Public security data collaborative analysis method based on differential privacy and block chain

Abstract

The invention discloses a public security data collaborative analysis method based on differential privacy and a block chain, and the method comprises the following steps: S1, constructing a disturbance configuration table by a first collaborative party, and configuring a disturbance trajectory compression Laplacian mechanism for trajectory data; s2, a second collaboration party formulates an analysis task request and determines a data type; s3, the third collaboration party writes the disturbance configuration table and the task request into the block chain through the task declaration smart contract; s4, each data owner performs differential disturbance on local data according to the disturbance configuration table, and writes the local data into the block chain through a data declaration smart contract; s5, the second collaboration party collects disturbance data and constructs a joint model, and submits a training intermediate result abstract to the block chain; and S6, the second collaboration party sends an encryption result to the first collaboration party after completing joint model training, and a third collaboration party records a result abstract and supports initiation of an on-chain correction request. According to the invention, differential privacy and block chain technologies are combined, and a cooperative analysis method oriented to public security data is provided.

Description

Technical Field

[0001] The present invention relates to the field of privacy protection technology, and in particular to a public security data collaborative analysis method based on differential privacy and blockchain. Background Art

[0002] With the continuous advancement of informatization in the public security sector, various departments (such as public security, fire protection, emergency response, and medical services) have generated a large amount of structured or semi-structured data. This data contains sensitive information such as the time, location, handling process, path trajectory, and post-event evaluation indicators of incidents. Joint analysis of this multi-source, heterogeneous data can assist decision-makers in situation assessment, risk warning, and resource allocation. However, due to the high sensitivity of this data, cross-agency data sharing and analysis are often faced with serious privacy risks, chaotic permission management, and uncontrollable data tampering, which seriously hinder the collaborative utilization of public security data.

[0003] To mitigate privacy risks in data sharing, differential privacy (DP), a mathematically formalized privacy protection mechanism, has been gradually introduced into public security data analysis in recent years. By injecting random noise into query results, differential privacy effectively prevents attackers from inferring individual privacy by comparing adjacent datasets. However, in traditional practice, differential privacy often relies on a centralized data processing architecture, where a single data aggregator uniformly adds perturbations and outputs analysis results. This model not only requires data owners to relinquish control of the original data but also poses the risk of single-point leakage, violating the requirements for distributed data governance across multiple institutions.

[0004] On the other hand, with the rapid development of blockchain technology, its immutable, traceable, and decentralized nature provides a trusted, neutral platform for multi-party data collaboration. Some research has attempted to apply blockchain to scenarios such as medical data sharing and traffic data trading, achieving a certain degree of trusted interaction through on-chain data registration, contract execution, and traceability control. However, existing blockchain-based data collaboration mechanisms still mostly focus on static data controls, such as data on-chain or data access authorization, and lack on-chain full-cycle audit capabilities for analytical tasks. This is particularly true for privacy-preserving modeling tasks, making it difficult to achieve on-chain binding and trusted oversight of perturbation strategies, model behavior, intermediate results, and final outputs.

[0005] Furthermore, among real-world public safety data types, trajectory data, as a typical high-dimensional spatiotemporal sequence, is not only extremely privacy-sensitive but also complex in dimensionality and involves a large number of raw points. Directly injecting noise can lead to a sharp decline in data quality, impacting subsequent analyses such as clustering and path prediction. Current differential privacy approaches for trajectory data often employ the standard Laplace mechanism, injecting independent noise at each trajectory point. While this mechanism meets theoretical protection requirements, it can easily lead to severe distortions in the trajectory structure, making it impossible to maintain the continuity and directional characteristics of the behavioral path. Furthermore, trajectory compression and differential perturbation have yet to form an efficient synergistic mechanism, lacking a systematic optimization framework between compression rate control, perturbation intensity adaptation, and trajectory reconstructability.

[0006] In terms of joint analysis models, existing approaches often employ centralized modeling, relying on central nodes to aggregate data or parameters. This not only increases the risk of data leakage but also violates privacy protection requirements. While some federated learning frameworks support multi-party collaborative training, they lack mechanisms for identifying and adapting to perturbed data in public safety scenarios. The model training process lacks the ability to track and adjust perturbation budget usage, model behavior, and the credibility of results. Furthermore, existing training processes mostly run off-chain, making the intermediate training process opaque and difficult to audit for subsequent decision-making.

[0007] Regarding contract mechanisms, while existing literature has proposed using smart contracts for task registration and access control, specific sub-processes within differential privacy task execution, such as perturbation strategy registration, task data declaration, model intermediate state writing, and result credibility feedback, lack standardized design. This fails to meet the multi-role, multi-state, and multi-constraint data collaboration requirements of real-world public safety analysis processes. Furthermore, existing contract structures generally lack task number binding mechanisms and perturbation label traceability mechanisms, making it difficult to accurately trace back process accountability for analysis results.

[0008] Overall, existing public safety data collaborative analysis technologies still have significant shortcomings in terms of privacy protection, trajectory data structure fidelity, modeling process auditing, and full-process chain management of tasks. Differential privacy mechanisms and blockchain technology have yet to achieve deep integration, lacking collaborative control capabilities at the structural, process, and outcome levels, and thus cannot effectively support the practical needs of secure multi-party data sharing and trusted joint analysis.

