Improved K clustering LBS privacy protection method based on dynamic feedback mechanism

Abstract

The invention provides an improved K clustering LBS privacy protection method based on a dynamic feedback mechanism, and belongs to the technical field of Internet of Vehicles security. The security problems that in a traditional method, the privacy protection intensity is low due to the fact that the number of regional users is too small, and position privacy is leaked due to the fact that the K value of a clustering algorithm is unreasonably set are solved. According to the technical scheme, the method comprises the following steps that S1, an RSU carries out region initial division; s2, the RSU carries out sparse state feedback adjustment according to the condition of the space where the vehicle is located, and region re-division is completed; s3, the RSU dynamically adjusts the value of the parameter K, a K-means clustering algorithm is executed, and a proper clustering value n is generated; s4, the RSU sends the clustering center point and the real position of the replacement vehicle to an LBS server; s5, the RSU forwards the response of the LBS server to the corresponding OBU; the method has the beneficial effect that the LBS location privacy security protection with the same intensity in the vehicle sparse user space state and the conventional user space state is realized.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking security technology, and in particular to an improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism. Background Art

[0002] The Internet of Vehicles refers to a technology that realizes the intelligence and networking of vehicles through the exchange and sharing of information between vehicles and the outside world. It combines technologies such as the Internet of Things, mobile Internet, artificial intelligence, and big data, enabling vehicles to communicate and collaborate with other vehicles, transportation infrastructure, Internet services, and applications to improve traffic safety, driving efficiency, and user experience.

[0003] With the development and widespread use of location-based services and location navigation technologies in the Internet of Vehicles, a large amount of location data with user location information has been generated. Location-based services (LBS) refer to obtaining the location information of mobile users through the Global Positioning System (GPS) or other technologies, and using the Geographic Information System (GIS) to provide end users with personalized services related to their current spatial location. In the Internet of Vehicles environment, vehicle users are increasingly paying attention to the perception of the vehicle and the surrounding geographical environment in order to obtain mobile online social services. Mobile online social services refer to providing users with location-based nearby place recommendations and neighbor friend search services through the vehicle positioning function. For example, a vehicle can request the LBS server to find the nearest gas station or the nearest restaurant to the vehicle through the roadside unit (RSU).

[0004] However, while providing users with location-based neighbor search services, the LBS service of the Internet of Vehicles also poses a serious risk of user privacy leakage. According to the National Mobile APP Security Research Report, 35 of the 50 platforms that provide LBS network services leak user location data to advertisers or data analysis service providers. At the same time, with the rapid development of data mining technology, it is becoming increasingly convenient for LBS service providers and malicious adversaries to obtain people's private information in social networks (such as mobile users' locations, personal data, and communication details, etc.), making it easier to infer, track, and predict the daily activities of vehicle users. By mining the location information of vehicle users, the adversary can infer with a certain probability that the user is attending an important ceremony in a certain place, thereby obtaining the user's occupation and social relations, and even predicting the religious beliefs of the user and his friends, as well as inferring the individual's whereabouts from the search history of the vehicle user.

[0005] Therefore, the application of LBS services in the Internet of Vehicles faces multiple security issues such as privacy leakage, data tampering, location deception, system attacks and data abuse, which have a serious negative impact on users and even the entire Internet of Vehicles system. The primary reason for location privacy leakage is that the security management of the LBS server is not in place, resulting in external illegal users attacking the server data, or the abuse of authority or malicious destruction by administrators with higher authority within the LBS service provider; on the other hand, the service provider defaults to obtaining the user's current location for tracking without obtaining special permission from the user. In order to prevent service providers and hackers from accessing location information without authorization, vehicle users can use pseudonyms to cut off the connection between the vehicle and the location. Representative solutions include the dynamic pseudonym technology mentioned by researcher Yang Min in the document "Research on Vehicle Location Privacy Protection Method Based on Dynamic Pseudonym Exchange Zone in Internet of Vehicles" and the k-anonymity technology mentioned by Li Yongjun et al. in the document "Enhanced Location k-anonymity Privacy Protection Scheme Based on Geohash". Among them, the application of k-anonymity technology is the most widely used. This scheme constructs an anonymous zone by selecting k-1 users similar to real users to achieve the purpose of privacy protection. In addition, vehicle users can obfuscate or encrypt location information locally before uploading it to the LBS server. Representative solutions include the encryption privacy protection technology proposed in the document "Data Mining Method for Privacy Protection of Internet of Vehicles Based on Homomorphic Encryption" and the differential privacy technology mentioned in the document "Research on Information Privacy Protection Technology Based on Location Services". For example, in 2021, Yaqiong Ma et al. proposed a protection scheme based on shadow vehicles in the document "Research on Trajectory Privacy Protection Technology for VANET Nodes". The scheme first uses a genetic algorithm to predict the trajectory and selects a suitable shadow vehicle based on the deviation of the predicted trajectory; then some false queries are inserted into the queries of the real vehicle and the shadow vehicle, and finally a query matrix is ​​formed and sent to the LBS server for query. Pseudo-queries make the paths of vehicles intersect, which can prevent the reverse mapping from vehicle identity to query content, making it impossible for attackers to distinguish the complete trajectory of the target vehicle. This scheme can prevent continuous query attacks. However, the scheme will also affect the accuracy of service provision.

