Federal learning city information model system information security protection method based on block chain and differential privacy

Abstract

The invention provides a federated learning city information model system information security protection method based on a block chain and differential privacy, and the method comprises the steps: combining the decentralization, non-tampering and traceability characteristics of the block chain with a mathematical privacy protection mechanism of the differential privacy; according to the method, the privacy and safety problems caused by data sharing and model training in a city information model system are effectively solved. Specifically, the block chain technology is used for recording and verifying the model update of each node in the federal learning process, and ensuring the transparency of the process and the integrity of data; according to the differential privacy, noise is added when a local model is updated, so that private data of each participating node is prevented from being inferred or leaked. According to the method, dependence on a central trusted party is eliminated through a decentralized architecture, the security and autonomy of the system are improved, and the method is suitable for data security protection and efficient cooperative calculation of the city information model system under the background of a smart city.

Description

Technical Field

[0001] The present invention relates to the field of urban information management and data security, and particularly to an information security protection method for a federated learning urban information model system based on blockchain and differential privacy. Background Art

[0002] With the rapid advancement of smart city construction, the City Information Modeling (CIM) system has become a core tool to support urban planning, management, and operation. The CIM system provides a comprehensive and accurate digital model for urban management by integrating multi-source heterogeneous data. However, the CIM system faces serious information security and privacy protection challenges in the process of realizing data integration and sharing.

[0003] Traditional urban information management systems usually rely on centralized data processing methods. Such methods are not only vulnerable to attacks but may also lead to security problems such as data leakage and tampering. At the same time, urban management departments and other relevant agencies usually possess a large amount of sensitive data, such as traffic flow, environmental monitoring, energy consumption, etc. These data face the risk of privacy leakage when shared across departments and collaboratively analyzed. Therefore, how to achieve efficient urban information collaborative management while ensuring data privacy has become an urgent problem to be solved.

[0004] Federated Learning, as an emerging distributed machine learning framework, allows participating parties to jointly train a global model without exchanging local data, thus solving the data privacy problem to a certain extent. However, there are still potential risks of data leakage in federated learning, especially during the process of sharing model parameters. Attackers may obtain the original data of participating nodes through reverse inference. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides an information security protection method for a federated learning urban information model system based on blockchain and differential privacy. This method aims to solve the information security hidden dangers of traditional federated learning methods in the urban information model system by combining decentralized blockchain technology and advanced differential privacy protection mechanisms.

[0006] The technical solution of the present invention includes the following aspects: 1. Introduction of blockchain technology: Use blockchain to record the contributions and parameter update processes of each node during the federated learning process, so that after each participating node completes local calculations, it submits the model update parameters to the blockchain to ensure the transparency and immutability of the data and model update processes.

[0007] Automatically execute the parameter aggregation process in federated learning through smart contracts to ensure model updates and sharing without a central trusted party.

[0008] 2. Differential privacy protection mechanism: During the local computing process of each node, apply the differential privacy mechanism to the model update parameters to protect the privacy of training data by adding noise. The differential privacy parameters (such as noise intensity) can be adjusted according to the actual needs of the CIM system to achieve a balance between privacy protection and model performance.

[0009] Ensure that the model parameters publicly available on the blockchain do not reveal the private data of any single node through differential privacy, thereby improving the overall security of federated learning.

[0010] 3. System architecture design: The overall system architecture includes multiple distributed nodes, a blockchain network, and a federated learning coordination server. Each node represents a data owner, such as a city management department, an infrastructure operator, etc.

[0011] The federated learning coordination server is responsible for coordinating the federated learning process of each node and realizing the aggregation and update of model parameters through the blockchain network.

