Vehicle position privacy protection method and system based on local differential privacy

Abstract

The invention relates to the field of privacy protection, in particular to a vehicle position privacy protection method and system based on local differential privacy, and the method comprises the steps: dividing a target region where a vehicle runs to obtain at least one large region, and dividing each large region to obtain a plurality of small regions; obtaining a vehicle number estimation total variance of each small area in the large area based on a disturbance overturning probability of epsilon-local differential privacy, and adding the vehicle number estimation total variances of all the small areas in the large area to obtain a comprehensive total variance; obtaining an optimization function based on the comprehensive total variance, and solving the optimization function to obtain an optimal disturbance flipping probability; and the optimal disturbance overturning probability is used as the disturbance overturning probability of epsilon-local differential privacy to disturb the position information of the vehicle, and privacy protection of the vehicle position is realized. According to the invention, the optimal balance between privacy protection and data availability can be realized, and the evaluation accuracy of vehicle distribution is improved.

Description

Technical Field

[0001] The present invention relates to the field of privacy protection, and in particular to a vehicle location privacy protection method and system based on local differential privacy. Background Art

[0002] With the development of vehicle-road-cloud multi-network integration Internet of Vehicles technology, the location information of vehicles is interactively shared between the vehicle side, road side and cloud side, and hackers intercept a large amount of vehicle location data through network attacks. On the one hand, the location information of connected vehicles contains key national basic information, which may be mastered by criminals. For example, by analyzing the density and flow of vehicles, the geographical location of sensitive infrastructure can be inferred, so it may be used for precise strikes for malicious purposes; on the other hand, the time-series location data of the same vehicle constitutes trajectory information, and trajectory information can reveal sensitive information such as personal living habits, whereabouts and even home addresses. Once this data is illegally obtained and used, it will cause serious infringement on personal privacy. For example, some criminals use trajectory information to commit theft, kidnapping or other criminal activities.

[0003] Differential privacy technology does not rely on the background knowledge of the attacker, and provides a strict and quantifiable level of privacy protection. The relationship between privacy and data effects can be balanced by adjusting the privacy budget. Differential privacy provides a powerful privacy protection mechanism that can resist differential attacks, background knowledge attacks, connection attacks and other privacy attacks. It is especially suitable for location data privacy protection in data mining and multi-network fusion scenarios in the big data era. Local differential privacy is the application of differential privacy technology in local scenarios. It does not need to rely on a trusted third party. Users can independently perturb their own data locally and then submit the perturbed data to the service provider. The local processing mechanism has a good protection effect on the user's personal privacy. For example, Chinese patent CN111915891A discloses a real-time road condition detection method based on local differential privacy. It performs ε-local differential privacy protection on the real location data of the vehicle, that is, perturbing the real location data of the vehicle, so that the server cannot know the real location of the vehicle, but can still build a real-time road condition map and provide accurate real-time road conditions to the outside world. However, the disturbance probability of this method is completely based on experience or manual settings, which makes it difficult to achieve the optimal balance between privacy protection and data availability, and thus it is difficult to ensure the accuracy of the final vehicle distribution assessment. Summary of the invention

[0004] In order to solve the problem that local differential privacy is difficult to achieve the optimal balance between privacy protection and data availability, and thus it is difficult to ensure the accuracy of the final vehicle distribution assessment, the present invention proposes a vehicle location privacy protection method based on local differential privacy, including:

[0005] Dividing the target area of ​​vehicle operation into at least one large area, and dividing each large area into a plurality of small areas;

[0006] The total variance of the estimated number of vehicles in each small area in the large area is obtained based on the perturbation flip probability of ε-local differential privacy, and the total variance of the estimated number of vehicles in all small areas in the large area is summed up to obtain the comprehensive total variance;

[0007] Obtaining an optimization function based on the comprehensive total variance, and solving the optimization function to obtain an optimal disturbance flipping probability;

[0008] The optimal perturbation flip probability is used as the perturbation flip probability of ε-local differential privacy to perturb the vehicle's location information, thereby achieving privacy protection of the vehicle's location.