[0006] The above schemes all have the problem of reduced privacy protection effect in sparse user space. For example, when applying k-anonymity technology, there are certain requirements for the scale of the space where the user is located. If the number of users is small, there is a problem of low security. For the protection scheme based on shadow vehicles, if high requirements are placed on the accuracy of vehicle trajectory prediction, a large number of historical trajectories are required, which is generally difficult to meet in actual scenarios. At the same time, if the attacker also has a large number of historical trajectories of vehicles, false queries and real vehicle information can be identified. In response to this problem, the research direction of previous scholars is mainly to add virtual users. For example, in response to the problem of increased correlation between user personal information and anonymous locations in sparse environments, researcher Liu Heng mentioned in the document "Research on Several Technologies for Privacy Protection Based on Location Services in Ubiquitous Computing Environments" that the problem can be solved by adding several virtual users to dilute this correlation. Wu Xu et al. proposed a virtual user generation scheme that combines spatial regional characteristics as a way to improve the authenticity of information in the document "Privacy Protection Method Based on Location Services in Sparse Environments", further improving the correlation strength in the anonymization process, thereby achieving the purpose of privacy protection. These solutions focus on the definition of sparse environments and the generation of false locations. The overhead of making the virtual information generated by the solutions highly practical is relatively high, and they are not suitable for environments with limited computing and communication resources of IoV nodes.

[0007] Based on a thorough investigation of the defects of current LBS privacy protection methods, the present invention proposes an improved K clustering LBS privacy protection method based on a dynamic feedback mechanism. The scheme uses sparse user space judgment and a dynamic K clustering algorithm. In order to clearly explain the principles and steps of the present invention, the relevant background technology is first explained.

[0008] 1. Prefix Tree

[0009] A prefix tree is a tree data structure that aims to achieve efficient search and prefix matching operations on a string set. Since the prefix tree has obvious tree structure advantages, it is suitable for storing string sets, especially for strings with a long common prefix.

[0010] The prefix tree has the characteristic of efficient search process, especially for searching strings with specified prefixes, such as the application in the paper "Research on Privacy Protection of Search Logs Based on Differential Privacy and Prefix Tree". Since strings with a common prefix share the same prefix path in the prefix tree, the search process only needs to start from the root node of the empty string and traverse downward until a target string is found, without comparing each character of the string one by one like other data structures, which greatly improves the search efficiency. In addition, the prefix tree also supports prefix matching operations, that is, it can quickly find all strings with a specified prefix. This is very useful for application scenarios such as search engines and automatic completion, and can quickly provide relevant search results or completion options.

[0011] 2. K-means clustering algorithm

[0012] K-means clustering algorithm is one of the classic unsupervised learning algorithms, which is used to divide a set of data into K disjoint clusters, each cluster consisting of the data points closest to the centroid. From a mathematical point of view, K-means clustering is a special case of the Gaussian mixture model solved by the maximum expectation algorithm when the covariance of the normal distribution is the unit matrix and the posterior distribution of the latent variable is a set of Dirac delta functions.

[0013] The basic process of the algorithm is:

[0014] First, the algorithm is initialized and K data points are randomly selected as the initial cluster centers at the beginning of the algorithm. Then, samples are assigned to each cluster and the Euclidean distance from each sample point to each cluster center is calculated, and the distance is used as an indicator to determine whether it is assigned to a cluster.

[0015]

[0016] Among them, x i represents the i-th sample point, c j represents the jth cluster center, and n represents the number of sample features.

[0017] Then, for each classified cluster, the mean of its internal samples needs to be calculated as the new cluster center and updated.

[0018]

[0019] Among them, S j represents the sample set in the jth cluster, |S j | represents the number of samples in the set.

[0020] Repeat the above process until the stop condition is met. The stop condition can be set as:

[0021] ⑴The cluster center no longer changes

[0022] ⑵ Reach the maximum number of iterations

[0023] ⑶ The change of sample points in each cluster is less than a certain threshold. The final K-means clustering center point can be obtained.

