Enterprise digital collaborative management system and method

Abstract

The invention relates to the technical field of workflow reconstruction management, and particularly discloses an enterprise digital collaborative management system, which comprises a data acquisition module, a multi-dimensional collaborative engine, a dynamic authority management module, an intelligent decision support module, a security protection module, a digital twin collaborative module, a visual interaction module and a block chain evidence storage unit, the data acquisition module performs real-time preprocessing and standardization; the multi-dimensional collaborative engine realizes collaborative scene matching and conflict automatic solution; the dynamic authority management module allocates authority, and the block chain evidence storage unit realizes authority operation non-tampering evidence storage and compliance auditing; the intelligent decision support module generates decision support information; the digital twinning collaboration module realizes virtual simulation deduction; the security protection module guarantees data security; and the visual interaction module supports multi-form data display. The system realizes an efficient, safe, intelligent and collaborative management method inside and outside an enterprise, and is suitable for a plurality of vertical industries such as manufacturing, supply chains, project management and the like.

Description

Technical Field

[0001] This invention relates to the field of workflow refactoring management technology, specifically to an enterprise digital collaborative management system and method. Background Technology

[0002] In the current era of accelerated digital transformation, enterprise collaborative management has become a core element for improving operational efficiency and enhancing market competitiveness. Whether it's internal departmental collaboration, cross-project resource allocation, or joint operations with upstream and downstream enterprises in the supply chain, higher demands are placed on the real-time nature, accuracy, and security of collaborative management. However, existing enterprise collaborative management models still face many pressing technical bottlenecks that need to be addressed: First, traditional collaborative management systems often employ a single-dimensional collaborative architecture, relying on centralized processing for data collection. This leads to high data transmission latency at edge nodes and excessive load on the core network. Particularly in cross-enterprise collaboration scenarios, inconsistent interface protocols and heterogeneous data formats among various business systems result in poor information flow and delayed synchronization, severely impacting project efficiency. For example, in manufacturing supply chain collaboration, inventory data and logistics information between manufacturers and suppliers cannot be exchanged in real time, often leading to resource shortages or backlogs. In project management scenarios, progress data between departments is isolated, making it difficult to achieve a holistic perspective for collaborative scheduling. Secondly, the lack of intelligent means for resolving collaborative conflicts means that existing systems largely rely on manual negotiation. Conflict prioritization is subjective, and solution selection is based on experience, which is not only time-consuming and labor-intensive but also prone to efficiency losses due to decision-making errors. Furthermore, access control often uses a static allocation model, failing to dynamically adapt to changes in employee positions or project roles. This creates security risks due to mismatches between data access permissions and collaborative scenarios, and insufficient traceability of access operations in cross-enterprise collaborations, making it difficult to meet compliance requirements such as the Data Security Law and the Personal Information Protection Law. Furthermore, intelligent decision support capabilities are weak. Traditional systems' data analysis often remains at the level of superficial statistics, lacking the ability to capture long-term and short-term dependencies in time-series data, dynamically optimize resource allocation, and detect anomalies across enterprise data with privacy protection. In addition, collaborative solutions lack virtual simulation and deduction mechanisms, making it difficult to identify potential risks in advance. Resource allocation optimization often relies on ex-post adjustments, resulting in persistently high collaboration costs. Finally, the existing collaborative management system has a poor visual interaction experience, a single data display format, and lacks 3D visualization and natural language interaction capabilities, making it difficult for employees to operate. At the same time, permission operation logs and audit data are easily tampered with, cross-enterprise collaborative compliance audits lack a reliable evidence storage carrier, and data traceability is difficult. Summary of the Invention

[0003] In view of the shortcomings of existing technologies, the purpose of this invention is to provide an enterprise digital collaborative management system that integrates cutting-edge technologies such as edge computing, AI deep learning, blockchain notarization, and digital twins, and can realize multi-dimensional collaboration, intelligent conflict resolution, dynamic access control, and security and compliance auditing.

[0004] Another objective of this invention is to provide a method for enterprise digital collaborative management using the aforementioned system.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an enterprise digital collaborative management system, comprising: The data acquisition module is used to collect business data, employee operation data, and external related data from various business systems within the enterprise. The business data includes project progress data, resource allocation data, and process approval data. The external related data includes supply chain data and customer demand data. The data acquisition module integrates an edge computing unit to perform real-time data preprocessing at edge nodes close to the data source, reducing the transmission pressure on the core network. The preprocessing includes data filtering, key information extraction, and real-time caching, and outputs standardized raw data after preprocessing. A multi-dimensional collaboration engine, communicating with the data acquisition module, is used to establish multi-dimensional collaboration models for project collaboration, departmental collaboration, and cross-enterprise collaboration. These models, based on preset collaboration rules, perform real-time association, field mapping, logical verification, and cross-scenario synchronization processing on standardized raw data output by the data acquisition module. The preset collaboration rules include data permission rules, process flow rules, and responsibility allocation rules. The multi-dimensional collaboration engine incorporates an AI large-scale model inference unit. This unit uses an industry-specific large-scale model generated by fine-tuning a general large-scale model base. This industry-specific large-scale model is adapted to vertical industries such as manufacturing, supply chain, project management, and financial services. Customized optimization is achieved through a LoRA lightweight fine-tuning method combined with a cloud-edge collaborative training and push architecture, enabling intelligent matching of collaboration scenarios, cross-role intent understanding, and automated resolution of collaboration conflicts. The conflict resolution employs conflict weight calculation and optimal solution selection formulas. The integrated data generated after association, synchronization, and conflict resolution by the multi-dimensional collaboration engine constitutes the collaborative data. The dynamic permission management module is connected to the multi-dimensional collaboration engine and is used to dynamically allocate data access permissions and operation permissions according to employee positions, project roles, and collaboration scenarios. The permission allocation includes fine-grained data viewing permissions, process approval permissions, and collaborative operation permissions. The intelligent decision support module, based on the collaborative data processed by the multi-dimensional collaborative engine, generates project progress analysis reports, resource optimization suggestions, and risk warning information through a data analysis model that integrates AI algorithms. The security protection module is used to encrypt and protect the data transmission, storage, and operation during the collaboration process. It includes a data transmission encryption unit, a storage encryption unit, and an operation log auditing unit. The operation log auditing unit records the log information of all collaborative operations and supports traceability query. The security protection module also integrates a privacy computing unit, which uses federated learning and differential privacy technology to achieve cross-enterprise data usability-invisibility collaborative analysis. The digital twin collaboration module communicates with the multi-dimensional collaboration engine and intelligent decision support module to build digital twins of enterprise business processes and project scenarios. It maps collaborative data to the twin for virtual simulation and deduction, and supports collaborative scheme pre-playing, risk early identification and resource optimization configuration simulation. The visualization and interaction module is used to display collaborative data, digital twin simulation results, project progress, permission allocation and decision support results in the form of a 3D visualization dashboard and dynamic flowchart. It supports employees to complete collaborative task creation, progress updates, process approval and information feedback through natural language interaction and voice commands. The blockchain evidence storage unit, as the core security component of the dynamic permission management module, communicates and connects with the dynamic permission management module, the security protection module, and the visual interaction module respectively. It is used to store the permission allocation records, permission change logs, and collaborative operation audit data of the dynamic permission management module on the blockchain, forming an immutable permission traceability chain. At the same time, it provides a compliance audit basis for cross-enterprise collaborative scenarios. Its evidence storage data is synchronized in real time with the operation log audit unit of the security protection module, and supports permission traceability query and audit report generation through the visual interaction module.

