Big data sharing method and sharing system

Abstract

The invention provides a big data sharing method and system, and the method comprises the steps: S1, obtaining real-time business data and environment dynamic data of a target sharing scene, and enabling the real-time business data to comprise multi-source heterogeneous to-be-shared data and data interaction features; the environment dynamic data comprises scene business rule change data, data feature distribution change data and security level change data; s2, carrying out standardized preprocessing on the real-time service data, carrying out quantitative coding processing on the environment dynamic data, and respectively extracting a service feature vector and an environment dynamic feature vector; and S3, based on a preset initial parameter set, the service feature vector and the environment dynamic feature vector, calculating through a dynamic parameter optimization algorithm to obtain a target parameter set adaptive to the current scene. The change rate of service data and environment data is monitored in real time, the self-adaptive adjustment process is triggered, and it is ensured that the system is stably adapted to a dynamic scene for a long time.

Description

Technical Field

[0001] This invention relates to the field of big data processing and information sharing technology, and in particular to a big data sharing method and system. Background Technology

[0002] With the rapid development of artificial intelligence and the Internet of Things, big data has become a core production factor for collaborative development across various industries. Big data sharing can break down data silos, enabling optimized resource allocation and improved decision-making efficiency. Currently, big data sharing technology has been widely applied in scenarios such as smart supply chains and e-commerce user feedback, resulting in a variety of data sharing methods and systems.

[0003] For example, patent CN117235181A discloses a smart supply chain big data sharing method and system. It constructs a blockchain network and designs smart contracts such as regional allocation contracts, data encryption contracts, and sharing contracts to achieve data storage and sharing among all members of the supply chain. Simultaneously, it introduces an interest-based calculation model to achieve automatic data sharing based on parameters such as transaction frequency and sharing frequency. While this technology ensures transparency in data sharing through the distributed nature of blockchain and enhances data security through encryption contracts, it still has significant limitations in practical applications.

[0004] Another patent, CN119917890A, discloses a user information sharing method and system based on big data. It achieves reliable sharing of user review information by performing feature clustering analysis on false review information from a target platform and similar platforms, integrating abnormal review indicators, and filtering reliable review data based on a dynamic credibility threshold. This technology focuses on filtering the authenticity of review data but does not address the issues of parameter adaptation and model expansion in dynamic scenarios.

[0005] While the aforementioned existing technologies have achieved big data sharing in specific scenarios, both employ a core architecture of preset parameters and fixed algorithms, lacking the ability to adapt to dynamic business scenarios. Specific shortcomings are as follows: In comparison document 1, the interest calculation model relies on preset weight parameters. When the transaction frequency and data sharing patterns of supply chain members change dynamically (e.g., a surge in transactions due to promotional activities or the addition of new long-term partners), the preset weights cannot adapt to the changes in the scenario in real time, requiring manual intervention for fine-tuning. This not only results in a delayed response but is also prone to decreased sharing accuracy due to human error. In Comparison Document 2, the fusion weights of abnormal evaluation indicators adopt a fixed allocation method (0.7 for the first abnormal evaluation indicator and 0.3 for related abnormal evaluation indicators), failing to consider the dynamic changes in the correlation between similar platforms (e.g., a similar platform's data reference value decreases due to the proliferation of false evaluations). Fixed weights cause the fusion results of abnormal evaluation indicators to deviate from the actual scenario requirements, thus affecting the accuracy of reliable evaluation data selection. This leads to parameters relying on manual adjustment, resulting in extremely poor flexibility.

[0006] The smart contract logic in Comparison Document 1 (including regional allocation rules and automatic sharing trigger conditions) is embedded in the blockchain network. If new business links are added to the supply chain (such as cross-border logistics tracking and after-sales service data synchronization), the smart contract code needs to be modified and redeployed to the blockchain node. This not only results in high development costs and long deployment cycles but may also affect the continuity of existing shared business. The evaluation trend analysis plugin in Comparison Document 2 is trained based on a BP neural network. The training data only covers existing features of false evaluations (such as short text positive reviews and low-quality image reviews). When new features of false evaluations emerge (such as AI-generated batch homogeneous reviews and image-text separation false reviews), the original model cannot identify the new features. It requires re-collecting sample data for model retraining, making it unable to respond in real time to the evolution trend of false evaluations, leading to the omission of unreliable evaluation data. This results in weak model generalization ability, making it difficult to cope with new scenarios and new features. The core flaw of existing technologies lies in the fact that intelligent algorithms lack the closed-loop capability of dynamic perception, parameter self-optimization, and model self-iteration. They rely excessively on manual intervention to complete parameter adjustments and model updates, making it difficult to adapt to dynamic changes in business scenarios (such as the addition of new business processes and rule adjustments) and the evolution of data characteristics (such as changes in data distribution and the emergence of new features). As a result, the efficiency, accuracy, and reliability of big data sharing are all limited by manual operation, and cannot meet the sharing needs in complex and dynamic scenarios. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to overcome the defects of the existing technology. The present invention proposes a big data sharing method and sharing system.

[0008] To address the core shortcomings of existing technologies, namely the lack of a closed-loop capability of dynamic perception, parameter self-optimization, and model self-iteration in intelligent algorithms, their over-reliance on manual intervention for parameter adjustment and model updates, and their inability to adapt to dynamic changes in business scenarios and the evolution of data characteristics, the present invention adopts the following technical solution: A big data sharing method and system, comprising: S1: Obtain real-time business data and environmental dynamic data of the target shared scenario. The real-time business data includes multi-source heterogeneous data to be shared and data interaction characteristics. The environmental dynamic data includes scenario business rule change data, data feature distribution change data, and security level change data. S2: Perform standardized preprocessing on the real-time business data, perform quantitative encoding on the environmental dynamic data, and extract business feature vectors and environmental dynamic feature vectors respectively; S3: Based on the preset initial parameter set and the business feature vector and the environmental dynamic feature vector, a target parameter set adapted to the current scenario is calculated through a dynamic parameter optimization algorithm. The target parameter set includes data sharing weight parameters, model decision threshold parameters, and encryption strength parameters. S4: Based on the security level change data and encryption strength parameters, an adaptive encryption algorithm is used to perform hierarchical encryption processing on the standardized preprocessed real-time business data to obtain encrypted data and plaintext data. S5: Input the encrypted data, plaintext data and business feature vector into a preset sharing decision model, and use the target parameter set to drive the sharing decision model to output preliminary sharing results. The preliminary sharing results include a shared data candidate set, a data receiver list and a sharing interface identifier. S6: Collect feedback data of the preliminary sharing results, calculate the real-time performance evaluation index of the sharing decision model, and determine whether the real-time performance evaluation index meets the preset performance threshold. S7: If the real-time performance evaluation index does not meet the preset performance threshold, then based on the environmental dynamic feature vector and the error data of the preliminary shared results, the network parameters of the shared decision model are updated through the model self-iteration algorithm, and the process returns to step S5; if the real-time performance evaluation index meets the preset performance threshold, then step S8 is executed. S8: Based on the data receiver list and the sharing interface identifier, the shared data candidate set is output to the corresponding receiver through a cross-scenario compatible sharing protocol to complete data sharing; S9: Monitor the change rate of business data and the change rate of environmental dynamic data in the target shared scenario in real time. If either change rate exceeds the preset change threshold, return to step S1 and re-execute the parameter optimization and model adaptation process.

