preliminary training is adjusted by the gradient penalty optimization method, and a gradient constraint term is introduced into the loss function to force the model parameters to maintain a smooth transition during the optimization process and avoid overfitting to specific attack patterns. The finally formed optimized model can effectively distinguish normal files from abnormal files in a mixed attack scenario, and improve the adaptability to new attacks while maintaining detection stability.

[0050] Compared with the prior art, traditional methods usually use samples of a single attack type for training, resulting in a significant decline in the detection performance of the model when facing mixed attacks. In the prior art, the adversarial training process lacks a dynamic optimization mechanism and is prone to reducing the generalization ability due to overfitting to specific attack patterns. In the prior art, the gradient optimization method does not consider the perturbation characteristics of adversarial samples and it is difficult to effectively balance the detection accuracy and the false positive rate.

[0051] Through the above technical solutions, this application effectively solves the problem of insufficient detection coverage in a mixed attack scenario, and improves the model's recognition ability for complex attacks through combined training with multi-type attack samples. By adopting a dynamic false positive rate threshold control mechanism, the false interception rate of normal files is reduced while ensuring the detection accuracy. The introduction of the gradient penalty optimization method significantly improves the generalization performance of the model, enabling it to adapt to evolving attack means.

[0052] The present application further proposes a dynamic threshold adjustment method for the anti-detection model. A Bayesian probability distribution model of the four-property indicators is established based on historical detection data. The KL divergence between the detection results of the current batch and the probability distribution is calculated in real time. When the divergence value exceeds the warning threshold, the adversarial data generation module is triggered to update the training set, and the weight coefficients of each detection dimension are adjusted to form a closed-loop feedback and dynamic optimization.

[0053] Among them, the Bayesian probability distribution model refers to a probability distribution model of the four-property indicators constructed based on historical detection data, which is realized by, for example, the kernel density estimation method and is used to dynamically model the data distribution of the four-property detection results. The KL divergence calculation refers to calculating the difference degree between the detection results of the current batch and the historical distribution, which is realized by, for example, the information entropy difference quantization method and is used to detect the degree of data distribution deviation. The update of the training set by the adversarial sample data generation module means automatically generating new adversarial samples when data deviation is detected, which is realized by, for example, the generative adversarial network technology and is used to expand the diversity of training data. The adjustment of the weight coefficients refers to reallocating the decision weights of the four-property detection dimensions according to the data deviation direction, which is realized by, for example, the gradient backpropagation algorithm and is used to balance the sensitivity of the anti-detection model to different indicators.

[0054] Specifically, the Bayesian probability distribution model transforms the historical four-property detection results into a probability distribution through the kernel density estimation method to form a dynamic benchmark. The KL divergence calculation module continuously monitors the difference between the current detection results and the benchmark distribution. When the difference exceeds the preset warning threshold, the adversarial data generation module is triggered to generate adversarial samples containing new attack patterns. The new samples are mixed with the original data to update the training set and retrain the anti-detection model. At the same time, the gradient backpropagation algorithm dynamically adjusts the weight coefficients of each detection dimension according to the performance of the model on the new data. For example, when it is detected that the timestamp forgery attack increases, the weight ratio of the time window matching degree is increased. Thus, a closed-loop feedback mechanism from data monitoring to model update is formed, enabling the detection threshold to be automatically optimized as the data distribution changes.

[0055] Compared with the prior art, the traditional method using a fixed threshold cannot adapt to the data deviation of the four-property detection caused by the change of engineering file types or new attack means. For example, when the existing static threshold model encounters an unknown signature forgery attack, it may produce misjudgments because the detection rules do not cover the new attack features. This solution can perceive the change of data distribution in real time through dynamic probability modeling and KL divergence quantization, actively expand the detection boundary of the model using the adversarial sample generation mechanism, and optimize the decision balance between detection dimensions by combining weight adjustment, effectively coping with the detection deviation brought by the dynamic evolution of data.