[0024] 3. Linear interpolation algorithm

[0025] The linear interpolation algorithm is a simple and common interpolation method, which is often used to estimate the value of new data points between known data points. The paper "Application of Linear Interpolation in Improving GPS Common View Time Comparison Performance" describes a typical application of the linear interpolation algorithm. Linear interpolation is based on the following assumption: between two adjacent known data points, the change between the data points is linear. Specifically, if there are two known data points (x1, y1) and (x2, y2), the linear interpolation algorithm assumes that at any point x between these two points, its corresponding y value can be calculated by the following formula:

[0026]

[0027] This formula is based on the equation of a line y=mx+b, where m is the slope, given by (y2-y1) / (x2-x1), and b is the y-intercept, given by y1-(m*x1).

[0028] In the LBS scenario of the Internet of Vehicles, the linear interpolation method mainly connects the real user location information contained in the request packets sent at different times to obtain the real user's driving trajectory. For malicious LBS service providers, this method can be used to interpolate points that are not smooth in time and space one by one, and then analyze the user's travel patterns. Summary of the invention

[0029] In view of the above-mentioned technical problems, the present invention provides an improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism. By defining the regular user environment and the sparse user environment, a threshold judgment and feedback mechanism is established, and an improved K-clustering algorithm with region division and dynamic K value is used to realize group segmentation of the regular user space. For the sparse user space, the feedback mechanism is used to adjust the region division to adapt to actual needs, which can simultaneously adapt to the security strength requirements of the regular user space and the sparse user space in the LBS privacy protection of the Internet of Vehicles.

[0030] The idea of ​​the present invention is: comprehensively consider the two elements of safety and efficiency, respond to the requirements of anonymity, high efficiency and high service quality in the LBS environment, hide the real position of the target vehicle through the improved K clustering technology, and realize anonymous service. At the same time, a sparse user space judgment and feedback algorithm with low computational overhead is adopted to solve the problem of weak privacy protection when there are too few users. Innovative improvements are made when using the K-means clustering algorithm. The reason is that although the traditional K-means clustering algorithm is feasible in each operation, under LBS, even if the requirements of the conventional user space are met, the value of K in the traditional K-means clustering algorithm may not be reasonable. For example, there are more real users requesting LBS services in a certain area, and the number of real users requesting LBS services in another area is small, but the requirements of the sparse user space are not met. At this time, it is impossible to obtain a more suitable clustering result through the mean clustering algorithm with a fixed K value. Therefore, in order to meet the actual needs of LBS services, the present invention further proposes a dynamic K-means clustering algorithm, which aims to use the dynamic K value to make the clustering results focus more on service quality in areas with a large number of users, and focus more on protection strength in areas with a small number of users. The improved algorithm is described as: confirming the initial K value, determining the upper and lower limits of the cluster size n, the lower limit min is the same as the trigger condition of the feedback mechanism, that is, whether it is a sparse user space threshold, and the upper limit can be adjusted, generally twice the lower limit. Then compare n in the target area with the upper and lower limits, and adjust it if the n value is inappropriate. The improved algorithm can obtain a relatively suitable value for the number of cluster center points, while improving the privacy protection strength and ensuring the quality of LBS services. The specific description of the implementation steps is described in detail below.

[0031] The present invention is achieved by the following measures: an improved K clustering LBS privacy protection method based on a dynamic feedback mechanism, comprising the following steps:

[0032] S1, RSU performs initial regional division;

[0033] S2. After the vehicle sends the LBS request, the RSU performs sparse state feedback adjustment based on the space conditions of the vehicle and completes the area redivision;

[0034] S3, RSU dynamically adjusts the value of parameter K, executes K-means clustering algorithm, and generates appropriate cluster value n;

[0035] S4, RSU replaces the real location of the vehicle with the cluster center point and sends it to the LBS server;

[0036] S5. RSU forwards the response result returned by the LBS server to the corresponding OBU;

[0037] Furthermore, the S1 step includes:

[0038] S11, RSU first takes the distance between the farthest on-board unit data packet and the partition subject as the optimal communication distance R max , or preset a conservative value before the algorithm starts, and construct a geographical proximity area with the optimal communication distance as the radius;

[0039] S12, RSU determines relevant parameters, such as: the number of innermost area divisions, the quantitative relationship between the inner and outer layers and the number of division layers, calculates the radius of the inner circle in each layer and the number of area divisions in the ring, and completes the initial division of the area;

