Enterprise intelligent decision management system based on big data

Abstract

The invention particularly relates to an enterprise intelligent decision management system based on big data, which relates to the technical field of big data processing, and comprises the following steps: initiating and executing a secure multi-party computing process to realize secure model aggregation and malicious update detection; and recording the global model weight version of each round of training, and distributing the updated global model weight version to all enterprise clients. According to the invention, multiple modules cooperatively construct a four-level privacy protection system, so that the risk of commercial secret leakage is effectively avoided and the data compliance requirement is met; the local model is updated through homomorphic encryption or secret sharing encryption, and whole-course ciphertext transmission and aggregation are ensured; the privacy protection decision-making module deploys the global model locally, and sensitive data is prevented from being transmitted outwards during decision making; according to the system, commercial confidentials such as purchase cost and core sales data are prevented from being reversely deduced through model updating, the requirements of laws and regulations such as GDPR are completely met, meanwhile, the data training value is guaranteed, and the privacy concern of enterprises participating in cross-collaboration is eliminated.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to an enterprise intelligent decision management system based on big data. Background Technology

[0002] In cross-enterprise supply chain collaborative decision-making, the data of each participant (supplier, manufacturer, distributor) is scattered and stored in their own ERP, SCM, CRM and other systems, forming data silos. This results in insufficient accuracy of decision-making models for demand forecasting and procurement optimization by individual enterprises due to the limited data dimensions. Directly sharing raw data to build a global model faces the risk of leakage of trade secrets and makes it difficult to meet the requirements of data privacy regulations such as GDPR for the protection of sensitive information.

[0003] In existing technologies, some cross-enterprise data collaboration solutions adopt a centralized data storage architecture, which requires data from each enterprise to be transmitted to a unified platform, posing a significant risk of privacy leakage. Although a few solutions based on federated learning achieve data non-movement, they suffer from problems such as inconsistent training benchmarks (e.g., inconsistent model architecture and parameters), weak resistance to malicious attacks (inability to identify model poisoning), and poor system stability (training interruption due to node offline), making it difficult to support the long-term and reliable intelligent decision-making needs in supply chain scenarios.

[0004] To address these issues, this invention proposes an enterprise intelligent decision management system based on big data, which addresses the aforementioned problems through multi-module collaborative design. Summary of the Invention

[0005] The purpose of this invention is to propose an enterprise intelligent decision-making management system based on big data in order to solve the above-mentioned problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The enterprise intelligent decision management system based on big data includes a secure data acquisition and collaborative access module, a distributed data storage and computing engine module, a privacy-preserving decision management and application module, and a federated intelligent analysis and security modeling module, configured as follows: Broadcast the task configuration to all certified enterprise clients; Initiate and execute secure multi-party computation processes to achieve secure model aggregation and malicious update detection; Record the global model weights for each training round. Version, and will be updated Distribute to all enterprise clients; The enterprise client's local modeling agent receives and parses the task configuration from the central server, and then processes the locally trained data. After encryption / splitting, press Threshold rules are sent to other enterprises and the central server, and participate in the SMPC collaborative decryption process, providing local shards for global update decryption.

[0007] Preferably, the secure data acquisition and collaborative access module is configured for local compliant acquisition and secure access preparation of cross-enterprise data, providing a standardized and privacy-enhanced data source for federated learning.

[0008] Preferably, the distributed data storage and computing engine module is configured to deploy a distributed storage architecture locally within the enterprise to store preprocessed standardized data, local training logs, model update backups, and encrypted components for updating the model after building a local model training engine.

[0009] Preferably, the process of training the local model and updating the encryption component is as follows: Training process: Download the latest global model weights distributed by the central server. ; Use local dataset conduct =5 rounds of local training, with local model updates calculated using the gradient descent algorithm. ; Calculate the model update quality on the local validation set as a weight for subsequent contribution. The basis for the calculation; Updating the encryption component: Integrate homomorphic encryption libraries or secret sharing tools to update the local model. Perform encryption / splitting operations: If homomorphic encryption is used, an encrypted model update will be generated. ; If using secret sharing, press... Threshold rules will Split into Slices .