[0056] Through the above technical solution, this application solves the problem of the increasing false positive rate caused by the data distribution shift in the static threshold model. Through dynamic probability modeling and closed-loop feedback mechanism, the detection model can automatically adapt to scenarios such as changes in engineering file types and the introduction of new attack methods, improving the accuracy and robustness of the four-property detection. For example, when a new electronic signature format is added to the engineering file management system, the model can automatically identify the change in the distribution of the signature timestamp, generate corresponding adversarial samples to update the training set, and adjust the weight parameters of the time window matching degree, thereby maintaining the detection accuracy.

[0057] This application further proposes that the electronic file package for archiving and handover includes encrypted original electronic files, timestamped complete file streams, and verification matrices.

[0058] Among them, the encrypted original electronic file refers to the encryption of the content of the electronic file through cryptographic algorithms. For example, it is implemented by using the AES-256 encryption algorithm in combination with the digital envelope technology, and access control is realized through the encryption key management system. This feature is used to prevent the file from being illegally stolen or tampered with during transmission or storage, ensuring data confidentiality. The timestamped complete file stream refers to the sequential record data containing all operation nodes of the electronic file. For example, an immutable timestamp sequence is generated by using blockchain technology and calibrated with the standard time source through the time synchronization server. This feature is used to establish a traceable operation evidence chain and provide time dimension support for authenticity verification. The verification matrix refers to a data structure composed of file metadata, hash values, and verification rules. For example, it is implemented by using sparse matrix encoding technology, and a field association relationship graph is generated through a preset verification algorithm. This feature is used to quickly locate abnormal associations between file content and metadata, improving the integrity verification efficiency.

[0059] Specifically, the encrypted original electronic file is stored in different media through a key splitting mechanism during archiving, and the data is only restored through multi-party collaborative decryption during the handover process. The timestamped file stream solidifies the operation time of each signature node to the blockchain distributed ledger through the timestamp server, forming a time series evidence chain covering the entire approval process cycle. The verification matrix generates a hash fingerprint by extracting key fields of the file metadata, constructs a matrix index according to the logical dependency relationship, and compares the mapping relationship between the hash value and the metadata through matrix operations during the verification stage to quickly locate abnormal fields.

[0060] Compared with the prior art, the traditional electronic file package only contains the original file and simple metadata, lacking cryptographic protection measures and time dimension verification mechanisms, and the integrity verification requires comparing hash values item by item. This solution constructs a three-in-one security verification system by structurally encapsulating encrypted files, timestamped file streams, and verification matrices, forming a synergistic effect in three dimensions: data encrypted storage, operation time sequence solidification, and verification efficiency optimization.

[0061] Through the above technical solutions, the present application strengthens the anti-tampering ability of electronic documents during the transmission process, reduces the time cost of authenticity verification through the timestamp evidence chain, and uses a matrix data structure to reduce the computational complexity of integrity verification from linear level to logarithmic level. The combination of the encryption mechanism and the verification matrix effectively blocks unauthorized access paths, and the timestamp file stream provides a verifiable time reference point for audit traceability.

[0062] Please refer to Figure 5 , the present application further proposes an electronic file sorting and single-set file transfer system under the four-nature environment, including a receiving module, an approval module, an electronic signature module, an adversarial detection model, an electronic file storage module, and an archiving and transfer module. The receiving module is used to receive electronic files during the project construction period; the approval module initiates an approval process for the electronic files for hierarchical approval; the electronic signature module is communicatively connected to the Contract Lock platform to electronically sign the electronic files: extract the keywords of the electronic files that need to be signed, determine the signature position based on the keywords; determine the signature rules according to the current review process; merge and transmit the corresponding signature nodes, signature positions, signature files, and signature rules to the Contract Lock platform for signature; the adversarial detection model is based on a neural network structure, and determines its weights and parameters through preliminary adversarial training and subsequent dynamic threshold adjustment, and is used to perform four-nature detection on electronic files; the electronic file storage module includes an ordinary storage unit and a stream file storage unit. The ordinary storage unit is used to store the original electronic files, and the stream file is used to store the file stream containing the signature stream data; the archiving and transfer module is used to archive and transfer the electronic files that pass the four-nature detection.