[0006] To further improve the system in this invention, the AI ​​large model inference unit of the multi-dimensional collaborative engine first calculates the conflict weight using the following formula when resolving collaborative conflicts: In the formula, The overall weight of the conflict is [0,1]. The higher the weight, the higher the priority of the conflict. The number of business dimensions involved in the conflict (such as schedule, resources, and permissions). For the first i The weighting coefficients for each business dimension satisfy... Configuration based on industry standards and enterprise historical data; For the first i The degree of conflict impact across each business dimension; This represents the maximum value of the degree of conflict impact in this dimension; j The hierarchy of collaborative roles involved in the conflict; For the first i The weighting coefficients for each business dimension satisfy... , m Total number of character levels; For the first j The urgency of conflict resolution at each role level. This represents the maximum urgency level. Based on conflict weight The optimal conflict resolution method is selected using the following formula: In the formula, The optimal conflict resolution solution; S For the set of all candidate solutions; c This is a tradeoff coefficient, with a value range of [0,1], used to balance conflict weights and solution costs; The cost of implementing the solution; The collaborative efficiency loss value of solution s. This represents the average efficiency loss value across all candidate schemes.

[0007] To further improve the system in this invention, the data acquisition module supports multiple data interface protocols, including RESTful API, WebService, direct database connection protocol and industrial Internet protocol. The edge computing unit supports millisecond-level data preprocessing response, and the cached data is automatically synchronized to the cloud database after the core network is restored. To further improve the system in this invention, the blockchain evidence storage unit adopts a consortium blockchain architecture. The consortium nodes include internal enterprise management nodes, cross-enterprise collaboration nodes, and third-party audit nodes. Specifically: internal enterprise management nodes are responsible for initial evidence storage and node management; cross-enterprise collaboration nodes only have the right to query evidence storage data and have no right to modify it; and third-party audit nodes are used to verify evidence storage data during compliance audits to ensure the credibility of the evidence storage data.

[0008] To further improve the system in this invention, the virtual simulation of the digital twin collaboration module calculates the similarity of the collaboration scene to achieve accurate mapping between the twin and the actual scene. The specific calculation formula is as follows: In the formula, Sim The similarity between the digital twin scenario and the actual collaborative scenario is denoted by a value in the range [0,1]. The closer the value is to 1, the higher the mapping accuracy. p The number of dimensions for scene features; w k For the first k The weights of each feature dimension are calculated based on the entropy weight method; For actual collaborative scenarios k 3D feature vector; For digital twin scenarios k 3D feature vector; Let cosine similarity be the similarity between the two vectors. Simultaneously, the following formula is used to correct the derivation error: In the formula, E corr This is the corrected inference error; E raw This represents the original deduction error; l This is the error correction coefficient, with a value range of [0.1, 0.5]. t This is the current simulation time; Sim ( t )for t Scene similarity at any given moment; d This is the attenuation coefficient, used to control the degree of influence of historical similarity on the current error correction.

[0009] To further improve the system in this invention, the data analysis model integrating AI algorithms in the intelligent decision support module includes a trend prediction model based on deep learning, a resource matching model driven by reinforcement learning, and a cross-enterprise anomaly detection model under a federated learning architecture. Specifically, in the federated learning architecture of the cross-enterprise anomaly detection model under the federated learning architecture, the following formula is used to achieve encrypted aggregation of model parameters: In the formula, i global These are the aggregated global model parameters; K The number of enterprise nodes participating in federated learning; N k For the first k Local data sample size for each enterprise node; i k For the first k Local model parameters for each enterprise node; For public key based pub _ key The homomorphic encryption function ensures privacy protection during parameter transmission; Decryption is performed using the private key. pri_key right i global Decrypt to obtain usable global model parameters: .

[0010] A digital collaborative management method for enterprises includes the following steps: Step S1: The edge computing unit of the data acquisition module collects and preprocesses business data, employee operation data and external related data from various business systems within the enterprise at the edge node. The preprocessed data is transmitted to the core system through encryption. At the same time, the collected raw data is cleaned to obtain standardized data. Step S2: The multi-dimensional collaboration engine establishes a multi-dimensional collaboration model for project collaboration, departmental collaboration, and cross-enterprise collaboration based on preset collaboration rules. The standardized data obtained in Step S1 is imported into the collaboration model. The AI ​​large model inference unit calls the conflict weight calculation and optimal solution selection formula to realize intelligent matching of collaboration scenarios and automatic resolution of collaboration conflicts. Real-time data association and synchronization are completed to generate collaboration data. Step S3: The dynamic permission management module dynamically allocates access permissions and operation permissions for collaborative data based on employee positions, project roles, and collaboration scenarios. At the same time, it encrypts and transmits the permission operation records to the blockchain evidence storage unit for on-chain evidence storage. After the blockchain evidence storage unit generates an evidence storage receipt, it is synchronized to the operation log auditing unit of the security protection module to ensure that collaborators can only access and operate collaborative data within their authorized scope. Step S4: The digital twin collaboration module constructs a digital twin of the enterprise's business processes or project scenarios based on collaborative data. It calls the scenario similarity calculation and inference error correction formula to perform virtual simulation inference, pre-simulate the collaboration plan and identify potential risks, and output optimization suggestions. The visualization interaction module displays the collaborative data and twin inference results in a 3D visualization form. Collaborators complete collaborative operations through natural language and voice commands. The multi-dimensional collaboration engine pushes updated information to relevant terminals in real time. Step S5: The intelligent decision support module analyzes the collaborative data; Step S6: The security protection module ensures data transmission security through SSL / TLS encryption protocol, and storage security through AES encryption algorithm. The privacy computing unit uses differential privacy technology to process sensitive data. When compliance audit or authorization traceability is required, the administrator can retrieve the evidence storage data of the blockchain evidence storage unit through the visual interaction module, generate an audit report, and achieve full-process traceability.