[0012] The specific technical solution of the present invention is as follows: An information security protection method for a federated learning urban information model system based on blockchain and differential privacy. The overall system architecture includes multiple distributed nodes, a blockchain network, and a federated learning coordination server. Each node represents a data owner. The federated learning coordination server is responsible for coordinating the federated learning process of each node and realizing the aggregation and update of model parameters through the blockchain network. At the same time, the differential privacy mechanism is used to process the node model parameters. The system federated learning process is as follows: Initialize the local federated learning model at each node in the urban information model system and perform local model training based on the local urban information data set held; After each node completes local model training, apply the differential privacy mechanism to perform noise processing on the model update parameters, specifically including calculating the gradient of the model parameters and adding noise to the gradient to prevent the leakage of original data during the model update process; Each node submits the model parameters processed by differential privacy to the federated learning coordination server through the blockchain network. The blockchain network records and verifies the submitted model parameters to ensure the integrity and immutability of the data; The federated learning coordination server automatically executes the aggregation of model parameters through smart contracts and generates a global model; The aggregated global model is distributed back to each node by the federated learning coordination server through the blockchain network. Each node updates its local model and conducts a new round of local model training, aggregation of model parameters, and update and distribution of the global model until the global model converges.

[0013] Furthermore, the blockchain network records the hash values, timestamps, and node identity information of the differentially private processed model parameters submitted by each node to ensure data transparency and traceability.

[0014] Furthermore, a privacy-utility trade-off mechanism is adopted during the noise addition process of differential privacy technology: when adding noise, the system weighs the relationship between privacy protection and model utility, and adjusts the noise intensity according to the requirements of the task to achieve a balance between privacy protection and model accuracy, preventing excessive noise from degrading the model performance.

[0015] Furthermore, smart contracts are pre-written with the logic for aggregating the model parameters of each node, and aggregation methods including weighted average and federated average are applied to the verified model parameters to generate the global model.

[0016] Furthermore, smart contracts adopt an incentive mechanism: rewards are distributed to honest nodes according to their participation and the quality of the submitted model parameters, encouraging nodes to actively participate in federated learning.

[0017] Furthermore, before submitting the differentially private processed model parameters to the blockchain network, the node encrypts the submitted data.

[0018] Furthermore, multiple training cycles are set in the federated learning process until the global model meets the preset convergence criteria. After that, the model update and distribution process is stopped, and the final global model is output by the federated learning coordination server.

[0019] Compared with the prior art, the beneficial effects of the present invention are: By leveraging the decentralized, immutable, and traceable characteristics of blockchain and combining them with the mathematical privacy protection mechanism of differential privacy, the present invention addresses the information security risks in traditional federated learning methods within urban information model systems. The method provided by the present invention is applicable to large-scale urban information model systems in the context of smart cities, and is used to achieve data collaboration and information sharing among urban management departments while ensuring data privacy and system security. That is, by combining the decentralized, immutable, and traceable characteristics of blockchain with the digital privacy protection mechanism of differential privacy, this method effectively solves the privacy and security issues brought about by data sharing and model training in urban information model systems. Specifically, blockchain technology is used to record and verify the model updates of each node during the federated learning process, ensuring the transparency of the process and the integrity of the data; differential privacy protects the private data of each participating node from being inferred or leaked by adding noise during local model updates. This method eliminates the dependence on a central trusted party through a decentralized architecture, improving the security and autonomy of the system, and is applicable to the data security protection and efficient collaborative computing of urban information model systems in the context of smart cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the overall architecture of the urban information model system of the present invention; Figure 2 It is a flowchart of the federated learning process of the present invention; Figure 3 It is a schematic diagram of the differential privacy protection mechanism of the present invention; Figure 4 It is an interaction schematic diagram between the blockchain and federated learning of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0022] The present invention proposes an information security protection method for a federated learning urban information model system based on blockchain and differential privacy. Figure 1 The figure shows the overall architecture of the federated learning urban information model system of the present invention based on blockchain and differential privacy. This architecture diagram mainly includes the following several parts: Node layer: Multiple distributed nodes, each node representing a different urban information management agency or department. Each node has a local dataset for local model training.

[0023] Federated learning coordination server: The coordination server communicates with each node through the blockchain network, collects local training model updates, and performs model aggregation operations.

[0024] Blockchain network: Connects all nodes and the coordination server, records the model updates and participation of each node, and ensures the transparency and immutability of data and model updates.

[0025] Global model: A global model generated after aggregating the local models of multiple nodes and distributed to each node.

[0026] Figure 2 Shows the main steps of federated learning for the information security protection method of the urban information model system, as follows: Step 1: Each node performs local model training.