[0009] Optionally, obtaining an optimization function based on the comprehensive total variance includes:

[0010] Integrating the comprehensive total variance and discarding constant coefficients to obtain a primary optimization function;

[0011] Based on the unary encoding mechanism of ε-local differential privacy, the correlation between the perturbation flip probability and the privacy budget of large regions in ε-local differential privacy is obtained;

[0012] The primary optimization function is simplified by using the association relationship to obtain an optimization function.

[0013] Optionally, the privacy budget of a large region in the ε-local differential privacy is obtained by the following steps:

[0014] Obtain the correlation level score between each small area in the large area and personal privacy;

[0015] The correlation level scores corresponding to each small area are summed up to obtain the comprehensive correlation level score of the corresponding large area;

[0016] A privacy budget for a large area is determined based on the comprehensive level score of the association degree.

[0017] Optionally, solving the optimization function to obtain an optimal disturbance flip probability includes:

[0018] The optimized function is solved by using a simulated annealing algorithm to obtain an optimal disturbance flip probability, wherein the state variable of the simulated annealing algorithm is the disturbance flip probability, and the optimization goal of the simulated annealing algorithm is to minimize the optimized function.

[0019] Optionally, the acceptance probability P of changing from the state q(n) of the nth iteration to the state q(n+1) of the n+1th iteration in the simulated annealing algorithm is:

[0020]

[0021] Among them, F(n) is the value of the optimization function obtained in the nth iteration, F(n + 1) is the value of the optimization function obtained in the (n + 1)th iteration, and T is the temperature.

[0022] Optionally, the step of using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to perturb the location information of the vehicle includes:

[0023] Performing unary coding on the location information of the vehicle to obtain a one-dimensional vector;

[0024] Using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to flip the elements in the one-dimensional vector to achieve perturbation of the vehicle location information.

[0025] Optionally, the optimal perturbation flipping probability includes a first optimal perturbation flipping probability and a second optimal perturbation flipping probability. After using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy, the perturbation flipping probability of ε-local differential privacy is:

[0026]

[0027] Among them, p* is the first optimal perturbation flipping probability, q* is the second optimal perturbation flipping probability, A[j] is the value of the jth element in vector A, B[j] is the value of the jth element after perturbation, Pr[·] is the probability function, 0 < q* < p* < 1, and ε is the privacy budget.

[0028] Optionally, the step of performing unary coding on the location information of the vehicle to obtain a one-dimensional vector includes:

[0029] Numbering the large area and the small areas in the large area in sequence;

[0030] Obtaining the GPS location information of the vehicle;

[0031] Judging the large area and the corresponding small area where the GPS location is located;

[0032] Assigning the value 1 to the small area where the vehicle is located in the large area, and assigning the value 0 to the remaining small areas, then a one-dimensional vector is obtained.

[0033] Optionally, after perturbing the location information of the vehicle based on the optimal perturbation flipping probability, it further includes:

[0034] Transmitting the location vector obtained after perturbation to the cloud management center or roadside infrastructure device to perform aggregation analysis on the location vector to obtain the estimated number of vehicles corresponding to each small area;

[0035] The estimated number of vehicles is corrected by using maximum likelihood estimation to obtain an unbiased estimate of the number of vehicles corresponding to each small area, and the vehicle distribution of the entire large area is obtained based on the unbiased estimate of the number of vehicles corresponding to each small area.

[0036] On the other hand, the present application proposes a vehicle location privacy protection system based on local differential privacy, comprising:

[0037] An area division module is used to divide the target area of ​​vehicle operation into at least one large area, and divide each large area into multiple small areas;

[0038] The variance acquisition module is used to obtain the total variance of the estimated number of vehicles in each small area in the large area based on the perturbation flip probability of ε-local differential privacy, and to sum the total variance of the estimated number of vehicles in all small areas in the large area to obtain the comprehensive total variance;

[0039] An optimal disturbance flip probability acquisition module is used to obtain an optimization function based on the comprehensive total variance, and solve the optimization function to obtain an optimal disturbance flip probability;

[0040] The perturbation module is used to perturb the vehicle's location information by using the optimal perturbation flip probability as the perturbation flip probability of ε-local differential privacy, so as to achieve privacy protection of the vehicle's location.