[0010] Preferably, the step of broadcasting the task configuration to all authenticated enterprise clients is as follows: Distribute a unified model architecture; Distribute initial global model weights ; Set training parameters; Inform them of the safe aggregation rules.

[0011] Preferably, the model aggregation and malicious update detection specifically include: Receive model update shards sent by all enterprise clients; Weights are calculated based on the quality of model updates uploaded by each company. High-quality data corresponds to higher ; Calculate all in encrypted state Cosine similarity is used to identify malicious model updates; After removing the top 15% and bottom 15% of abnormal updates, calculate a weighted average for the remaining updates. ; A global update that gathers ≥t enterprise clients to collaboratively decrypt and encrypt data. Obtain plaintext Update the global model .

[0012] Preferably, the privacy-protected decision management and application module is configured to receive the final global model distributed by the central server when the federated training reaches convergence. It is deployed on the enterprise's local decision engine to generate decisions for supply chain scenarios, and collects actual data after the decisions are implemented as incremental training data for labeling.

[0013] Preferably, the decision generation for the supply chain scenario specifically includes: Demand Forecast: Input local historical sales and regional promotional plans, and output the product demand forecast for a future preset time period; Procurement optimization: Combine demand forecasts with raw material inventory data to generate a procurement plan with the lowest cost; Logistics scheduling: Optimize warehouse allocation and transportation routes based on the regional distribution of demand.

[0014] Preferably, the method further includes: The collaborative management and feedback optimization module is configured to ensure the stability, reliability, and continuous evolution of the entire distributed federation system.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention constructs a four-level privacy protection system through multi-module collaboration, effectively avoiding the risk of leakage of trade secrets and meeting data compliance requirements; local model updates are encrypted with homomorphic encryption or secret sharing encryption to ensure that the transmission and aggregation are encrypted throughout; the privacy protection decision module deploys the global model locally to avoid the transmission of sensitive data during decision-making; this system not only prevents the inference of trade secrets such as procurement costs and core sales data through model updates, but also fully complies with regulations such as GDPR, while ensuring the value of data training and dispelling the privacy concerns of enterprises participating in cross-collaboration.

[0016] 2. This invention improves model accuracy and enhances practical application value by addressing the Non-IID data problem. The federated intelligent analysis module distributes a unified model architecture and pre-trained initial weights to avoid training benchmark confusion. It assigns higher weights to high-quality data through contribution-weighted calculation and uses the TrimmedMean algorithm to eliminate abnormal updates, resulting in a global model with significantly higher accuracy than a single enterprise's local model. This effectively solves the performance degradation caused by the non-independent and identically distributed nature of supply chain data. The decision-making module generates practical decisions for demand forecasting, procurement optimization, and logistics scheduling, reducing enterprise inventory costs and logistics losses. The system collaborative management module ensures stable system operation and allows the model to dynamically optimize with business needs through node monitoring, contribution incentives, and automatic iteration triggering, effectively improving the efficiency of supply chain collaborative decision-making and long-term adaptability. Attached Figure Description

[0017] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0018] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0019] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0020] Example 1

[0021] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0022] Appendix Figure 1 The block diagram of the enterprise intelligent decision-making management system based on big data provided in the embodiments of the present invention shows the connection relationship between the secure data acquisition and collaborative access module and the collaborative management and feedback optimization module, and marks the main functional interaction flow of each module.

