Intelligent Collaboration-Based Digital Management System and Methods for Enterprise End-to-End Operations

Abstract

This invention relates to the field of enterprise operation management technology, specifically to a digital management and control system and method for the entire enterprise operation chain based on intelligent collaboration. The system's end-to-end intelligent collaborative management and control module, combined with a constructed operation data platform, utilizes a rule engine and machine learning algorithms to initiate customized operation process templates, achieving automated process flow and cross-departmental collaboration. It automatically pushes pending tasks and triggers alerts and intercepts abnormal nodes. This invention achieves data interoperability across multiple enterprise operation modules through multi-source data adaptive fusion, solving the problems of fragmented and cumbersome data integration in traditional operations, providing solid data support for end-to-end management and collaboration. Furthermore, through the end-to-end intelligent collaborative mechanism and customized templates, it automates the operation process, reduces manual intervention, and effectively shortens the process processing cycle. It also supports multi-industry and multi-scenario adaptation, meeting the personalized operational needs of enterprises of different sizes and types.

Description

Technical Field

[0001] This invention relates to the field of enterprise operation management technology, specifically to a digital control system and method for the entire enterprise operation chain based on intelligent collaboration. Background Technology

[0002] Currently, SMEs face significant challenges in their operational management, including low levels of digitalization and fragmented processes, hindering operational efficiency and compliance. Existing technological shortcomings primarily manifest in the fragmented storage of data across different operational modules (business, finance, and human resources) in various systems, resulting in severe data silos, cumbersome integration processes, and information asynchrony, preventing end-to-end data connectivity. Furthermore, operational processes rely heavily on manual intervention, leading to low cross-departmental collaboration efficiency, insufficient process standardization, and poor flexibility to adapt to diverse industries and scenarios. Simultaneously, SMEs exhibit weak compliance and risk control capabilities, lacking dynamic verification and early warning mechanisms. The identification of compliance risks related to finance, taxation, and internal control lags behind, easily triggering administrative penalties and operational losses. Moreover, existing operational management systems suffer from functional redundancy and high barriers to entry, failing to meet the lightweight, low-cost needs of SMEs and lacking sufficient remote collaboration capabilities. Therefore, the industry urgently requires a digital management system and methodology for the entire enterprise operational chain based on intelligent collaboration. Summary of the Invention

[0003] The purpose of this invention is to provide a digital management and control system and method for the entire enterprise operation based on intelligent collaboration, so as to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a digital management and control system for the entire enterprise operation chain based on intelligent collaboration, the system comprising: The multi-module operational data adaptive fusion module is used to access data from multiple internal and external systems and identify the data types; it performs cross-modal feature extraction on the data identification results and performs semantic alignment and fusion processing on the feature extraction results to build an operational data platform; The end-to-end intelligent collaborative management module, combined with the constructed operational data platform, uses a rule engine and machine learning algorithms to launch custom operational process templates, realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; The dynamic compliance risk control and early warning module has a built-in dynamically updated compliance rule library. Combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered early warnings. The big data analytics module constructs multi-dimensional analysis and feedback information based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, and displays and pushes this information through a visual dashboard. The multi-terminal collaboration and adaptation module builds a four-terminal collaboration platform for enterprises, management, employees, and services. It receives data pushed by the big data analysis module and dynamically filters and displays the pushed visual data in the corresponding visual dashboard of the platform according to the permissions of different port platforms. The encrypted storage module uses encryption algorithms to securely store operational data and operation logs, supporting end-to-end operation traceability.

[0005] Furthermore, the data from various internal and external systems includes internal ERP, CRM, financial, and human resources systems, as well as external industry policies and credit data. The types of data that can be identified include unstructured image data, semi-structured time-series data, and structured tabular data. In the process of cross-modal feature extraction of the data recognition results, the CNN algorithm is used to extract the spatial features of the corresponding image-type unstructured data by using the pixel matrix of the image as input, the LSTM algorithm is used to extract the temporal features of the corresponding temporal data by using the temporal data as input, and the Transformer algorithm is used to extract the association features of the tabular structured data.

[0006] This invention can connect with internal enterprise ERP, CRM, financial systems (accounting, taxation, reimbursement), human resources systems, and external data (industry policies, credit information), automatically identifying structured, unstructured, and semi-structured data, and completing semantic alignment and fusion; it builds a unified operational data platform, which supports dynamic data updates and flexible interface expansion, and has a high fusion accuracy, breaking down enterprise operational data silos; at the same time, it solves the problems of data fragmentation, cumbersome connection, and information asynchrony among modules in traditional enterprise operations, realizing the same source and interoperability of operational data, laying the foundation for full-link management and control.