[0041] Optionally, the optimal disturbance flip probability acquisition module obtains an optimization function based on the comprehensive total variance, including:

[0042] Integrating the comprehensive total variance and discarding constant coefficients to obtain a primary optimization function;

[0043] Based on the unary encoding mechanism of ε-local differential privacy, the correlation between the perturbation flip probability and the privacy budget of large regions in ε-local differential privacy is obtained;

[0044] The primary optimization function is simplified by using the association relationship to obtain an optimization function.

[0045] Optionally, the privacy budget of a large region in the ε-local differential privacy in the variance acquisition module is obtained by the following steps:

[0046] Obtain the correlation level score between each small area in the large area and personal privacy;

[0047] The correlation level scores corresponding to each small area are summed up to obtain the comprehensive correlation level score of the corresponding large area;

[0048] A privacy budget for a large area is determined based on the comprehensive level score of the association degree.

[0049] Optionally, the optimal perturbation flipping probability obtaining module solves the optimization function to obtain the optimal perturbation flipping probability, including:

[0050] Solving the optimization function by using a simulated annealing algorithm to obtain the optimal perturbation flipping probability, where the state variable of the simulated annealing algorithm is the perturbation flipping probability, and the optimization objective of the simulated annealing algorithm is to minimize the optimization function.

[0051] Optionally, in the simulated annealing algorithm of the optimal perturbation flipping probability obtaining module, the acceptance probability P for changing from the state q(n) in the nth iteration to the state q(n + 1) in the (n + 1)th iteration is:

[0052]

[0053] where F(n) is the value of the optimization function obtained in the nth iteration, F(n + 1) is the value of the optimization function obtained in the (n + 1)th iteration, and T is the temperature.

[0054] Optionally, the perturbation module uses the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to perturb the location information of the vehicle, including:

[0055] Performing one-hot encoding on the location information of the vehicle to obtain a one-dimensional vector;

[0056] Using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to flip the elements in the one-dimensional vector to achieve perturbation of the vehicle location information.

[0057] Optionally, the optimal perturbation flipping probability in the perturbation module includes a first optimal perturbation flipping probability and a second optimal perturbation flipping probability. After using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy, the perturbation flipping probability of ε-local differential privacy is:

[0058]

[0059] where p* is the first optimal perturbation flipping probability, q* is the second optimal perturbation flipping probability, A[j] is the value of the jth element in vector A, B[j] is the value of the jth element after perturbation, Pr[·] is the probability function, 0 < q* < p* < 1, and ε is the privacy budget.

[0060] Optionally, the perturbation module performs one-hot encoding on the location information of the vehicle to obtain a one-dimensional vector, including:

[0061] Numbering large regions and small regions within the large regions in sequence;

[0062] Obtaining the GPS location information of the vehicle;

[0063] Determine the large area where the GPS location is located and the corresponding small area;

[0064] In the large area, the small area where the vehicle is located is assigned a value of 1, and the other small areas are assigned a value of 0, and a one-dimensional vector is obtained.

[0065] Optionally, after the disturbance module disturbs the position information of the vehicle based on the optimal disturbance flip probability, the method further includes:

[0066] The position vector obtained after the disturbance is transmitted to the cloud management center or the roadside infrastructure equipment to perform aggregation analysis on the position vector to obtain the estimated number of vehicles corresponding to each small area;

[0067] The estimated number of vehicles is corrected by using maximum likelihood estimation to obtain an unbiased estimate of the number of vehicles corresponding to each small area, and the vehicle distribution of the entire large area is obtained based on the unbiased estimate of the number of vehicles corresponding to each small area.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] The present invention provides a vehicle location privacy protection method and system based on local differential privacy, the method comprising dividing a target area for vehicle operation to obtain at least one large area, and dividing each large area to obtain a plurality of small areas; obtaining the estimated total variance of the number of vehicles in each small area in the large area based on the perturbation flip probability of ε-local differential privacy, and summing up the estimated total variance of the number of vehicles in all small areas in the large area to obtain a comprehensive total variance; obtaining an optimization function based on the comprehensive total variance, and solving the optimization function to obtain an optimal perturbation flip probability; using the optimal perturbation flip probability as the perturbation flip probability of ε-local differential privacy to perturb the location information of the vehicle, so as to achieve privacy protection of the vehicle location; the present application divides the target area into a large area and a small area, constructs an optimization function with the flip probability as a parameter based on the small area, and obtains the perturbation flip probability of ε-local differential privacy by solving the optimization function, thereby avoiding the subjective problem caused by artificial setting, achieving the optimal balance between privacy protection and data availability, and improving the accuracy of vehicle distribution assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A schematic diagram of the steps of the vehicle location privacy protection method based on local differential privacy proposed in the present invention;