[0009] Preferably, the calculation process of the dynamic parameter optimization algorithm in step S3 includes: S31: Extract the key influencing factors from the business feature vector. The key influencing factors include the data interaction frequency change rate Δf, the data interaction amount change rate ΔC, and the data sharing demand intensity S. S32: Extract the scene adaptation factor from the dynamic feature vector of the environment, the scene adaptation factor including the business rule change coefficient. Data feature evolution coefficient ; S33: For each parameter in the initial parameter set The target parameters are calculated using the following formula. : in, For the preset adjustment factor, satisfy ,and ; =(Current cycle interaction frequency - Historical average interaction frequency) / Historical average interaction frequency =(Current period interaction limit - Historical average interaction limit) / Historical average interaction limit It is calculated by weighting the frequency and urgency of the data recipient's sharing requests; When the business rules remain unchanged When business rules are completely updated It is calculated from the cosine similarity change of the data feature distribution.

[0010] Preferably, the specific implementation process of step S2 includes: S21: Standardize and preprocess real-time business data: For numerical business data, the Z-score standardization formula is used for processing: in, The data is in the original numerical format. This is the historical average of this type of data. This represents the historical standard deviation of this type of data; For text-based business data, the TF-IDF algorithm is used to extract text features and standardize them into vector form: in, For the term In the document word frequency in For the term In the document The number of times it appears in This is the sum of the occurrences of all terms in document d; For the term Inverse document frequency, Total number of documents For included terms The number of documents; for image-type business data, normalization is used to map pixel values ​​to the [0,1] interval: ; S22: Perform quantization and encoding processing on dynamic environmental data: map the scenario business rule change data into discrete values ​​(no change: 0, partial change: 0.5, complete change: 1); calculate the distribution difference value of the data feature distribution change data through KL divergence and map it to the [0,1] interval as the data feature evolution coefficient; map the security level change data into quantization values ​​of 1-5 levels, corresponding to the encryption strength level.

[0011] Preferably, the adaptive encryption algorithm in step S4 includes: S41: Determine the encryption level L (L∈{1,2,3,4,5}) based on the security level quantification value. S42: When L=1, the sensitive fields are encrypted using the AES-128 algorithm, while the non-sensitive fields are stored in plaintext. When L=2, the AES-192 algorithm is used to encrypt all business data, and the key length is 192 bits. When L=3, AES-256 algorithm is used for encryption, combined with hash verification (SHA-256) to ensure data integrity; When L=4, the AES-256 algorithm + homomorphic encryption algorithm is used, which supports finite operations in the ciphertext state; When L=5, the AES-256 algorithm + federated learning encryption protocol is used to make the data usable but invisible; S43: Optimize the key distribution mechanism based on the Ethereum interaction protocol, and generate dynamic keys using the following formula: in, The base key is T, and T is the timestamp (accurate to the second). As a unique identifier for the data recipient, It uses the SHA-256 hash function; the key validity period is positively correlated with the security level. When L=1, the validity period is 24 hours. For each level increase of L, the validity period is shortened by 50%.

[0012] Preferably, the real-time performance evaluation metrics in step S6 include shared accuracy. Shared recall rate Shared latency Data security rate The specific calculation method is as follows: in, This represents the number of data entries in the initial shared results; The preset performance thresholds are: Acc≥95%, Rec≥90%, Lat≤100ms, Sec=100%. When all indicators meet the threshold requirements, it is determined that the preset performance thresholds are met.

[0013] Preferably, the model self-iterative algorithm in step S7 adopts an adaptive gradient descent algorithm, specifically including: S71: Construct the loss function L, which is a multi-objective optimization function: in, For loss weights, satisfying ,and ; Maximum allowed sharing latency (default is 500ms); S72: Calculate the gradient ∇L of the loss function L with respect to the network parameters W of the shared decision model. in, This refers to the number of network parameters. S73: Update network parameters using adaptive learning rate. ,in, The initial learning rate (default is 0.001). This is the attenuation coefficient (default is 0.1). The initial loss value, 'This refers to the updated network parameters; S74: Repeat steps S71-S73 until the loss function L ≤ the preset loss threshold (preset to 0.05), or the number of iterations reaches the maximum iteration threshold (preset to 100), then stop the iteration and save the updated model parameters.

[0014] Preferably, the cross-scenario compatible sharing protocol mentioned in step S8 includes: S81: Determine the interface type (REST interface, WebSocket interface, blockchain smart contract interface) based on the shared interface identifier. S82: Use corresponding data packet formats for different interface types: For REST interfaces, shared data is encapsulated in JSON format, with fields including data_id (unique data identifier), data_type (data type), data_content (data content, encrypted or plaintext), timestamp, and signature (digital signature). For the WebSocket interface, binary stream format is used for transmission, and the data packet header includes interface identifier, data length, and checksum. For blockchain smart contract interfaces, data is encapsulated according to the contract ABI specification, and the contract's shared execution function is called to complete the data upload to the chain and authorization by the recipient; S83: Establish interface adaptation middleware to automatically identify the receiver's interface type and convert the data packet format, ensuring cross-regional and cross-platform data interoperability and compatibility.

[0015] Preferably, the calculation methods for the business data change rate R1 and the environmental dynamic data change rate R2 in step S9 are as follows: in, This is the feature vector of the current cycle's business. This represents the average of the business feature vectors over three historical periods. This represents the dynamic feature vector of the current periodic environment. The mean of the environmental dynamic feature vectors over three historical periods is given by ||・||2, which is the L2 norm. The preset change threshold is R1≥30% or R2≥20%. When this condition is met, it is determined that the scene has changed significantly, triggering the parameter re-optimization and model adaptation process.

[0016] A big data sharing system, applied to the above-mentioned big data sharing method, includes: Data acquisition module: used to acquire real-time business data and environmental dynamic data of the target shared scenario. The real-time business data includes multi-source heterogeneous data to be shared and data interaction characteristics. The environmental dynamic data includes scenario business rule change data, data feature distribution change data, and security level change data. Data preprocessing module: used to perform standardized preprocessing on the real-time business data, quantitative encoding on the environmental dynamic data, and extract business feature vectors and environmental dynamic feature vectors respectively; Parameter optimization module: Based on a preset initial parameter set and the business feature vector and environmental dynamic feature vector, it calculates a target parameter set that is suitable for the current scenario through a dynamic parameter optimization algorithm. The target parameter set includes data sharing weight parameters, model decision threshold parameters, and encryption strength parameters. Adaptive encryption module: used to perform hierarchical encryption processing on standardized preprocessed real-time business data according to the security level change data and encryption strength parameters, and to obtain encrypted data and plaintext data; Model Decision Module: This module is used to input the encrypted data, plaintext data, and business feature vectors into a preset sharing decision model, and to drive the sharing decision model to output preliminary sharing results using the target parameter set. The preliminary sharing results include a candidate set of shared data, a list of data recipients, and a sharing interface identifier. Performance evaluation module: used to collect feedback data of the preliminary sharing results, calculate the real-time performance evaluation index of the sharing decision model, and determine whether the real-time performance evaluation index meets the preset performance threshold; Model Iteration Module: When the real-time performance evaluation index does not meet the preset performance threshold, it updates the network parameters of the shared decision model based on the environmental dynamic feature vector and the error data of the preliminary shared results through a model self-iteration algorithm. Shared execution module: When the real-time performance evaluation index meets the preset performance threshold, it outputs the shared data candidate set to the corresponding receiver based on the data receiver list and the shared interface identifier, and completes data sharing through a cross-scenario compatible sharing protocol; Dynamic monitoring module: Used to monitor the change rate of business data and the change rate of environmental dynamic data in the target shared scenario in real time. If either change rate exceeds the preset change threshold, the data acquisition module is triggered to reacquire data and start a new round of parameter optimization and model adaptation process.