[0009] Therefore, how to provide a collaborative analysis method for public security data based on differential privacy and blockchain is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0010] One purpose of the present invention is to propose a collaborative analysis method for public safety data based on differential privacy and blockchain. The present invention integrates differential privacy and blockchain technology to propose a collaborative analysis method for multi-source public safety data. It adopts a perturbation configuration table and a task declaration contract to realize the on-chain registration of the entire analysis task process, combines trajectory compression and noise addition mechanisms to improve the availability of trajectory data, and realizes efficient training and trusted reasoning of perturbation data through a joint model. It has the advantages of strong privacy protection, reconstructable trajectory, auditable analysis process, and trusted and correctable results. It is suitable for joint modeling and intelligent analysis scenarios of sensitive data with the participation of multiple institutions.

[0011] The public security data collaborative analysis method based on differential privacy and blockchain according to an embodiment of the present invention includes the following steps:

[0012] S1. The first collaborator sets the differential privacy parameters and perturbation adjustment factors, builds a perturbation configuration table, and configures the perturbation trajectory compressed Laplace mechanism for trajectory data.

[0013] S2. The second collaborator formulates an analysis task request, determines the data type, and submits it to the third collaborator;

[0014] S3. The third party will write the perturbation configuration table and task request into the blockchain through the task declaration smart contract to complete the task registration.

[0015] S4. Each data owner performs differential perturbation on the local data according to the perturbation configuration table. Trajectory data is compressed and noise is added to the start and end points. A data summary with perturbation labels is generated and written to the blockchain through the data declaration smart contract.

[0016] S5. The second collaborator collects perturbation data and builds a joint model, and submits a summary of the intermediate training results to the blockchain;

[0017] S6. After completing the joint model training, the second collaborator sends the encrypted result to the first collaborator, who decrypts it and generates the final result. The third collaborator records the result summary and supports initiating an on-chain correction request.

[0018] Optionally, the first collaborator sets differential privacy parameters and a disturbance adjustment factor, and constructing a disturbance configuration table specifically includes:

[0019] The first collaborator evaluates the sensitivity of each type of data to be analyzed and defines a local sensitivity upper bound Δ for each type of data, which is used to represent the maximum output change caused by any two adjacent input records under the same query function;

[0020] Set the differential privacy parameter ε>0, and set the perturbation adjustment factor α∈[0.05,0.5] according to the data type and sensitivity level to adjust the intensity of the noise injected into each type of data during perturbation;

[0021] The local sensitivity upper bound Δ, differential privacy parameter ε, and perturbation adjustment factor α are combined into a triplet and mapped with the corresponding data types to construct a perturbation configuration table.

[0022] Optionally, configuring a perturbation trajectory compressed Laplace mechanism for trajectory data in S1 specifically includes the following steps:

[0023] S11. Receive the original trajectory data sequence and represent it as a set of trajectory points T arranged in ascending time order:

[0024] T={(x1,y1,t1),(x2,y2,t2),…,(x n ,y n ,t n )};

[0025] Among them, (x i ,y i ) represents the two-dimensional coordinates of the i-th trajectory point, t i represents the timestamp of the i-th trajectory point, n represents the total number of trajectory points, and n≥3;

[0026] S12. Set three trajectory compression control parameters, including the Euclidean distance threshold θ>0, the direction change threshold φ∈(0,π), and the sliding window length w∈N + , which are used to limit the maximum offset, maximum direction change angle and window step size of the trajectory points respectively;

[0027] S13. Perform a sliding window scan on the trajectory point set T, extract any three consecutive trajectory points to determine the trajectory stability section, and calculate the direction change angle of the intermediate trajectory point:

[0028]

[0029] Among them, ∠ i Indicates the direction change angle of the intermediate trajectory point, (x i-1 ,y i-1 ) represents the two-dimensional coordinates of the i-1th trajectory point, (x i+1 ,y i+1 ) represents the two-dimensional coordinates of the i+1th trajectory point, arccos represents the inverse cosine function;

[0030] If the conditions are met∠ i <φ and It is determined that the compression condition is met;

[0031] S14, for the continuous trajectory point sequence that meets the compression condition, retain the trajectory point (x' k ,y' k ,t' k), where (x′ k ,y′ k ) represents the starting or ending coordinates of the continuous trajectory point sequence that meets the compression condition, t′ k The timestamps of the continuous trajectory point sequences that meet the compression conditions constitute the compressed trajectory set T':

[0032] T′={(x′1,y′1,t′1),(x′2,y′2,t′2),…,(x′ m ,y′ m ,t′ m )};

[0033] Among them, m represents the total number of trajectory points after compression, satisfying m <n;

[0034] S15. Calculate the trajectory compression ratio r, which is used to measure the compression effect:

[0035]

[0036] S16, for each trajectory point (x' k ,y' k )Inject differential privacy Laplace noise to generate perturbed coordinates:

[0037]

[0038] in, and represents the trajectory coordinates after perturbation, Δ represents the upper bound of local sensitivity, Lap represents the Laplace distribution random variable, ε represents the differential privacy parameter, and α represents the perturbation adjustment factor;

[0039] S17. Generate a meta-information label package for the perturbed trajectory set.

[0040] Optionally, the meta-information tag package includes an alpha-tag, a ratio-tag, and a hash-track, wherein the alpha-tag is used to record the value of the disturbance adjustment factor, the ratio-tag is used to record the value of the trajectory compression ratio, and the hash-track is used to generate a complete trajectory summary for the perturbed trajectory set using a hash function, which is used for result verification and non-repudiation auditing.