[0011] To better implement the method of the present invention, further, in step S3, the interaction process between the blockchain evidence storage unit and the dynamic permission management module is as follows: Step S31: After the dynamic permission management module performs permission allocation or change operations, it generates operation records in real time; Step S32: The operation record is encrypted through the transmission encryption unit of the security protection module and then sent to the blockchain evidence storage unit; Step S33: The blockchain evidence storage unit uses a hash algorithm to generate a unique hash value for the operation record, and writes it into the consortium blockchain along with the operation record; Step S34: After the evidence storage is completed, a successful evidence storage receipt is returned to the dynamic permission management module, and simultaneously synchronized to the operation log audit unit; Step S35: When traceability is required, a query request is initiated through the visual interaction module. The blockchain evidence storage unit verifies the data integrity based on the hash value and then returns the query result.

[0012] To better implement the method of the present invention, further, in step S5, the specific process of the intelligent decision support module analyzing the collaborative data includes the following steps: Step S51: Call the deep learning-based trend prediction model that integrates AI algorithms in the intelligent decision support module, train it, input historical collaborative time series data, and output the future trend prediction results of project progress and resource consumption. Step S52: Call the reinforcement learning-driven resource matching model that integrates AI algorithms in the intelligent decision support module, train it based on the current resource status and task requirements, and after training, optimize the resource allocation scheme through the DDPG algorithm to generate resource optimization suggestions. Step S53: Call the cross-enterprise anomaly detection model under the federated learning architecture that integrates AI algorithms in the intelligent decision support module, train it, and after training, work together with cross-enterprise nodes to complete anomaly scoring and global threshold aggregation, identify abnormal information in collaborative data and trigger risk warnings.

[0013] To better implement the method of this invention, the specific process for training the deep learning-based trend prediction model, the reinforcement learning-driven resource matching model, and the cross-enterprise anomaly detection model under the federated learning architecture that integrate AI algorithms in the intelligent decision support module is as follows: The training process of the deep learning-based trend prediction model is as follows: historical collaborative data is divided into training and test sets in a 7:3 ratio, and training samples are generated using the sliding window method; the model is trained using the mean squared error as the loss function and the Adam optimizer is used for at least 100 iterations; the model performance is verified using the test set, and the prediction error is considered as a result of these iterations. MAE Stop training and save model parameters when the percentage is less than 5%. The training process of the reinforcement learning-driven resource matching model is as follows: Initialize an experience replay buffer with a capacity of 105. D Randomly initialize the parameters of the Actor and Critic networks; in each training episode, the agent adjusts its parameters according to the current state. S Selecting actions using the Actor network A Receive a reward after performing the action. r With the new state S ',Will( S , A, r , S ′) Save D ;from D Randomly sample batches of samples, update the parameters of the Critic and Actor networks, and synchronize the parameters of the target network; repeat the training for 500 episodes, and stop training when the average reward converges; The training process of the cross-enterprise anomaly detection model under the federated learning architecture is as follows: each enterprise node trains an autoencoder model based on local collaborative data for 50 training rounds, with the reconstruction error as the loss function; each node calculates the 95th percentile of its local anomaly score as τ. k After encryption, the data is uploaded to the aggregation node; the aggregation node calculates τ using the encrypted aggregation formula based on the parameters. global After decryption, the data is distributed to each node; each node uses τ global Perform anomaly detection, periodically update the local model, and re-aggregate global thresholds.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention integrates an edge computing unit through a data acquisition module to complete real-time preprocessing at edge nodes close to the data source, effectively reducing the transmission pressure of the core network and achieving millisecond-level data response; the multi-dimensional collaboration engine constructs a three-level collaboration model of project, department, and cross-enterprise, completes data association, field mapping and logic verification through preset rules, and combines the AI ​​large model inference unit to realize intelligent matching of collaboration scenarios and automatic resolution of conflicts, greatly reducing manual intervention and improving collaboration efficiency; the application of conflict weight calculation and optimal solution selection formula ensures priority ranking and scheme optimization of conflict handling, and reduces collaboration efficiency loss; (2) In this invention, the state permission management module dynamically allocates permissions based on job position, role, and scenario. Combined with the consortium blockchain architecture of the blockchain evidence storage unit, permission allocation records, change logs, and operation audit data are stored on the blockchain, achieving immutability and full-process traceability of permission operations. The consortium blockchain nodes are set up with internal enterprise management nodes, cross-enterprise collaboration nodes, and third-party audit nodes. This ensures secure data access for cross-enterprise collaboration, meets compliance audit requirements, effectively avoids the risks of data leakage and permission abuse, and complies with regulatory provisions such as the Data Security Law. (3) In this invention, the intelligent decision support module integrates three AI algorithm models. The deep learning-based trend prediction model adopts the Transformer encoder-decoder architecture to accurately capture the long-term and short-term dependencies of time series data and reduce the prediction error of project progress and resource consumption. The reinforcement learning-driven resource matching model achieves a dynamic balance between resource utilization and task completion efficiency through the MDP framework and DDPG algorithm, thereby improving the resource allocation optimization rate. The cross-enterprise anomaly detection model under the federated learning architecture adopts encrypted aggregation technology to achieve accurate identification of cross-enterprise anomaly data while ensuring that the data is available but not visible, thus shortening the risk warning response time. (4) In this invention, the digital twin collaboration module constructs a digital twin of the enterprise's business processes and project scenarios. Through scenario similarity calculation and deduction error correction formula, it achieves high-precision mapping between the twin and the actual scenario, supports pre-simulation of collaborative solutions and early identification of risks. The virtual simulation deduction function can discover potential problems such as resource allocation conflicts and schedule delays in advance, providing a scientific basis for collaborative decision-making and reducing the risk of project failure and collaborative costs. (5) In this invention, the visualization interaction module displays collaborative data and inference results in the form of a three-dimensional visualization dashboard and dynamic flowchart, supports natural language interaction and voice command operation, and reduces the operating threshold for employees; at the same time, it supports administrators to retrieve blockchain evidence data to generate standardized audit reports, shorten the audit process, and greatly improve the efficiency of compliance audit. (6) The data acquisition module in this invention supports multiple interface protocols such as RESTfulAPI, WebService, and OPCUA, which can be adapted to various existing business systems of enterprises and reduce system integration costs. The cloud-edge collaborative architecture and modular design enable the system to flexibly expand functional modules according to enterprise size and industry characteristics, and adapt to the collaborative management needs of different vertical industries such as manufacturing, supply chain, and financial services. Attached Figure Description