[0071] Figure 2 The present invention proposes Figure 1 A schematic diagram of the step of obtaining an optimization function based on the comprehensive total variance in step S3;

[0072] Figure 3 The present invention proposes Figure 2 Schematic diagram of the steps for obtaining the privacy budget of a large region in ε-local differential privacy in step S32;

[0073] Figure 4 The present invention proposes Figure 1 Detailed step diagram of step S4;

[0074] Figure 5 The present invention proposes Figure 4 Detailed step diagram of step S41;

[0075] Figure 6 A further step of the vehicle location privacy protection method based on local differential privacy proposed by the present invention;

[0076] Figure 7 A schematic diagram of the working principle of the vehicle location privacy protection method based on local differential privacy proposed in the present invention for vehicle-road-cloud multi-network integration;

[0077] Figure 8 A detailed application flow diagram of the vehicle location privacy protection method based on local differential privacy proposed in the present invention;

[0078] Fig. 9 This is a schematic diagram of the structure of the vehicle location privacy protection system based on local differential privacy proposed in the present invention. DETAILED DESCRIPTION

[0079] The present invention proposes a vehicle location privacy protection method based on local differential privacy, which performs regionalized coding processing on the target area of ​​vehicle operation, optimizes the differential privacy mechanism through a simulated annealing algorithm, solves the problem of vehicle location data privacy, and is particularly suitable for effective privacy protection of location data in a vehicle-road-cloud multi-network scenario.

[0080] Embodiment 1:

[0081] A vehicle location privacy protection method based on local differential privacy, such as Figure 1 As shown, the process includes the following steps S1 to S4.

[0082] S1: Divide the target area for vehicle operation into at least one large area, and divide each large area into a plurality of small areas.

[0083] The target area for vehicle operation is divided into multiple large areas according to traffic conditions and natural conditions. Based on the granularity required by statistics, the large area is further divided into multiple small areas.

[0084] The M large areas are assigned unique numbers, such as 1, 2, ..., M, and the H small areas in each large area are also numbered, such as 1, 2, ..., H.

[0085] The number corresponding to each small area and the GPS location of the corresponding area are recorded in the vehicle's on-board computer, cloud management center and roadside infrastructure equipment respectively.

[0086] S2: Based on the perturbation flip probability of ε-local differential privacy, the estimated total variance of the number of vehicles in each small area in the large area is obtained, and the estimated total variance of the number of vehicles in all small areas in the large area is added up to obtain the comprehensive total variance.

[0087] The perturbation probability formula of ε-Local Differential Privacy (LDP) is:

[0088]

[0089] Where p is the flip probability of the first perturbation, q is the flip probability of the second perturbation, A[j] is the value of the jth element in vector A, B[j] is the value of the jth element after perturbation, Pr[·] is the probability function, 0 <q<p<1。

[0090] Therefore, the total variance of the estimated number of vehicles in small area i is for:

[0091]

[0092] Among them, G is the total number of vehicles in the large area, f i (·) is the ratio of the number of vehicles in small area i to the total number of vehicles in the large area G, then the comprehensive total variance of all small areas in the large area is for:

[0093]

[0094] S3: Obtaining an optimization function based on the comprehensive total variance, and solving the optimization function to obtain an optimal disturbance flipping probability.

[0095] In a further preferred embodiment, Figure 2 As shown, the optimization function is obtained based on the comprehensive total variance, including S31-S33:

[0096] S31: Integrate the comprehensive total variance and discard constant coefficients to obtain a primary optimization function;

[0097] Since G is the total number of vehicles in a large area, the total number G is fixed under the condition that the area division is determined, so the primary optimization function f(q) is:

[0098]

[0099] S32: The relationship between the perturbation flipping probability and the privacy budget is obtained based on the unary encoding mechanism of ε-local differential privacy.