[0023] In this embodiment, it includes: The secure data acquisition and collaborative access module is configured for local compliant acquisition and secure access of cross-enterprise data, providing standardized, privacy-enhanced data sources for federated learning, specifically including: Extract heterogeneous data from enterprise ERP (procurement data), SCM (logistics data), CRM (sales data) and other systems, and perform format unification and feature alignment (such as unifying the input features of the demand forecasting model: timestamp, product SKU, and region code). It has a built-in secure communication component that establishes an encrypted channel with the central coordination server and completes identity verification through an enterprise-specific digital certificate. Only after successful verification can it access the federated training cluster; this prevents man-in-the-middle attacks and the addition of fake nodes (such as clients forged by competitors), ensuring the legitimacy of the federated training cluster. It also includes: injecting calibrated Gaussian noise into the local standardized data according to the privacy protection requirements issued by the central server, adding the first layer of privacy barrier to the original data.

[0024] The core functions of the secure data acquisition and collaborative access module are clearly defined, providing a standardized and highly secure data source foundation for cross-enterprise federated learning. After extracting heterogeneous data from enterprise ERP, SCM, CRM, and other systems, it performs format unification and feature alignment (such as unifying the input features of the demand forecasting model), effectively solving the federated training adaptation problem caused by differences in data formats among different enterprises. This ensures that the data from all participants can meet the training requirements of a unified model, avoiding training interruptions or accuracy deviations caused by inconsistencies in data dimensions and types. Meanwhile, the built-in secure communication components construct an encrypted channel and complete identity verification through enterprise-specific digital certificates. This can accurately intercept man-in-the-middle attacks and fake nodes forged by competitors, ensuring the legitimacy of the federated training cluster and the security of data transmission from the source of access, thus building a solid first line of defense for subsequent cross-enterprise data collaboration.

[0025] The design of injecting calibrated Gaussian noise into locally standardized data further enhances data privacy protection capabilities, adding a crucial first layer of privacy barrier to the original data. This differential privacy preprocessing operation effectively prevents the risk of inferring the company's original business data (such as procurement costs and sales data of core customers) through subsequent model updates. It meets the requirements of data privacy regulations such as GDPR for the protection of sensitive information without sacrificing the training value of the data. The injected noise is calibrated to achieve a balance between privacy protection and data availability, ensuring that the preprocessed data can still effectively support local model training.

[0026] The distributed data storage and computing engine module is configured to deploy a distributed storage architecture on the enterprise's local machine to store pre-processed standardized data, local training logs, and model update backups. Data is not transferred across enterprises throughout the process. Only read-only permissions are granted to the local training module, which is an encrypted component for updating the model after building the local model training engine.

[0027] Deploying a distributed storage architecture on-premises within an enterprise limits the pre-processed standardized data, local training logs, and model update backups entirely to internal enterprise storage. This physically eliminates the risk of data transmission across enterprises and effectively avoids the potential leakage of original business data (such as procurement costs, production schedules, and core inventory data) during transmission. It perfectly meets the privacy protection requirements of data preservation and value sharing in cross-enterprise collaboration.

[0028] Meanwhile, the design of granting read-only access only to the local training module further restricts the scope of data use, prevents data from being illegally tampered with or misused, and provides a safe, reliable and complete data source for subsequent local model training, ensuring the authenticity and availability of training data.

[0029] The design of an encrypted component for updating the model after building a local model training engine lays a crucial foundation for the local training and secure update process of federated learning.

[0030] The local model training engine can directly execute training tasks based on high-quality data stored locally, without relying on external data transmission, which greatly reduces training latency and improves the efficiency of generating local model updates. The configuration of the model update encryption component provides functional support for subsequent model update encryption operations in advance, such as homomorphic encryption or secret sharing, ensuring that locally generated model updates have secure encryption capabilities from the source, connecting with subsequent upload and aggregation processes, and strengthening the end-to-end security of the entire federated learning system.

[0031] The process of training the local model and updating the encrypted components is as follows: Training process: Download the latest global model weights distributed by the central server. ; Use local dataset conduct =5 rounds of local training, calculating local model updates using gradient descent algorithms (such as the Adam optimizer). ; Calculate the model update quality on the local validation set as a weight for subsequent contribution. The basis for the calculation; Updating the encryption component: Integrate homomorphic encryption libraries or secret sharing tools to update the local model. Perform encryption / splitting operations: If homomorphic encryption is used, an encrypted model update will be generated. ; If using secret sharing, press... Threshold rules (such as) =8 participating companies =5) Split into Slices .