[0015] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the system framework in this invention; Figure 2 This is a flowchart of the method in this invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0018] Example 1: This embodiment provides a workflow refactoring management technology, the main structure of which is as follows: Figure 1 As shown, it includes: The data acquisition module is used to collect business data, employee operation data, and external related data from various business systems within the enterprise. The business data includes project progress data, resource allocation data, and process approval data. The external related data includes supply chain data and customer demand data. The data acquisition module integrates an edge computing unit to perform real-time data preprocessing at edge nodes close to the data source, reducing the transmission pressure on the core network. The preprocessing includes data filtering, key information extraction, and real-time caching, and outputs standardized raw data after preprocessing. A multi-dimensional collaboration engine, communicating with the data acquisition module, is used to establish multi-dimensional collaboration models for project collaboration, departmental collaboration, and cross-enterprise collaboration. These models, based on preset collaboration rules, perform real-time association, field mapping, logical verification, and cross-scenario synchronization processing on standardized raw data output by the data acquisition module. The preset collaboration rules include data permission rules, process flow rules, and responsibility allocation rules. The multi-dimensional collaboration engine incorporates an AI large-scale model inference unit. This unit uses an industry-specific large-scale model generated by fine-tuning a general large-scale model base. This industry-specific large-scale model is adapted to vertical industries such as manufacturing, supply chain, project management, and financial services. Customized optimization is achieved through a LoRA lightweight fine-tuning method combined with a cloud-edge collaborative training and push architecture, enabling intelligent matching of collaboration scenarios, cross-role intent understanding, and automated resolution of collaboration conflicts. The conflict resolution employs conflict weight calculation and optimal solution selection formulas. The integrated data generated after association, synchronization, and conflict resolution by the multi-dimensional collaboration engine constitutes the collaborative data. The dynamic permission management module is connected to the multi-dimensional collaboration engine and is used to dynamically allocate data access permissions and operation permissions according to employee positions, project roles, and collaboration scenarios. The permission allocation includes fine-grained data viewing permissions, process approval permissions, and collaborative operation permissions. The intelligent decision support module, based on the collaborative data processed by the multi-dimensional collaborative engine, generates project progress analysis reports, resource optimization suggestions, and risk warning information through a data analysis model that integrates AI algorithms. The security protection module is used to encrypt and protect the data transmission, storage, and operation during the collaboration process. It includes a data transmission encryption unit, a storage encryption unit, and an operation log auditing unit. The operation log auditing unit records the log information of all collaborative operations and supports traceability query. The security protection module also integrates a privacy computing unit, which uses federated learning and differential privacy technology to achieve cross-enterprise data "usable but not visible" collaborative analysis. The digital twin collaboration module communicates with the multi-dimensional collaboration engine and intelligent decision support module to build digital twins of enterprise business processes and project scenarios. It maps collaborative data to the twin for virtual simulation and deduction, and supports collaborative scheme pre-playing, risk early identification and resource optimization configuration simulation. The visualization and interaction module is used to display collaborative data, digital twin simulation results, project progress, permission allocation and decision support results in the form of a 3D visualization dashboard and dynamic flowchart. It supports employees to complete collaborative task creation, progress updates, process approval and information feedback through natural language interaction and voice commands. The blockchain evidence storage unit, as the core security component of the dynamic permission management module, communicates and connects with the dynamic permission management module, the security protection module, and the visual interaction module respectively. It is used to store the permission allocation records, permission change logs, and collaborative operation audit data of the dynamic permission management module on the blockchain, forming an immutable permission traceability chain. At the same time, it provides a compliance audit basis for cross-enterprise collaborative scenarios. Its evidence storage data is synchronized in real time with the operation log audit unit of the security protection module, and supports permission traceability query and audit report generation through the visual interaction module.

[0019] Methods for enterprise digital collaborative management based on the above system, such as Figure 2 As shown, it includes the following steps: Step S1: The edge computing unit of the data acquisition module collects and preprocesses business data, employee operation data and external related data from various business systems within the enterprise at the edge node. The preprocessed data is transmitted to the core system through encryption. At the same time, the collected raw data is cleaned to obtain standardized data. Step S2: The multi-dimensional collaboration engine establishes a multi-dimensional collaboration model for project collaboration, departmental collaboration, and cross-enterprise collaboration based on preset collaboration rules. The standardized data obtained in Step S1 is imported into the collaboration model. The AI ​​large model inference unit calls the conflict weight calculation and optimal solution selection formula to realize intelligent matching of collaboration scenarios and automatic resolution of collaboration conflicts. Real-time data association and synchronization are completed to generate collaboration data. Step S3: The dynamic permission management module dynamically allocates access permissions and operation permissions for collaborative data based on employee positions, project roles, and collaboration scenarios. At the same time, it encrypts and transmits the permission operation records to the blockchain evidence storage unit for on-chain evidence storage. After the blockchain evidence storage unit generates an evidence storage receipt, it is synchronized to the operation log auditing unit of the security protection module to ensure that collaborators can only access and operate collaborative data within their authorized scope. Step S4: The digital twin collaboration module constructs a digital twin of the enterprise's business processes or project scenarios based on collaborative data. It calls the scenario similarity calculation and inference error correction formula to perform virtual simulation inference, pre-simulate the collaboration plan and identify potential risks, and output optimization suggestions. The visualization interaction module displays the collaborative data and twin inference results in a 3D visualization form. Collaborators complete collaborative operations through natural language and voice commands. The multi-dimensional collaboration engine pushes updated information to relevant terminals in real time. Step S5: The intelligent decision support module analyzes the collaborative data; Step S6: The security protection module ensures data transmission security through SSL / TLS encryption protocol, and storage security through AES encryption algorithm. The privacy computing unit uses differential privacy technology to process sensitive data. When compliance audit or authorization traceability is required, the administrator can retrieve the evidence storage data of the blockchain evidence storage unit through the visual interaction module, generate an audit report, and achieve full-process traceability.

