Differential privacy location protection method based on dynamic attenuated noise field

Abstract

The application relates to a differential privacy location protection method based on a dynamic attenuation noise field, which comprises the following steps: obtaining a plurality of candidate regions based on user trajectories, identifying different sensitive regions from the plurality of candidate regions based on the behavior characteristics of the user, and calculating the dynamic sensitivity of each sensitive region; wherein the behavior characteristics of the user include the stay identifier and the access time of the user in the candidate region; a three-dimensional dynamic noise model for generating a dynamic attenuation space-time noise field is constructed; the position coordinates and the corresponding time stamp of a target position in the sensitive region are obtained, and the corresponding noise amplitude is generated by adopting the three-dimensional dynamic noise model; and based on the noise amplitude, a two-dimensional position offset is generated by adopting a Laplace mechanism, and the position coordinates of the target position are disturbed based on the two-dimensional position offset. The scheme can realize adaptive adjustment of the noise amplitude, and realizes the collaborative optimization of strong protection of a high-sensitivity region and high precision of a low-sensitivity region.

Description

Technical Field

[0001] This invention relates to the field of information privacy protection technology, and in particular to a differential privacy location protection method based on a dynamically attenuated noise field. Background Technology

[0002] In location-based applications, such as ride-hailing routes and food delivery, traditional differential privacy technologies typically employ a fixed-intensity noise addition mechanism. This approach indiscriminately adds the same level of disturbance to all geographic locations, resulting in insufficient noise intensity in sensitive areas and failing to effectively prevent the leakage of sensitive personal locations such as home addresses and frequently visited medical facilities. Conversely, excessive noise addition in high-activity areas such as commercial districts and transportation hubs leads to chaotic route planning and decreased scheduling efficiency. This "one-size-fits-all" noise strategy struggles to balance the strength of privacy protection with the usability of the data.

[0003] Secondly, traditional solutions neglect the impact of data timeliness, and excessive protection of historical data reduces its long-term analytical value. Static noise models cannot adapt to dynamic changes in cities. The emergence of sensitive areas such as new residential areas renders existing noise configurations ineffective, while manual adjustment mechanisms are slow to respond, creating privacy vulnerabilities. Finally, in large cities, fixed noise parameters are difficult to adapt to differences in regional population density, resulting in insufficient noise intensity in high-density areas.

[0004] Currently, most location-based services rely on large-scale user location data to support business decisions. However, raw location data contains sensitive user information, and direct publication can lead to user privacy leaks. To protect user privacy, services use differential privacy techniques to add noise when publishing data to protect individual privacy. However, in high-dimensional, dynamic location data scenarios, traditional differential privacy is insufficient in protecting sensitive areas. Using the same noise intensity for highly sensitive and low-sensitive areas fails to balance privacy and utility. Secondly, there is a conflict in data timeliness. Using the same privacy budget to protect both new and old data results in excessive noise addition to historical data, reducing its long-term analytical value, while insufficient protection for new data leads to real-time privacy leaks. Finally, there is a lack of dynamic adjustment. Most current location services rely on manually configured static noise parameters, which cannot adapt to the emergence of new sensitive locations.

[0005] Traditional solutions employ uniform noise injection, adding fixed-intensity noise to location coordinates using Laplace or Gaussian mechanisms. This fails to balance privacy protection strength and utility, easily leading to insufficient protection for highly sensitive areas while causing excessive distortion of data in less sensitive areas. Other solutions use Locality Sensitive Hash (LSH) indexing, mapping location data to Hamming space and constructing a single hash table index. However, this single hash table structure results in insufficient hash collision probability, significantly reducing the recall rate of nearest neighbor queries. The index update mechanism is rigid; when user behavior patterns change (e.g., adding frequently visited locations), a full index rebuild is required, with computational costs increasing quadratically with data size, making it unsuitable for dynamic scenarios. Still other solutions utilize Bayesian networks to model mutual information dependencies between attributes, allocating differentiated privacy budgets based on association strength. However, their sensitive area identification relies on pre-defined semantic labels and cannot automatically discover unlabeled emerging sensitive areas. Furthermore, the budget allocation strategy ignores the decay requirements over time, applying equal protection strength to historical and new data, resulting in unnecessary loss of the analytical value of historical data.