[0007] The specific steps for extracting spatial features of corresponding image-type unstructured data in this invention are as follows: S111. Convert the image into a single-channel grayscale image and construct the pixel matrix of the corresponding image. Each element in the pixel matrix represents the quotient of the corresponding grayscale value in the image divided by 255. S112. Input the constructed image pixel matrix into the CNN algorithm. In the convolutional layer, capture the local spatial features of the corresponding image pixel matrix with a predefined convolutional kernel and stride, and perform dimensionality reduction processing in the pooling layer. S113. In the fully connected layer, the two-dimensional feature map after multi-layer convolution and pooling is flattened into a one-dimensional vector and mapped to a feature vector of a preset dimension to obtain the spatial features of the corresponding image.

[0008] The specific steps for extracting temporal features from temporal unstructured data according to this invention are as follows: S121. Arrange the time-series unstructured data in chronological order according to the timestamp to generate corresponding time-series data. In the time-series data, non-numerical fields are converted into One-Hot codes, and numerical fields are normalized to the interval [0, 1]. S122. The temporal dependencies between data in the corresponding time series data are captured by the LSTM algorithm. The feature vector output by the hidden layer of the LSTM algorithm corresponding to the last time step is used as the temporal feature of the corresponding time series data. The temporal feature is 256-dimensional, and each dimension corresponds to a temporal association feature.

[0009] The specific steps for extracting association features from tabular structured data according to this invention are as follows: S131. Map the numerical / category value of each field in the tabular structured data to an embedding vector of the first preset dimension; S132. Add the corresponding position information to each field in the embedding vector and concatenate them to obtain the input matrix of the Transformer algorithm; S133. An 8-head attention mechanism is adopted, and each attention head in the Transformer algorithm focuses on different field association combinations; S134. Execute the algorithm configuration of S133 on the input matrix obtained in S132 to output the quantized value of the association strength between different field association combinations corresponding to each attention head; and through the fully connected feedforward network in the Transformer algorithm, map the matrix corresponding to the quantized value of the association strength between different field association combinations output by the attention layer to a fixed-dimensional feature vector to obtain the association features of the corresponding tabular structured data. Furthermore, in the process of semantic alignment and fusion of the feature extraction results in S1, a pre-trained BERT-based semantic encoder maps the feature extraction results with different dimensions corresponding to the recognition results of different data types to the same semantic space for semantic alignment, transforming them into semantic vectors of equal dimensions. The cosine similarity of any two types of semantic vectors is calculated to quantify the semantic correlation between different feature extraction results. A unified fusion feature is generated through weighted fusion to obtain the enterprise operation data platform. The calculation formulas involved are as follows: Here, the number of data types identified is denoted as N, where i and j both belong to [1, N] and i ≠ j; h i h represents the semantic vector mapped in semantic space to the feature extraction result corresponding to the recognition result of the i-th data type; j This represents the semantic vector mapped in the semantic space to the feature extraction result corresponding to the j-th data type identification result; h j with h iSemantic similarity between; ||h i || represents h i The corresponding Euclidean norm; ||h j || represents h j The corresponding Euclidean norm; Where F represents the fusion feature matrix of the feature extraction results corresponding to the recognition results of each data type; w i represents the fusion weight of the feature extraction result corresponding to the identification result of the i-th data type; SU represents the sum of the semantic similarities between any two different semantic vectors in the semantic vector mapped in the semantic space for each data type identification result and the feature extraction result.

[0010] Furthermore, the end-to-end intelligent collaborative management and control module includes: The custom operation process template is a customized configuration of operation process nodes, permissions and corresponding flow rules based on the enterprise's industry and business needs. The system retrieves a fusion feature matrix from the operational data platform and extracts feature dimensions associated with process nodes from this matrix to form an operational status vector. These feature dimensions include approval sequence-related features, data consistency-related features, and process-related features. The rule engine transforms the operational process flow rules into machine-executable logic, initiating automated process flow. The resulting operational status vector is input into a pre-trained XGBoost anomaly prediction model for cross-dimensional process anomaly identification. Tasks are automatically pushed to the corresponding terminals, and alerts and interceptions are triggered for anomaly nodes, including approval timeouts and data errors. The XGBoost anomaly prediction model formula is as follows: Where X represents the operational status vector; the fused feature matrix retrieved from the operational data platform is denoted as FD. p This represents the approval sequence association feature dimension associated with process nodes in FD; FD t This represents the data consistency association feature dimension associated with process nodes in FD; FD e η represents the process association feature dimension associated with process nodes in FD; γ represents the preset learning rate; λ represents the preset split loss threshold; y represents the L2 regularization coefficient; pred This represents the anomaly prediction result corresponding to X, and y pred The value of y is 0 or 1. pred If the value of y is 0, then the anomaly prediction result corresponding to X is determined to be that there is no process anomaly; if y predIf the value of X is 1, then the abnormal prediction result corresponding to X is determined to be that there is an abnormal situation in the process.