[0040] S121, define the number of innermost area divisions Num_1, and the linear relationship between the number of inner and outer area divisions: num_outer == N*num_inner+M, calculate the number of area divisions of each layer num_i, i∈[2,Layer_num];

[0041] S122. Calculate Num_i=Num_(i-1)+num_i according to num_i obtained in S121, where i∈[2,Layer_num], and obtain the total number of area divisions of the first i layers;

[0042] S123. Calculate the radius of the inner circle on each layer according to the optimal communication distance r_max and the total number of the first i-layer area divisions obtained in S122. The calculation method is: r_m = sqrt((r_max**2)*Num_m / Num_4), m∈[1,Layer_num];

[0043] S124, generating the initial state division of this region according to the relevant parameters generated in the above steps;

[0044] S13, the RSU determines the grouping of target vehicles according to the radius of the inner circle in each layer and the angle between the vehicle and the division subject calculated in step S12;

[0045] S131, RSU obtains the real position locu (lu_x, lu_y) of the target vehicle, and uses the formula distance = sqrt ((lu_x-lu_a) ** 2 + (lu_y-lu_b) ** 2), angle = atan2 (lu_y-lu_b, lu_x-lu_a) to calculate the distance and horizontal angle between the target vehicle and the divided subject respectively;

[0046] S132, using the calculated distance and angle to determine the target vehicles by area grouping;

[0047] Furthermore, the S2 step includes:

[0048] S21, RSU first determines the grouping ID of the area where the current user is located, and then determines whether the number of real users in the target area meets the data demand L given by the K-means clustering algorithm. If not, it can be determined that the current area is in a sparse user space state and needs to enter step S22 to re-divide the area. If it is satisfied, skip step S2;

[0049] S22: If the user is determined to be in a sparse user space, feedback adjustment is performed under the sparse user space state to complete region redivision;

[0050] In step S22, the specific process of feedback adjustment in the sparse user space state includes the following steps:

[0051] S221: The linear relationship between the number of outer regions num_outer and the number of inner regions num_inner is defined as:

[0052] num_outer = num_inner*N+M, where N and M are adjustable parameters, and their values ​​are both non-negative integers. By adjusting the values ​​of N and M and the number of division layers, the area where the user is located is a non-sparse user space.

[0053] In step S221, the specific process includes the following steps:

[0054] S2211. First, try to adjust the value of M so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space.

[0055] S2222. If the M value has been adjusted to 0 and the space is still sparse, start adjusting the parameter N so that its value decreases by 1 each time.

[0056] Synchronously adjust M again to ensure that the number of regions is adjusted to 1 each time until the area where the user is located is a non-sparse user space.

[0057] S2223. If the above steps cannot achieve the normalization of the sparse space, further adjust the number of area division layers so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space;

[0058] This determination process can realize a feedback mechanism, thereby adjusting the region division algorithm in real time according to the regular or sparse user space appearing in a certain area;

[0059] Furthermore, the S3 step includes:

[0060] S31. First, define the upper and lower limits (max, min) of the initial K value and the cluster size n in the K-means clustering algorithm, where the lower limit min of the cluster size n is the threshold for judging the sparse user space, that is, the trigger condition of the feedback mechanism, and the upper limit max can be adjusted, generally taking a value at least twice greater than min;

[0061] S32. Dynamically adjust the value of K to control the size of the clusters, and execute the K-means clustering algorithm each time to control the size of the cluster n. Until the cluster size n meets the conditions. The process is: if the current result n≤min, it means that for the selected K value, the clusters in the area themselves have sparse user space, that is, the divided clusters contain too few real users. The value of K needs to be reduced by 1 each time; if the current result n≥min, it means that for the selected K value, the clusters in the area themselves are too large, and the size of the clusters can be controlled by increasing the value of K, adding 1 each time;

[0062] Furthermore, the S4 step includes:

[0063] S41. The RSU uses the cluster center point data in each cluster generated in step S32 to replace the real location in the user's LBS service request packet, and forwards the replaced request packet to the LBS server.

[0064] Furthermore, the step S5 includes:

[0065] S51. The LBS server returns a response result to the RSU, and the RSU forwards it to the corresponding OBU.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] (1) The present invention proposes an improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism, and proposes a region division algorithm based on the prefix tree concept. The algorithm comprehensively considers the actual needs and the region determination problem in the simple calculation process. Based on the proposed region division algorithm, the present invention proposes a single-time point location protection scheme in combination with the clustering algorithm in machine learning. The scheme can be applied to both sparse user space and conventional user space, solving the problem of low security when the number of users is too small. The main process is to formulate a set of sparse user space judgment criteria and feedback adjustment mechanism. It aims to deal with the problem that when users request LBS services in real time, the privacy protection strength is low, which may lead to the collection and leakage of real locations due to the small number of users in the area. This scheme generates cluster center points through the real-time location of the real vehicle, which guarantees the service quality.