[0017] Preferably, the hardware architecture of the system includes: Edge acquisition layer: Composed of data acquisition terminals, sensors, and API gateways. The data acquisition terminals include industrial-grade IoT terminals and server clusters, used to collect real-time business data from multiple heterogeneous sources. The API gateway supports multiple protocols such as HTTP, HTTPS, and MQTT, and is used to connect to data source systems in the target shared scenario. Edge computing layer: Computing servers deployed on edge nodes, configured with CPUs and GPUs, used to perform low-latency computing tasks such as data preprocessing and dynamic parameter optimization; The cloud platform layer includes a cloud server cluster, a distributed database, and a blockchain node cluster. The cloud server cluster is used to deploy shared decision-making models and perform self-iterative training of the models. The distributed database is used to store real-time business data, dynamic environmental data, model parameters, and shared logs. The blockchain node cluster is used to store encrypted data hash values ​​and shared authorization records to ensure data traceability. Application Interface Layer: Composed of interface adaptation middleware and load balancer. The interface adaptation middleware supports adaptive conversion of REST, WebSocket, and blockchain smart contract interfaces; the load balancer is used to distribute shared requests and avoid single points of failure.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention proposes a parameter self-optimization algorithm that integrates business characteristics and environmental characteristics. It quantifies the impact of scene changes on parameters through mathematical formulas, replacing manual adjustments and achieving real-time adaptation of parameters to the scene. Based on the security level, it dynamically adjusts the encryption algorithm and key validity period to balance data security and sharing efficiency, solving the problem of fixed encryption strategies in existing technologies. This invention constructs a loss function that integrates accuracy, recall, latency, and security. It employs an adaptive gradient descent algorithm to automatically update model parameters, thereby improving the model's generalization ability to new scenarios and features. Through interface adaptation middleware, it enables adaptive conversion of various interface types, solving the interface compatibility problem of cross-platform and cross-regional data sharing. It monitors the rate of change of business data and environmental data in real time, triggering an adaptive adjustment process to ensure the system's long-term stable adaptation to dynamic scenarios. Attached Figure Description

[0019] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein: Figure 1 This is a flowchart illustrating the steps of the big data sharing method of the present invention; Figure 2 This is a diagram showing the overall architecture of the big data sharing system of this invention. Detailed Implementation

[0020] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0021] Specific embodiments of the present invention are described below with reference to the accompanying drawings.

[0022] Please see Figures 1-2 This embodiment proposes a big data sharing system, which includes a data acquisition module, a data preprocessing module, a parameter optimization module, an adaptive encryption module, a model decision-making module, a performance evaluation module, a model iteration module, a sharing execution module, and a dynamic monitoring module. The specific contents are as follows: Data acquisition module: used to acquire real-time business data and environmental dynamic data of the target shared scenario. Real-time business data includes multi-source heterogeneous data to be shared and data interaction characteristics. Environmental dynamic data includes scenario business rule change data, data feature distribution change data, and security level change data. Data preprocessing module: used to standardize and preprocess real-time business data, quantize and encode environmental dynamic data, and extract business feature vectors and environmental dynamic feature vectors respectively; Parameter optimization module: Based on a preset initial parameter set and business feature vectors and environmental dynamic feature vectors, it calculates a target parameter set that is suitable for the current scenario through a dynamic parameter optimization algorithm. The target parameter set includes data sharing weight parameters, model decision threshold parameters, and encryption strength parameters. Adaptive encryption module: Used to perform hierarchical encryption processing on standardized preprocessed real-time business data based on the security level changes and encryption strength parameters, and to obtain encrypted data and plaintext data. Model Decision Module: This module is used to input encrypted data, plaintext data, and business feature vectors into a preset sharing decision model. It uses a target parameter set to drive the sharing decision model to output preliminary sharing results, which include a candidate set of shared data, a list of data recipients, and a sharing interface identifier. Performance evaluation module: used to collect feedback data of preliminary sharing results, calculate real-time performance evaluation indicators of the sharing decision model, and determine whether the real-time performance evaluation indicators meet the preset performance thresholds; Model Iteration Module: When the real-time performance evaluation metrics do not meet the preset performance threshold, the network parameters of the shared decision model are updated through a model self-iteration algorithm based on the environmental dynamic feature vector and the error data of the preliminary shared results. Shared execution module: When the real-time performance evaluation indicators meet the preset performance threshold, it outputs the candidate set of shared data to the corresponding receiver based on the list of data receivers and the shared interface identifier, and completes the data sharing by using a cross-scenario compatible sharing protocol. Dynamic monitoring module: Used to monitor the rate of change of business data and the rate of change of dynamic environmental data in the target shared scenario in real time. If either rate of change exceeds the preset change threshold, the data acquisition module is triggered to reacquire data and start a new round of parameter optimization and model adaptation process.

[0023] A big data sharing method includes a data acquisition stage, a data processing stage, a model decision-making and judgment stage, and a result output and dynamic monitoring stage. The specific content is as follows: 1. Data Acquisition Stage The execution entity is the data acquisition module, and the specific implementation process is as follows: Real-time business data acquisition: Collect transaction data (order amount, transaction frequency) and production data (production plan, inventory quantity) in the supply chain through industrial-grade IoT terminals, and collect user evaluation data (evaluation text, images, ratings) through web crawlers and API interfaces. It supports the collection of structured data (MySQL database tables), semi-structured data (JSON / XML files), and unstructured data (text, images, audio). The collection frequency is configurable (adjustable from 100ms to 1s). Environmental dynamic data collection: Obtain data on changes in business rules (such as rule documents for new cross-border logistics links and notifications of updated evaluation and review rules) by connecting to business systems in the target scenario (such as supply chain management systems and e-commerce platform backends); analyze data distribution changes and calculate data feature evolution coefficients by using data feature monitoring tools (based on Python sklearn); and obtain data on changes in security levels (such as network attack events and data leakage risk warnings) by using security situation awareness systems (such as Qi An Xin Tian Qing).