[0041] Optionally, the S2 specifically includes:

[0042] S21. The second collaborator determines the analysis target based on the public safety application requirements and generates the task request content;

[0043] S22. The second collaborating party identifies different data types and determines the data type;

[0044] S23, encapsulating the task request content and the corresponding data type tag into a task structure to be registered, and attaching a digital signature to confirm the legal initiation identity of the second collaborating party for the task;

[0045] S24. Submit the signed task structure to the third collaborator, who then writes the task request into the blockchain to form a task statement.

[0046] Optionally, the S3 specifically includes:

[0047] S31. The third collaborator calls the initialization interface of the task declaration smart contract, and loads the perturbation configuration table set by the first collaborator and the analysis task request submitted by the second collaborator as parameters into the contract execution environment;

[0048] S32. The task declaration smart contract verifies the integrity of the task content according to the preset logic, including whether the task identifier is repeated, whether the configuration policy is missing, and whether the task status is in the registration stage;

[0049] S33. The task declaration smart contract automatically generates a task declaration record after passing verification and updates the task status from pending registration to registered;

[0050] S34. After the task declaration smart contract is updated, the task declaration record is written into the blockchain ledger, and the on-chain evidence operation is performed to generate a unique task registration hash and timestamp;

[0051] S35. After the task declaration smart contract is executed, the status information of task registration completion is returned.

[0052] Optionally, the S4 specifically includes:

[0053] S41. After completing the local differential perturbation processing, the data owner constructs a data summary;

[0054] S42: Call the data declaration smart contract deployed by the third party, and submit the data summary as a parameter to the data declaration smart contract call entry;

[0055] S43. After receiving the call request, the data declaration smart contract verifies whether the data summary structure is complete, whether the caller has registration authority, and checks whether the task binding status matches the current task declaration;

[0056] S44. After verification, the data declaration smart contract creates an on-chain declaration record, sets the record status to valid, and binds the data declaration to the unique identifier of the registered task;

[0057] S45. After the data declaration smart contract is successfully written, the declaration number, writing timestamp and blockchain transaction index information are automatically generated;

[0058] S46. Return the confirmation information of the registration completion to the calling data owner, and synchronize the registration result to the third-party collaborator for audit verification and task traceability.

[0059] Optionally, the data summary structure includes a data disturbance label, a data type identifier, a data source identifier and processing time information.

[0060] Optionally, the joint model is a data collaborative training model that supports differential privacy, and has distributed perturbation data access capabilities and parameter update mechanisms. The input end of the joint model includes multiple perturbation data channels, each channel corresponds to a data source of a data owner, and the channel supports unified access and format standardization processing of structured data and unstructured data.

[0061] Optionally, the joint model includes a privacy control module, a training control module and a parameter aggregation module. The privacy control module is used to identify perturbation labels in the input data and automatically adapt the perturbation intensity. The training control module is used to allocate the privacy budget, schedule the number of training rounds and the step size. The parameter aggregation module is used to jointly update the local model gradients of each perturbation data channel.

[0062] The beneficial effects of the present invention are:

[0063] First, the present invention implements a differentiated perturbation control strategy for multiple types of data by setting differential privacy parameters and perturbation adjustment factors through a first collaborating party and constructing a perturbation configuration table. Specifically, the perturbation trajectory compression Laplace mechanism proposed for trajectory data effectively reduces noise redundancy and trajectory deformation in trajectory perturbations through trajectory point compression, direction angle control, and start and end point noise addition, while preserving the continuity and main behavioral trends of the original trajectory. The perturbation adjustment label, compression ratio label, and trajectory perturbation summary generated on this basis not only improve the usability of trajectory data but also provide an audit basis and structural verification capabilities for perturbation behavior.

[0064] Secondly, during the analysis and modeling phase, this invention constructs a joint model through a second collaborator, introducing a model structure that supports multi-channel input, perturbation label adaptation, training control, and parameter aggregation, thus achieving differentially private joint modeling of multi-party data. During training, the model dynamically adjusts its training behavior based on the perturbation labels of different data sources, and generates summaries of the perturbation budget usage, model error changes, and local contributions during each training round. These intermediate training summaries are promptly submitted to a third collaborator for registration on the blockchain, ensuring the verifiability and non-repudiation of the entire modeling process, providing a technical foundation for trustworthy analysis in the public security sector.

[0065] In addition, the present invention constructs a full-process collaborative mechanism on the chain for task initialization, data submission, model process audit and result feedback through the task declaration smart contract and data declaration smart contract deployed by the third collaborator. The task declaration smart contract realizes the binding registration of the analysis task and the privacy policy, ensuring that all subsequent data and model operations are carried out based on a unified privacy policy; the data declaration smart contract records the metadata of each piece of disturbed data, and clarifies the boundaries of data use and responsibility. At the same time, after the model is completed, the first collaborator decrypts and outputs the analysis results, and the result summary is registered on the chain by the third collaborator. If the data owner has any objection to the result, he can initiate a correction request on the chain and start the trusted reconstruction process, further enhancing the closed-loop control capability of the system and the multi-party trust foundation. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0067] Figure 1 This is the overall flow chart of the public security data collaborative analysis method based on differential privacy and blockchain proposed in this invention;

[0068] Figure 2 A flowchart of constructing a perturbation configuration table and configuring a trajectory data perturbation mechanism for the first collaborative party proposed in the present invention;

[0069] Figure 3 Flowchart of the process of executing the perturbed trajectory compressed Laplace mechanism for trajectory data proposed in the present invention. DETAILED DESCRIPTION

[0070] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0071] refer to Figure 1 , a collaborative analysis method for public security data based on differential privacy and blockchain, including the following steps:

[0072] S1. The first collaborator sets the differential privacy parameters and perturbation adjustment factors, builds a perturbation configuration table, and configures the perturbation trajectory compressed Laplace mechanism for trajectory data.