[0006] Therefore, the main drawbacks of existing technologies are as follows:

[0007] 1. Low spatial sensitivity discrimination: High-sensitivity areas are not adequately protected while low-sensitivity areas are excessively distorted.

[0008] 2. Reliance on manually preset geofences or semantic tags: Identification of emerging sensitive areas is lagging and maintenance costs are high.

[0009] 3. Ignoring the impact of data timeliness: Applying high-intensity noise to both historical and new data indiscriminately leads to unnecessary loss of the value of historical data.

[0010] Therefore, in the field of location data privacy protection, traditional differential privacy methods, due to their use of fixed noise intensity, result in insufficient protection for highly sensitive areas such as residential areas and excessive distortion for low-sensitivity areas such as commercial areas. Furthermore, existing solutions rely on manually pre-setting sensitive area boundaries, making them difficult to adapt to dynamic urban development. Summary of the Invention

[0011] The technical problem to be solved by the present invention is to provide a differential privacy location protection method based on a dynamically attenuated noise field.

[0012] To achieve the above-mentioned objectives, this invention provides a differential privacy location protection method based on a dynamically attenuated noise field, comprising the following steps:

[0013] S1. Obtain multiple candidate regions based on user trajectories, and identify different sensitive regions from the multiple candidate regions based on user behavior characteristics, and calculate the dynamic sensitivity of each sensitive region; wherein, user behavior characteristics include: user dwell time and access time in the candidate regions;

[0014] S2. Construct a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field; wherein, the three-dimensional dynamic noise model is used to generate different noise amplitudes for different location coordinates and different timestamps in the sensitive region;

[0015] S3. Obtain the location coordinates and corresponding timestamp of the target location in the sensitive area and generate the corresponding noise amplitude using the three-dimensional dynamic noise model. Based on the noise amplitude, generate a two-dimensional position offset using the Laplace mechanism, and perturb the location coordinates of the target location based on the two-dimensional position offset, and output the perturbed location coordinates.

[0016] According to one aspect of the present invention, step S1, the step of obtaining multiple candidate regions based on user trajectories, includes:

[0017] Collect a series of trajectory points in the user trajectory and obtain the spatiotemporal feature vector of each trajectory point;

[0018] Based on the spatiotemporal feature vector, the trajectory points are clustered using a clustering method to generate multiple candidate regions.

[0019] According to one aspect of the present invention, step S1, which involves identifying different sensitive regions from a plurality of candidate regions based on user behavior characteristics and calculating the dynamic sensitivity of each sensitive region, includes:

[0020] Select the time period to be judged, and obtain the user's dwell time ratio and access stability index in the candidate area based on the user's behavioral characteristics; wherein, the dwell time ratio is based on the user's behavior in the candidate area. The number of trajectory points and candidate regions within a preset time period The access stability index is obtained by the ratio of the number of all trajectory points in the candidate region. The time spent accessing the site;

[0021] Determine whether the dwell rate index and the access stability index meet preset conditions. If they do, determine that the candidate area is a sensitive area, obtain the center coordinates and sensitivity score of the sensitive area, and construct a set of sensitive areas.

[0022] According to one aspect of the present invention, in the step of selecting a time period to be judged and obtaining a user's dwell time ratio and access stability index in a candidate area based on user behavioral characteristics, the dwell time ratio is expressed as:

[0023] ;

[0024] in, Indicates the percentage of stay. Represents the trajectory points, Indicates the candidate region. This represents the total number of trajectory points across all candidate regions. Indicates an indicator function, The timestamp representing the trajectory point , This indicates the start and end times of the selected time period.