[0024] 2. Data processing stage This stage includes three core steps: data preprocessing, parameter optimization, and adaptive encryption. All of these are executed by the execution entity (main control server) by scheduling resources from the edge computing layer and cloud platform layer. (1) Data preprocessing Standardized preprocessing: Different processing methods are adopted for different types of data to ensure that the data format is uniform and the magnitude is consistent, laying the foundation for subsequent parameter optimization and model input. For example, inventory quantities in the supply chain (numerical type) are standardized using Z-score to eliminate the impact of differences in inventory units among different suppliers; user review texts (text type) use the TF-IDF algorithm to extract features, converting unstructured text into structured vectors; and review images (image type) are normalized to unify the pixel magnitude.

[0025] Quantization encoding processing: Converting non-numerical dynamic environmental data into computable numerical values, for example, mapping the change status of business rules (no change / partial change / complete change) to discrete values ​​of 0 / 0.5 / 1; calculating the difference between the current data feature distribution and the historical distribution using KL divergence, as shown in the following formula: in Given the current data feature distribution, For the distribution of historical data characteristics, The divergence is normalized to the [0,1] interval and used as the data feature evolution coefficient. The security levels (low / medium / high / very high / extreme) are mapped to quantitative values ​​of 1-5 levels.

[0026] (2) Dynamic parameter optimization The core of this step is to automatically adjust model parameters and shared configuration parameters based on real-time data characteristics and environmental changes, without the need for manual intervention.

[0027] Key influencing factor extraction: Extract the data interaction frequency change rate Δf, data interaction amount change rate ΔC, and data sharing demand intensity S from the business feature vector. For example, in the supply chain scenario, Δf = (number of transactions in the current month - average number of transactions in the past 6 months) / average number of transactions in the past 6 months, ΔC = (total transaction amount in the current month - average total transaction amount in the past 6 months) / average total transaction amount in the past 6 months; S is calculated by weighting the number of sharing requests from the recipient (weight 0.6) and the urgency of the request (weight 0.4, urgency is 1, normal is 0.5, low priority is 0.2).

[0028] Scenario adaptation factor extraction: The business rule change coefficient K1 is determined based on the range of rule changes. If only the parameter threshold is modified, K1=0.3; if a new business step is added, K1=0.8; if the business process is completely restructured, K1=1. The data feature evolution coefficient K2 is obtained by KL divergence normalization.

[0029] Target parameter calculation: Calculate the target parameters based on the initial parameter set. For example, initial weights. Then the target parameter It adapts to scenarios where both transaction frequency and amount are increasing.

[0030] (3) Adaptive encryption The encryption strategy is dynamically adjusted based on the security level to balance security and sharing efficiency. Encryption level determination: The security level is output by the security situation awareness system. For example, it is level 1 when there is no security risk, level 2 when there is a minor network attack, and level 5 after a data breach. Encryption algorithm execution: Level 1 encryption only uses AES-128 encryption for sensitive fields (such as customer ID number and bank card number), while non-sensitive fields are transmitted in plaintext; Level 5 encryption uses AES-256 + federated learning encryption protocol, and the data completes model inference in the encrypted state. The receiver can only obtain the inference result and cannot obtain the original data. Dynamic key generation and distribution: The base key Key_base is initialized by the system administrator and stored in the Hardware Security Module (HSM). Each time a key is distributed, a dynamic key is generated by combining the current timestamp (accurate to the second) with the unique identifier of the recipient. The key validity period is adjusted according to the security level to ensure key security.

[0031] 3. Model Decision-Making and Judgment Stage (1) Construction of shared decision-making model The shared decision-making model employs a hybrid deep learning architecture. Its inputs include encrypted data, plaintext data, and business feature vectors, and its output is the preliminary sharing result. The model structure is as follows: Input layer: Receives standardized business feature vectors (dimensions are adjusted according to the scenario; 64 dimensions for supply chain scenarios and 128 dimensions for user evaluation scenarios). Feature fusion layer: Employs an attention mechanism to fuse encrypted data features and plaintext data features, with weights dynamically adjusted by the encryption strength parameter; Decision layer: It consists of a 3-layer fully connected neural network (with 128, 64 and 32 hidden nodes respectively), with ReLU activation function, and outputs a shared data candidate set, a list of data receivers and a shared interface identifier; Output layer: The Sigmoid function is used to output the data sharing probability. Data with a probability greater than or equal to the target decision threshold (from the target parameter set) is considered shareable data.

[0032] (2) Generation of preliminary shared results The executing entity inputs preprocessed encrypted data, plaintext data, and business feature vectors into the shared decision model, and drives the model operation using weight parameters and decision threshold parameters from the target parameter set. For example, in a user review scenario, the model uses the TF-IDF vector of the review text and image quality features, combined with the target decision threshold (0.85), to filter out reliable review data with a sharing probability ≥ 0.85, generating a candidate set of shared data; based on the associated platforms and user groups of the review data, it determines the list of data recipients (such as e-commerce platform operation systems and third-party review and monitoring agencies); and determines the shared interface identifier (REST interface) according to the interface type of the recipient.

[0033] (3) Performance evaluation and judgment The executing entity collects feedback data on the initial sharing results (such as confirmation receipts from the recipients and data usage logs) and calculates four core performance indicators: Accuracy of sharing (Acc): This is the percentage of correctly shared data (data that meets the recipient's needs and is error-free) out of the total amount of shared data. For example, if 1000 evaluation data points are shared, and 960 of them are reliable data that meets the recipient's needs, then Acc = 96%. Shared Recall (Rec): This is the percentage of the target data that was correctly shared (data that the recipient explicitly needs) out of the total target data that should be shared. For example, if the recipient needs 100 high-quality review data points, and the model successfully filters and shares 92 of them, then Rec = 92%. Shared latency (Lat): Calculates the average time from receiving a share request to outputting each piece of data to the receiver. For example, the average latency for 1000 pieces of data is 80ms. Data security rate (Sec): This is the percentage of shares that did not result in unauthorized access or data breaches out of the total number of shares. If there are no security incidents in 100 shares, then Sec = 100%. The executing entity compares the above indicators with preset thresholds (Acc≥95%, Rec≥90%, Lat≤100ms, Sec=100%). If all indicators are met, the shared execution phase begins; if any indicator is not met (e.g., Rec=88%<90%), the model self-iteration process is triggered.

[0034] (4) Model self-iteration The executing entity updates the model parameters using an adaptive gradient descent algorithm based on dynamic environmental feature vectors and error data. Loss function construction: A multi-objective optimization function is adopted, which comprehensively considers accuracy, recall, latency, and safety, with weight allocation tilted towards safety and accuracy. ), to ensure the reliability and security of shared data; Gradient calculation and parameter update: The gradient of the loss function with respect to the model network parameters is calculated through the backpropagation algorithm. The parameters are updated using an adaptive learning rate (which decays as the loss value decreases) to avoid model oscillation and accelerate convergence. Iteration termination condition: When the loss function L≤0.05 or the number of iterations reaches 100, stop the iteration, save the updated model parameters, and return to step S5 to regenerate the preliminary shared results.