[0073] S2. The second collaborator formulates an analysis task request, determines the data type, and submits it to the third collaborator;

[0074] S3. The third party will write the perturbation configuration table and task request into the blockchain through the task declaration smart contract to complete the task registration.

[0075] S4. Each data owner performs differential perturbation on the local data according to the perturbation configuration table. Trajectory data is compressed and noise is added to the start and end points. A data summary with perturbation labels is generated and written to the blockchain through the data declaration smart contract.

[0076] S5. The second collaborator collects perturbation data and builds a joint model, and submits a summary of the intermediate training results to the blockchain;

[0077] S6. After completing the joint model training, the second collaborator sends the encrypted result to the first collaborator, who decrypts it and generates the final result. The third collaborator records the result summary and supports initiating an on-chain correction request.

[0078] This method combines differential privacy with blockchain technology to construct a complete process system for collaborative analysis of multi-source public safety data, supporting reliable model training and result acquisition without exposing the original data. By setting up three types of collaborators, the division of labor and coordination among privacy policy formulation, data registration, task scheduling, model execution, and audit feedback is achieved, improving the controllability and auditability of the system. The differential privacy mechanism is used to protect local data from disturbances, while the blockchain ensures that the data declaration, task registration, and result feedback processes cannot be tampered with. This method has high credibility and traceability, making it particularly suitable for public safety scenarios such as public security, emergency response, and medical care, which have high requirements for data security and result verifiability.

[0079] In this embodiment, the first collaborator sets the differential privacy parameters and the perturbation adjustment factor, and constructs the perturbation configuration table, specifically including:

[0080] The first collaborator evaluates the sensitivity of each type of data to be analyzed and defines a local sensitivity upper bound Δ for each type of data, which is used to represent the maximum output change caused by any two adjacent input records under the same query function;

[0081] Set the differential privacy parameter ε>0, and set the perturbation adjustment factor α∈[0.05,0.5] according to the data type and sensitivity level to adjust the intensity of the noise injected into each type of data during perturbation;

[0082] The local sensitivity upper bound Δ, differential privacy parameter ε, and perturbation adjustment factor α are combined into a triplet and mapped with the corresponding data types to construct a perturbation configuration table.

[0083] By having the first collaborator set differential privacy parameters and perturbation adjustment factors and construct a perturbation configuration table, it is possible to achieve policy centralization and dynamic parameter management of the differential privacy mechanism. This configuration table clearly calibrates the upper bounds of the sensitivity of different data types and maps differential privacy parameters and perturbation adjustment factors for each type of data, ensuring that each type of data is protected against differential perturbations based on its sensitivity. In a federated data modeling environment, this approach can balance model accuracy and privacy strength while providing consistency guarantees for subsequent data perturbation processing, model adaptation, and trusted labeling. The standardization of this policy configuration enables multiple parties to complete perturbation consistency negotiations before participating in data collaboration, providing an institutional foundation for system-level privacy control and auditing.

[0084] In this embodiment, the perturbation trajectory compressed Laplace mechanism is configured for trajectory data in S1, which specifically includes the following steps:

[0085] S11. Receive the original trajectory data sequence and represent it as a set of trajectory points T arranged in ascending time order:

[0086] T={(x1,y1,t1),(x2,y2,t2),…,(x n ,y n ,t n )};

[0087] Among them, (x i ,y i ) represents the two-dimensional coordinates of the i-th trajectory point, t i represents the timestamp of the i-th trajectory point, n represents the total number of trajectory points, and n≥3;

[0088] S12. Set three trajectory compression control parameters, including the Euclidean distance threshold θ>0, the direction change threshold φ∈(0,π), and the sliding window length w∈N + , which are used to limit the maximum offset, maximum direction change angle and window step size of the trajectory points respectively;

[0089] S13. Perform a sliding window scan on the trajectory point set T, extract any three consecutive trajectory points to determine the trajectory stability section, and calculate the direction change angle of the intermediate trajectory point:

[0090]

[0091] Among them, ∠ i Indicates the direction change angle of the intermediate trajectory point, (x i-1 ,y i-1 ) represents the two-dimensional coordinates of the i-1th trajectory point, (x i+1 ,y i+1 ) represents the two-dimensional coordinates of the i+1th trajectory point, arccos represents the inverse cosine function;

[0092] If the conditions are met∠ i <φ and It is determined that the compression condition is met;

[0093] S14. For the continuous trajectory point sequence that meets the compression condition, retain the trajectory point (x′ k ,y′ k ,t′ k ), where (x′ k ,y′ k ) represents the starting or ending coordinates of the continuous trajectory point sequence that meets the compression condition, t′ k The timestamps of the continuous trajectory point sequences that meet the compression conditions constitute the compressed trajectory set T′:

[0094] T′={(x′1,y′1,t′1),(x′2,y′2,t′2),…,(x′ m ,y′ m ,t′ m )};

[0095] Among them, m represents the total number of trajectory points after compression, satisfying m <n;

[0096] S15. Calculate the trajectory compression ratio r, which is used to measure the compression effect:

[0097]

[0098] S16, for each trajectory point (x′ k ,y′ k )Inject differential privacy Laplace noise to generate perturbed coordinates:

[0099]

[0100] in, and represents the trajectory coordinates after perturbation, Δ represents the upper bound of local sensitivity, Lap represents the Laplace distribution random variable, ε represents the differential privacy parameter, and α represents the perturbation adjustment factor;

[0101] S17. Generate a meta-information label package for the perturbed trajectory set.