[0100] Since the unary encoding mechanism satisfies the conditions for ε-local differential privacy:

[0101]

[0102] That is, the relationship between ε, p and q is:

[0103]

[0104] Among them, ε is the privacy budget.

[0105] S33: Using the association relationship to simplify the primary optimization function to obtain an optimization function.

[0106] The above Substituting the primary optimization function f(q) into the optimization function f(q)* is:

[0107]

[0108] In a further preferred embodiment, Figure 3 As shown, the privacy budget of the large area in the ε-local differential privacy in step 32 is obtained through the following steps S321 to S323:

[0109] S321: Obtain the correlation level score between each small area in the large area and personal privacy.

[0110] If there are a large number of sensitive points (for example, schools, hospitals, etc.) in a large area, since hospitals reflect a person's physical condition and schools reflect a person's family situation, which are highly correlated with personal privacy, a higher correlation level score will be set for the small area containing these locations.

[0111] S322: Add up the correlation level scores corresponding to the small areas to obtain the comprehensive correlation level score of the corresponding large area.

[0112] The correlation level scores corresponding to each small area are summed up, because the more sensitive points related to personal privacy in the large area, the larger the corresponding comprehensive correlation level score.

[0113] S323: Determine a privacy budget for the large area based on the comprehensive level score of the degree of association.

[0114] Regions with larger comprehensive correlation scores require more stringent privacy protection, and therefore set a larger privacy budget.

[0115] In a further preferred solution, in step S3, solving the optimization function to obtain the optimal disturbance flip probability includes:

[0116] The optimized function is solved by using a simulated annealing algorithm to obtain an optimal disturbance flip probability, wherein the state variable of the simulated annealing algorithm is the disturbance flip probability, and the optimization goal of the simulated annealing algorithm is to minimize the optimized function.

[0117] The simulated annealing algorithm consists of two parts, namely the Metropolis algorithm and the annealing process, which correspond to the inner loop and the outer loop respectively.

[0118] The external cycle is the annealing process, which brings the solid to a higher temperature (initial temperature T0), and then reduces the temperature in a certain proportion according to the cooling coefficient α. When the end temperature T0 is reached, f When the cooling is finished, the annealing process is finished. In this embodiment, the initial temperature T0 is 100, the temperature drop coefficient α is 0.99, and the end temperature T f is 0.01.

[0119] The Metropolis algorithm is an inner loop, that is, at each temperature, iterates L times (for example, 100 times) to find the minimum value of the energy (optimization function) at that temperature (that is, the corresponding optimal solution). Corresponding to this embodiment, the perturbation flip probability is used as the state variable, and the optimization function is minimized as the optimization goal. The acceptance probability P of the state q(n) of the nth iteration to the state q(n+1) of the n+1th iteration in the corresponding simulated annealing algorithm is:

[0120]

[0121] Wherein, F(n) is the value of the optimization function obtained at the nth iteration, F(n+1) is the value of the optimization function obtained at the n+1th iteration, and T is the temperature.

[0122] Based on the above simulated annealing algorithm, the optimal perturbation flipping probabilities p* and q* are obtained.

[0123] S4: Using the optimal disturbance flipping probability as the disturbance flipping probability of ε-local differential privacy to perturb the vehicle's location information, thereby achieving privacy protection of the vehicle's location.

[0124] In a further preferred solution, the vehicle position information is disturbed based on the optimal disturbance flip probability, such as Figure 4 As shown, the following steps S41 and S42 are included:

[0125] S41: Perform unary encoding on the vehicle's position information to obtain a one-dimensional vector.

[0126] In a further preferred embodiment, Figure 5 As shown, step S41 includes the following steps S411 to S414, wherein:

[0127] S411: Sequentially number the large area and the small areas in the large area. The large area and the small areas are numbered in the manner of step S1.

[0128] S412: Obtain the GPS location information of the vehicle; use the GPS positioning system to track the location of the vehicle in real time.

[0129] S413: Determine the large area where the GPS location is located and the corresponding small area.

[0130] The vehicle's on-board computing unit can determine which large area the vehicle's GPS location belongs to and mark the vehicle. This data is public, so the vehicle's large area is exposed, and then determine which small area of ​​the large area m the vehicle's location belongs to.