[0068] (2) The improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of the present invention is to make the clustering results focus more on service quality in areas with a large number of users and more on protection strength in areas with a small number of users. A dynamic K-means clustering algorithm is proposed to adjust the inappropriate K value in the conventional user space to obtain a K value that will make the clustering result better. This innovative solution provides a safe, efficient and reliable location protection solution for LBS privacy protection.

[0069] (3) The improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of the present invention takes into account both privacy protection and LBS service quality. Each vehicle can obtain corresponding services without exposing its real location. The sparse users and feedback mechanism adopted by the scheme solve the problem of the strength of privacy protection when there are too few users. The feasibility of the scheme is proved by simulation trajectory, and the performance comparison with other schemes also shows that the present invention guarantees LBS service quality while being safe. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0071] Figure 1 The present invention provides an overall flow chart of an improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism.

[0072] Figure 2 It is a schematic diagram of the service process in the present invention.

[0073] Figure 3 It is a schematic diagram of the LBS system architecture in the present invention.

[0074] Figure 4 This is a schematic diagram of the region division algorithm provided by the present invention.

[0075] Figure 5 This is a schematic diagram of the tree structure of the region partitioning algorithm provided by the present invention.

[0076] Figure 6 A schematic diagram of the central source trajectory containing the region division algorithm provided by the present invention.

[0077] Figure 7 A representation diagram of the simulated trajectory provided by the present invention on a tree structure.

[0078] Figure 8 A schematic diagram of an LBS location protection solution based on area division and K clustering algorithm provided by the present invention.

[0079] Fig. 9 A schematic diagram of the visualization of initial area division provided by the present invention.

[0080] Fig.10 This is a schematic diagram after the random vehicle positions provided by the present invention are generated.

[0081] Fig.11 This is a visualized area division diagram adjusted by the feedback mechanism provided by the present invention.

[0082] Fig.12 This is a diagram showing the final clustering results provided by the present invention.

[0083] Fig.13 This is a simulated trajectory diagram of the random source used for testing provided by the present invention.

[0084] Fig.14 A schematic diagram of the simulated trajectory and cluster points in the nearby area provided by the present invention.

[0085] Fig.15 This is a comparison chart between the cluster point linear interpolation result and the original trajectory provided by the present invention.

[0086] Fig.16 This is a comparison diagram between the linear interpolation results of clustered points provided by the present invention and the original trajectory, and a comparison diagram between the linear interpolation results of clustered points provided by other schemes and the original trajectory. DETAILED DESCRIPTION

[0087] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0088] Example 1

[0089] This embodiment 1 provides an improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism, such as Figure 1 As shown, the following steps are included:

[0090] S1, RSU performs initial regional division;

[0091] S2. After the vehicle sends the LBS request, the RSU performs sparse state feedback adjustment based on the space conditions of the vehicle and completes the area redivision;

[0092] S3, RSU dynamically adjusts the value of parameter K, executes K-means clustering algorithm, and generates appropriate cluster value n;

[0093] S4, RSU replaces the real location of the vehicle with the cluster center point and sends it to the LBS server;

[0094] S5. RSU forwards the response result returned by the LBS server to the corresponding OBU;

[0095] like Figure 4 Visualize the region division results. S1 specifically includes the following steps:

[0096] S11, RSU determines the optimal communication distance r_max of the partition subject, takes the distance between the farthest acceptable vehicle unit data packet and the partition subject center_location (lu_a, lu_b) as the radius, or presets a conservative value before the algorithm starts, and constructs the geographical proximity area R max ;

[0097] S12, RSU determines the number of innermost area divisions Num_1, the quantitative relationship Qr of the area divisions (the quantitative relationship between the inner layer and the outer layer) and the number of division layers Layer_num and other parameters, and calculates the radius r_j of the inner circle, j∈[1,Layer_num], and the number num_i of the area divisions in each layer of the ring, i∈[1,Layer_num] according to the determined quantitative relationship;

[0098] S121, Qr means the relationship between the number of outer regions and the number of inner regions: num_outer == N*num_inner+M. Num_1 is a known condition. According to the known condition and the quantitative relationship, the number of regions divided in each layer is calculated, num_i, i∈[1,Layer_num];

[0099] S122, define Num_i,i∈[1,Layer_num] as the total number of area divisions of the first i layers, the RSU obtains num_i obtained in step S121, and calculates the total number of area divisions of the first i layers using the formula: Num_i=Num_(i-1)+num_i,i∈[2,Layer_num];