[0102] By introducing a perturbed trajectory compression Laplace mechanism into trajectory data, a structurally optimized perturbation strategy is implemented for high-frequency trajectory data. This mechanism not only effectively reduces the volume of perturbed data by combining sliding windows and directional angles to compress trajectory points, but also preserves the main trend of trajectory structure. Furthermore, Laplace noise is injected into the start and end points of the compressed segments to further enhance privacy protection. This method significantly reduces the cumulative perturbation error of trajectory data and improves the trainability and stability of trajectory models. It is particularly suitable for public safety scenarios involving continuous spatiotemporal data processing, such as traffic flow analysis and path prediction, and has strong engineering applicability.

[0103] In this embodiment, the meta-information tag package includes alpha-tag, ratio-tag, and hash-track. The alpha-tag is used to record the value of the disturbance adjustment factor, the ratio-tag is used to record the value of the trajectory compression ratio, and the hash-track is used to generate a complete trajectory summary for the perturbed trajectory set using a hash function for result verification and non-repudiation audit.

[0104] By introducing three types of metadata tags—alpha-tag, ratio-tag, and hash-track—the system not only transparently identifies the source of disturbances for each segment of perturbed trajectory data but also enhances trustworthy matching capabilities in subsequent model processing. The alpha-tag records the perturbation intensity configuration, the ratio-tag reflects the degree of compression, and the hash-track verifies trajectory integrity. This allows analysts to accurately assess data quality and perturbation risk during model training and result backtracking, improving model credibility and interpretability. Furthermore, the hash-track provides data non-repudiation evidence for subsequent on-chain audits, preventing data tampering or intermixing, and enhancing data compliance and accountability throughout the data lifecycle.

[0105] In this embodiment, S2 specifically includes:

[0106] S21. The second collaborator determines the analysis target based on the public safety application requirements and generates the task request content;

[0107] S22. The second collaborating party identifies different data types and determines the data type;

[0108] S23, encapsulating the task request content and the corresponding data type tag into a task structure to be registered, and attaching a digital signature to confirm the legal initiation identity of the second collaborating party for the task;

[0109] S24. Submit the signed task structure to the third collaborator, who then writes the task request into the blockchain to form a task statement.

[0110] Through the process design in which the second collaborator encapsulates the analysis task request and submits it to the third collaborator, the structured expression and trusted registration of task objectives, data types, and model expectations are achieved. This process not only clarifies the responsibilities of the task initiator, but also ensures that the task source is verifiable and the content cannot be tampered with through a signature binding mechanism. In the context of multi-agency collaboration in public security, it can effectively prevent task conflicts, duplicate registrations, and excessive authority, while providing data owners with clear task objectives and call scopes, which is conducive to the consistency control of subsequent data declarations and perturbation execution, and significantly improves the organizational efficiency and system transparency of multi-agency collaborative analysis.

[0111] In this embodiment, S3 specifically includes:

[0112] S31. The third collaborator calls the initialization interface of the task declaration smart contract, and loads the perturbation configuration table set by the first collaborator and the analysis task request submitted by the second collaborator as parameters into the contract execution environment;

[0113] S32. The task declaration smart contract verifies the integrity of the task content according to the preset logic, including whether the task identifier is repeated, whether the configuration policy is missing, and whether the task status is in the registration stage;

[0114] S33. The task declaration smart contract automatically generates a task declaration record after passing verification and updates the task status from pending registration to registered;

[0115] S34. After the task declaration smart contract is updated, the task declaration record is written into the blockchain ledger, and the on-chain evidence operation is performed to generate a unique task registration hash and timestamp;

[0116] S35. After the task declaration smart contract is executed, the status information of task registration completion is returned.

[0117] By introducing a task declaration smart contract mechanism executed by a third-party collaborator, the trusted binding and on-chain registration of analysis tasks and differential privacy policies is achieved. This contract not only performs logical verification of task identification, status, policy validity, and other information, but also automatically updates task status and writes declaration information to the blockchain ledger, ensuring the uniqueness, legitimacy, and non-repudiation of tasks within the system. Its execution process is standardized and proceduralized, avoiding information inconsistencies or compliance vulnerabilities caused by manual intervention. The on-chain evidence generated after task registration is completed can serve as the primary index field for subsequent data declaration, result auditing, and anomaly correction, helping to build a complete task lifecycle tracking system that meets the stringent public safety requirements for controllable tasks, traceable processes, and auditable behaviors.

[0118] In this embodiment, the S4 specifically includes:

[0119] S41. After completing the local differential perturbation processing, the data owner constructs a data summary;

[0120] S42: Call the data declaration smart contract deployed by the third party, and submit the data summary as a parameter to the data declaration smart contract call entry;

[0121] S43. After receiving the call request, the data declaration smart contract verifies whether the data summary structure is complete, whether the caller has registration authority, and checks whether the task binding status matches the current task declaration;

[0122] S44. After verification, the data declaration smart contract creates an on-chain declaration record, sets the record status to valid, and binds the data declaration to the unique identifier of the registered task;

[0123] S45. After the data declaration smart contract is successfully written, the declaration number, writing timestamp and blockchain transaction index information are automatically generated;

[0124] S46. Return the confirmation information of the registration completion to the calling data owner, and synchronize the registration result to the third-party collaborator for audit verification and task traceability.