[0063] Among them, the receiving module is used to receive data during the project construction period. The project construction period includes the entire process from before construction to during construction and the post-completion period. The data before construction includes the certification of the project departments of the construction / supervision / agency units, personnel certification, five-party declaration materials, and all approved data; the data for construction process control includes all data in the progress control, quality management, safety management, and quality and safety inspection processes; the post-completion data includes all data in the completion management and defect liability period. The data during the project construction period received by the receiving module is sorted and merged according to the project number and cycle. When all the data during the same project construction period is complete, it is sent to the approval module, and the approval module initiates an approval process. The approval process is carried out in sequence according to the set rules. When the approval is passed, the electronic signature module is started for electronic signature in the process that requires electronic signature, and the process that does not require signature automatically flows to the next approval process until all approvals are completed.

[0064] Specifically, after the receiving module receives an electronic document, it first classifies the electronic document, including structured data and unstructured data. For structured data, data cleaning is performed, and for unstructured data, preprocessing is carried out. The processed data is uniformly verified, coded with project numbers, signed, and then stored in the storage module in cycles. When the entire project cycle is completed, it enters the approval module. The approval module mainly operates based on an intelligent approval engine and follows a preset approval process. When each approval process passes, it determines whether a signature is required. When a signature is required, the electronic signature module is called; otherwise, it is automatically archived according to the project number. The position recognition of the electronic signature can be determined based on the combination of OCR and keyword matching.

[0065] Among them, the communication connection between the electronic signature module and the contract lock platform means that data intercommunication between the system and a third-party electronic signature platform is achieved through a standard interface protocol. For example, an integration method based on RESTful API is adopted to realize the automatic transmission of signature requests and the return of results. The adversarial detection model based on a neural network structure refers to a hybrid architecture that combines a multi-layer convolutional network and a graph neural network. The training of the adversarial detection model can be achieved through the TensorFlow framework, enabling it to process both structured metadata and unstructured file content simultaneously. Dynamic threshold adjustment means using a Bayesian probability model to model the distribution of detection results. For example, the KL divergence is used to calculate the deviation between real-time data and historical distributions, triggering the update of model parameters. The stream file storage unit for storing the file stream containing signature stream data means that the timestamp, node order, and certificate chain information of the signature operation are saved in the form of incremental logs.

[0066] Specifically, the electronic signature module identifies the key clause positions in the electronic document that need to be signed through a keyword extraction algorithm. For example, semantic matching technology based on regular expressions is used to locate fields such as contract amount and signatory name. The signature rule generation module automatically matches the preset permission matrix according to the sequence of links in the current approval process. For example, in a three-level approval process, the electronic signature nodes of the department head, legal specialist, and project director are triggered in sequence. During the training phase, the adversarial detection model injects adversarial samples such as forged certificate chains and abnormal timestamps to enable the model to learn to distinguish between real signature data and attack data. For example, 20% of abnormal samples are mixed in the training set to improve the robustness of the model. While saving the original electronic document, the stream file storage unit records the time series data of all signature operations. For example, the action type, executor identity, and timestamp of each signature are appended to the file stream in JSON format. After passing the detection, the archiving and handover module packages the encrypted original file, signature stream data, and verification matrix into a standardized file package. For example, the AES-256 algorithm is used to encrypt the file, and a verification matrix containing a hash check code is generated.

[0067] Compared with the prior art, traditional engineering management systems rely on manual configuration of signature rules, resulting in permission conflicts and process breaks during cross-platform integration. In this solution, signature rules are dynamically generated and deeply connected with the Qiyuesuo platform to achieve automated matching of the signature process. Most existing detection methods use static rule libraries for four-property verification and cannot identify new attack methods. This solution enables the model to continuously adapt to changes in attack patterns through adversarial training. At the same time, streaming storage technology is used to retain complete signature process data to provide traceability support for post-event auditing. Existing file storage solutions usually separate metadata from entity files for storage, making integrity verification difficult. This solution adopts a dual-storage unit structure to ensure the relevance and consistency of the original file and signature stream data.

[0068] Through the above technical solutions, this application realizes the integrated integration of the engineering electronic file signature process and the management system, solves the problem of inconsistent cross-platform signature permissions through keyword positioning and dynamic rule matching; uses an adversarial training neural network model to improve the detection accuracy of four-property abnormal features, and ensures the complete traceability of file full-life cycle data through a streaming storage structure; the standardized archiving process ensures that the transferred file package meets compliance requirements and avoids omissions or format errors caused by manual operations.