[0027] Data preparation: Each node first performs data preprocessing on the urban information dataset held locally, including steps such as data cleaning, feature selection, and normalization. These datasets may include traffic data, energy consumption data, environmental monitoring data, etc., depending on the functions of the urban management department represented by each node.

[0028] Model training: After data preprocessing is completed, each node independently trains a machine learning model. Common model types include regression models, decision trees, neural networks, etc. Model training is only carried out locally to ensure that the original data does not leave the control range of the node.

[0029] Step 2: Apply differential privacy technology to add noise to the local model parameters.

[0030] Calculate model gradients: After training is completed, each node calculates the gradients or parameter update values of the model, which will be used for model optimization. The gradient information contains the direction and magnitude of how the model adjusts the parameters to reduce the error.

[0031] Noise addition: To protect the privacy of the local data of the node, differential privacy technology is applied to the calculated gradients or parameters. During this process, the system adds noise to the gradients or parameters according to a preset privacy budget (usually represented by the parameter ε). The addition of noise ensures that even if the model parameters are leaked, attackers cannot easily infer the information of the original data.

[0032] Step 3: Each node submits the processed model parameters to the coordination server through the blockchain network.

[0033] Data Encryption: Before submitting the differentially private processed model parameters to the blockchain network, the node first encrypts the data. The encryption operation ensures that the data remains unreadable even if intercepted during data transmission.

[0034] Data Submission: The node submits the encrypted model parameters to the blockchain network. During the submission process, the node interacts with other nodes on the blockchain network to ensure the legality and integrity of the submission operation.

[0035] Step 4: The blockchain network records the submission process and verifies the integrity of the data.

[0036] Block Generation: The model parameters submitted by each node are recorded on the blockchain, generating a new block. This block contains the submitted data, the identity of the submitting node, the timestamp, and the hash value of the data.

[0037] Consensus Verification: The blockchain network verifies the submitted model parameter data through a consensus mechanism. With the support of consensus mechanisms such as PoW (Proof of Work), PoS (Proof of Stake), or PBFT (Practical Byzantine Fault Tolerance), it ensures that only the verified and legitimate data will be recorded on the blockchain and become part of the global model update.

[0038] Step 5: The coordination server aggregates the model parameters through a smart contract to generate a global model.

[0039] The generation process of the global model does not rely on any single node or central server, but achieves decentralized management through the blockchain network and smart contracts.

[0040] Parameter Aggregation: When the blockchain network completes the verification of the model parameters submitted by each node, the federated learning coordination server starts the smart contract to aggregate all valid model parameters. The aggregation operation can be a simple weighted average or a more complex algorithm, depending on the specific requirements of the federated learning task.

[0041] Global Model Generation: By aggregating the model parameters of all participating nodes, a globally representative model is generated. This global model synthesizes the local data characteristics of each node and can more accurately reflect the state or trend of the entire urban information system.

[0042] Model Storage and Distribution: The generated global model is stored in the blockchain to ensure its immutability and distributed to each node through the blockchain network. After each node receives the global model, it continues the next round of local model training to further optimize the model performance.

[0043] Figure 3Shows the differential privacy protection mechanism of the urban information model system security protection method in the present invention, which specifically includes the following content: 1. Original data Node data holding: Each node represents a different urban information management agency or department and holds an urban information dataset in a specific domain. These datasets may include traffic flow data, air quality data, energy consumption data, population density data, etc. These data are only stored within the local node and are not shared externally to ensure the initial privacy of the data.

[0044] Data format: The original data usually exists in a structured or semi-structured format, including forms such as tables, database entries, sensor data, etc. The data format of each node may vary depending on the type of urban information it manages.

[0045] 2. Model training Local model training: Each node uses its local dataset to train an independent machine learning model. The model types can include but are not limited to linear regression, decision tree, support vector machine (SVM), neural network, etc. The training process is carried out within the node, and the model is optimized according to the local data characteristics, aiming to analyze and predict urban management tasks (such as predicting traffic flow and monitoring air quality).

[0046] Model parameter generation: During the training process, the node generates a series of model parameters (such as weights, biases, gradients, etc.), which reflect how the model makes predictions or classifications based on the input data. The model parameters are an important basis for aggregating the global model later.