[0125] By invoking the data declaration smart contract and registering the perturbed data summary on-chain, the data owner achieves identity binding and full-process traceability for data submission. This mechanism ensures the legitimacy and structural integrity of all data declaration operations by verifying call permissions and verifying the consistency of the data summary structure with the task declaration, preventing the intrusion of illegal data or data-task mismatches. The declaration number and transaction index generated after writing to the blockchain provide an immutable timestamp record, providing a basis for verifying the legitimacy, integrity, and original appearance of input data for analytical models. This mechanism is particularly suitable for data interoperability and collaboration between public security, fire protection, and other agencies, providing a structured solution for trusted access and responsibility delineation of multi-source data.

[0126] In this embodiment, the data summary structure includes a data disturbance tag, a data type identifier, a data source identifier, and processing time information.

[0127] By defining a standardized data summary structure, including perturbation labels, data type identifiers, data source identifiers, and processing time information, structural constraints and semantic transparency are achieved for perturbation data declarations. This structure not only provides parameter specifications for third-party collaborators to perform contract verification, but also provides a basis for perturbation parameter adaptation, data weight adjustment, and source mapping during the model training phase. Especially in multi-institutional joint analysis, the data sensitivity, perturbation methods, and timeliness attributes of each party vary. A unified data summary structure can effectively promote input consistency processing in the modeling phase, improve model fusion quality and the level of trusted analysis. This structure also facilitates on-chain result auditing and rapid location of abnormal data, enhancing the stability and interpretability of the system.

[0128] In this embodiment, the joint model is a data collaborative training model that supports differential privacy, and has distributed perturbation data access capabilities and parameter update mechanisms. The input end of the joint model includes multiple perturbation data channels, each channel corresponds to a data source of a data owner, and the channel supports unified access and format standardization processing of structured data and unstructured data.

[0129] By constructing a joint model structure with multi-channel input capabilities, it is able to uniformly receive and standardize perturbation data submitted by multiple data owners, effectively resolving the incompatibility issues of heterogeneous data sources in collaborative training. The model supports unified format alignment processing for structured and unstructured data, and is designed to be compatible with data differences after multi-party perturbation processing, ensuring consistency and privacy adaptability of input distribution during the model training phase. In the context of cross-departmental collaboration in public security, this model structure can improve modeling efficiency, reduce upfront manual intervention and data cleaning costs, and provide a stable modeling framework for joint learning and intelligent analysis of large-scale perturbation data.

[0130] In this embodiment, the joint model includes a privacy control module, a training control module and a parameter aggregation module. The privacy control module is used to identify perturbation labels in the input data and automatically adapt the perturbation intensity. The training control module is used to allocate the privacy budget, schedule the number of training rounds and step size, and the parameter aggregation module is used to jointly update the local model gradients of each perturbation data channel.

[0131] By integrating a privacy control module, a training control module, and a parameter aggregation module into a joint model structure, a controllable modeling structure for multi-source perturbation data is formed. The privacy control module identifies and adjusts the strength of perturbation labels to avoid model misjudgment or training failure; the training control module implements dynamic budget scheduling and training step management to optimize model convergence speed and stability; and the parameter aggregation module supports multi-channel model fusion through distributed gradient aggregation, improving global model accuracy. This modular structure enhances the model's convergence stability, training accuracy, and task adaptability under differential privacy constraints, providing a privacy-preserving data intelligence modeling infrastructure for high-reliability scenarios such as public security and emergency response.

[0132] Example 1:

[0133] To verify the feasibility of this invention, we applied it to a scenario involving joint analysis of public safety emergency data during a major event by the Public Security Bureau and the Emergency Management Bureau of a provincial capital city. The city was hosting a five-day, large-scale international event, which involved key tasks such as managing the dense flow of people across regions, real-time resource response, and public security situation analysis. Because this task involved sharing multiple sensitive data types, including trajectory data, police dispatch records, equipment alarms, and medical admission records, across multiple government departments, it placed extremely high demands on data privacy protection, secure sharing, and traceability of analysis results.

[0134] Traditional data sharing methods rely on manually signed agreements or the upload of desensitized data for centralized modeling, resulting in low efficiency, uneven desensitization, and a high risk of privacy leaks. The collaborative public security data analysis method based on differential privacy and blockchain, proposed in this paper, precisely meets the combined needs of such tasks, which require high-frequency data flows, strict confidentiality requirements, and high analytical accuracy.

[0135] In this example, the Public Security Bureau's Big Data Command served as the first collaborator, responsible for setting the differential privacy parameters and perturbation adjustment factors for this collaborative task. By analyzing historical dispatch trajectory data, personnel information, and incident response records, the Public Security Bureau assessed the local sensitivity upper bound of trajectory data to be 220 meters. They set a differential privacy budget of 1.2, set the perturbation adjustment factor between 0.15 and 0.3 based on the data dimension, and specified the use of the trajectory compressed Laplace perturbation mechanism for trajectory data. Once configured, a task-specific perturbation configuration table was created.