[0025] According to one aspect of the present invention, in the step of selecting a time period to be judged and obtaining a user's dwell time ratio and access stability index in a candidate area based on user behavioral characteristics, the access stability index is obtained based on the following steps, which include:

[0026] Extract candidate regions within a preset time period The sequence of access times for user trajectory points in the middle, and the candidate regions. Sort all access times by timestamps and represent them as follows:

[0027] ;

[0028] in, Indicates the user's preference for candidate regions A timestamp sequence of all access times. , , , Indicates the use of candidate regions The 1st, 2nd, ..., 1st The timestamp of the first visit;

[0029] Based on user's selection of candidate regions Calculate the adjacent access interval using timestamp sequences:

[0030] ;

[0031] in, Indicates the interval between adjacent visits. Indicates the first Next and first The time interval between visits, and , ;

[0032] The access stability index is obtained by calculating the reciprocal of the standard deviation of the adjacent access intervals, wherein the access stability index is expressed as:

[0033] ;

[0034] in, This indicates an index of access stability.

[0035] According to one aspect of the present invention, the step of obtaining the center location coordinates and sensitivity score of the sensitive region and constructing a set of sensitive regions includes:

[0036] Obtain the sensitivity score of the sensitive region, wherein the sensitivity score is expressed as:

[0037] ;

[0038] in, Indicates sensitivity score;

[0039] The set of sensitive regions is constructed based on the center coordinates and sensitivity scores of the acquired sensitive regions, and is represented as follows:

[0040] ;

[0041] ;

[0042] ;

[0043] in, Represents a set of sensitive regions. , Indicates the coordinates of the center location of the sensitive area. This indicates the sensitivity score of the sensitive area. Indicates the sensitivity threshold. Represents trajectory points longitude coordinates Represents trajectory points Latitude coordinates.

[0044] According to one aspect of the present invention, in step S2, the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field, wherein the three-dimensional dynamic noise model is used to generate a dynamically decaying spatiotemporal noise field in spatial, temporal, and density dimensions, wherein, in the spatial dimension, the noise parameter is dynamically changed based on the distance from the center of the sensitive area, and if it is far from the center of the sensitive area, the noise parameter is controlled to decay exponentially; in the temporal dimension, the noise parameter is dynamically changed based on the timeliness of the data, and if the data is older than the present time, the noise parameter is controlled to decay exponentially; in the density dimension, the noise parameter is dynamically changed based on the regional population density, and if the regional population density decreases, the noise parameter is controlled to decay linearly.

[0045] According to one aspect of the present invention, in step S2, in the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field, the density dimension is used as the basic noise term of the three-dimensional dynamic noise model, and the spatial dimension and the time dimension are used as the decay term of the three-dimensional dynamic noise model.

[0046] According to one aspect of the present invention, in step S2, the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field, wherein the three-dimensional dynamic noise model is represented as:

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] in, Based on the minimum noise amplitude, Population density scaling factor Given the current population density of the area, For reference population density, For regional sensitivity, The spatial attenuation coefficient, Distance to the sensitive center The time decay magnitude, The decay rate over time. For data timeliness, Based on the attenuation coefficient, The building density coefficient, For regional building density, Based on the decay rate, For safety level coefficient, For data security level, The current system time. This is the original timestamp of the location data with perturbations.

[0052] According to one aspect of the present invention, in step S3, in the step of generating a two-dimensional position offset using a Laplace mechanism based on the noise amplitude, the two-dimensional position offset is expressed as:

[0053] ;

[0054] in, This represents the longitude coordinate offset value in the two-dimensional position offset. This represents the latitude coordinate offset value in the two-dimensional position offset. Indicates the Laplace mechanism, This represents the noise amplitude generated by the three-dimensional dynamic noise model;

[0055] In step S3, the perturbation of the target position coordinates based on the two-dimensional position offset is represented as follows:

[0056] ;

[0057] ;

[0058] in, , This represents the coordinate values ​​after the disturbance. The timestamp representing the trajectory point.

[0059] According to one aspect of the present invention, the present invention aims to establish a dynamic correlation mechanism between noise intensity and geographical location sensitivity, so that the noise amplitude can be adaptively adjusted, thereby achieving synergistic optimization of strong protection in highly sensitive areas and high precision in low sensitive areas.

[0060] According to one aspect of the present invention, this approach proposes an automatic detection method for sensitive areas based on trajectory behavior characteristics. By using a data-driven algorithm to capture the spatiotemporal distribution of sensitive locations in real time, the method eliminates reliance on manual configuration and improves the dynamic adaptability of the system.