[0069] In some other preferred embodiments of the present invention, a computer-readable storage medium is provided, storing a computer program, which when executed by a processor, causes the processor to execute the steps of the method as described in the above embodiments.

[0070] Among them, the storage medium refers to the physical carrier for storing the computer program. For example, it is implemented using a solid-state drive, disk array, or cloud storage server to ensure that program instructions can be called and executed by the processor.

[0071] Among them, the computer program refers to a set of instructions containing data processing logic. For example, it is written in an object-oriented programming language to implement the automated control of the electronic file signature process and the invocation of four-property detection algorithms.

[0072] Among them, the execution by the processor means that the program instructions are processed by a central processing unit or a graphics processing unit. For example, a multi-threaded parallel computing architecture is adopted to synchronously process the signature verification and four-property detection tasks of electronic files.

[0073] Specifically, when the computer program stored in the storage medium is loaded and executed, it establishes a communication connection with the e-signature platform through the e-signature module, automatically matches the signature position based on the file keyword extraction technology, and dynamically generates signature rules according to the review process, realizing the deep integration of the project management system and the signature platform. The anti-detection model conducts adversarial training on the injected forged certificate chain, abnormal timestamp, and metadata conflict samples through the neural network structure, establishes a dynamic threshold adjustment mechanism, and optimizes the detection weight coefficient in real time to cope with new attack patterns. The streaming file storage unit and the ordinary storage unit work together to record the original electronic file and the signature process data containing the timestamp sequence respectively, forming a complete evidence chain. The archiving and handover module seals and packages the detected electronic file to generate a standardized data packet containing a verification matrix, ensuring the long-term verifiability of the transferred file.

[0074] Compared with the prior art, most existing storage media adopt a single program to execute fixed detection rules and cannot adapt to the compatibility problems caused by the interface differences between the project management system and the signature platform. This solution solves the problems of misaligned signature positions and interrupted processes caused by inconsistent interface protocols in traditional systems through dynamic signature rule matching and the streaming file dual-track storage structure. Most existing anti-detection models rely on static sample library training, while this solution constructs a dynamic training set by injecting virtual signature nodes and metadata conflict samples, enabling the model to have the ability to continuously evolve for anomaly detection. The existing archiving and handover process lacks the association mechanism between the timestamp file stream and the verification matrix. This solution improves the traceability after file handover through the composite packaging method of encrypting the original file and generating the verification matrix.

[0075] Through the above technical solutions, this application realizes the seamless docking between the project management system and the e-signature platform, eliminating the problem of interrupted signature processes caused by interface protocol differences; through adversarial training and the dynamic threshold adjustment mechanism, the four-nature detection model can identify new forgery attacks and format tampering behaviors in real time; through the dual-track storage mode of streaming files and ordinary files, the full life cycle data of electronic files from generation to archiving is completely recorded; through the sealed packaging method including the timestamp file stream and the verification matrix, the verifiability and anti-tampering ability of the transferred archives during long-term storage are ensured.

[0076] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0077] The above description is a detailed description of the preferred and feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. Any equivalent changes or modifications made under the technical spirit disclosed by the present invention shall fall within the scope of the patent covered by the present invention.

Claims

1. Method for sorting electronic files and transferring single-set files under the four-property environment, characterized in that, Including: Collecting electronic documents during the engineering construction period; Adding the electronic documents to the approval process and performing electronic signature stamping to generate a file stream containing signature stream data; Pre-archiving and classifying the approved electronic documents; Adopting a detection strategy based on an adversarial detection model to perform four-nature detection on the pre-archived documents, where the four natures include authenticity, integrity, availability, and security; Registering and warehousing the electronic documents during the engineering construction period after passing the detection, and performing archiving and transfer; The adversarial detection model adopts an adversarial training strategy: Creating a virtual signature node to generate adversarial stream data containing forged certificate chains and abnormal timestamps; Constructing metadata conflict samples, including abnormal sample data such as out-of-bounds enumeration values, format misalignment, and duplicate fields; Mixing the adversarial stream data, abnormal sample data, and real data in a preset ratio to generate a training set, and using the training set to perform multiple adversarial trainings on the adversarial detection model until the misjudgment rate of the adversarial detection model for the input data is lower than the threshold.