[0020] Example 2: This embodiment, based on the above embodiments, further defines the AI ​​large-scale model inference unit to achieve intelligent matching of collaborative scenarios, cross-role intent understanding, and automated resolution of collaborative conflicts. The AI ​​large-scale model inference unit of the multi-dimensional collaborative engine supports collaborative dialogue generation, intelligent decision-making in scenarios with unclear rules, and cross-language collaborative communication. The industry-specific large model is fine-tuned and trained based on enterprise historical collaborative data and industry standard process data to adapt to the collaborative scenario requirements of specific industries. Its specific content is as follows: The general-purpose large model base of the industry-specific large model is selected from at least one of LLAMA2-13B, Qwen2.5vl-32B, and DeepSeek-14B. The general-purpose large model base has multimodal understanding, long text processing and logical reasoning capabilities, and is adapted to the complex instruction parsing needs of enterprise collaborative management scenarios.

[0021] The LoRA lightweight fine-tuning method is implemented through the KTransformers heterogeneous inference framework. It only performs incremental training on the attention layer weights of the general large model base, occupies no more than 41GB of GPU memory and no more than 2TB of RAM, and has a fine-tuning throughput of no less than 46.55 tokens / s. It supports training on a single consumer-grade GPU (RTX4090 and above) or Ascend NPU, which greatly reduces the hardware cost of fine-tuning.

[0022] The cloud-edge collaborative training and push architecture adopts a storage-computing separation technology solution. Edge nodes retain sensitive collaborative data of enterprises (including historical conflict cases, process specifications, and permission rule data), while the cloud provides intelligent computing resources. The model parameters are transmitted across domains through intelligent IP wide area network (AIWAN), reducing the degradation of computing efficiency in scenarios with a distance of 100KM-400KM and ensuring the security of fine-tuning requirements without leaving the domain.

[0023] The fine-tuning data for the industry-specific large model includes an industry collaboration scenario corpus and an enterprise-specific dataset. The industry collaboration scenario corpus covers process specifications, collaboration standards, and conflict resolution cases for vertical industries (such as manufacturing supply chain collaboration rules and financial service approval process specifications). The enterprise-specific dataset includes enterprise historical collaboration logs, organizational structure data, permission allocation rules, and project collaboration cases.

[0024] The fine-tuned data is processed by Huawei Cloud Flexus data cleaning and extraction tool, which automatically completes format standardization, redundant data removal and high-quality question-answer pair generation, reducing the traditional manual document processing work of several months to days and improving the accuracy of the generated training data.

[0025] The fine-tuning process of the industry-specific large model also includes a model splitting and learning step: the general large model base is split into a core layer and an adaptation layer according to the security level. The core layer is deployed on cloud intelligent computing nodes, and the adaptation layer is deployed on edge nodes and trained based on enterprise personalized data. The low latency and high reliability of cross-node data transmission are ensured through precise flow control technology and SRv6 slicing function.

[0026] When resolving collaborative conflicts, the AI ​​large-model inference unit of the multi-dimensional collaborative engine first calculates the conflict weight using the following formula: In the formula, The overall weight of the conflict is [0,1]. The higher the weight, the higher the priority of the conflict. The number of business dimensions involved in the conflict (such as schedule, resources, and permissions). For the first i The weighting coefficients for each business dimension satisfy... Configuration based on industry standards and enterprise historical data; For the first i The degree of conflict impact across each business dimension; This represents the maximum value of the degree of conflict impact in this dimension; j The hierarchy of collaborative roles involved in the conflict; For the first i The weighting coefficients for each business dimension satisfy... , mTotal number of character levels; For the first j The urgency of conflict resolution at each role level. This represents the maximum urgency level.

[0027] in, and The analytic hierarchy process (AHP) is used to determine the following: construct a judgment matrix for business dimensions and role levels, calculate eigenvectors and perform consistency checks, and use the eigenvectors that pass the checks as weight coefficients.

[0028] Based on conflict weight The optimal conflict resolution method is selected using the following formula: In the formula, The optimal conflict resolution solution; S For the set of all candidate solutions; c This is a tradeoff coefficient, with a value range of [0,1], used to balance conflict weights and solution costs; The cost of implementing the solution; The collaborative efficiency loss value of solution s. This represents the average efficiency loss value across all candidate schemes.

[0029] Example 3: Based on the above embodiments, this embodiment further defines the data acquisition module, which supports multiple data interface protocols, including RESTful API, WebService, database direct connection protocol and industrial Internet protocol. The edge computing unit supports millisecond-level data preprocessing response, and the cached data is automatically synchronized to the cloud database after the core network is restored.

[0030] Example 4: Based on the above embodiments, this embodiment further defines the blockchain evidence storage unit. The blockchain evidence storage unit adopts a consortium blockchain architecture. The consortium nodes include internal enterprise management nodes, cross-enterprise collaboration nodes, and third-party audit nodes. Among them, the internal enterprise management nodes are responsible for the initial evidence storage and node management of permission operations; the cross-enterprise collaboration nodes only have the right to query the evidence storage data and have no right to modify it; the third-party audit nodes are used to verify the evidence storage data during compliance audits to ensure the credibility of the evidence storage data.

[0031] The consortium blockchain nodes of the blockchain evidence storage unit adopt the Byzantine fault-tolerant algorithm to ensure that the permission records remain consistent and traceable even in the event of partial node failure or malicious attack.

[0032] The core functions of the blockchain evidence storage unit include: 1) providing full-process on-chain evidence storage for the allocation, modification, and revocation of permissions in the dynamic permission management module. Each operation record includes the operator, operation time, operation content, and the state before and after the permission change, ensuring that permission operations are traceable and non-repudiable; 2) serving as a trusted third-party evidence storage carrier in cross-enterprise collaboration scenarios, recording the permission acquisition and operation trajectory of collaborators from different enterprises, meeting compliance requirements such as the Data Security Law and the Personal Information Protection Law; 3) linking with the operation log audit unit and visual interaction module of the security protection module, enabling rapid retrieval of on-chain evidence storage data and generation of standardized audit reports when compliance audits or security incident tracing are required.

[0033] The interaction process between the blockchain evidence storage unit and the dynamic permission management module is as follows: Step S31: After the dynamic permission management module performs permission allocation or change operations, it generates operation records in real time; Step S32: The operation record is encrypted through the transmission encryption unit of the security protection module and then sent to the blockchain evidence storage unit; Step S33: The blockchain evidence storage unit uses a hash algorithm to generate a unique hash value for the operation record, and writes it into the consortium blockchain along with the operation record; Step S34: After the evidence storage is completed, a successful evidence storage receipt is returned to the dynamic permission management module, and simultaneously synchronized to the operation log audit unit; Step S35: When traceability is required, a query request is initiated through the visual interaction module. The blockchain evidence storage unit verifies the data integrity based on the hash value and then returns the query result.