[0035] 4. Results Output and Dynamic Monitoring (1) Data sharing execution Based on the list of data recipients and the sharing interface identifier in the preliminary sharing results, the executing entity outputs data through a cross-scenario compatible sharing protocol: Interface type adaptation: The interface adaptation middleware automatically identifies the interface type of the recipient. If the recipient is an e-commerce platform backend (REST interface), the shared data is encapsulated in JSON format; if the recipient is a blockchain node (smart contract interface), the data is encapsulated according to the contract ABI specification and the contract's shareData function is called to complete the data upload to the blockchain. Load balancing and fault tolerance: The load balancer distributes shared requests to idle application interface nodes. If a node fails, it automatically switches to a backup node to ensure the continuity of shared services. Shared log recording: The executing entity records information such as the unique identifier of the shared data, the recipient's identifier, the sharing time, and the encryption level to the distributed database and blockchain nodes to ensure that data sharing is traceable.

[0036] (2) Dynamic monitoring and adaptive adjustment The implementing entity perceives scene changes in real time through the dynamic monitoring module: Change rate calculation: The L2 norm is used to calculate the change rate R1 of business data and the change rate R2 of environmental dynamic data. For example, if the L2 norm ratio of the current business feature vector to the historical mean vector is 0.35, then R1 = 35%. Threshold trigger: If R1=35%≥30% or R2=22%≥20%, it is determined that the scene has changed significantly. The execution subject triggers the data acquisition module to re-collect data and start a new round of parameter optimization and model adaptation process to ensure that the model and parameters are always adapted to the current scene.

[0037] System hardware deployment 1. Edge acquisition layer deployment Data acquisition terminals: Deploy 10 industrial-grade IoT terminals (Huawei IoTGatewayAR502H) to connect to the supply chain's ERP system (SAPS / 4HANA) and warehouse management system (WMS) to collect production plans, inventory quantities, transaction data, etc.; Deploy 5 web crawler servers (Intel Xeon E5-2699v4) to collect user review data from e-commerce platforms. API Gateway: Deploy 2 Nginx API Gateway servers, supporting HTTP, HTTPS, and MQTT protocols, configuring request rate limiting (maximum 100 requests per second per IP), permission verification functions, and connecting to data source system interfaces (such as SAPOData interface, e-commerce platform OpenAPI).

[0038] 2. Edge computing layer deployment Computing Servers: Deploy 4 edge computing servers, configured with CPU (Intel Xeon Platinum 8375C, 32 cores and 64 threads), GPU (NVIDIA A100, 40GB VRAM), memory (256GB DDR4), and hard disk (4TB SSD) to perform data preprocessing and dynamic parameter optimization tasks; Software configuration: Install Ubuntu 20.04LTS operating system, deploy Python 3.9 environment, and install libraries such as scikit-learn (data preprocessing), NumPy (numerical computation), and TensorFlowLite (lightweight model inference).

[0039] 3. Cloud platform layer deployment Cloud server cluster: Deploy 8 cloud servers (Alibaba Cloud ECSg7.xlarge), each configured with CPU (16 cores), memory (64GB), and GPU (NVIDIA T4) to deploy a shared decision model and execute model self-iterative training; Distributed database: Deploy a Hadoop HDFS cluster (3 NameNodes + 10 DataNodes) to store real-time business data and dynamic environmental data (100TB storage capacity); deploy a MongoDB cluster (3 shards, 3 replicas per shard) to store unstructured data such as model parameters and shared logs; Blockchain node cluster: Deploy 4 blockchain nodes (based on Ethereum Quorum), distributed in different regions (Beijing, Shanghai, Guangzhou, Shenzhen), used to store encrypted data hash values ​​and share authorization records, with a block generation time of 10 seconds.

[0040] 4. Application Interface Layer Deployment Interface adaptation middleware: Deploy 2 middleware servers (Intel Xeon E5-2678v3) and install self-developed interface conversion software to support adaptive conversion of REST, WebSocket and blockchain smart contract interfaces; Load balancer: Deploy 2 Nginx+Keepalived load balancer servers, configure round-robin scheduling algorithm to distribute shared requests, and ensure that the load on each application interface server is even (CPU utilization ≤70%).

[0041] Software module implementation The data acquisition module software implementation is based on the Spring Boot framework and developed using the Java language.

[0042] The data preprocessing module was implemented using Python.

[0043] The parameter optimization module is implemented using Python.

[0044] Adaptive encryption module software implementation: developed in Java and based on the BouncyCastle encryption library.

[0045] Shared decision model implementation: Built using TensorFlow 2.10.

[0046] The model iteration module software implementation is developed using the Python language.

[0047] The dynamic monitoring module software is developed using Java and based on the Prometheus+Grafana monitoring framework.

[0048] Example 1: Supply Chain Big Data Sharing Scenario 1. Scene Background A large manufacturing supply chain includes 10 suppliers, 5 manufacturers, 3 logistics companies, 2 warehousing companies, and 8 distributors. It needs to achieve real-time sharing of information such as production plans, inventory quantities, logistics progress, and transaction data. The scenario is characterized by transaction frequency and amount fluctuating with the seasons, and occasional new business links (such as cross-border logistics).

[0049] 2. System Deployment and Parameter Configuration Initial parameter set: Interest level calculation weights , Shared decision threshold Initial value of encryption strength level ; Preset performance thresholds: Acc ≥ 95%, Rec ≥ 90%, Lat ≤ 100ms, Sec = 100%; Change thresholds: R1≥30%, R2≥20%.

[0050] 3. Specific Implementation Steps (1) Data acquisition The data acquisition module collects real-time business data from each supply chain member through IoT terminals: Supplier A's transaction data: Monthly transaction frequency = 12 times / month, historical average = 8 times / month, Δf = (12-8) / 8 = 0.5; Monthly transaction amount = 1.2 million yuan, historical average = 1 million yuan, ΔC = (1.2 million-1 million) / 1 million = 0.2; Manufacturer B's production data: Production plan completion rate = 95%, Inventory quantity = 5000 units; Environmental dynamic data: No change in business rules (K1=0), KL divergence of data feature distribution change = 0.15 (K2=0.18), security level = level 2 (no security events).

[0051] (2) Data preprocessing Standardized preprocessing: Supplier A's transaction amount (1.2 million yuan) is standardized using Z-score ( Ten thousand yuan, (ten thousand yuan), we get x'=(120-100) / 15≈1.33; Quantitative coding: Environmental dynamic data is coded as K1=0, K2=0.18, and security level code=2.

[0052] (3) Dynamic parameter optimization Data sharing demand intensity S: Distributor C requests data sharing from supplier A 5 times / week, urgency level = normal (0.5), S = 0.6 × (5 / 7) + 0.4 × 0.5 ≈ 0.43 + 0.2 = 0.63; Target parameter calculation: ; ; Shared decision threshold (Since the upper limit of the threshold is 1, we take...) =1.0); Encryption strength level L=2 (security level unchanged).

[0053] (4) Adaptive encryption The AES-192 algorithm is used to encrypt transaction data and inventory data. The key length is 192 bits and the key validity period is 12 hours (when L=2, the validity period is 24 hours × 50% = 12 hours).

[0054] (5) Model decision The encrypted transaction data, production data, and business feature vectors ( Input the shared decision model, using the objective parameters ( The driving model outputs preliminary shared results: Shared data candidate set: Transaction data of supplier A, and inventory data of manufacturer B; Data recipient list: Distributor C, Logistics Company D; Shared interface identifier: REST interface.