[0100] S123, using r_m=sqrt((r_max**2)*Num_m / Num_4), m∈[1,Layer_num], calculate the radius of the inner circle on each layer;

[0101] S124. According to the relevant parameters generated in the above steps, the RSU divides the initial state of the region;

[0102] S13, RSU according to r_j and num_i obtained in S12, the current geographical adjacent area R max The vehicles in the area are grouped according to the area division results. When grouping the area to which the vehicle belongs, the main parameters are the distance to the main body and the horizontal angle;

[0103] S131, RSU obtains the real position locu(lu_x,lu_y) of the target vehicle, and uses distance=sqrt((lu_x-lu_a)**2+(lu_y-lu_b)**2) and angle=atan2(lu_y-lu_b,lu_x-lu_a) to calculate the distance and horizontal angle between the target vehicle and the divided subject respectively;

[0104] S132, using the calculated distance and angle to determine the target vehicles by area grouping;

[0105] The complete algorithm implementation code of S1 is as follows:

[0106]

[0107]

[0108] The specific contents of S2 include the following steps:

[0109] S21, RSU first determines the grouping ID of the area where the current user is located, and then determines whether the number of real users in the target area meets the data demand L given by the K-means clustering algorithm. If not, it can be determined that the current area is in a sparse user space state and needs to enter step S22 to re-divide the area. If it is satisfied, skip step S2;

[0110] The specific implementation code of S21 is described as follows:

[0111]

[0112] S22: If the user is determined to be in a sparse user space, feedback adjustment is performed under the sparse user space state to complete region redivision;

[0113] In step S22, the specific process of feedback adjustment in the sparse user space state includes the following steps:

[0114] S221: Define the number of inner regions as num_inner, then the linear relationship between the number of outer regions num_outer and num_inner is defined as: num_outer = num_inner*N+M, where N and M are adjustable parameters, and their values ​​are both non-negative integers. By adjusting the values ​​of N and M and the number of division layers, the area where the user is located is a non-sparse user space.

[0115] In step S221, the specific process includes the following steps:

[0116] S2211. First, try to adjust the value of M so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space.

[0117] S2222. If the M value has been adjusted to 0 and the space is still sparse, start adjusting the parameter N so that its value decreases by 1 each time.

[0118] Synchronously adjust M again to ensure that the number of regions is adjusted to 1 each time until the area where the user is located is a non-sparse user space.

[0119] S2223. If the above steps cannot achieve the normalization of the sparse space, further adjust the number of area division layers so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space;

[0120] Through this determination process, a feedback mechanism can be implemented, thereby adjusting the region partitioning algorithm in real time according to the regular or sparse user space appearing in a certain area.

[0121] The specific contents of S3 include the following steps:

[0122] S31. First, define the upper and lower limits (max, min) of the initial K value and the cluster size n in the K-means clustering algorithm, where the lower limit min of the cluster size n is the threshold for judging the sparse user space, that is, the triggering condition of the feedback mechanism, and the upper limit max can be adjusted, generally taking a value at least twice greater than min; N represents the total number of users;

[0123] S32. Dynamically adjust the value of K to control the size of the clusters, and execute the K-means clustering algorithm each time to control the size of the cluster n. Until the cluster size n meets the conditions. The process is: if the current result n≤min, it means that for the selected K value, the clusters in the area themselves have sparse user space, that is, the divided clusters contain too few real users. The value of K needs to be reduced by 1 each time; if the current result n≥min, it means that for the selected K value, the clusters in the area themselves are too large, and the size of the clusters can be controlled by increasing the value of K, adding 1 each time;

[0124] The specific implementation code of S3 is described as follows:

[0125] Pseudo code for dynamic K value:

[0126]

[0127] like Figure 8 As shown, the present invention takes into account that too little user space or an unreasonable K value will have a negative impact on LBS privacy protection, and proposes an improved K clustering LBS privacy protection method based on a dynamic feedback mechanism.

[0128] The specific contents of S4 include the following steps:

[0129] S41, the RSU uses the cluster center point data in each cluster generated in step S32 to replace the real location in the user's LBS service request packet, and forwards the replaced request packet to the LBS server to request LBS service;

[0130] The specific contents of S5 include the following steps:

[0131] S51. The LBS server returns a result based on the content of interest in the data packet, and the RSU then forwards the result of the location service to all real users in the cluster where the vehicle is located.