[0047] 3. Noise addition Differential privacy mechanism: To protect the privacy of local data, before submitting the model parameters, the node first processes these parameters using differential privacy technology. Differential privacy prevents the leakage of sensitive information by adding noise. The magnitude of the noise is determined by the privacy budget ε value. The smaller the ε value, the stronger the privacy protection, and vice versa.

[0048] Noise addition process: The differential privacy mechanism is usually implemented by adding Laplace noise or Gaussian noise to each model parameter. After model training, the node calculates the gradient or parameter update value of the model and adds random noise to these values. The purpose of noise addition is to ensure that even if the model parameters are leaked, the attacker cannot infer the specific content of the original data.

[0049] Privacy-utility trade-off: When adding noise, the system will balance the relationship between privacy protection and model utility. Although noise can protect privacy, too much noise may lead to a decline in model performance. Therefore, the system will adjust the noise intensity according to the requirements of the task to achieve a balance between privacy protection and model accuracy.

[0050] 4. Secure Transmission Data Encryption: After differential privacy processing, the model parameters will be encrypted to ensure that they are not stolen or tampered with during transmission. Encryption techniques may include symmetric encryption (such as AES) or asymmetric encryption (such as RSA) to ensure the security of data during network transmission.

[0051] Blockchain Network Transmission: The encrypted model parameters are transmitted to the federated learning coordination server through the blockchain network. In the blockchain network, the submission of each node is recorded in the blockchain, generating a block. The distributed ledger technology of the blockchain ensures the transparency and immutability of data transmission, and all nodes can verify the legality of the data.

[0052] Data Integrity Verification: The blockchain network uses consensus mechanisms (such as PoW, PoS or PBFT) to verify whether the submitted data is complete and authentic. After verification, the data is written into the blockchain and used for the aggregation of the global model.

[0053] Figure 4 It shows the interaction process between the blockchain and federated learning in the urban information model system. The entire system combines blockchain technology and federated learning to ensure efficient collaborative computing on the premise of data privacy and security. The following details the key interaction steps in the figure: 1. Block Generation Model Update Submission: After each node independently conducts local model training, it submits the differentially private processed model parameters to the blockchain network. These parameters are uploaded in encrypted form to ensure that the data will not be leaked or tampered with during transmission.

[0054] Block Recording: The model updates submitted by each node will generate a new block. The block contains model parameters, timestamps, node identifiers, and other relevant metadata. The distributed ledger technology of the blockchain network will record these blocks to ensure the immutability and traceability of the data.

[0055] Blockchain Connection: The generated blocks are connected together through a chain structure to form a complete blockchain. This chain structure ensures the front-to-back relevance of each block. If any block is attempted to be tampered with, the integrity of the entire chain will be damaged, ensuring the security and transparency of the system.

[0056] 2. Consensus Mechanism Model Update Verification: When a node submits a model update, the blockchain network will start the consensus mechanism to verify the submitted data. Consensus mechanisms can use algorithms such as PoW (Proof of Work), PoS (Proof of Stake) or PBFT (Practical Byzantine Fault Tolerance) to ensure that only verified and legitimate data will be included in the blockchain.

[0057] Consensus node participation: Multiple consensus nodes (validators) participate in the verification of model updates and decide whether to write the submitted data to the blockchain through voting or computing power competition. The existence of the consensus mechanism ensures the decentralized operation of the system. Even if some nodes fail or act maliciously, the entire system can still operate stably.

[0058] Data integrity and security: After the consensus is passed, the submitted model parameters are written to the blockchain and a new block is officially generated. Through this verification mechanism, it is ensured that all model updates submitted by participating nodes are legal and trustworthy, preventing tampering and the introduction of malicious data.

[0059] 3. Global model update Smart contract execution: Once the blockchain network completes the verification of the model parameters submitted by all nodes, the federated learning coordination server will perform the aggregation operation of the model parameters through a smart contract. The smart contract is automatically executed according to pre-set rules without manual intervention, ensuring the transparency and automation of the entire process.