[0061] According to one aspect of the present invention, a mathematical model is designed to autonomously decay noise intensity over time, thereby ensuring full privacy protection for new data while gradually releasing the analytical potential of historical data and realizing full lifecycle value management of data.

[0062] According to one aspect of the present invention, a novel mechanism is designed to dynamically adjust noise intensity based on geographical location, time decay, and data value. This achieves a balance between high-intensity protection of highly sensitive locations and high data availability in low-sensitive areas while ensuring strict differential privacy requirements are met. Simultaneously, it constructs automated sensitive area detection capabilities, eliminating reliance on manual configuration and adapting to dynamic urban development.

[0063] According to one aspect of the present invention, this solution achieves a balance between privacy protection and data utility. It processes the original trajectory data through a spatiotemporal density clustering algorithm, automatically identifies sensitive areas such as residential areas and hospitals, and achieves accurate and comprehensive analysis of user behavior characteristics based on dwell time ratio and access stability. This allows for the generation of regional sensitivity scores without the need for manual geofencing.

[0064] According to one aspect of the present invention, the method automatically identifies the sensitivity of each location based on features, and then generates a spatiotemporal noise field based on the sensitivity to construct a dynamic noise model, thereby achieving the effect that the offset of the high-sensitivity area is significantly greater than that of the low-sensitivity area.

[0065] According to one aspect of the present invention, this approach automatically identifies highly sensitive areas by analyzing user behavior characteristics such as dwell time percentage and access frequency, and calculates the dynamic sensitivity of each area. The entire process does not require manual pre-setting of geofences or semantic tags, thereby converting raw trajectory data into geographic block information with sensitivity markers.

[0066] According to one aspect of the present invention, considering that fixed noise intensity leads to insufficient protection in highly sensitive areas such as residential areas and excessive distortion in low-sensitive areas such as commercial areas, a dynamic noise field attenuation method is proposed. Based on the output results of sensitive areas, a three-dimensional dynamic noise model is constructed. In the spatial dimension, the noise intensity is adjusted according to the distance between the location and the sensitive center, with stronger noise the closer the distance; in the temporal dimension, noise is automatically attenuated according to the timeliness of the data, with higher protection intensity for new data and weaker disturbance for historical data; in the density dimension, the base noise amplitude is dynamically adjusted in conjunction with the regional population size. These three aspects are integrated to generate a continuously distributed noise field function, providing customized noise parameters for each location point.

[0067] According to one embodiment of the present invention, the core advantage of this invention compared to existing technologies lies in solving three major bottleneck problems in location data privacy protection: spatial indistinguishability, strong reliance on manual intervention, and conflicting timeliness. Firstly, through a sensitive area detection module, the nighttime activity patterns and access stability characteristics in user dwell patterns are analyzed, identifying highly sensitive areas without any manual annotation, significantly reducing the lag in protecting emerging sensitive points. Compared to the limitations of traditional Bayesian network solutions that rely on preset semantic labels, this method fully realizes data-driven dynamic perception, greatly reducing system maintenance costs and improving response speed.

[0068] According to one aspect of the present invention, regarding noise control mechanisms, this invention proposes a spatiotemporal noise field model, breaking through the rigid mode of traditional fixed noise. This model automatically adjusts the noise intensity based on the distance between the location and the sensitive center, causing high-intensity disturbances around residential areas while maintaining a slight shift in commercial areas, successfully achieving differentiated protection in the spatial dimension. Simultaneously, a time decay factor is introduced, ensuring sufficient privacy for new data while the noise intensity of historical data automatically decreases with time, effectively releasing the value of long-term data. This fundamentally avoids the problem of excessive distortion of historical data caused by neglecting timeliness in existing technologies. Attached Figure Description

[0069] Figure 1 This is a flowchart illustrating the steps of a differential privacy location protection method based on a dynamically attenuated noise field according to one embodiment of the present invention.

[0070] Figure 2 This is a schematic diagram illustrating a scenario application of the differential privacy location protection method based on a dynamically attenuated noise field, according to one embodiment of the present invention. Detailed Implementation

[0071] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The embodiments cannot be described in detail here, but the embodiments of the present invention are not limited to the following embodiments.