[0011] This module of the present invention can automatically allocate and remind cross-departmental and cross-position tasks based on a rule engine and intelligent push, thereby improving collaboration efficiency; and it can also support enterprises to quickly configure processes according to industry characteristics, adapting to the operational needs of multiple industries and multiple scenarios.

[0012] Furthermore, the compliance rule base in the dynamic compliance risk control and early warning module includes various financial and tax regulations, industry regulatory requirements, and enterprise internal control standards; the compliance rule base is updated once every preset time interval. The dynamic compliance risk control and early warning module monitors multi-dimensional risks in real time and triggers tiered early warnings. During this process, it retrieves the integrated feature matrix from the operation data platform, extracts the cash flow feature dimension, compliance feature dimension, and operational efficiency feature dimension from the integrated feature matrix, maps them to different risk scoring indicators, and constructs a multi-dimensional risk weighted assessment model based on the mapped risk scoring indicators. Based on the output results of the multi-dimensional risk weighted assessment model, it triggers different levels of risk warnings, simultaneously locks the relevant warning-corresponding processes, and generates a compliance report. The risk warning levels include general, relatively severe, and serious. The formula for the multi-dimensional risk-weighted assessment model is as follows: Among them, R total This represents the comprehensive risk assessment value corresponding to the fused feature matrix; g k Let represent the weight coefficient corresponding to the k-th risk scoring indicator, and ; indicates; R k ∈{R1, R2, R3}, R k Let R1 represent the risk scoring index corresponding to the k-th risk dimension feature submatrix in {R1, R2, R3}; R1 represents the risk scoring index corresponding to the cash flow feature dimension feature submatrix in the fused feature matrix; R2 represents the risk scoring index corresponding to the compliance feature dimension feature submatrix in the fused feature matrix; and R3 represents the risk scoring index corresponding to the operational efficiency feature dimension feature submatrix in the fused feature matrix. FE k ∈{FE1, FE2, FE3}, where FE1 represents the feature submatrix of the capital flow feature dimension in the fusion feature matrix; FE2 represents the feature submatrix of the compliance feature dimension in the fusion feature matrix; and FE3 represents the feature submatrix of the operational efficiency feature dimension in the fusion feature matrix. represents the Sigmoid function used for normalizing risk scores; MLP() represents a multilayer perceptron used to map the corresponding feature submatrix to a risk score index of 0 to 100. If R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the first threshold, then a general-level risk warning is triggered; if R... total Greater than the first threshold and R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the second threshold, then the output of the model is deemed to trigger a more severe risk warning; if R... total If the result exceeds the second threshold, the output of the multi-dimensional risk weighted assessment model is determined to trigger a severe risk warning.

[0013] Furthermore, after a severe risk warning is triggered, the dynamic compliance risk control and early warning module automatically locks the fund payment function of the relevant account. After the risk is handled and verified, the administrator can manually unlock the fund payment function of the relevant account.

[0014] Furthermore, the cloud backup of the encrypted storage module adopts an off-site disaster recovery deployment strategy, and the full-link operation traceability function supports accurate querying by operation type or time range.

[0015] Furthermore, the encrypted storage module uses the AES-256 encryption algorithm to encrypt and store data.

[0016] A digital management and control method for the entire enterprise operation chain based on intelligent collaboration, the method including: S1. Access data from multiple internal and external systems and identify data types; extract cross-modal features from the data identification results, and perform semantic alignment and fusion processing on the feature extraction results to build an operational data platform; S2. Combined with the constructed operational data platform, based on the rule engine and machine learning algorithms, launch custom operational process templates to realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; S3. Through a built-in dynamically updated compliance rule library, combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered warnings. S4. Based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, construct multi-dimensional analysis and feedback information, and display and push it through a visual dashboard; S5 establishes a collaborative platform for four terminals: enterprise, management, employee, and service. Each terminal receives the display and push data obtained from S4 and dynamically filters and displays the pushed visual data in the corresponding visual dashboard based on the permissions of different terminal platforms.