[0060] Global model aggregation: Under the control of the smart contract, the server performs a weighted average or other specified aggregation algorithm on all verified local model parameters to generate a global model. This global model synthesizes the local data characteristics of each node and can provide more accurate and extensive analysis results for urban information management.

[0061] Global model storage: The generated global model is stored through the blockchain. Due to the immutability of the blockchain, once the global model is generated and stored, no one can modify it. The record of the global model ensures the transparency and traceability of the results of each round of model training, providing a basis for future data auditing and system verification.

[0062] 4. Global model distribution Model update broadcast: When the global model is generated, the federated learning coordination server distributes the global model to each participating node through the blockchain network. Each node receives and applies the latest global model to continue the next round of local model training and update, ensuring the synchronization and collaboration of the system.

[0063] Local model update: After receiving the global model, the node updates it with the local model to further improve the accuracy of the local model. Through multiple rounds of model training and aggregation, the global model of the entire system will gradually converge, ultimately achieving a highly accurate urban information prediction or analysis model.

[0064] In specific implementation, the method of the present invention can be realized through the following steps: Step 1: Build the system architecture, specifically including: Node initialization: Each city information management node initializes its own federated learning model and trains the model based on the local dataset. Each node registers with the blockchain network during system initialization to generate a unique public-private key pair for subsequent identity authentication and data encryption.

[0065] Federated learning coordination server: The federated learning coordination server serves as the central coordination entity and communicates with all nodes through the blockchain network. It is responsible for collecting the model update parameters of each node and performing parameter aggregation through smart contracts to generate the global model. This server does not directly access the raw data of any node, ensuring data privacy.

[0066] During the aggregation process, the federated learning coordination server does not directly access or store the raw data of any node. All data exchanges are carried out through the differentially private processed model parameters to further enhance the privacy protection ability of the system.

[0067] The functions of the federated learning coordination server include: receiving and storing the model parameters submitted by each node; using smart contracts to perform the aggregation operation of the model parameters; generating and distributing the global model to each node.

[0068] Application of the blockchain network: The blockchain network connects all nodes and the coordination server and records the model update process of each node. Each block in the blockchain contains the differentially private processed model parameters, timestamps, and hash values uploaded by the nodes to ensure data transparency and immutability.

[0069] Global model update and distribution: After the coordination server completes the aggregation of the global model, the global model is distributed to all nodes through the blockchain network. Each node receives the updated global model and continues the next round of training based on the local data. This process continues until the model reaches the preset convergence standard.

[0070] Step 2: Add a differential privacy protection mechanism, specifically as follows: Local model training and noise addition: Each node independently trains the model based on the local dataset. To protect data privacy, after training is completed, differential privacy processing is applied to the model parameters. In the differential privacy mechanism, first, the gradient of the model parameters is calculated, and then noise is added to the gradient according to the differential privacy algorithm. The intensity of the added noise can be adjusted according to the preset privacy budget to ensure that privacy is protected without significantly affecting the accuracy of the model.

[0071] Secure Submission of Model Parameters: The differentially private processed model parameters are submitted to the coordination server through the blockchain network. During the submission process, the blockchain network records and verifies the submitted data to ensure the integrity and legality of the data. The model parameters submitted by each node are stored in a new block of the blockchain and jointly maintained and verified by all participating nodes.

[0072] Step 3: Enable the smart contract to generate the global model, as follows: Smart Contract Executes Parameter Aggregation: The smart contract is pre-written with the logic for aggregating the model parameters of each node. When the model parameters submitted by each node are verified on the blockchain, the smart contract will automatically execute, performing weighted averaging or other specified aggregation methods on the model parameters of all nodes to generate the global model.

[0073] Update and Broadcast of the Global Model: The aggregated global model is broadcast by the coordination server to all nodes through the blockchain network. After receiving the new global model, the nodes update their local models and start the next round of training. This decentralized model update mechanism ensures that the system can continue to operate stably even if some nodes fail or drop out.

[0074] Security and Privacy Protection: Throughout the process, the differential privacy mechanism ensures that the local data of each node will not be leaked due to the submission of model parameters. Blockchain technology guarantees the transparency, integrity, and immutability of the model update process. In addition, the automatic execution of the smart contract eliminates potential security risks brought by human operations, further improving the overall security of the system.