[0131] S414: In the large area, a value of 1 is assigned to the small area where the vehicle is located, and values ​​of 0 are assigned to the other small areas, thereby obtaining a one-dimensional vector.

[0132] If the vehicle is in small area i, it is encoded as a one-dimensional vector A (0, 0, 0, ..., 1, 0), where the i-th bit in vector A is 1 and the others are 0.

[0133] S42: Using the optimal disturbance flipping probability as the disturbance flipping probability of ε-local differential privacy to flip the elements in the one-dimensional vector to achieve disturbance of the vehicle position information.

[0134] The optimal perturbation flip probability includes a first optimal perturbation flip probability and a second optimal perturbation flip probability. After the optimal perturbation flip probability is used as the perturbation flip probability of ε-local differential privacy, the perturbation flip probability of ε-local differential privacy is:

[0135]

[0136] Among them, p* is the first optimal perturbation flip probability, and q* is the second optimal perturbation flip probability.

[0137] In a further preferred embodiment, Figure 6 As shown, step S4 also includes S5 and S6:

[0138] S5: The position vector obtained after the disturbance is transmitted to the cloud management center or the roadside infrastructure equipment to perform aggregation analysis on the position vector to obtain the estimated number of vehicles corresponding to each small area.

[0139] By counting the total number of vectors in each large area, we can get the total number of vehicles in the large area Q, and the number of all vectors in each large area whose i-th position is 1 n′ i , which is the estimated number of vehicles corresponding to small area i.

[0140] S6: Using maximum likelihood estimation to correct the estimated number of vehicles to obtain an unbiased estimate of the number of vehicles corresponding to each small area, and based on the unbiased estimate of the number of vehicles corresponding to each small area, obtain the vehicle distribution of the entire large area.

[0141] Since the estimated number of vehicles is perturbed data, the maximum likelihood estimation is used to correct the estimated number of vehicles to obtain an unbiased estimate of the number of vehicles.

[0142]

[0143] At this point, we can get an accurate estimate of the number of vehicles in the large area G and the unbiased estimate of the number of vehicles in each small area.

[0144] like Figure 7 As shown, this is a schematic diagram of the working principle of the method of the present application for the multi-network integration of vehicle, road and cloud. First, the GPS position information (x, y) of the vehicle is obtained based on the GPS system, and the GPS position information (x, y) is converted into a one-dimensional vector in the vehicle-side computing unit. The one-dimensional vector is differentially privacy processed to obtain a disturbed vector, and the disturbed vector is transmitted to the cloud management center and the roadside infrastructure. Aggregate analysis is performed in the cloud management center and the roadside infrastructure to obtain estimated data of the vehicle in each small area, and then the vehicle distribution in the entire large area is obtained. The cloud management center and the roadside infrastructure share information, and the vehicle position data is shared in the vehicle-road-cloud system.

[0145] The specific applications are as follows Figure 8As shown in the figure, firstly, the area within the target area is divided into large areas and small areas based on traffic conditions and natural conditions and numbered; the privacy budget of each large area is determined according to the number of sensitive locations in the entire large area; the vehicle location information is unary encoded, the vehicle GPS location is converted into a data format that can be processed for differential privacy, a one-dimensional vector A is generated, the total method for estimating the number of vehicles in all small areas is calculated, the optimized function f(q)* is determined, the optimized function is optimized using the simulated annealing algorithm, and the parameter q that minimizes the total error, that is, the optimization function, is obtained, that is, the flipping probability of local differential privacy, and the unary encoding mechanism of local differential privacy is used to perturb each data in vector A to generate vector B. The vector B after differential privacy processing is transmitted to the cloud management center and roadside infrastructure. The cloud brain and roadside center obtain more accurate small area data through aggregation analysis. The cloud center and roadside infrastructure share information, and the vehicle location data is shared in the vehicle-road cloud system.