[0017] Furthermore, the AES-256 encryption algorithm is used to securely store operational data and operation logs.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are: (1) This invention achieves data sharing between multiple modules of enterprise operation through multi-source adaptive fusion, effectively improves the accuracy of data fusion, solves the problems of fragmented and cumbersome data connection in traditional operation, provides solid data support for full-link management and collaboration, and significantly improves cross-departmental collaboration efficiency; (2) This invention achieves automated operation process flow through a full-process intelligent collaboration mechanism and custom templates, reduces manual intervention, and effectively shortens the process processing cycle; at the same time, it supports multi-industry and multi-scenario adaptation, which can meet the personalized operation needs of enterprises of different sizes and types, and significantly improves flexibility and applicability. (3) This invention can realize a closed loop of full-process compliance risk control, realize dynamic risk monitoring and graded early warning, effectively avoid risks such as financial and tax penalties and operating losses, improve the enterprise's compliance level and stable operation capabilities, and meet the compliance management needs of small and micro enterprises. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the enterprise end-to-end digital management and control method based on intelligent collaboration, as described in this invention. Figure 2 This is a schematic diagram of the structure of the enterprise full-chain digital management and control system based on intelligent collaboration, which is the subject of this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-2 The present invention provides a technical solution: As shown in the figure, this embodiment provides a method for digital management and control of the entire enterprise operation based on intelligent collaboration, the method including: S1. Access data from multiple internal and external systems and identify data types; extract cross-modal features from the data identification results, and perform semantic alignment and fusion processing on the feature extraction results to build an operational data platform; S2. Combined with the constructed operational data platform, based on the rule engine and machine learning algorithms, launch custom operational process templates to realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; S3. Through a built-in dynamically updated compliance rule library, combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered warnings. S4. Based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, construct multi-dimensional analysis and feedback information, and display and push it through a visual dashboard; S5 establishes a collaborative platform for four terminals: enterprise, management, employee, and service. Each terminal receives the display and push data obtained from S4 and dynamically filters and displays the pushed visual data in the corresponding visual dashboard based on the permissions of different terminal platforms.

[0022] In the process of implementing the enterprise full-chain digital management and control method based on intelligent collaboration, the AES-256 encryption algorithm is used to securely store operational data and operation logs.

[0023] This embodiment adopts a "cloud deployment + local backup" mode, configuring 4 high-performance servers (2 application servers, 1 database server, and 1 backup server), running the CentOS operating system, using the Spark big data framework, Transformer algorithm engine, and MySQL database, and integrating multi-terminal adaptation components; it connects to the enterprise's existing ERP system, financial software, and human resources management system, synchronizes external data such as industry policies and credit information, and builds a four-terminal collaborative platform to achieve flexible access from multiple terminals.

[0024] like Figure 2 As shown, this embodiment provides a digital management and control system for the entire enterprise operation chain based on intelligent collaboration. The system includes: The multi-module operational data adaptive fusion module is used to access data from multiple internal and external systems and identify the data types; it performs cross-modal feature extraction on the data identification results and performs semantic alignment and fusion processing on the feature extraction results to build an operational data platform; The data from various internal and external systems include internal ERP, CRM, finance, and human resources systems, as well as external industry policies and credit data. The types of data that can be identified include unstructured image data, semi-structured time-series data, and structured tabular data. In the process of cross-modal feature extraction of the data recognition results, the CNN algorithm is used to extract the spatial features of the corresponding image-type unstructured data by using the pixel matrix of the image as input, the LSTM algorithm is used to extract the temporal features of the corresponding temporal data by using the temporal data as input, and the Transformer algorithm is used to extract the association features of the tabular structured data.

[0025] The cross-modal coding algorithm described in this embodiment combines the advantages of CNN and LSTM networks. In the CNN algorithm, the convolutional kernel size is 3×3 and the stride is 1; in the LSTM algorithm, the hidden layer dimension is 256 and the number of layers is 2; and the feature dimension corresponding to the output result of the Transformer algorithm is 512 and the number of multi-head attention heads is 8. The specific steps for extracting spatial features of corresponding image-type unstructured data in this invention are as follows: S111. Convert the image into a single-channel grayscale image and construct the pixel matrix of the corresponding image. Each element in the pixel matrix represents the quotient of the corresponding grayscale value in the image divided by 255. S112. Input the constructed image pixel matrix into the CNN algorithm. In the convolutional layer, capture the local spatial features of the corresponding image pixel matrix with a predefined convolutional kernel and stride, and perform dimensionality reduction processing in the pooling layer. S113. In the fully connected layer, the two-dimensional feature map after multi-layer convolution and pooling is flattened into a one-dimensional vector and mapped to a feature vector of a preset dimension to obtain the spatial features of the corresponding image. In this example, a document image is used as an example. The spatial features of the corresponding image have a feature dimension of 256, and each dimension corresponds to a spatial feature. Among them, dimensions 1 to 50 correspond to the position and shape features of the document border, dimensions 51 to 150 correspond to the arrangement direction and spacing features of the text lines in the document, and dimensions 151 to 256 correspond to the texture features of the handwritten signature in the document.