[0055] (6) Performance evaluation Shared accuracy (Acc): Of 100 shared data points, 98 meet the recipient's requirements, resulting in an Acc=98%. Shared Recall (Rec): The receiver needs 50 target data points, and the model successfully shares 48, Rec = 96%; Shared latency (Lat): Average latency = 75ms; Data security rate Sec=100%; All metrics meet the preset thresholds (Acc≥95%, Rec≥90%, Lat≤100ms, Sec=100%).

[0056] (7) Data sharing The interface adaptation middleware encapsulates the shared data into JSON format and sends it to the systems of distributor C and logistics company D via REST interface. The load balancer distributes the requests to idle interface nodes, and the shared logs are recorded in the distributed database and blockchain nodes.

[0057] (8) Dynamic monitoring One month later, a new cross-border logistics link was added to the supply chain (business rules completely changed, K1=1). The environmental dynamic data change rate R2=0.25≥20%. The dynamic monitoring module triggered the data acquisition module to re-collect data and start a new round of parameter optimization and model adaptation process. The target parameter α' was updated to 0.6×[1+0.4×0.3+0.3×0.1+0.3×0.7×(1+0.2)]=0.6×1.522≈0.913. After the model self-iterative update, it successfully adapted to the data sharing requirements of the cross-border logistics link.

[0058] 4. Implementation Results In this embodiment, the system does not require manual parameter adjustment and automatically adapts to the changes in the scenario of increased transaction frequency and new business links. The sharing accuracy reaches 98%, the sharing latency is 75ms, and the data security rate is 100%. Compared with the existing technology (manual parameter adjustment, accuracy of 90%, latency of 200ms), the sharing efficiency is improved by 62.5% and the accuracy is improved by 8.9 percentage points.

[0059] Example 2: User Review Information Sharing Scenario 1. Scene Background An e-commerce platform needs to share user review data with a third-party review monitoring agency. It needs to screen reliable review data (filtering out fake positive reviews and malicious negative reviews). The scenario is characterized by the dynamic changes in the features of fake reviews (such as the appearance of AI-generated reviews), and the correlation between similar platforms fluctuates over time.

[0060] 2. System Deployment and Parameter Configuration Initial parameter set: weighted fusion of anomaly evaluation metrics Shared decision threshold Initial value of encryption strength level ; Preset performance thresholds: Acc ≥ 96%, Rec ≥ 92%, Lat ≤ 90ms, Sec = 100%; Change thresholds: R1≥30%, R2≥20%.

[0061] 3. Specific Implementation Steps (1) Data acquisition Real-time business data: Collect 1,000 user review data (text + images) from the target platform, including 200 fake positive reviews, 100 malicious negative reviews, and 700 reliable reviews; collect 500 review data from each of the three similar platforms. Environmental dynamics data: The target platform's evaluation and review rules remain unchanged (K1=0), but false reviews exhibit AI-generated characteristics (KL divergence of data feature evolution = 0.3, K2 = 0.35), and the security level is 1.

[0062] (2) Data preprocessing Standardized preprocessing: The evaluation text was processed using the TF-IDF algorithm to extract features (max_features=1000), resulting in a 1000-dimensional feature vector; the evaluation images were normalized to the [0,1] interval. Quantitative coding: Environmental dynamic data is coded as K1=0, K2=0.35, and security level code=1.

[0063] (3) Dynamic parameter optimization Business Feature Extraction: Data interaction frequency change rate Δf: The frequency of requests from third-party evaluation and monitoring agencies to share evaluation data of the target platform. The number of requests in the current period (1 week) is 30 times, the historical average number of requests is 20 times, Δf=(30-20) / 20=0.5; Data interaction limit change rate ΔC: Current period valid evaluation data volume = 900 (after filtering out incomplete data), historical average valid data volume = 750, ΔC = (900-750) / 750 = 0.2; Shared demand intensity S: The request frequency weight of third-party organizations is 0.6, and the current request frequency is 30 times / week (the baseline value is 20 times / week, which is 30 / 20=1.5 after normalization, and the upper limit is 1.0); the urgency level is high (weight 0.4, quantified value 1.0), so S=0.6×1.0+0.4×1.0=1.0; Scenario adaptation factor: K1=0 when business rules remain unchanged, and K2=0.35 when data feature evolution coefficient remains unchanged (obtained by KL divergence normalization). Target parameter calculation (initial fusion weights) Regulatory factors : Because the fusion weights need to meet the following requirements The results were normalized. Shared decision threshold Take the upper limit Encryption strength level L=1 (security level remains unchanged).

[0064] (4) Adaptive encryption Encryption strategy: Security level L=1, using the AES-128 algorithm to encrypt sensitive fields (user nickname, mobile phone number data before anonymization) in the evaluation data, and storing non-sensitive fields (evaluation text, rating, image features) in plaintext. Dynamic key generation: Base key Key_base="a8f5d2e9c3b7a16d4f8e2c9d3a7b1f0" (stored in HSM), receiver unique identifier ID_user=third_party_001, current timestamp T=1718000000 (seconds), then: Key validity period: When L=1, the key is valid for 24 hours and will be automatically regenerated upon expiration.

[0065] (5) Model decision Model input: TF-IDF feature vectors (1000-dimensional) of 1000 evaluation data points, normalized image features (256-dimensional), concatenated into a 1256-dimensional business feature vector; encrypted sensitive field features (32-dimensional, AES-128 encrypted data hash value). Feature fusion: Plaintext features and encrypted features are fused through an attention mechanism. The attention weights are dynamically allocated by the encryption strength parameter L=1 (plaintext feature weight 0.8, encrypted feature weight 0.2). Model computation: Shared decision models use objective parameters The sharing probability of each evaluation data is predicted, and evaluation data with a probability ≥ 1.0 are selected. Preliminary shared results: Shared data candidate set: 720 reliable evaluation data (including the original 700 reliable evaluations + 20 evaluations that were misjudged as false and identified as true evaluations after model optimization); Data recipient list: Third-party evaluation and monitoring agencies (interface type: WebSocket); Shared interface identifier: WebSocket interface (port 8080, protocol version 13).

[0066] (6) Performance evaluation Shared accuracy (Acc): Of the 720 shared data entries, 710 were manually reviewed and confirmed as reliable, Acc = 710 / 720 ≈ 98.61%; Shared recall Rec: There are 730 actual reliable reviews in the target platform (including the original 700 reviews + 30 newly added reliable reviews that were not initially labeled), and the model successfully shared 710 reviews. Rec = 710 / 730 ≈ 97.26%; Shared latency (Lat): Average transmission latency for 720 data entries = 82ms (the real-time transmission advantage of the WebSocket protocol); Data security rate Sec: No unauthorized access during the sharing process, sensitive fields are encrypted and not leaked, Sec=100%; Threshold judgment: All indicators (Acc≈98.61%≥96%, Rec≈97.26%≥92%, Lat=82ms≤90ms, Sec=100%) meet the preset performance thresholds, and there is no need to trigger model self-iteration.