[0146] Embodiment 2:

[0147] The present invention based on the same inventive concept also provides a vehicle location privacy protection system based on local differential privacy, such as Fig. 9 As shown, including:

[0148] An area division module is used to divide the target area of ​​vehicle operation into at least one large area, and divide each large area into multiple small areas;

[0149] The variance acquisition module is used to obtain the total variance of the estimated number of vehicles in each small area in the large area based on the perturbation flip probability of ε-local differential privacy, and to sum the total variance of the estimated number of vehicles in all small areas in the large area to obtain the comprehensive total variance;

[0150] An optimal disturbance flip probability acquisition module is used to obtain an optimization function based on the comprehensive total variance, and solve the optimization function to obtain an optimal disturbance flip probability;

[0151] The perturbation module is used to perturb the vehicle's location information by using the optimal perturbation flip probability as the perturbation flip probability of ε-local differential privacy, so as to achieve privacy protection of the vehicle's location.

[0152] In a further preferred solution, the optimal disturbance flip probability acquisition module obtains an optimization function based on the comprehensive total variance, including:

[0153] Integrating the comprehensive total variance and discarding constant coefficients to obtain a primary optimization function;

[0154] Based on the unary encoding mechanism of ε-local differential privacy, the correlation between the perturbation flip probability and the privacy budget of large regions in ε-local differential privacy is obtained;

[0155] The primary optimization function is simplified by using the association relationship to obtain an optimization function.

[0156] In a further preferred solution, the privacy budget of the large region in the ε-local differential privacy in the variance acquisition module is obtained by the following steps:

[0157] Obtain the correlation level score between each small area in the large area and personal privacy;

[0158] The correlation level scores corresponding to each small area are summed up to obtain the comprehensive correlation level score of the corresponding large area;

[0159] A privacy budget for a large area is determined based on the comprehensive level score of the association degree.

[0160] In a further preferred solution, the optimal disturbance flip probability acquisition module solves the optimization function to obtain the optimal disturbance flip probability, including:

[0161] The optimized function is solved by using a simulated annealing algorithm to obtain an optimal disturbance flip probability, wherein the state variable of the simulated annealing algorithm is the disturbance flip probability, and the optimization goal of the simulated annealing algorithm is to minimize the optimized function.

[0162] In a further preferred solution, the acceptance probability P of the state q(n) of the nth iteration changing to the state p(n+1) of the n+1th iteration in the simulated annealing algorithm in the optimal disturbance flip probability acquisition module is:

[0163]

[0164] Wherein, F(n) is the value of the optimization function obtained at the nth iteration, F(n+1) is the value of the optimization function obtained at the n+1th iteration, and T is the temperature.

[0165] In a further preferred solution, the perturbation module uses the optimal perturbation flip probability as the perturbation flip probability of ε-local differential privacy to perturb the location information of the vehicle, including:

[0166] Perform unary encoding on the vehicle's location information to obtain a one-dimensional vector;

[0167] The optimal perturbation flipping probability is used as the perturbation flipping probability of ε-local differential privacy to flip the elements in the one-dimensional vector to achieve perturbation of the vehicle position information.

[0168] In a further preferred solution, the optimal perturbation flipping probability in the perturbation module includes a first optimal perturbation flipping probability and a second optimal perturbation flipping probability. After using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy, the perturbation flipping probability of ε-local differential privacy is as follows:

[0169]

[0170] Where p* is the first optimal perturbation flipping probability, q* is the second optimal perturbation flipping probability, A[j] is the value of the j-th element in vector A, B[j] is the value of the j-th element after perturbation, Pr[·] is the probability function, 0 < q* < p* < 1, and ε is the privacy budget.

[0171] In a further preferred solution, the perturbation module performs one-hot encoding on the position information of the vehicle to obtain a one-dimensional vector, including:

[0172] Number the large area and the small areas within the large area in sequence;

[0173] Obtain the GPS position information of the vehicle;

[0174] Determine the large area and the corresponding small area where the GPS position is located;

[0175] Assign a value of 1 to the small area where the vehicle is located in the large area, and assign a value of 0 to the remaining small areas, then a one-dimensional vector is obtained.

[0176] In a further preferred solution, after the perturbation module perturbs the position information of the vehicle based on the optimal perturbation flipping probability, it further includes:

[0177] Transmit the perturbed position vector to the cloud management center or roadside infrastructure to perform aggregation analysis on the position vector to obtain the estimated number of vehicles corresponding to each small area;

[0178] Use maximum likelihood estimation to correct the estimated number of vehicles to obtain an unbiased estimate of the number of vehicles corresponding to each small area, and obtain the vehicle distribution of the entire large area based on the unbiased estimate of the number of vehicles corresponding to each small area.