[0026] The specific steps for extracting temporal features from temporal unstructured data according to this invention are as follows: S121. Arrange the time-series unstructured data in chronological order according to the timestamp to generate corresponding time-series data. In the time-series data, non-numerical fields are converted into One-Hot codes, and numerical fields are normalized to the interval [0, 1]. S122. The temporal dependencies between data in the corresponding time series data are captured by the LSTM algorithm. The feature vector output by the hidden layer of the LSTM algorithm corresponding to the last time step is used as the temporal feature of the corresponding time series data. The temporal feature is 256-dimensional, and each dimension corresponds to a temporal association feature. In this example, taking an order as an example, in the temporal feature of the corresponding order, dimensions 1 to 80 correspond to the trend feature of the change of order amount, dimensions 81 to 160 correspond to the flow speed feature of order status, and dimensions 161 to 256 correspond to the identification feature of an abnormal order determined by multiple status changes within a preset unit of time. The specific steps for extracting association features from tabular structured data according to this invention are as follows: S131. Map the numerical / category value of each field in the tabular structured data to an embedding vector of the first preset dimension; S132. Add the corresponding position information to each field in the embedding vector and concatenate them to obtain the input matrix of the Transformer algorithm; S133. An 8-head attention mechanism is adopted, and each attention head in the Transformer algorithm focuses on different field association combinations; S134. Execute the algorithm configuration of S133 on the input matrix obtained in S132 to output the quantized value of the association strength between different field association combinations corresponding to each attention head; and through the fully connected feedforward network in the Transformer algorithm, map the matrix corresponding to the quantized value of the association strength between different field association combinations output by the attention layer to a fixed-dimensional feature vector to obtain the association features of the corresponding tabular structured data. This embodiment takes a monthly sales statistics table as an example. During the process of capturing field associations in the multi-head attention layer, such as: Head 1 focuses on the association between order amount and customer level; Head 2 focuses on the association between order amount and payment method; Head 3 focuses on the association between inventory quantity and sales price; Head 4 focuses on the association between order amount and payment cycle, etc.

[0027] In the process of semantic alignment and fusion of feature extraction results, the multi-module operational data adaptive fusion module uses a pre-trained BERT-based semantic encoder to map feature extraction results with different dimensions corresponding to different data types to the same semantic space for semantic alignment, transforming them into semantic vectors of equal dimensions. The cosine similarity of any two semantic vectors is calculated to quantify the semantic correlation between different feature extraction results. A unified fusion feature is generated through weighted fusion to obtain the enterprise operational data platform. The calculation formulas involved are as follows: Here, the number of data types identified is denoted as N, where i and j both belong to [1, N] and i ≠ j; h ih represents the semantic vector mapped in semantic space to the feature extraction result corresponding to the recognition result of the i-th data type; j This represents the semantic vector mapped in the semantic space to the feature extraction result corresponding to the j-th data type identification result; h j with h i Semantic similarity between; ||h i || represents h i The corresponding Euclidean norm; ||h j || represents h j The corresponding Euclidean norm; Where F represents the fusion feature matrix of the feature extraction results corresponding to the recognition results of each data type; w i represents the fusion weight of the feature extraction result corresponding to the identification result of the i-th data type; SU represents the sum of the semantic similarities between any two different semantic vectors in the semantic vector mapped in the semantic space for each data type identification result and the feature extraction result.

[0028] The end-to-end intelligent collaborative management module, combined with the constructed operational data platform, uses a rule engine and machine learning algorithms to launch custom operational process templates, realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; In this embodiment, when building a custom operation process template, the needs of commercial and trade enterprises are combined to customize three core process templates: procurement approval, sales order processing, and expense reimbursement. Approval nodes and permissions are configured. The system starts automated workflow. After the procurement order is submitted, it is automatically pushed to the procurement manager and finance for approval. The sales order is synchronized to the inventory management module. The expense reimbursement automatically verifies the invoice information. The process processing cycle is shortened from an average of 3 days to 1 day, and manual intervention is reduced by 70%.