[0067] (7) Data sharing Interface adaptation: The interface adaptation middleware identifies the receiver's interface type as WebSocket, encapsulates the shared data into a binary stream format, and the data packet header includes the interface identifier (0x02), data length (720 data packets × 1KB = 720KB), and checksum (CRC32 = 0x12345678). Load balancing: The load balancer distributes shared requests to application interface node 2 (CPU utilization 65%, in an idle state). Data transmission: The binary stream data is pushed to the receiving server of the third-party evaluation and monitoring agency via a WebSocket long connection. The transmission rate is 10MB / s, and the total transmission time is approximately 720KB / 10MB / s ≈ 72ms. Log recording: The shared log content includes data_id (a list of unique identifiers for evaluation data), recipient ID (third_party_001), sharing time (2025-06-10 14:30:00), encryption level (L=1), transmission delay (82ms), and is synchronously stored in the MongoDB cluster and blockchain node (block height=10086).

[0068] (8) Dynamic monitoring and adaptive adjustment Scenario change: 3 months later, a large number of AI-generated fake reviews appeared on the target platform (new features: text similarity ≥90%, no actual semantics, images obtained by web crawlers). The KL divergence of the data feature distribution change was 0.45, and the normalized K2 was 0.48. Rate of change calculation: Given the current environmental dynamic feature vector U_curr=[0 (K1=0), 0.48 (K2), 1 (security level)], and the historical three-period mean vector U_hist=[0, 0.22, 1], then: Adaptive triggering: The dynamic monitoring module determines that the scene has changed significantly, triggering the data acquisition module to re-collect data (adding 500 AI-generated false evaluation data), and start a new round of parameter optimization and model self-iteration; Model self-iteration: Error data: The initial model's recognition accuracy for AI-generated evaluations was only 65%, and the error data consisted of 35% of misjudged sample features; Loss function calculation: L = 0.35 × (1 - 0.65) + 0.25 × (1 - 0.60) + 0.1 × (82 / 500) + 0.3 × (1 - 1.0) = 0.35 × 0.35 + 0.25 × 0.4 + 0.1 × 0.164 + 0 = 0.1225 + 0.1 + 0.0164 = 0.2389 > 0.05; Parameter update: Adaptive gradient descent algorithm is used. After 60 iterations, the loss function L=0.048≤0.05, the iteration stops, and the model network parameters are updated. New parameter adaptation: Updated target fusion weights Shared decision threshold The accuracy rate of identifying AI-generated fake reviews has been improved to 93%, successfully filtering 465 AI-generated reviews, and the reliability of the shared data candidate set remains above 98%.

[0069] 4. Implementation Results In this embodiment, the system automatically adapts to the new features of AI-generated false reviews through dynamic parameter optimization and model self-iteration, without the need for manual model retraining; the sharing accuracy reaches 98.61%, the sharing latency is 82ms, and the data security rate is 100%. Compared with the existing technology (60% accuracy in identifying AI-generated reviews, with a latency of 250ms), the accuracy of the new feature identification in this invention is improved by 55%, and the sharing efficiency is improved by 67.2%, effectively ensuring the authenticity and real-time nature of user review information sharing.

[0070] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A big data sharing method, characterized in that, include: S1: Obtain real-time business data and environmental dynamic data of the target shared scenario. The real-time business data includes multi-source heterogeneous data to be shared and data interaction characteristics. The environmental dynamic data includes scenario business rule change data, data feature distribution change data, and security level change data. S2: Perform standardized preprocessing on the real-time business data, perform quantitative encoding on the environmental dynamic data, and extract business feature vectors and environmental dynamic feature vectors respectively; S3: Based on the preset initial parameter set and the business feature vector and the environmental dynamic feature vector, a target parameter set adapted to the current scenario is calculated through a dynamic parameter optimization algorithm. The target parameter set includes data sharing weight parameters, model decision threshold parameters, and encryption strength parameters. S4: Based on the security level change data and encryption strength parameters, an adaptive encryption algorithm is used to perform hierarchical encryption processing on the standardized preprocessed real-time business data to obtain encrypted data and plaintext data. S5: Input the encrypted data, plaintext data and business feature vector into a preset sharing decision model, and use the target parameter set to drive the sharing decision model to output preliminary sharing results. The preliminary sharing results include a shared data candidate set, a data receiver list and a sharing interface identifier. S6: Collect feedback data of the preliminary sharing results, calculate the real-time performance evaluation index of the sharing decision model, and determine whether the real-time performance evaluation index meets the preset performance threshold. S7: If the real-time performance evaluation index does not meet the preset performance threshold, then based on the environmental dynamic feature vector and the error data of the preliminary shared results, the network parameters of the shared decision model are updated through the model self-iteration algorithm, and the process returns to step S5; if the real-time performance evaluation index meets the preset performance threshold, then step S8 is executed. S8: Based on the data receiver list and the sharing interface identifier, the shared data candidate set is output to the corresponding receiver through a cross-scenario compatible sharing protocol to complete data sharing; S9: Monitor the change rate of business data and the change rate of environmental dynamic data in the target shared scenario in real time. If either change rate exceeds the preset change threshold, return to step S1 and re-execute the parameter optimization and model adaptation process.

2. The big data sharing method according to claim 1, characterized in that, The calculation process of the dynamic parameter optimization algorithm in step S3 includes: S31: Extract the key influencing factors from the business feature vector. The key influencing factors include the data interaction frequency change rate Δf, the data interaction amount change rate ΔC, and the data sharing demand intensity S. S32: Extract the scene adaptation factor from the dynamic feature vector of the environment, the scene adaptation factor including the business rule change coefficient. Data feature evolution coefficient ; S33: For each parameter in the initial parameter set The target parameters are calculated using the following formula. : in, For the preset adjustment factor, satisfy ,and ; =(Current cycle interaction frequency - Historical average interaction frequency) / Historical average interaction frequency =(Current period interaction limit - Historical average interaction limit) / Historical average interaction limit It is calculated by weighting the frequency and urgency of the data recipient's sharing requests; When the business rules remain unchanged When business rules are completely updated It is calculated from the cosine similarity change of the data feature distribution.

3. The big data sharing method according to claim 1, characterized in that, The specific implementation process of step S2 includes: S21: Standardize and preprocess real-time business data: For numerical business data, the Z-score standardization formula is used for processing: in, The data is in the original numerical format. This is the historical average of this type of data. This represents the historical standard deviation of this type of data; For text-based business data, the TF-IDF algorithm is used to extract text features and standardize them into vector form: in, For the term In the document word frequency in For the term In the document The number of times it appears in This is the sum of the occurrences of all terms in document d; For the term Inverse document frequency, Total number of documents For included terms The number of documents; for image-type business data, normalization is used to map pixel values ​​to the [0,1] interval: ； S22: Perform quantization and encoding processing on dynamic environmental data: map the scenario business rule change data into discrete values ​​(no change: 0, partial change: 0.5, complete change: 1); calculate the distribution difference value of the data feature distribution change data through KL divergence and map it to the [0,1] interval as the data feature evolution coefficient; map the security level change data into quantization values ​​of 1-5 levels, corresponding to the encryption strength level.