[0136] The second collaborator, the Smart Analysis Center of the Emergency Management Bureau, will generate analysis requests based on event security tasks. They plan to use clustering algorithms to analyze the concentration of key dispatch areas and resource response latency models. This task will utilize four types of data: dispatch trajectories, force deployment plans, call records, and camera monitoring data. All data types are provided by multiple units, labeled with data classifications, and then packaged and submitted to the third collaborator for registration.

[0137] The third collaborator, the technical operator and maintenance provider of the city government blockchain platform, is responsible for writing the perturbation configuration table and task request to the consortium blockchain via the task declaration smart contract. This registers the task number, security parameters, and call time window, establishing an immutable on-chain basis for subsequent data perturbation and use. Once registration is complete, the task hash and on-chain index are returned to the first and second collaborators.

[0138] Subsequently, the public security, traffic police, and fire departments, as data owners, each invoked a perturbation configuration table in their local systems to compress the trajectory data and inject Laplace noise at the start and end points. For example, a set of road patrol trajectory data provided by the traffic police contained 2,532 original track points, which were compressed to 412 segments. Each segment was independently perturbed to create a perturbation-labeled data summary with a compression ratio of 0.162, a perturbation intensity of 0.25, and a hash summary length of 64 bits. The data was then submitted to the data declaration smart contract, registered on-chain, and bound to the task.

[0139] After the second collaborator collected all the data, they initiated the joint modeling task locally. During the modeling process, trajectory data was first reconstructed using B-spline interpolation for timeline reconstruction, followed by density clustering analysis based on perturbation label weights. The joint model was trained for 60 epochs, with a learning rate of 0.01 and a perturbation budget allocated within 0.4. The model converged at epoch 37 with an accuracy of 86.3% and an error convergence rate within 3.5%. Intermediate process summaries, including loss curves, perturbation utilization distribution, and model structure identification, were recorded and submitted to the blockchain, forming an on-chain trusted log of the model training phase.

[0140] The final model output is encrypted and sent to the first collaborator. The Public Security Bureau decrypts the output and, after removing any residual disturbances from the original configuration, generates a cluster heat map and a statistical table of concentrated dispatch periods, which are then returned to the Emergency Management Bureau. This result is used within the emergency task deployment system to optimize police dispatch and allocation plans. If any data owner later deems the results to be biased, the third collaborator can initiate an on-chain correction request, allowing the system to fine-tune the calculations for certain data dimensions.

[0141] This application demonstrates that this method can effectively achieve the combined goals of data preservation, controllable analysis, trustworthy models, and auditable results in cross-agency, highly sensitive, and real-time public security missions. The entire process does not rely on a centralized data platform, significantly reducing the risk of data leakage and enhancing technical trust between institutions.

[0142] Table 1 Comparison of disturbance trajectory compression processing effects

[0143]

[0144] As can be seen from the data in Table 1 above, the present invention has demonstrated good trajectory compression and disturbance processing capabilities in actual deployment by multiple data owners. Taking the First Squadron of the Chengnan Traffic Police as an example, the original number of trajectory points is 2532, and only 412 segments are retained after compression, with a compression ratio of 0.162, which greatly reduces the number of data points that need to be noised, effectively reducing the total amount of disturbance while ensuring the integrity of the trajectory structure. The disturbance intensity is controlled between 0.20 and 0.30 in each group of data, which is in line with the setting range of the differential privacy policy, indicating that the disturbance process is controllable and adjustable. All data summaries successfully generated 64-bit hash summaries and completed on-chain submission within 2 seconds, indicating that the system has good real-time performance and computational efficiency. Comprehensive comparison results show that the perturbation trajectory compression Laplace mechanism adopted by the present invention not only improves the processing efficiency of trajectory data, but also significantly reduces data redundancy and transmission pressure while maintaining privacy protection, and has strong engineering applicability and practical promotion value.

[0145] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A collaborative analysis method for public security data based on differential privacy and blockchain, characterized by: The steps include: S1. The first collaborator sets the differential privacy parameters and perturbation adjustment factors, builds a perturbation configuration table, and configures the perturbation trajectory compressed Laplace mechanism for trajectory data. S2. The second collaborator formulates an analysis task request, determines the data type, and submits it to the third collaborator; S3. The third party will write the perturbation configuration table and task request into the blockchain through the task declaration smart contract to complete the task registration. S4. Each data owner performs differential perturbation on the local data according to the perturbation configuration table. Trajectory data is compressed and noise is added to the start and end points. A data summary with perturbation labels is generated and written to the blockchain through the data declaration smart contract. S5. The second collaborator collects perturbation data and builds a joint model, and submits a summary of the intermediate training results to the blockchain; S6. After completing the joint model training, the second collaborator sends the encrypted result to the first collaborator, who decrypts it and generates the final result. The third collaborator records the result summary and supports initiating an on-chain correction request.

2. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The first collaborator sets differential privacy parameters and perturbation adjustment factors, and constructs a perturbation configuration table, specifically including: The first collaborator evaluates the sensitivity of each type of data to be analyzed and defines a local sensitivity upper bound Δ for each type of data, which is used to represent the maximum output change caused by any two adjacent input records under the same query function; Set the differential privacy parameter ε>0, and set the perturbation adjustment factor α∈[0.05,0.5] according to the data type and sensitivity level to adjust the intensity of the noise injected into each type of data during perturbation; The local sensitivity upper bound Δ, differential privacy parameter ε, and perturbation adjustment factor α are combined into a triplet and mapped with the corresponding data types to construct a perturbation configuration table.

3. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: In S1, a perturbation trajectory compressed Laplace mechanism is configured for trajectory data, which specifically includes the following steps: S11. Receive the original trajectory data sequence and represent it as a set of trajectory points T arranged in ascending time order: T={(x1,y1,t1),(x2,y2,t2),…,(x n ,y n ,t n )}; Among them, (x i ,y i ) represents the two-dimensional coordinates of the i-th trajectory point, t i represents the timestamp of the i-th trajectory point, n represents the total number of trajectory points, and n≥3; S12. Set three trajectory compression control parameters, including the Euclidean distance threshold θ>0, the direction change threshold φ∈(0,π), and the sliding window length w∈N + , which are used to limit the maximum offset, maximum direction change angle and window step size of the trajectory points respectively; S13. Perform a sliding window scan on the trajectory point set T, extract any three consecutive trajectory points to determine the trajectory stability section, and calculate the direction change angle of the intermediate trajectory point: Among them, ∠ i Indicates the direction change angle of the intermediate trajectory point, (x i-1 ,y i-1 ) represents the two-dimensional coordinates of the i-1th trajectory point, (x i+1 ,y i+1 ) represents the two-dimensional coordinates of the i+1th trajectory point, arccos represents the inverse cosine function; If the conditions are met∠ i <φ and It is determined that the compression condition is met; S14. For the continuous trajectory point sequence that meets the compression condition, retain the trajectory point (x′ k ,y′ k ,t′ k ), where (x′ k ,y′ k ) represents the starting or ending coordinates of the continuous trajectory point sequence that meets the compression condition, t′ k The timestamps of the continuous trajectory point sequences that meet the compression conditions constitute the compressed trajectory set T′: T′={(x′1,y′1,t′1),(x′2,y′2,t′2),…,(x′ m ,y′ m ,t′ m )}; Among them, m represents the total number of trajectory points after compression, satisfying m <n; S15. Calculate the trajectory compression ratio r, which is used to measure the compression effect: S16, for each trajectory point (x′ k ,y′ k )Inject differential privacy Laplace noise to generate perturbed coordinates: in, and represents the trajectory coordinates after perturbation, Δ represents the upper bound of local sensitivity, Lap represents the Laplace distribution random variable, ε represents the differential privacy parameter, and α represents the perturbation adjustment factor; S17. Generate a meta-information label package for the perturbed trajectory set.

4. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 3 is characterized in that: The meta-information tag package includes alpha-tag, ratio-tag and hash-track, where alpha-tag is used to record the value of the disturbance adjustment factor, ratio-tag is used to record the value of the trajectory compression ratio, and hash-track is used to generate a complete trajectory summary for the perturbed trajectory set using a hash function, which is used for result verification and non-repudiation audit.

5. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The S2 specifically includes: S21. The second collaborator determines the analysis target based on the public safety application requirements and generates the task request content; S22. The second collaborating party identifies different data types and determines the data type; S23, encapsulating the task request content and the corresponding data type tag into a task structure to be registered, and attaching a digital signature to confirm the legal initiation identity of the second collaborating party for the task; S24. Submit the signed task structure to the third collaborator, who then writes the task request into the blockchain to form a task statement.

6. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The S3 specifically includes: S31. The third collaborator calls the initialization interface of the task declaration smart contract, and loads the perturbation configuration table set by the first collaborator and the analysis task request submitted by the second collaborator as parameters into the contract execution environment; S32. The task declaration smart contract verifies the integrity of the task content according to the preset logic, including whether the task identifier is repeated, whether the configuration policy is missing, and whether the task status is in the registration stage; S33. The task declaration smart contract automatically generates a task declaration record after passing verification and updates the task status from pending registration to registered; S34. After the task declaration smart contract is updated, the task declaration record is written into the blockchain ledger, and the on-chain evidence operation is performed to generate a unique task registration hash and timestamp; S35. After the task declaration smart contract is executed, the status information of task registration completion is returned.

7. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The S4 specifically includes: S41. After completing the local differential perturbation processing, the data owner constructs a data summary; S42: Call the data declaration smart contract deployed by the third party, and submit the data summary as a parameter to the data declaration smart contract call entry; S43. After receiving the call request, the data declaration smart contract verifies whether the data summary structure is complete, whether the caller has registration authority, and checks whether the task binding status matches the current task declaration; S44. After verification, the data declaration smart contract creates an on-chain declaration record, sets the record status to valid, and binds the data declaration to the unique identifier of the registered task; S45. After the data declaration smart contract is successfully written, the declaration number, writing timestamp and blockchain transaction index information are automatically generated; S46. Return the confirmation information of the registration completion to the calling data owner, and synchronize the registration result to the third-party collaborator for audit verification and task traceability.

8. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 7 is characterized in that: The data summary structure includes a data disturbance label, a data type identifier, a data source identifier and processing time information.

9. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The joint model is a data collaborative training model that supports differential privacy and has distributed perturbation data access capabilities and a parameter update mechanism. The input end of the joint model includes multiple perturbation data channels, each channel corresponds to a data source of a data owner, and the channel supports unified access and format standardization processing of structured data and unstructured data.

10. The public security data collaborative analysis method based on differential privacy and blockchain according to claim 1 is characterized in that: The joint model includes a privacy control module, a training control module, and a parameter aggregation module. The privacy control module is used to identify perturbation labels in the input data and automatically adapt the perturbation intensity. The training control module is used to allocate the privacy budget, schedule the number of training rounds and the step size. The parameter aggregation module is used to jointly update the local model gradients of each perturbation data channel.

Images