[0034] Example 5: Based on the above embodiments, this embodiment further specifies that in the virtual simulation and deduction of the digital twin collaborative module, the similarity of the collaborative scene is calculated to achieve accurate mapping between the twin and the actual scene. The specific calculation formula is as follows: In the formula, Sim The similarity between the digital twin scenario and the actual collaborative scenario is denoted by a value in the range [0,1]. The closer the value is to 1, the higher the mapping accuracy. p The number of dimensions for scene features; w k For the first k The weights of each feature dimension are calculated based on the entropy weight method; For actual collaborative scenarios k 3D feature vector; For digital twin scenarios k 3D feature vector; Let cosine similarity be the similarity between the two vectors. Simultaneously, the following formula is used to correct the derivation error: In the formula, E corr This is the corrected inference error; E raw This represents the original deduction error; l This is the error correction coefficient, with a value range of [0.1, 0.5]. t This is the current simulation time; Sim ( t )for t Scene similarity at any given moment; d This is the attenuation coefficient, used to control the degree of influence of historical similarity on the current error correction.

[0035] The virtual simulation and deduction of the digital twin collaboration module supports real-time access to actual collaboration data and is dynamically updated through scene similarity calculation formulas. Sim Value, when Sim When the value is less than 0.85, the feature vector calibration is automatically triggered to ensure dynamic synchronization between the twin and the physical world. When the actual collaborative progress deviates from the simulation plan by more than the preset threshold (e.g., 10%), the simulation parameters are adjusted through the error correction formula, and adjustment suggestions are output.

[0036] Example 6: Based on the above embodiments, this embodiment further defines the data analysis model that integrates AI algorithms in the intelligent decision support module, specifically including a trend prediction model based on deep learning, a resource matching model driven by reinforcement learning, and a cross-enterprise anomaly detection model under a federated learning architecture.

[0037] The specific architecture of the deep learning-based trend prediction model is as follows: the input layer receives time-series collaborative data (including project progress values, resource consumption values, and process completion rates), with dimensions [T, D] (T is the time step, and D is the feature dimension); the embedding layer maps the input features to a high-dimensional space through linear transformation, with an output dimension [T, D]. embed ] (D embed =128); the encoder uses a 6-layer Transformer encoder, each containing a multi-head self-attention mechanism (number of heads h=8), a layer normalization and feedforward neural network (hidden layer dimension 2048), and the attention weights are calculated using the following formula: In the formula, Q, K, and V are the query matrix, key matrix, and value matrix, respectively, all with dimensions [T, d]. k ](d k =D embed / h=16); M is a mask matrix used to mask data from future time steps; d kScaling factor to avoid gradient vanishing; the decoder employs a 4-layer Transformer decoder, which associates the encoder output through a cross-attention mechanism, and finally outputs the future t through a linear layer and a sigmoid activation function. pred Predicted values ​​at each time step: In the formula, s Use the Sigmoid activation function; W out This is the output layer weight matrix; b out For output layer bias terms; X Input time series data.

[0038] The specific training method for this deep learning-based trend prediction model is as follows: (1) Divide the historical collaborative data into training and testing sets in a 7:3 ratio, and use the sliding window method (window size) T =30) Generate training samples; (2) The model is trained using the mean squared error (MSE) as the loss function and the Adam optimizer (learning rate 0.0001, decay rate 0.9) for at least 100 iterations. (3) Verify the model performance using the test set, when the prediction error MAE Training is stopped when the percentage is less than 5%, and the model parameters are saved.

[0039] The specific architecture of the reinforcement learning-driven resource matching model is as follows: state space S Defined as S ={ R , T , P},in R This is a resource inventory vector (including human, material, and financial resources). T The vector of tasks to be assigned. P Task priority matrix; action space A Defined as a resource allocation matrix A ∈[0,1] m×n ( m For the number of resource types, n (Number of tasks) A i,j Indicates the first i Class resources are allocated to the first j The proportion of tasks; reward function r ( S , A The following formula is used to calculate and balance resource utilization and task completion efficiency: In the formula, oh 1, oh 2 is the weighting coefficient, which satisfies oh 1+ oh 2 = 1; α The efficiency coefficient has a value range of [0.8, 1.2]. T j For the first j Total resource requirements for each task; Optimization algorithm: DDPG algorithm is used, including an Actor network (outputting the optimal action). A *) and the Critic network (evaluating action value) update the Actor network parameters using the following formula: In the formula, or The learning rate, with a value ranging from [0.001, 0.01]. J ( i π ) is the objective function of the Actor network; D This serves as a buffer for experience replay. Q ( s , A | i Q ) represents the action value output by the Critic network.

[0040] The specific training method for this reinforcement learning-driven resource matching model is as follows: (1) Initialize the experience replay buffer D (capacity 105), and randomly initialize the parameters of the Actor network and Critic network; (2) In each training episode, the agent selects action A through the Actor network according to the current state S, and after executing the action, it obtains reward r and a new state S′, and stores (S,A,r,S′) into D; (3) Randomly sample a batch of samples from D (batch size 64), update the parameters of the Critic network and the Actor network, and synchronize the target network parameters (soft update coefficient τ=0.001). (4) Repeat the training for 500 episodes, and stop training when the average reward converges.

[0041] The specific architecture of the cross-enterprise anomaly detection model under the federated learning architecture is as follows: Local Model: Each enterprise node deploys an autoencoder (AE) model, with an architecture of input layer → encoding layer (a 3-layer fully connected network, with dimensions as follows). D → D / 2→ D / 4) → Decoding layer (3-layer fully connected network) D / 4→ D / 2→ DAnomaly score: The local model calculates the sample reconstruction error using the following formula as the anomaly score: In the formula, x For local collaborative data samples, Reconstruct samples for the model; D For sample feature dimensions; Global threshold aggregation: The federated averaging algorithm is used to encrypt and aggregate the local anomaly score quantiles of each node to generate a global anomaly threshold. In the formula, t k For the first k The local anomaly threshold for each node (taken as the 95th percentile of the local score); Enc( t k , pub _ key ) is a homomorphic encryption function; anomaly detection: each node scores its local sample and compares it with the decrypted τ. global In comparison, when score local (x)>τ global When the data is identified as abnormal, it is synchronized to the collaboration engine to trigger an alert.