[0072] like Figure 1 As shown, according to one embodiment of the present invention, a differential privacy location protection method based on a dynamically attenuated noise field includes the following steps:

[0073] S1. Obtain multiple candidate regions based on user trajectories, and identify different sensitive regions from the multiple candidate regions based on user behavior characteristics, and calculate the dynamic sensitivity of each sensitive region; wherein, user behavior characteristics include: user dwell time and access time in the candidate regions;

[0074] S2. Construct a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field; wherein, the three-dimensional dynamic noise model is used to generate different noise amplitudes for different location coordinates and different timestamps in the sensitive region;

[0075] S3. Obtain the location coordinates and corresponding timestamp of the target location in the sensitive area and generate the corresponding noise amplitude using a three-dimensional dynamic noise model. Based on the noise amplitude, generate a two-dimensional position offset using the Laplace mechanism, and perturb the location coordinates of the target location based on the two-dimensional position offset, and output the perturbed location coordinates.

[0076] According to one embodiment of the present invention, step S1, which involves obtaining multiple candidate regions based on user trajectories, includes:

[0077] A series of trajectory points are collected from the user's trajectory, and the spatiotemporal feature vector of each trajectory point is obtained. In this embodiment, a corresponding fixed block is selected, and the user trajectory within the fixed block is collected. The obtained user trajectory is discretized to obtain a series of continuous trajectory points. The obtained series of trajectory points is represented by a trajectory point set, which is:

[0078] ;

[0079] in, Represents a set of trajectory points. Represents the trajectory points, Represents trajectory points longitude coordinates Represents trajectory points latitude coordinates The timestamps represent the trajectory points;

[0080] Based on spatiotemporal feature vectors, a clustering method is used to cluster trajectory points to generate multiple candidate regions. In this embodiment, the clustering method employed can be an improved DBSCAN clustering algorithm, thereby realizing the effect of spatiotemporal density clustering to process the original user trajectory data and automatically identify the corresponding candidate regions. In this embodiment, spatiotemporal clustering is performed based on spatiotemporal feature vectors, merging regions within a distance radius. Proximity points within the time window form candidate regions. Then it is represented as:

[0081] ;

[0082] in, Indicates the distance radius. and Indicates nearby trajectory points, and Indicates nearby trajectory points and The corresponding timestamp.

[0083] According to one embodiment of the present invention, step S1, which involves identifying different sensitive regions from a plurality of candidate regions based on user behavior characteristics and calculating the dynamic sensitivity of each sensitive region, includes:

[0084] Select the time period to be judged, and obtain the user's dwell time ratio and access stability index in the candidate area based on user behavior characteristics; among them, the dwell time ratio is based on the user's behavior in the candidate area. Number of trajectory points and candidate regions within a preset time period The access stability metric is obtained by comparing the number of all trajectory points in the candidate region. The time spent accessing the site;

[0085] In this embodiment, the dwell rate index is expressed as:

[0086] ;

[0087] in, Indicates the percentage of stay. Represents the trajectory points, Indicates the candidate region. This represents the total number of trajectory points across all candidate regions. Indicates an indicator function, , This indicates the start and end times of the selected time period.

[0088] Taking nighttime as an example, to analyze the nighttime stay ratio of each candidate area, the start and end times of the corresponding time period can be set specifically. For example, if the start time of nighttime is set to 22:00 and the end time is 06:00, then the stay ratio can be transformed into a formula specific to that time period, namely:

[0089] .

[0090] Furthermore, access stability metrics are obtained based on the following steps:

[0091] Extract candidate regions within a preset time period The sequence of access times for user trajectory points in the middle, and the candidate regions. Sort all access times by timestamps and represent them as follows:

[0092] ;

[0093] in, Indicates the user's preference for candidate regions A timestamp sequence of all access times. , , , Indicates the use of candidate regions The 1st, 2nd, ..., 1st The timestamp of the first visit;

[0094] Based on user's selection of candidate regions Calculate the adjacent access interval using timestamp sequences:

[0095] ;

[0096] in, Indicates the interval between adjacent visits. Indicates the first Next and first The time interval between visits, and , ;

[0097] The access stability index is obtained by calculating the reciprocal of the standard deviation of adjacent access intervals, where the access stability index is expressed as:

[0098] ;

[0099] in, This indicates an index of access stability.