[0032] By clarifying the complete process of local model training, a standardized execution basis is provided for federated learning, effectively ensuring the consistency between local training and global model iteration.

[0033] It stipulates that enterprise clients must first download the latest global model weights distributed by the central server. This ensures that the training starting point of each participant is synchronized with the global objective; then, a local dataset is used. Perform 5 epochs of training and compute local model updates using gradient descent algorithm. This approach fully leverages the value of local enterprise data while avoiding the privacy risks associated with the external transmission of raw data; simultaneously, it calculates the quality of model updates on a local validation set to serve as contribution weights. Based on this, the subsequent aggregation process can accurately identify high-quality training results, laying the foundation for improving the accuracy of the global model and solving the problem of model performance degradation caused by simple averaging in Non-IID data scenarios.

[0034] The detailed process of encrypting the model update component establishes a core security barrier for the model update transmission, eliminating the risk of update data being stolen or tampered with during transmission. It supports two encryption methods: integrating a homomorphic encryption library or a secret sharing tool. Homomorphic encryption can generate encrypted model updates. This ensures that updated data is always in encrypted form during transmission and subsequent aggregation; Secret sharing is done by Threshold rules will The system is split into N fragments, and decryption can only be performed when a sufficient number of fragments are collected. This prevents a single node (including the central server) from obtaining the complete update alone, and also avoids the loss of updates due to single point of failure.

[0035] Both encryption methods meet data privacy regulations, significantly reducing security concerns for enterprises participating in federated learning and ensuring the compliance and reliability of cross-enterprise collaboration.

[0036] The Federated Intelligence Analysis and Security Modeling module is configured as follows: Broadcast the task configuration to all certified enterprise clients; Initiate and execute secure multi-party computation (SMPC) processes to achieve secure model aggregation and malicious update detection; Record the global model weights for each training round. Version, and will be updated Distribute to all enterprise clients; The enterprise client's local modeling agent receives and parses the task configuration from the central server, and then processes the locally trained data. After encryption / splitting, press Threshold rules are sent to other enterprises and the central server, and participate in the SMPC collaborative decryption process, providing local shards for global update decryption.

[0037] Broadcast the task configuration to all authenticated enterprise clients, as follows: Distribute a unified model architecture; Distribute initial global model weights (Pre-trained based on publicly available industry data); Set training parameters (e.g., learning rate η=0.001, total number of communication rounds T=20); Inform about the safe aggregation rules (contribution-based weighted average and Byzantine robust algorithm).

[0038] By explicitly broadcasting the specific content of the task configuration to the certified enterprise client, a unified standard is provided for cross-enterprise federated learning training, thus avoiding the problem of collaboration failure caused by inconsistent training benchmarks among enterprises from the source. It distributes a unified model architecture (such as a demand prediction neural network structure) to ensure that the local training models of all participants are fully aligned in terms of input layer, hidden layer, and output layer design, so as to avoid the situation where the model structure is incompatible and cannot be aggregated. Simultaneously distribute initial global model weights pre-trained based on publicly available industry data. This significantly reduces the cold start cost of federated learning: Enterprises no longer need to train a basic model from scratch; they can directly iterate and optimize on pre-trained weights, significantly shortening the first training cycle and improving overall collaborative training efficiency.