2. The method for sorting electronic files and transferring single-set files under the four-nature environment according to claim 1, characterized in that, The specific detection of the authenticity by the detection strategy includes: Extracting the timestamp sequence in the file stream and constructing a time series feature vector through an LSTM network; Comparing the matching degree of the actual signature node with the time window of the approval process; Calculating the cryptographic strength indicators of the certificate chain, including key length, hash collision probability, and signature timeliness; Inputting the time series feature vector, time window matching degree, and cryptographic strength indicators into the adversarial detection model after adversarial training, and outputting a credibility score.

3. The method for organizing electronic files and transferring single-set files under the four-property environment according to claim 1, characterized in that The specific detection of the integrity by the detection strategy includes: Constructing a metadata topology graph and analyzing the logical dependency relationship between fields; Detecting abnormal connection edges and missing nodes through a graph neural network; Calculating the semantic similarity between the file entity and the metadata description; Dynamically generating adversarial texts containing randomly missing fields to verify the misjudgment rate of the adversarial detection model.

4. The method for organizing electronic files and transferring single-set files under the four-property environment according to claim 1, wherein The specific detection of the availability by the detection strategy includes: Judging whether the format of the pre-archived electronic document meets the requirements for long-term storage; Judging whether the metadata of the pre-archived electronic document is accessible and whether the metadata in the information package is readable.

5. The method for sorting electronic files and transferring single-set archives in the four-property environment according to claim 1, wherein The specific detection of the security by the detection strategy includes: Generating attack simulation data for data tampering attacks, signature forgery attacks, and structure destruction attacks, mixing the attack simulation data and real data in a preset ratio into the training set of the adversarial detection model, and training the adversarial detection model until the misjudgment rate of the adversarial detection model for the input data is lower than the threshold; Optimizing the generated adversarial detection model using gradient penalty, and using the optimized adversarial detection model to perform security detection on the electronic documents.

6. The method for organizing electronic files and transferring single-set files under the four-nature environment according to claim 1, characterized in that, It also includes dynamic threshold adjustment of the adversarial detection model: Establishing a Bayesian probability distribution model of four-nature indicators based on historical detection data; Calculating the KL divergence between the detection results of the current batch and the probability distribution in real time; When the divergence value exceeds the warning threshold, triggering the adversarial data generation module to update the training set; Adjusting the weight coefficients of each detection dimension to form a closed-loop feedback and dynamic optimization of the adversarial detection model.

7. The method for sorting electronic files and transferring single-set files under the four-property environment according to claim 1, characterized in that, The archived and transferred electronic document package includes the encrypted original electronic document, the complete file stream with timestamps, and the verification matrix.

8. An electronic file sorting and single-set file transfer system in a four-property environment, which is used to implement the electronic file sorting and single-set file transfer method in a four-property environment described in any one of claims 1-7, characterized in that It includes a receiving module, an approval module, an electronic signature module, an adversarial detection model, an electronic document storage module, and an archiving and handover module. The receiving module is used to receive electronic documents during the engineering construction period. The approval module initiates an approval process for the electronic documents and conducts step-by-step approval. The electronic signature module is communicatively connected to the contract lock platform to electronically sign the approved electronic documents. Extract the keywords of the electronic document that needs to be signed and determine the signature position based on the keywords. Determine the signature rules according to the current review process. Merge and transmit the corresponding signature nodes, signature positions, signature files, and signature rules to the contract lock platform for signature. The adversarial detection model is based on a neural network structure and determines its weights and parameters through preliminary adversarial training and subsequent dynamic threshold adjustment, and is used to perform four-nature detection on the electronic documents. The electronic document storage module includes a general storage unit and a streaming file storage unit. The general storage unit is used to store the original electronic documents, and the streaming file is used to store the file stream containing the signature stream data. The archiving and handover module is used to archive and hand over the electronic documents that have passed the four-nature detection.

Images