[0042] The specific training method for the cross-enterprise anomaly detection model under this federated learning architecture is as follows: (1) Each enterprise node trains an autoencoder model based on local collaborative data, with 50 training rounds and the loss function being the reconstruction error; (2) Each node calculates the 95th percentile of its local anomaly score as... t k After encryption, it is uploaded to the aggregation node; (3) Aggregation nodes calculate the aggregation formula by encrypting parameters. t global After decryption, the data is distributed to each node. (4) Each node adopts t global Perform anomaly detection, update the local model periodically (e.g., daily), and re-aggregate global thresholds.

[0043] Furthermore, in the federated learning architecture of the cross-enterprise anomaly detection model, the following formula is used to achieve encrypted aggregation of model parameters: In the formula, i global These are the aggregated global model parameters; K The number of enterprise nodes participating in federated learning; N k For the first kLocal data sample size for each enterprise node; i k For the first k Local model parameters for each enterprise node; For public key based pub _ key The homomorphic encryption function ensures privacy protection during parameter transmission; Decryption is performed using the private key. pri_key right i global Decrypt to obtain usable global model parameters: .

[0044] The intelligent decision support module analyzes collaborative data in the following specific processes: Step S51: Call the deep learning-based trend prediction model that integrates AI algorithms in the intelligent decision support module, train it, input historical collaborative time series data, and output the future trend prediction results of project progress and resource consumption. Step S52: Call the reinforcement learning-driven resource matching model that integrates AI algorithms in the intelligent decision support module, train it based on the current resource status and task requirements, and after training, optimize the resource allocation scheme through the DDPG algorithm to generate resource optimization suggestions. Step S53: Call the cross-enterprise anomaly detection model under the federated learning architecture that integrates AI algorithms in the intelligent decision support module, train it, and after training, work together with cross-enterprise nodes to complete anomaly scoring and global threshold aggregation, identify abnormal information in collaborative data and trigger risk warnings.

[0045] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An enterprise digital collaborative management system, characterized in that, The application comprises: a data acquisition module for acquiring business data, employee operation data and external associated data of various business systems in an enterprise; the data acquisition module integrates an edge computing unit for completing real-time data preprocessing at the edge node close to the data source, reducing the transmission pressure of the core network, and outputting standardized raw data after preprocessing; a multi-dimensional collaboration engine for establishing a multi-dimensional collaboration model of project collaboration, department collaboration and cross-enterprise collaboration, which performs real-time association, field mapping, logical verification and cross-scene synchronization processing on the standardized raw data based on preset collaboration rules; the multi-dimensional collaboration engine has an AI large model reasoning unit built-in to realize intelligent matching of collaboration scenarios, cross-role intent understanding and automatic resolution of collaboration conflicts; the integrated data generated after association, synchronization and conflict resolution by the multi-dimensional collaboration engine is collaborative data; a dynamic permission management module for dynamically assigning data access permissions and operation permissions according to employee positions, project roles and collaboration scenarios; an intelligent decision support module for generating project progress analysis reports, resource optimization suggestions and risk warning information through a data analysis model fused with AI algorithms; a security protection module for encrypting and protecting data transmission, storage and operation during collaboration; a digital twin collaboration module for constructing digital twins of enterprise business processes and project scenarios, mapping collaborative data into the digital twins for virtual simulation and deduction, supporting collaborative scenario rehearsal, risk identification in advance and resource optimization configuration simulation; a visual interaction module for displaying collaborative data, digital twin deduction results, project progress, permission allocation and decision support results in the form of three-dimensional visual dashboards and dynamic process diagrams, and supporting employees to complete collaborative task creation, progress update, process approval and information feedback through natural language interaction and voice commands; a blockchain storage unit for recording permission allocation, permission change logs and collaborative operation audit data on the chain to form an unalterable permission traceability chain, and providing compliance audit basis for cross-enterprise collaboration scenarios, and the storage data is synchronized in real time with the operation log audit unit of the security protection module to support permission traceability query and audit report generation through the visual interaction module.

2. The enterprise digital collaboration management system of claim 1, wherein, The AI large model reasoning unit of the multi-dimensional collaboration engine first calculates the conflict weight by the following formula when resolving collaboration conflicts: ； In the formula, is the comprehensive weight of the conflict, and the value range is [0, 1], the higher the weight, the higher the priority of the conflict; is the number of business dimensions involved in the conflict (such as progress, resource, and permission dimensions); is the weight coefficient of the first i business dimension, satisfying , and being configured based on industry standards and enterprise historical data; is the conflict influence degree value of the first i business dimension; is the maximum value of the conflict influence degree of the dimension; j is the role level involved in the conflict; is the weight coefficient of the first i business dimension, satisfying , m is the total number of role levels; is the conflict handling urgency of the first j role level, is the maximum value of the urgency; Based on conflict weight The optimal conflict resolution is selected by the following equation: ； In the formula, is the optimal conflict resolution solution; S is the set of all candidate resolution solutions; γ is a balancing coefficient, taking value in the range [0, 1], for balancing the conflict weight and the solution cost; is the execution cost of the solution; is the synergy efficiency loss value of the solution s, is the average efficiency loss value of all candidate solutions. 3.The enterprise digital collaborative management system according to claim 1 or 2, characterized in that, The data acquisition module supports multiple data interface protocols, including RESTful API, WebService, database direct connection protocol and industrial internet protocol, and the edge computing unit supports millisecond-level data preprocessing response, and cached data is automatically synchronized to the cloud database after recovery in the core network.

4. The enterprise digital collaboration management system of claim 1 or 2, wherein, The blockchain storage unit adopts a consortium chain architecture, and the consortium nodes include enterprise internal management nodes, cross-enterprise collaboration nodes and third-party audit nodes, wherein: the enterprise internal management nodes are responsible for initial storage of permission operations and node management; the cross-enterprise collaboration nodes only have permission to query the stored data and have no modification permission; and the third-party audit nodes are used for verification of stored data for compliance audit to ensure the credibility of the stored data.