[0100] The system determines whether the dwell rate and access stability metrics meet preset conditions. If they do, the candidate region is identified as a sensitive region. The center coordinates and sensitivity score of the sensitive region are then obtained, and a set of sensitive regions is constructed. (See [link to relevant documentation]). Figure 2 As shown. In this embodiment, the preset condition for determining the dwell rate indicator is a preset threshold. Therefore, by comparing the value of the dwell rate indicator with the set preset threshold, it can be determined whether the dwell rate indicator meets the standard for a sensitive area. Similarly, the preset condition for determining the access stability indicator is also a preset threshold. Therefore, by comparing the value of the access stability indicator with the set preset threshold, it can be determined whether the access stability indicator meets the standard for a sensitive area.

[0101] Through the above settings, this solution utilizes the designed dwell time ratio and access stability indicators to accurately quantify and analyze uncertain behavioral characteristics, greatly improving the accuracy of the analysis of sensitive areas. In particular, the set access stability indicator can more accurately and reliably filter out strong correlations between user behavior characteristics and candidate areas, thereby enabling more accurate identification of candidate areas.

[0102] According to one embodiment of the present invention, the step of obtaining the center coordinates and sensitivity score of a sensitive region and constructing a set of sensitive regions includes:

[0103] Obtain the sensitivity score of the sensitive area, where the sensitivity score is expressed as:

[0104] ;

[0105] in, Indicates sensitivity score;

[0106] Based on the sensitivity scoring set above, the dwell rate index and access stability index are effectively considered together. This allows for a more accurate and comprehensive evaluation of the sensitivity of corresponding areas based on relevant behavioral characteristics, which is more beneficial for improving the accuracy of sensitive area delineation.

[0107] A set of sensitive regions is constructed based on the center coordinates and sensitivity scores of the acquired sensitive regions, and is represented as follows:

[0108] ;

[0109] ;

[0110] ;

[0111] in, Represents a set of sensitive regions. , Indicates the coordinates of the center location of the sensitive area. This indicates the sensitivity score of the sensitive area. Indicates the sensitivity threshold. Represents trajectory points longitude coordinates Represents trajectory points Latitude coordinates.

[0112] In this embodiment, the sensitivity threshold can be set to 0.7.

[0113] According to one embodiment of the present invention, in step S2, the three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field and the step of generating different noise field functions for different sensitive areas are used to generate a dynamically decaying spatiotemporal noise field in spatial, temporal, and density dimensions. In the spatial dimension, the noise parameter is dynamically changed based on the distance from the center of the sensitive area. If the distance is far from the center of the sensitive area, the noise parameter is controlled to decay exponentially. In the temporal dimension, the noise parameter is dynamically changed based on the timeliness of the data. If the data is older than the present time, the noise parameter is controlled to decay exponentially. In the density dimension, the noise parameter is dynamically changed based on the population density of the area. If the population density of the area decreases, the noise parameter is controlled to decay linearly.

[0114] In this embodiment, the density dimension is used as the basic noise term of the three-dimensional dynamic noise model, and the spatial and temporal dimensions are used as the attenuation terms of the three-dimensional dynamic noise model. That is, the overall configuration of the three-dimensional dynamic noise model is: basic noise term × spatial dimension attenuation term × temporal dimension attenuation term. Among them, the spatial dimension attenuation term has a higher priority than the temporal dimension attenuation term. Therefore, the attenuation ratio of the location in the sensitive area is given more priority, so that the privacy protection of the location is more important.

[0115] According to one embodiment of the present invention, in step S2, the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field, the three-dimensional dynamic noise model is represented as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] in, Based on the minimum noise amplitude, Population density scaling factor Given the current population density of the area, For reference population density, For regional sensitivity, The spatial attenuation coefficient, Distance to the sensitive center The time decay magnitude, The decay rate over time. For data timeliness, Based on the attenuation coefficient, The building density coefficient, For regional building density, Based on the decay rate, For safety level coefficient, For data security level, The current system time. This is the original timestamp of the location data with perturbations.