4. The big data sharing method according to claim 1, characterized in that, The adaptive encryption algorithm mentioned in step S4 includes: S41: Determine the encryption level L (L∈{1,2,3,4,5}) based on the security level quantification value. S42: When L=1, the sensitive fields are encrypted using the AES-128 algorithm, while the non-sensitive fields are stored in plaintext. When L=2, the AES-192 algorithm is used to encrypt all business data, and the key length is 192 bits. When L=3, AES-256 algorithm is used for encryption, combined with hash verification (SHA-256) to ensure data integrity; When L=4, the AES-256 algorithm + homomorphic encryption algorithm is used, which supports finite operations in the ciphertext state; When L=5, the AES-256 algorithm + federated learning encryption protocol is used to make the data usable but invisible; S43: Optimize the key distribution mechanism based on the Ethereum interaction protocol, and generate dynamic keys using the following formula: in, The base key is T, and T is the timestamp (accurate to the second). As a unique identifier for the data recipient, It uses the SHA-256 hash function; the key validity period is positively correlated with the security level. When L=1, the validity period is 24 hours. For each level increase of L, the validity period is shortened by 50%.

5. The big data sharing method according to claim 1, characterized in that, The real-time performance evaluation metrics mentioned in step S6 include shared accuracy. Shared recall rate Shared latency Data security rate The specific calculation method is as follows: in, This represents the number of data entries in the initial shared results; The preset performance thresholds are: Acc≥95%, Rec≥90%, Lat≤100ms, Sec=100%. When all indicators meet the threshold requirements, it is determined that the preset performance thresholds are met.

6. The big data sharing method according to claim 1, characterized in that, The model self-iterative algorithm described in step S7 employs an adaptive gradient descent algorithm, specifically including: S71: Construct the loss function L, which is a multi-objective optimization function: in, For loss weights, satisfying ,and ; Maximum allowed sharing latency (default is 500ms); S72: Calculate the gradient ∇L of the loss function L with respect to the network parameters W of the shared decision model. in, This refers to the number of network parameters. S73: Update network parameters using adaptive learning rate. ,in, The initial learning rate (default is 0.001). This is the attenuation coefficient (default is 0.1). The initial loss value, 'This refers to the updated network parameters; S74: Repeat steps S71-S73 until the loss function L ≤ the preset loss threshold (preset to 0.05), or the number of iterations reaches the maximum iteration threshold (preset to 100), then stop the iteration and save the updated model parameters.

7. The big data sharing system according to claim 1, characterized in that, The cross-scenario compatible sharing protocol mentioned in step S8 includes: S81: Determine the interface type (REST interface, WebSocket interface, blockchain smart contract interface) based on the shared interface identifier. S82: Use corresponding data packet formats for different interface types: For REST interfaces, shared data is encapsulated in JSON format, with fields including data_id (unique data identifier), data_type (data type), data_content (data content, encrypted or plaintext), timestamp, and signature (digital signature). For the WebSocket interface, binary stream format is used for transmission, and the data packet header includes interface identifier, data length, and checksum. For blockchain smart contract interfaces, data is encapsulated according to the contract ABI specification, and the contract's shared execution function is called to complete the data upload to the chain and authorization by the recipient; S83: Establish interface adaptation middleware to automatically identify the receiver's interface type and convert the data packet format, ensuring cross-regional and cross-platform data interoperability and compatibility.

8. The big data sharing system according to claim 1, characterized in that, The calculation methods for the business data change rate R1 and the environmental dynamic data change rate R2 mentioned in step S9 are as follows: in, This is the feature vector of the current cycle's business. This represents the average of the business feature vectors over three historical periods. This represents the dynamic feature vector of the current periodic environment. The mean of the environmental dynamic feature vectors over three historical periods is given by ||・||2, which is the L2 norm. The preset change threshold is R1≥30% or R2≥20%. When this condition is met, it is determined that the scene has changed significantly, triggering the parameter re-optimization and model adaptation process.

9. A big data sharing system, characterized in that, The big data sharing method applied to any one of claims 1-8 includes: Data acquisition module: used to acquire real-time business data and environmental dynamic data of the target shared scenario. The real-time business data includes multi-source heterogeneous data to be shared and data interaction characteristics. The environmental dynamic data includes scenario business rule change data, data feature distribution change data, and security level change data. Data preprocessing module: used to perform standardized preprocessing on the real-time business data, quantitative encoding on the environmental dynamic data, and extract business feature vectors and environmental dynamic feature vectors respectively; Parameter optimization module: Based on a preset initial parameter set and the business feature vector and environmental dynamic feature vector, it calculates a target parameter set that is suitable for the current scenario through a dynamic parameter optimization algorithm. The target parameter set includes data sharing weight parameters, model decision threshold parameters, and encryption strength parameters. Adaptive encryption module: used to perform hierarchical encryption processing on standardized preprocessed real-time business data according to the security level change data and encryption strength parameters, and to obtain encrypted data and plaintext data; Model Decision Module: This module is used to input the encrypted data, plaintext data, and business feature vectors into a preset sharing decision model, and to drive the sharing decision model to output preliminary sharing results using the target parameter set. The preliminary sharing results include a candidate set of shared data, a list of data recipients, and a sharing interface identifier. Performance evaluation module: used to collect feedback data of the preliminary sharing results, calculate the real-time performance evaluation index of the sharing decision model, and determine whether the real-time performance evaluation index meets the preset performance threshold; Model Iteration Module: When the real-time performance evaluation index does not meet the preset performance threshold, it updates the network parameters of the shared decision model based on the environmental dynamic feature vector and the error data of the preliminary shared results through a model self-iteration algorithm. Shared execution module: When the real-time performance evaluation index meets the preset performance threshold, it outputs the shared data candidate set to the corresponding receiver based on the data receiver list and the shared interface identifier, and completes data sharing through a cross-scenario compatible sharing protocol; Dynamic monitoring module: Used to monitor the change rate of business data and the change rate of environmental dynamic data in the target shared scenario in real time. If either change rate exceeds the preset change threshold, the data acquisition module is triggered to reacquire data and start a new round of parameter optimization and model adaptation process.

10. The big data sharing system according to claim 9, characterized in that, The hardware architecture of the system includes: Edge acquisition layer: Composed of data acquisition terminals, sensors, and API gateways. The data acquisition terminals include industrial-grade IoT terminals and server clusters, used to collect real-time business data from multiple heterogeneous sources. The API gateway supports multiple protocols such as HTTP, HTTPS, and MQTT, and is used to connect to data source systems in the target shared scenario. Edge computing layer: Computing servers deployed on edge nodes, configured with CPUs and GPUs, used to perform low-latency computing tasks such as data preprocessing and dynamic parameter optimization; The cloud platform layer includes a cloud server cluster, a distributed database, and a blockchain node cluster. The cloud server cluster is used to deploy shared decision-making models and perform self-iterative training of the models. The distributed database is used to store real-time business data, dynamic environmental data, model parameters, and shared logs. The blockchain node cluster is used to store encrypted data hash values ​​and shared authorization records to ensure data traceability. Application Interface Layer: Composed of interface adaptation middleware and load balancer. The interface adaptation middleware supports adaptive conversion of REST, WebSocket, and blockchain smart contract interfaces; the load balancer is used to distribute shared requests and avoid single points of failure.

Images