5. The enterprise digital collaboration management system of claim 1 or 2, wherein, In the virtual simulation deduction of the digital twin collaborative module, the similarity of the collaborative scene is calculated to realize accurate mapping of the twin and the actual scene, and the specific calculation formula is as follows: ； In the formula, Sim The similarity between the digital twin scenario and the actual collaborative scenario is denoted by a value in the range [0,1]. The closer the value is to 1, the higher the mapping accuracy. p The number of dimensions for scene features; w k For the first k The weights of each feature dimension are calculated based on the entropy weight method; For actual collaborative scenarios k 3D feature vector; For digital twin scenarios k 3D feature vector; Let cosine similarity be the similarity between the two vectors. At the same time, the error correction is deduced through the following formula: ； wherein, E corr is the corrected extrapolation error; E raw is the original extrapolation error; λ is the error correction coefficient, taking the value range [0.1, 0.5]; t is the current extrapolation time; Sim τ is the scene similarity at the time; τ δ is the decay coefficient, used to control the influence degree of the historical similarity on the current error correction.​​ 6. The enterprise digital collaboration management system of claim 1 or 2, wherein, The data analysis model of the intelligent decision support module fusing AI algorithms includes a trend prediction model based on deep learning, a resource matching model driven by reinforcement learning, and a cross-enterprise anomaly detection model under a federated learning architecture; in the federated learning architecture in the cross-enterprise anomaly detection model under the federated learning architecture, the following formula is used to realize model parameter encryption aggregation: ； In the formula, θ global is the global model parameter after aggregation; K is the number of enterprise nodes participating in federated learning; N k is the local data sample size of the k th enterprise node; θ k is the local model parameter of the k th enterprise node; is a homomorphic encryption function based on a public key pub ; key ensures privacy protection during parameter transmission; Decryption by private key pri_key To θ global Decryption, get the global model parameters available: 。 7. A method for managing enterprise digitization collaboration, characterized in that, The method comprises the following steps: Step S1: The edge computing unit of the data acquisition module collects and pre-processes the business data, employee operation data and external associated data of each business system in the enterprise at the edge node, and the pre-processed data is transmitted to the core system through encryption, and the original data collected is cleaned to obtain standardized data; Step S2: The multi-dimensional collaboration engine establishes a multi-dimensional collaboration model of project collaboration, department collaboration and cross-enterprise collaboration based on the preset collaboration rules, imports the standardized data obtained in step S1 into the collaboration model, calls the conflict weight calculation and optimal solution selection formula through the AI large model reasoning unit, realizes intelligent matching and automatic resolution of collaboration scenes, completes real-time association and synchronization of data, and generates collaboration data; Step S3: The dynamic permission management module dynamically allocates access permissions and operation permissions of collaboration data according to the employee post, project role and collaboration scene, and simultaneously encrypts the permission operation record and transmits it to the blockchain storage unit for on-chain storage. The blockchain storage unit generates a storage receipt and synchronizes it to the operation log auditing unit of the security protection module, so that the collaboration personnel can only access and operate the collaboration data within the permission range; Step S4: The digital twin collaborative module constructs a digital twin of the enterprise business process or project scene based on the collaboration data, calls the scene similarity calculation and deduction error correction formula for virtual simulation deduction, pre-plays the collaboration scheme and identifies potential risks, and outputs optimization suggestions; The visual interaction module displays the collaboration data and the twin deduction result in a three-dimensional visual form, and the collaboration personnel completes the collaboration operation through natural language and voice instructions. The multi-dimensional collaboration engine pushes the updated information to the related terminal in real time; Step S5: The intelligent decision support module analyzes the collaboration data; Step S6: The security protection module guarantees data transmission security through SSL / TLS encryption protocol and storage security through AES encryption algorithm. The privacy computing unit processes sensitive data by using differential privacy technology; When compliance audit or permission tracing is required, the administrator retrieves the storage data of the blockchain storage unit through the visual interaction module, generates an audit report, and realizes traceability of the whole process.

8. The enterprise digital collaborative management method of claim 7, wherein, In step S3, the interaction process between the blockchain storage unit and the dynamic permission management module is as follows: Step S31: After the dynamic permission management module executes the permission allocation or change operation, an operation record is generated in real time; Step S32: The operation record is sent to the blockchain storage unit after being encrypted by the transmission encryption unit of the security protection module; Step S33: The blockchain storage unit generates a unique hash value of the operation record using a hash algorithm, and writes the operation record and the hash value into the alliance chain; Step S34: After the storage is completed, a storage success receipt is returned to the dynamic permission management module, and is synchronized to the operation log audit unit; Step S35: When it is necessary to trace back, a query request is initiated through the visual interaction module, and the blockchain storage unit returns a query result after verifying the data integrity based on the hash value.

9. The enterprise digital collaborative management method of claim 7 or 8, wherein, In the step S5, the specific process of analyzing the collaborative data by the intelligent decision support module includes the following steps: Step S51: A trend prediction model based on deep learning with AI algorithm is called in the intelligent decision support module, and is trained. After the training is completed, historical collaborative time series data is input, and future trend prediction results of project progress and resource consumption are output; Step S52: A resource matching model driven by reinforcement learning with AI algorithm is called in the intelligent decision support module, and is trained based on the current resource state and task demand. After the training is completed, a resource allocation scheme is optimized through a DDPG algorithm, and a resource optimization suggestion is generated; Step S53: A cross-enterprise anomaly detection model under a federated learning architecture with AI algorithm is called in the intelligent decision support module, and is trained. After the training is completed, anomaly scoring and global threshold aggregation are completed jointly by cross-enterprise nodes, abnormal information in the collaborative data is identified, and a risk warning is triggered.

10. The method for managing enterprise digitization collaboration according to claim 9, characterized in that, The specific process of training the trend prediction model based on deep learning, the resource matching model driven by reinforcement learning, and the cross-enterprise anomaly detection model under the federated learning architecture with AI algorithm in the intelligent decision support module is as follows: The training process of the deep learning-based trend prediction model is as follows: historical collaborative data is divided into a training set and a test set in a 7:3 ratio, and a sliding window method is used to generate training samples; a mean square error is used as a loss function, an Adam optimizer is used to train the model, and the number of iterations is not less than 100 rounds; the model performance is verified by the test set, and when the prediction error MAE <5% is stopped, the model parameters are saved; The training process of the reinforcement learning driven resource matching model is: initializing an experience replay buffer with a capacity of 105 D , randomly initializing the parameters of the Actor network and the Critic network; in each training episode, the agent selects an action according to the current state S through the Actor network A , obtains a reward after executing the action r and a new state S ', stores ( S , A , r , S ') in D ; randomly samples a batch of samples from D , updates the parameters of the Critic network and the Actor network, and synchronizes the parameters of the target network; repeating the training for 500 episodes, and stopping the training when the average reward converges; The training process of the cross-enterprise anomaly detection model under the federated learning architecture is: each enterprise node trains an autoencoder model based on local collaborative data, the training round is 50 rounds, and the loss function is reconstruction error; each node calculates the 95th percentile of the local anomaly score as τ k , which is uploaded to the aggregation node after encryption; the aggregation node calculates τ global by parameter encryption aggregation formula, and then sends it to each node after decryption; each node uses τ global for anomaly detection, and updates the local model and re-aggregates the global threshold regularly.

Images