[0029] The end-to-end intelligent collaborative management and control module includes: The custom operation process template is a customized configuration of operation process nodes, permissions and corresponding flow rules based on the enterprise's industry and business needs. In this embodiment, the custom operation process template is adapted to the needs of multiple industries such as manufacturing, trade and circulation and service. Manufacturing enterprises focus on the configuration of production material requisition and finished product warehousing processes, while trade and circulation enterprises focus on the configuration of purchase, sales, inventory and tax compliance processes. The system retrieves a fusion feature matrix from the operational data platform and extracts feature dimensions associated with process nodes from this matrix to form an operational status vector. These feature dimensions include approval sequence-related features, data consistency-related features, and process-related features. The rule engine transforms the operational process flow rules into machine-executable logic, initiating automated process flow. The resulting operational status vector is input into a pre-trained XGBoost anomaly prediction model for cross-dimensional process anomaly identification. Tasks are automatically pushed to the corresponding terminals, and alerts and interceptions are triggered for anomaly nodes, including approval timeouts and data errors. The XGBoost anomaly prediction model formula is as follows: Where X represents the operational status vector; the fused feature matrix retrieved from the operational data platform is denoted as FD. p This represents the approval sequence association feature dimension associated with process nodes in FD; FD t This represents the data consistency association feature dimension associated with process nodes in FD; FD e η represents the process association feature dimension associated with process nodes in FD; γ represents the preset learning rate; λ represents the preset split loss threshold; y represents the L2 regularization coefficient; pred This represents the anomaly prediction result corresponding to X, and y pred The value of y is 0 or 1. pred If the value of y is 0, then the anomaly prediction result corresponding to X is determined to be that there is no process anomaly; if y pred If the value of X is 1, then the abnormal prediction result corresponding to X is determined to be that there is an abnormal situation in the process.

[0030] The dynamic compliance risk control and early warning module has a built-in dynamically updated compliance rule library. Combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered early warnings. The dynamic compliance risk control and early warning module includes a compliance rule base that comprises various financial and tax regulations, industry regulatory requirements, and corporate internal control standards; the compliance rule base is updated every preset time interval. The dynamic compliance risk control and early warning module monitors multi-dimensional risks in real time and triggers tiered early warnings. During this process, it retrieves the integrated feature matrix from the operation data platform, extracts the cash flow feature dimension, compliance feature dimension, and operational efficiency feature dimension from the integrated feature matrix, maps them to different risk scoring indicators, and constructs a multi-dimensional risk weighted assessment model based on the mapped risk scoring indicators. Based on the output results of the multi-dimensional risk weighted assessment model, it triggers different levels of risk warnings, simultaneously locks the relevant warning-corresponding processes, and generates a compliance report. The risk warning levels include general, relatively severe, and serious. The formula for the multi-dimensional risk-weighted assessment model is as follows: Among them, R total This represents the comprehensive risk assessment value corresponding to the fused feature matrix; g k Let represent the weight coefficient corresponding to the k-th risk scoring indicator, and ; indicates; R k ∈{R1, R2, R3}, R k Let R1 represent the risk scoring index corresponding to the k-th risk dimension feature submatrix in {R1, R2, R3}; R1 represents the risk scoring index corresponding to the cash flow feature dimension feature submatrix in the fused feature matrix; R2 represents the risk scoring index corresponding to the compliance feature dimension feature submatrix in the fused feature matrix; and R3 represents the risk scoring index corresponding to the operational efficiency feature dimension feature submatrix in the fused feature matrix. FE k ∈{FE1, FE2, FE3}, where FE1 represents the feature submatrix of the capital flow feature dimension in the fusion feature matrix; FE2 represents the feature submatrix of the compliance feature dimension in the fusion feature matrix; and FE3 represents the feature submatrix of the operational efficiency feature dimension in the fusion feature matrix. represents the Sigmoid function used for normalizing risk scores; MLP() represents a multilayer perceptron used to map the corresponding feature submatrix to a risk score index of 0 to 100. If R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the first threshold, then a general-level risk warning is triggered; if R... total Greater than the first threshold and R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the second threshold, then the output of the model is deemed to trigger a more severe risk warning; if R... total If the result exceeds the second threshold, the output of the multi-dimensional risk weighted assessment model is determined to trigger a severe risk warning.

[0031] Once a severe risk warning is triggered, the dynamic compliance risk control and early warning module automatically locks the fund payment function of the relevant account. After the risk is handled and verified, the administrator can manually unlock the fund payment function of the relevant account.

[0032] This embodiment relies on a compliance rule base to automatically verify the authenticity of purchase invoices, the reasonableness of reimbursement amounts, and the accuracy of tax declarations. It identifies two reimbursement expenses exceeding the standard and one invoice information anomaly, marks them as serious risks, and pushes them to the finance manager. The finance manager urges rectification, and after rectification, the system re-verifies and passes, generates a monthly compliance report for archiving, and no compliance risk events occur.

[0033] The big data analytics module constructs multi-dimensional analysis and feedback information based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, and displays and pushes this information through a visual dashboard. The multi-terminal collaboration and adaptation module builds a four-terminal collaboration platform for enterprises, management, employees, and services. It receives data pushed by the big data analysis module and dynamically filters and displays the pushed visual data in the corresponding visual dashboard of the platform according to the permissions of different port platforms. The encrypted storage module uses encryption algorithms to securely store operational data and operation logs, supporting end-to-end operation traceability.