[0121] According to one embodiment of the present invention, in step S3, where a two-dimensional position offset is generated based on the noise amplitude using a Laplace mechanism, the position coordinates of the target position are perturbed based on the two-dimensional position offset, and the perturbed position coordinates are output, the two-dimensional position offset is expressed as:

[0122] ;

[0123] in, This represents the longitude coordinate offset value in the two-dimensional position offset. This represents the latitude coordinate offset value in the two-dimensional position offset. Indicates the Laplace mechanism, This represents the noise amplitude generated by the three-dimensional dynamic noise model;

[0124] Furthermore, in step S3, the perturbation of the target position coordinates based on the two-dimensional position offset is represented as follows:

[0125] ;

[0126] ;

[0127] in, , This represents the coordinate values ​​after the disturbance. The timestamp representing the trajectory point.

[0128] According to this invention, by analyzing dwelling behavior in location data, highly sensitive areas such as residential areas and hospitals are automatically identified and dynamic scores are calculated without manual annotation. A spatiotemporal noise field generation module constructs a dynamic noise model based on this, generating customized noise parameters according to three factors: spatial distance, time period, and population density. The noise parameters are adaptively invoked to generate a differential privacy-preserving offset for each location point, outputting the perturbation location. The entire process achieves fully automated processing of sensitive area identification, dynamic noise modeling, and targeted location perturbation, ensuring strong offset protection for residential areas while maintaining high accuracy in commercial areas, thus achieving a balance between privacy and data utility.

[0129] The above description is merely an example of a specific solution of the present invention. For any devices and structures not described in detail herein, it should be understood that they are implemented using common devices and methods already available in the art.

[0130] The above description is merely one embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A differential privacy location protection method based on dynamically attenuated noise field, characterized in that, Includes the following steps: S1. Obtain multiple candidate regions based on user trajectories, and identify different sensitive regions from the multiple candidate regions based on user behavior characteristics, and calculate the dynamic sensitivity of each sensitive region; wherein, user behavior characteristics include: user dwell time and access time in the candidate regions; S2. Construct a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field; wherein, the three-dimensional dynamic noise model is used to generate different noise amplitudes for different location coordinates and different timestamps in the sensitive region; wherein, the three-dimensional dynamic noise model is represented as: in, Based on the minimum noise amplitude, Population density scaling factor Given the current population density of the area, For reference population density, For regional sensitivity, The spatial attenuation coefficient, Distance to the sensitive center The time decay magnitude, The decay rate over time. For data timeliness, Based on the attenuation coefficient, The building density coefficient, For regional building density, Based on the decay rate, For safety level coefficient, For data security level, The current system time. The original timestamp of the location data with perturbation; S3. Obtain the location coordinates and corresponding timestamp of the target location in the sensitive area and generate the corresponding noise amplitude using the three-dimensional dynamic noise model. Based on the noise amplitude, generate a two-dimensional position offset using the Laplace mechanism, and perturb the location coordinates of the target location based on the two-dimensional position offset, and output the perturbed location coordinates.

2. The differential privacy location protection method based on dynamically attenuated noise field according to claim 1, characterized in that, Step S1, which involves obtaining multiple candidate regions based on user trajectories, includes: Collect a series of trajectory points in the user trajectory and obtain the spatiotemporal feature vector of each trajectory point; Based on the spatiotemporal feature vector, the trajectory points are clustered using a clustering method to generate multiple candidate regions.

3. The differential privacy location protection method based on dynamically attenuated noise field according to claim 2, characterized in that, Step S1, which involves identifying different sensitive regions from multiple candidate regions based on user behavior characteristics and calculating the dynamic sensitivity of each sensitive region, includes: Select the time period to be judged, and obtain the user's dwell time ratio and access stability index in the candidate area based on the user's behavioral characteristics; wherein, the dwell time ratio is based on the user's behavior in the candidate area. The number of trajectory points and candidate regions within a preset time period The access stability index is obtained by the ratio of the number of all trajectory points in the candidate region. The time spent accessing the site; Determine whether the dwell rate index and the access stability index meet preset conditions. If they do, determine that the candidate area is a sensitive area, obtain the center coordinates and sensitivity score of the sensitive area, and construct a set of sensitive areas.