[0132] In order to verify the feasibility of the present invention, the effectiveness of this embodiment 1 is analyzed as follows:

[0133] 1. The visualization results after randomly generating users are as follows Fig.10 As shown in the figure, based on the RSU area division, the area division result of the third layer from the inside out is used as the restricted space for generating random points. The blue is the generated random point, and the green is the real position of all vehicles in the area where the simulated vehicle is located when requesting LBS service. After a simple statistics, the number of users in the current area of ​​the user is 6. According to the definition of sparse user space, the user is in the sparse user space and needs to enter the feedback mechanism for adjustment.

[0134] After the feedback mechanism is adjusted, the results are as follows Fig.11 The code for dividing the visualization area is shown below:

[0135]

[0136]

[0137] The visualization results after the feedback algorithm processing are as follows Fig.11 , where the green part is the location of all real users in the area where the user currently requesting LBS service is located. After statistics, the total number of users in the area is 24, which does not fall within the definition of sparse user space. However, when the number of mean cluster center points is 1, the number of members in the cluster is greater than the specified max value, so the value of K needs to be adjusted before clustering. After adjusting the K value, we get the final clustering result; clustering and coloring the location information of all users in the area, the result is as follows Fig.12 shown.

[0138] According to the specific point conditions, a total of 2 cluster centers are obtained, and each cluster represented by each cluster center is colored and marked. It is not difficult to see from this figure that the processing of cluster members is relatively uniform after adjustment according to actual needs. For the base station, the status of the user request data set it receives is the colored dot, and for the LBS third party, the location of the user request service it receives from the base station is the red cross. This meets the requirements of protecting the location of real users and privacy from being leaked in LBS services, while also taking into account the quality of LBS services obtained by real users.

[0139] 2. Under the random trajectory generation method, the solution in this embodiment is safe for users and takes into account the quality of service. ,The analysis is as follows:

[0140] First, we conduct experimental verification. The code for random trajectory generation is as follows:

[0141]

[0142] x=max(0,min(x,max_x))

[0143] y=max(0,min(y,max_y))

[0144] timestamp+=1

[0145] trajectory.append(TrajectoryNode(x,y,timestamp))

[0146] angle_history.append(current_angle)

[0147] trajectories.append(trajectory)

[0148] return trajectories

[0149] In order to restore the real vehicle trajectory as much as possible, the experiment divides the entire trajectory into points. When each point is generated, the angle change from the initial direction is automatically checked, and the standard deviation of the angle change and the maximum angle change of adjacent trajectory points are set. After testing, it is effectively restored that the real vehicle has a roughly consistent direction during driving, and is not as smooth as the function trajectory. The simulated trajectory of the vehicle is as follows Fig.13 shown.

[0150] according to Fig.13It is not difficult to see that it has passed through three areas. Assuming that it has requested LBS services four times in the three areas, the vehicle user trajectory information in the three areas is retrieved and clustered at the time of the four LBS services. The user group that the random trajectory of the vehicle generated this time has passed through is determined by the coloring result, and the cluster center point of the user group that has passed through is marked in red. The clustering result after marking is as follows: Fig.14 shown.

[0151] Will Fig.14 The red cross part is drawn by linear interpolation and compared with the original trajectory diagram. The comparison diagram is as follows Fig.15 As shown in the figure. The blue part is the real trajectory of the vehicle, and the red line is the location point collected by the LBS third party when the vehicle requested the LBS service. It can be seen that the trajectory data collected by the LBS third party is somewhat different from the actual situation, which to a certain extent guarantees the location privacy and security of users during driving. In addition, the generated trajectory has an intersection with the original trajectory and generally has the same direction. This feature also takes into account the quality of LBS services obtained by real users.

[0152] Therefore, this embodiment fully implements this solution and verifies its effectiveness and safety.

[0153] Example 2

[0154] To further verify the feasibility of the present invention, Example 2 compares the functionality and feasibility of the present invention with other LBS privacy protection methods, including the disturbance trajectory protection method proposed by Xu Chuan et al. in the document "Personalized Location Privacy Protection Based on Location Services in Internet of Vehicles". The comparison of the linear interpolation results of cluster points and the original trajectory in the present invention and the above method is shown in the figure below: Fig.16 shown.

[0155] The red triangle line segment is the trajectory result obtained after linear interpolation of discrete points without privacy protection, the green cross line segment is the trajectory result processed by the perturbation trajectory protection scheme, and the blue square line segment is the linear interpolation result of this method after processing the points at each service request.