[0034] The cloud backup of the encrypted storage module adopts an off-site disaster recovery deployment strategy, and the full-link operation traceability function supports accurate querying by operation type or time range.

[0035] The encrypted storage module uses the AES-256 encryption algorithm to encrypt and store data.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0037] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital management and control system for the entire enterprise operation chain based on intelligent collaboration, characterized in that: The system includes: The multi-module operational data adaptive fusion module is used to access data from multiple internal and external systems and identify the data types; it performs cross-modal feature extraction on the data identification results and performs semantic alignment and fusion processing on the feature extraction results to build an operational data platform; The end-to-end intelligent collaborative management module, combined with the constructed operational data platform, uses a rule engine and machine learning algorithms to launch custom operational process templates, realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; The dynamic compliance risk control and early warning module has a built-in dynamically updated compliance rule library. Combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered early warnings. The big data analytics module constructs multi-dimensional analysis and feedback information based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, and displays and pushes this information through a visual dashboard. The multi-terminal collaboration and adaptation module builds a four-terminal collaboration platform for enterprises, management, employees, and services. It receives data pushed by the big data analysis module and dynamically filters and displays the pushed visual data in the corresponding visual dashboard of the platform according to the permissions of different port platforms. The encrypted storage module uses encryption algorithms to securely store operational data and operation logs, supporting end-to-end operation traceability.

2. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 1, characterized in that: The data from various internal and external systems include internal ERP, CRM, finance, and human resources systems, as well as external industry policies and credit data. The types of data that can be identified include unstructured image data, semi-structured time-series data, and structured tabular data. In the process of cross-modal feature extraction of the data recognition results, the CNN algorithm is used to extract the spatial features of the corresponding image-type unstructured data by using the pixel matrix of the image as input, the LSTM algorithm is used to extract the temporal features of the corresponding temporal data by using the temporal data as input, and the Transformer algorithm is used to extract the association features of the tabular structured data.

3. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 1, characterized in that: In the process of semantic alignment and fusion of feature extraction results, the multi-module operational data adaptive fusion module uses a pre-trained BERT-based semantic encoder to map feature extraction results with different dimensions corresponding to different data types to the same semantic space for semantic alignment, transforming them into semantic vectors of equal dimensions. The cosine similarity of any two semantic vectors is calculated to quantify the semantic correlation between different feature extraction results. A unified fusion feature is generated through weighted fusion to obtain the enterprise operational data platform. The calculation formulas involved are as follows: Here, the number of data types identified is denoted as N, where i and j both belong to [1, N] and i ≠ j; h i h represents the semantic vector mapped in semantic space to the feature extraction result corresponding to the recognition result of the i-th data type; j This represents the semantic vector mapped in the semantic space to the feature extraction result corresponding to the j-th data type identification result; h j with h i Semantic similarity between; ||h i || represents h i The corresponding Euclidean norm; ||h j || represents h j The corresponding Euclidean norm; Where F represents the fusion feature matrix of the feature extraction results corresponding to the recognition results of each data type; w i represents the fusion weight of the feature extraction result corresponding to the identification result of the i-th data type; SU represents the sum of the semantic similarities between any two different semantic vectors in the semantic vector mapped in the semantic space for each data type identification result and the feature extraction result.

4. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 2, characterized in that: The end-to-end intelligent collaborative management and control module includes: The custom operation process template is a customized configuration of operation process nodes, permissions and corresponding flow rules based on the enterprise's industry and business needs. The system retrieves a fusion feature matrix from the operational data platform and extracts feature dimensions associated with process nodes from this matrix to form an operational status vector. These feature dimensions include approval sequence-related features, data consistency-related features, and process-related features. The rule engine transforms the operational process flow rules into machine-executable logic, initiating automated process flow. The resulting operational status vector is input into a pre-trained XGBoost anomaly prediction model for cross-dimensional process anomaly identification. Tasks are automatically pushed to the corresponding terminals, and alerts and interceptions are triggered for anomaly nodes, including approval timeouts and data errors. The XGBoost anomaly prediction model formula is as follows: Where X represents the operational status vector; the fused feature matrix retrieved from the operational data platform is denoted as FD. p This represents the approval sequence association feature dimension associated with process nodes in FD; FD t This represents the data consistency association feature dimension associated with process nodes in FD; FD e η represents the process association feature dimension associated with process nodes in FD; γ represents the preset learning rate; λ represents the preset split loss threshold; y represents the L2 regularization coefficient; pred This represents the anomaly prediction result corresponding to X, and y pred The value of y is 0 or 1. pred If the value of y is 0, then the anomaly prediction result corresponding to X is determined to be that there is no process anomaly; if y pred If the value of X is 1, then the abnormal prediction result corresponding to X is determined to be that there is an abnormal situation in the process.

5. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 1, characterized in that: The dynamic compliance risk control and early warning module includes a compliance rule base that comprises various financial and tax regulations, industry regulatory requirements, and corporate internal control standards; the compliance rule base is updated every preset time interval. The dynamic compliance risk control and early warning module monitors multi-dimensional risks in real time and triggers tiered early warnings. During this process, it retrieves the integrated feature matrix from the operation data platform, extracts the cash flow feature dimension, compliance feature dimension, and operational efficiency feature dimension from the integrated feature matrix, maps them to different risk scoring indicators, and constructs a multi-dimensional risk weighted assessment model based on the mapped risk scoring indicators. Based on the output results of the multi-dimensional risk weighted assessment model, it triggers different levels of risk warnings, simultaneously locks the relevant warning-corresponding processes, and generates a compliance report. The risk warning levels include general, relatively severe, and serious. The formula for the multi-dimensional risk-weighted assessment model is as follows: Among them, R total This represents the comprehensive risk assessment value corresponding to the fused feature matrix; g k Let represent the weight coefficient corresponding to the k-th risk scoring indicator, and ; indicates; R k ∈{R1, R2, R3}, R k Let R1 represent the risk scoring index corresponding to the k-th risk dimension feature submatrix in {R1, R2, R3}; R1 represents the risk scoring index corresponding to the cash flow feature dimension feature submatrix in the fused feature matrix; R2 represents the risk scoring index corresponding to the compliance feature dimension feature submatrix in the fused feature matrix; and R3 represents the risk scoring index corresponding to the operational efficiency feature dimension feature submatrix in the fused feature matrix. FE k ∈{FE1, FE2, FE3}, where FE1 represents the feature submatrix of the capital flow feature dimension in the fusion feature matrix; FE2 represents the feature submatrix of the compliance feature dimension in the fusion feature matrix; and FE3 represents the feature submatrix of the operational efficiency feature dimension in the fusion feature matrix. represents the Sigmoid function used for normalizing risk scores; MLP() represents a multilayer perceptron used to map the corresponding feature submatrix to a risk score index of 0 to 100. If R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the first threshold, then a general-level risk warning is triggered; if R... total Greater than the first threshold and R total If the output of the multi-dimensional risk weighted assessment model is less than or equal to the second threshold, then the output of the model is deemed to trigger a more severe risk warning; if R... total If the result exceeds the second threshold, the output of the multi-dimensional risk weighted assessment model is determined to trigger a severe risk warning.

6. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 5, characterized in that: Once a severe risk warning is triggered, the dynamic compliance risk control and early warning module automatically locks the fund payment function of the relevant account. After the risk is handled and verified, the administrator can manually unlock the fund payment function of the relevant account.

7. The enterprise end-to-end digital management and control system based on intelligent collaboration as described in claim 1, characterized in that: The cloud backup of the encrypted storage module adopts an off-site disaster recovery deployment strategy, and the full-link operation traceability function supports accurate querying by operation type or time range.

8. The enterprise end-to-end digital management and control system based on intelligent collaboration according to claim 1, characterized in that: The encrypted storage module uses the AES-256 encryption algorithm to encrypt and store data.

9. A method for digital management and control of enterprise end-to-end operations based on intelligent collaboration, applied to the enterprise end-to-end operations digital management and control system based on intelligent collaboration as described in any one of claims 1 to 8, characterized in that, The method includes: S1. Access data from multiple internal and external systems and identify data types; extract cross-modal features from the data identification results, and perform semantic alignment and fusion processing on the feature extraction results to build an operational data platform; S2. Combined with the constructed operational data platform, based on the rule engine and machine learning algorithms, launch custom operational process templates to realize automated process flow and cross-departmental collaboration, automatically push to-do tasks, and trigger reminders and interception for abnormal nodes; S3. Through a built-in dynamically updated compliance rule library, combined with the constructed operation data platform, it automatically verifies the compliance of operation processes, monitors multi-dimensional risks in real time, and triggers tiered warnings. S4. Based on the operational data platform, intercepted abnormal nodes, and multi-dimensional risk monitoring results, construct multi-dimensional analysis and feedback information, and display and push it through a visual dashboard; S5 establishes a collaborative platform for four terminals: enterprise, management, employee, and service. Each terminal receives the display and push data obtained from S4 and dynamically filters and displays the pushed visual data in the corresponding visual dashboard based on the permissions of different terminal platforms.

10. The enterprise end-to-end digital management and control method based on intelligent collaboration according to claim 9, characterized in that: The AES-256 encryption algorithm is used to securely store operational data and operation logs.

Images