[0039] Model aggregation and malicious update detection, specifically including: Shard collection: Receives model update shards sent by all enterprise clients (e.g., each client sends...). - One shard is distributed to other companies, and one shard is given to the central server); Contribution-weighted calculation: Weights are calculated based on the quality of model updates uploaded by each enterprise. High-quality data corresponds to higher ; Anomaly detection: Calculate all in encrypted state Cosine similarity (e.g., a similarity threshold of 0.7, with values ​​below the threshold being marked as abnormal) is used to identify malicious model updates (e.g., poisoning attacks). Robust aggregation: The TrimmedMean algorithm is used to remove the top 15% and bottom 15% of abnormal updates, and a weighted average is calculated for the remaining updates. ; Collaborative Decryption: Invite ≥t enterprise clients to collaboratively decrypt encrypted global updates. Obtain plaintext Update the global model .

[0040] The process of refining model aggregation and malicious update detection effectively improves the accuracy of the global model and the utilization rate of data value, and solves the problem of model performance degradation caused by non-independent and identically distributed (Non-IID) supply chain data.

[0041] Sharded data collection ensures that model updates from all enterprise clients participate in global aggregation, preventing local data gaps from affecting model integrity; contribution-weighted calculation allocates weights based on the quality of model updates uploaded by each enterprise. This allows enterprises with high-quality data and good update efficiency to contribute a larger proportion of their data in the global model, breaking the waste of high-quality data value caused by traditional simple average aggregation. This enables the global model to more accurately integrate cross-enterprise data advantages and improve the accuracy of decision-making scenarios such as demand forecasting and procurement optimization.

[0042] The privacy protection decision management and application module is configured to receive the final global model distributed by the central server when the federated training reaches the convergence condition (e.g., the global model accuracy improvement is <0.5% for 3 consecutive rounds). Deployed on the enterprise's local decision engine (such as Docker containerized deployment), the model never leaves the local environment; it generates decisions for supply chain scenarios and collects actual data after the decisions are implemented (such as the deviation between actual sales and predicted sales, and the execution rate of procurement plans) as incremental training data for labeling.

[0043] The core design of the decision management and application module for privacy protection is clearly defined, and the global model is deployed locally. This completely blocks the risk of data privacy leakage from the decision-making stage, while ensuring the security and compliance of the model application.

[0044] Once the federated training converges, the enterprise receives the final global model. And it is deployed on a local decision engine (such as a Docker container), so that the model never leaves the enterprise's local environment; This demonstrates that when enterprises generate decisions based on a global model, they do not need to upload business data to external platforms. This avoids the leakage of sensitive data (such as actual sales and inventory levels) during the decision-making process and complies with the data localization requirements of data privacy regulations such as GDPR. It eliminates enterprises' concerns about transmitting data when using the model and removes key obstacles to the implementation of cross-enterprise collaborative decision-making.

[0045] Decision generation for supply chain scenarios specifically includes: Demand Forecast: Input local historical sales and regional promotional plans, and output the product demand forecast for a future preset time period (e.g., one month); Procurement optimization: Combine demand forecasts with raw material inventory data to generate a procurement plan with the lowest cost (such as the EOQ economic order quantity model). Logistics scheduling: Optimize warehouse allocation and transportation routes based on the regional distribution of demand.

[0046] The collaborative management and feedback optimization module is configured to ensure the stability, reliability, and continuous evolution of the entire distributed federation system, specifically including: Federation node collaborative management (managing the entire lifecycle of enterprise clients): Verify the enterprise's digital certificate and confirm its supply chain role (supplier / manufacturer / distributor); only those who meet the requirements can access the platform. Real-time monitoring of client online status and computing resource usage (such as GPU utilization) triggers an alarm if a node goes offline (to avoid affecting step five: secure aggregation of shards). When an enterprise exits the market, the other enterprise model shards it stores are automatically destroyed to prevent data residue.

[0047] Contribution audit and incentive linkage: Recording each company's performance in each round of training (Contribution weight), data quality score, and generate a federal contribution audit report, which supports: Cross-enterprise settlement: For example, the top 3 contributing enterprises can receive supply chain collaborative discounts; Data quality improvement: Push data cleaning suggestions (such as supplementing missing historical sales data) to enterprises with low contribution.