[0156] Comparison of the linear interpolation results of cluster points in the present invention and the above method with the original trajectory Fig.16 It can be seen that compared with the perturbation trajectory privacy protection method, the linear interpolation results of this method for discrete real trajectory points have a relatively average correlation coefficient, and there are no points with prominent differences. While ensuring the quality of service, it also takes into account the requirements of privacy protection strength and the strong correlation is not easy to be identified.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An improved K-clustering LBS privacy protection method based on a dynamic feedback mechanism, comprising the following steps: S1, RSU performs initial regional division; S2. After the vehicle sends the LBS request, the RSU performs sparse state feedback adjustment according to the space conditions where the vehicle is located. Complete regional redistricting; S3, RSU dynamically adjusts the parameter K value, executes the K-means clustering algorithm, and generates an appropriate clustering value n; S4, RSU replaces the real location of the vehicle with the cluster center point and sends it to the LBS server; S5. The RSU forwards the LBS server response result to the corresponding OBU.

2. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism according to claim 1, the step S1 The following steps are involved: S11, RSU takes the distance between the farthest received on-board unit data packet and the divided subject as the optimal communication distance, or presets a conservative value before the algorithm starts, and constructs a geographical proximity area with the optimal communication distance as the radius; S12, RSU determines relevant parameters, the number of innermost area divisions, the quantitative relationship between the inner and outer layers and the number of division layers, calculates the radius of the inner circle in each layer and the number of area divisions in the ring, and completes the initial division of the area; S13. The RSU determines the grouping of target vehicles according to the radius of the inner circle in each layer and the angle between the vehicle and the divided body calculated in step S12.

3. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of claim 1, in step S12, the process of calculating the radius of the inner circle and the number of area divisions in each layer of the ring comprises the following steps: S121, defining a linear relationship between the number of inner and outer regions according to the number of innermost region divisions, and calculating the number of region divisions in each layer; S122, according to the number of regional divisions of each layer obtained in the previous step S121, the RSU calculates the total number of regional divisions of the first i layers respectively; S123, according to the total number of area divisions of the first i layers and the optimal communication distance, the RSU calculates the radius of the inner circle on each layer; S124. According to the relevant parameters generated in the above steps, the RSU performs initial state division of the area.

4. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of claim 1, in the step S13, the process of determining the area to which a certain vehicle belongs includes the following steps: S131, RSU obtains the real position of the target vehicle, and calculates the distance and horizontal angle between the target vehicle and the divided subject respectively; S132: grouping target vehicles into regions using distance and angle.

5. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism according to claim 1, step S2 The following steps are involved: S21, RSU determines whether the number of real users in the target area meets the data demand L given by the K-means clustering algorithm. If not, the current area is judged to be in a sparse user space state, and it is necessary to enter step S22 to re-divide the area. If it is satisfied, skip step S2; S22, performing feedback adjustment in the sparse user space state to complete region redivision; In step S22, the specific process of feedback adjustment in the sparse user space state includes the following steps: S221: Define the number of inner regions as num_inner, and the number of outer regions as num_outer. The linear relationship between the two is defined as: num_outer = num_inner*N+M, where N and M are adjustable parameters, and their values ​​are both non-negative integers. By adjusting the values ​​of N and M and the number of division layers, the area where the user is located is a non-sparse user space.

6. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of claim 5, the step S221 comprises the following steps: S2211, first try to adjust the value of M so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space; S2222. If the M value has been adjusted to 0 and the space is still sparse, start adjusting the parameter N so that its value decreases by 1 each time. Synchronously adjust M again to ensure that the number of regions is 1 each time, until the area where the user is located is a non-sparse user space; S2223. If the above steps cannot achieve the normalization of the sparse space, the number of area division layers is further adjusted so that its value decreases by 1 each time until the area where the user is located is a non-sparse user space.

7. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism according to claim 1, the step S3 The following steps are involved: S31, define the upper and lower limits (max, min) of the initial K value and the cluster size n in the K clustering algorithm, where the lower limit min of the cluster size n is the threshold for judging the sparse user space, that is, the triggering condition of the feedback mechanism, and the upper limit max can be adjusted to be at least twice greater than min; S32, dynamically adjust the value of K, and execute the K-means clustering algorithm to control the size of the cluster n. The process is: if the current n≤min, reduce the value of K by 1 each time; otherwise, increase the value of K by 1 each time until the cluster size n meets the conditions.

8. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism according to claim 1, the step S4 process The following steps are involved: S41. The RSU uses the cluster center point data in each cluster generated in step S32 to replace the real location in the user's LBS service request packet, and forwards the replaced request packet to the LBS server.

9. According to the improved K-clustering LBS privacy protection method based on dynamic feedback mechanism of claim 1, the step S5 The following steps are involved: S51. The LBS server returns a response result to the RSU, and the RSU forwards it to the corresponding OBU.

Images