[0075] Blockchain technology, with its characteristics of decentralization, immutability, and traceability, provides new guarantees for data security. By introducing blockchain technology into federated learning, it is possible to achieve transparent recording and verification of the model update process of each node, eliminate the dependence on a central trusted party, and prevent data tampering and forgery.

[0076] Differential privacy technology further enhances the level of data protection. By adding noise to the model parameters, differential privacy can effectively prevent attackers from inferring the original data by analyzing the model parameters. Combining differential privacy technology in the federated learning framework can significantly reduce the risk of data leakage and ensure the privacy of the data of each node.

[0077] Therefore, applying blockchain and differential privacy technologies to the federated learning urban information model system can not only enhance the security and privacy protection capabilities of the system but also enable secure sharing and collaborative computing of data among multiple urban management departments, providing strong technical support for the development of smart cities.

[0078] The present invention has been described in detail through the embodiments above. However, the above content is only an exemplary embodiment of the present invention and cannot be considered as defining the scope of implementation of the present invention. The protection scope of the present invention is defined by the claims. All those who utilize the technical solutions of the present invention, or those skilled in the art who, inspired by the technical solutions of the present invention, design similar technical solutions within the essence and protection scope of the present invention to achieve the above technical effects, or make equivalent changes and improvements to the scope of the application, etc., should still fall within the scope of patent protection covered by the present invention. It should be noted that, for the sake of clear expression, the description of the present invention omits the description of some components and processes that are not directly and obviously related to the protection scope of the present invention but are known to those skilled in the art.

Claims

1. A method for information security protection of a federated learning urban information model system based on blockchain and differential privacy, characterized in that, The overall system architecture includes multiple distributed nodes, a blockchain network, and a federated learning coordination server. Each node represents a data owner. The federated learning coordination server is responsible for coordinating the federated learning process among the nodes, and realizing the aggregation and update of model parameters through the blockchain network. At the same time, a differential privacy mechanism is adopted to process the node model parameters. The system's federated learning process is as follows: Each node in the city information model system initializes its local federated learning model and conducts local model training based on the local city information dataset it holds; After each node completes local model training, the differential privacy mechanism is applied to perform noise processing on the model update parameters, specifically including calculating the gradient of the model parameters and adding noise to the gradient to prevent the leakage of original data during the model update process; Each node submits the model parameters processed by differential privacy to the federated learning coordination server through the blockchain network. The blockchain network records and verifies the submitted model parameters to ensure the integrity and immutability of the data; The federated learning coordination server automatically executes the aggregation of model parameters through smart contracts and generates a global model; The aggregated global model is distributed back to each node by the federated learning coordination server through the blockchain network. Each node updates its local model and conducts a new round of local model training, aggregation of model parameters, and update and distribution of the global model until the global model converges.

2. The method according to claim 1, wherein The blockchain network records the hash values, timestamps, and node identity information of the model parameters processed by differential privacy submitted by each node to ensure the transparency and traceability of the data.

3. The method according to claim 1, characterized in that, A privacy-utility trade-off mechanism is adopted in the noise addition process of the differential privacy technology: when adding noise, the system will balance the relationship between privacy protection and model utility. The system will adjust the noise intensity according to the requirements of the task to achieve a balance between privacy protection and model accuracy, and prevent excessive noise from degrading the model performance.

4. The method according to claim 1, wherein The smart contract is pre-written with the logic for aggregating the model parameters of each node, and uses aggregation methods including weighted average and federated average for the verified model parameters to generate a global model.

5. The method according to claim 1, characterized in that, The smart contract adopts an incentive mechanism: according to the participation of nodes and the quality of the submitted model parameters, rewards are distributed to honest nodes to encourage nodes to actively participate in federated learning.

6. The method according to claim 1, characterized in that, Before submitting the model parameters processed by differential privacy to the blockchain network, the node encrypts the submitted data.

7. The method according to claim 1, characterized in that, The federated learning process sets multiple training cycles until the global model reaches the preset convergence standard. After that, the model update and distribution process is stopped, and the final global model is output by the federated learning coordination server.

Images