4. The differential privacy location protection method based on dynamically attenuated noise field according to claim 3, characterized in that, In the steps of selecting the time period to be judged and obtaining the user's dwell time ratio and access stability index in the candidate area based on the user's behavioral characteristics, the dwell time ratio index is expressed as: in, Indicates the percentage of stay. Represents the trajectory points, Indicates the candidate region. This represents the total number of trajectory points across all candidate regions. Indicates an indicator function, The timestamp representing the trajectory point , This indicates the start and end times of the selected time period.

5. The differential privacy location protection method based on dynamically attenuated noise field according to claim 4, characterized in that, In the steps of selecting the time period to be judged and obtaining the user's dwell time ratio and access stability index in the candidate area based on user behavior characteristics, the access stability index is obtained based on the following steps, which include: Extract candidate regions within a preset time period The sequence of access times for user trajectory points in the middle, and the candidate regions. Sort all access times by timestamps and represent them as follows: in, Indicates the user's preference for candidate regions A timestamp sequence of all access times. , , , Indicates the use of candidate regions The 1st, 2nd, ..., 1st The timestamp of the first visit; Based on user-selected candidate regions Calculate the adjacent access interval using timestamp sequences: in, Indicates the interval between adjacent visits. Indicates the first Next and first The time interval between visits, and , ; The access stability index is obtained by calculating the reciprocal of the standard deviation of the adjacent access intervals, wherein the access stability index is expressed as: in, This indicates an index of access stability.

6. The differential privacy location protection method based on dynamically attenuated noise field according to claim 5, characterized in that, The steps of obtaining the center coordinates and sensitivity scores of the sensitive regions and constructing a set of sensitive regions include: Obtain the sensitivity score of the sensitive region, wherein the sensitivity score is expressed as: in, Indicates sensitivity score; The set of sensitive regions is constructed based on the center coordinates and sensitivity scores of the acquired sensitive regions, and is represented as follows: in, Represents a set of sensitive regions. , Indicates the coordinates of the center location of the sensitive area. This indicates the sensitivity score of the sensitive area. Indicates the sensitivity threshold. Represents trajectory points longitude coordinates Represents trajectory points Latitude coordinates.

7. The differential privacy location protection method based on dynamically attenuated noise field according to claim 6, characterized in that, In step S2, the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field is described. This three-dimensional dynamic noise model is used to generate a dynamically decaying spatiotemporal noise field in spatial, temporal, and density dimensions. Specifically, in the spatial dimension, the noise parameters are dynamically changed based on the distance from the center of the sensitive area; if the distance is far from the center of the sensitive area, the noise parameters are controlled to decay exponentially. In the temporal dimension, the noise parameters are dynamically changed based on the timeliness of the data; if the data is older than the present, the noise parameters are controlled to decay exponentially. In the density dimension, the noise parameters are dynamically changed based on the regional population density; if the regional population density decreases, the noise parameters are controlled to decay linearly.

8. The differential privacy location protection method based on dynamically attenuated noise field according to claim 7, characterized in that, In step S2, the step of constructing a three-dimensional dynamic noise model for generating a dynamically decaying spatiotemporal noise field uses the density dimension as the basic noise term of the three-dimensional dynamic noise model, and the spatial dimension and the time dimension as the decay term of the three-dimensional dynamic noise model.

9. The differential privacy location protection method based on dynamically attenuated noise field according to claim 8, characterized in that, In step S3, the step of generating a two-dimensional position offset using the Laplace mechanism based on the noise amplitude, the two-dimensional position offset is expressed as: in, This represents the longitude coordinate offset value in the two-dimensional position offset. This represents the latitude coordinate offset value in the two-dimensional position offset. Indicates the Laplace mechanism, This represents the noise amplitude generated by the three-dimensional dynamic noise model; In step S3, the perturbation of the target position coordinates based on the two-dimensional position offset is represented as follows: in, , This represents the coordinate values ​​after the disturbance. The timestamp representing the trajectory point.

Images