[0048] Security Audit and Iteration Triggering: All key operations are recorded (such as model upload time, shard receiving node, SMPC decryption participants), and the logs are stored on the blockchain and are tamper-proof (meeting GDPR compliance requirements). When the error rate in decision feedback exceeds 15% or the enterprise data update volume exceeds 30%, a new round of federated training request is automatically initiated to the central coordination service to restart the iterative model training process.

[0049] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0050] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0051] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0052] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0053] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0055] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0056] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0058] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A big data-based enterprise intelligent decision management system, comprising a secure data acquisition and collaborative access module, a distributed data storage and computing engine module, and a privacy-protected decision management and application module, characterized in that: It also includes a federated intelligence analytics and security modeling module, configured as follows: Broadcast the task configuration to all certified enterprise clients; Initiate and execute secure multi-party computation processes to achieve secure model aggregation and malicious update detection; Record the global model weights for each training round. Version, and will be updated Distribute to all enterprise clients; The enterprise client's local modeling agent receives and parses the task configuration from the central server, and then processes the locally trained data. After encryption / splitting, press Threshold rules are sent to other enterprises and the central server, and participate in the SMPC collaborative decryption process, providing local shards for global update decryption.

2. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, The secure data acquisition and collaborative access module is configured for local compliant acquisition and secure access preparation of cross-enterprise data, providing a standardized and privacy-enhanced data source for federated learning.

3. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, The distributed data storage and computing engine module is configured to deploy a distributed storage architecture locally within the enterprise to store preprocessed standardized data, local training logs, model update backups, and encrypted components for updating the model after building a local model training engine.

4. The enterprise intelligent decision-making management system based on big data according to claim 3, characterized in that, The process of training the local model and updating the encrypted components is as follows: Training process: Download the latest global model weights distributed by the central server. ; Use local dataset conduct =5 rounds of local training, with local model updates calculated using the gradient descent algorithm. ; Calculate the model update quality on the local validation set as a weight for subsequent contribution. The basis for the calculation; Updating the encryption component: Integrate homomorphic encryption libraries or secret sharing tools to update the local model. Perform encryption / splitting operations: If homomorphic encryption is used, an encrypted model update will be generated. ; If using secret sharing, press... Threshold rules will Split into Slices .

5. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, The task configuration is broadcast to all authenticated enterprise clients, as follows: Distribute a unified model architecture; Distribute initial global model weights ; Set training parameters; Inform them of the safe aggregation rules.

6. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, Model aggregation and malicious update detection, specifically including: Receive model update shards sent by all enterprise clients; Weights are calculated based on the quality of model updates uploaded by each company. High-quality data corresponds to higher ; Calculate all in encrypted state Cosine similarity is used to identify malicious model updates; After removing the top 15% and bottom 15% of abnormal updates, calculate a weighted average for the remaining updates. ; A global update that gathers ≥t enterprise clients to collaboratively decrypt and encrypt data. Obtain plaintext Update the global model .

7. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, The privacy protection decision management and application module is configured to receive the final global model distributed by the central server when the federated training reaches convergence. Deployed on the enterprise's local decision engine; To generate decisions for supply chain scenarios, the actual data collected after the decisions are implemented will be used as incremental training data for labeling.

8. The enterprise intelligent decision-making management system based on big data according to claim 7, characterized in that, The decision generation for supply chain scenarios specifically includes: Demand Forecast: Input local historical sales and regional promotional plans, and output the product demand forecast for a future preset time period; Procurement optimization: Combine demand forecasts with raw material inventory data to generate a procurement plan with the lowest cost; Logistics scheduling: Optimize warehouse allocation and transportation routes based on the regional distribution of demand.

9. The enterprise intelligent decision-making management system based on big data according to claim 1, characterized in that, Also includes: The collaborative management and feedback optimization module is configured to ensure the stability, reliability, and continuous evolution of the entire distributed federation system.

Images