[0179] The input of this system is the real-time GPS position of each vehicle in the urban area, the large area division of the city, the small area division within each large area, and the privacy protection degree of each large area. The output result is the spatial distribution of the number of vehicles in the urban area at various granularities that meet local privacy requirements, that is, the distribution of vehicles within the divided cell range can be obtained. In the context of multi-network integration of vehicle, road and cloud, the transmitted data is the perturbed data that meets local differential privacy, which can resist various privacy attacks, and realize the privacy protection of the location data of the vehicle on the vehicle side, and use the vehicle's location data information in the cloud and roadside for real-time traffic status perception and supervision.

[0180] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A vehicle location privacy protection method based on local differential privacy, characterized in that: Including: Dividing the target area of vehicle operation to obtain at least one large area, and dividing each large area to obtain multiple small areas; Based on the perturbation flipping probability of ε-local differential privacy, obtaining the total variance estimate of the number of vehicles in each small area in the large area, and summing the total variance estimates of the number of vehicles in all small areas in the large area to obtain the comprehensive total variance; Obtaining an optimization function based on the comprehensive total variance, and solving the optimization function to obtain the optimal perturbation flipping probability; Using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to perturb the position information of the vehicle, so as to achieve privacy protection of the vehicle position.

2. The vehicle location privacy protection method based on local differential privacy according to claim 1 is characterized in that: The obtaining of the optimization function based on the comprehensive total variance includes: Integrating the comprehensive total variance and discarding the constant coefficient to obtain a primary optimization function; Based on the unary coding mechanism of ε-local differential privacy, obtaining the correlation relationship between the perturbation flipping probability and the privacy budget of the large area in ε-local differential privacy; Simplifying the primary optimization function by using the correlation relationship to obtain the optimization function.

3. The vehicle location privacy protection method based on local differential privacy according to claim 1 or 2, characterized in that: The privacy budget of the large area in the ε-local differential privacy is obtained through the following steps: Obtaining the correlation degree level score between each small area in the large area and personal privacy; Summing the correlation degree level scores corresponding to each small area to obtain the comprehensive correlation degree level score of the corresponding large area; Determining the privacy budget of the large area based on the comprehensive correlation degree level score.

4. The vehicle location privacy protection method based on local differential privacy according to claim 1, characterized in that: The solving of the optimization function to obtain the optimal perturbation flipping probability includes: Using the simulated annealing algorithm to solve the optimization function to obtain the optimal perturbation flipping probability, where the state variable of the simulated annealing algorithm is the perturbation flipping probability, and the optimization goal of the simulated annealing algorithm is to minimize the optimization function.

5. The vehicle location privacy protection method based on local differential privacy according to claim 4 is characterized in that: The acceptance probability P of the state q(n) in the nth iteration of the simulated annealing algorithm changing to the state q(n + 1) in the (n + 1)th iteration is: Where F(n) is the value of the optimization function obtained in the nth iteration, F(n + 1) is the value of the optimization function obtained in the (n + 1)th iteration, and T is the temperature.

6. The vehicle location privacy protection method based on local differential privacy according to claim 1, characterized in that: The using of the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to perturb the position information of the vehicle includes: Performing unary coding on the position information of the vehicle to obtain a one-dimensional vector; Using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy to flip the elements in the one-dimensional vector to achieve perturbation of the vehicle position information.

7. The vehicle location privacy protection method based on local differential privacy according to claim 6 is characterized in that: The optimal perturbation flipping probability includes a first optimal perturbation flipping probability and a second optimal perturbation flipping probability. After using the optimal perturbation flipping probability as the perturbation flipping probability of ε-local differential privacy, the perturbation flipping probability of the ε-local differential privacy is: Where p* is the first optimal perturbation flipping probability, q* is the second optimal perturbation flipping probability, A[j] is the value of the jth element in vector A, B[j] is the value of the jth element after perturbation, Pr[·] is the probability function, 0 < q* < p* < 1, and ε is the privacy budget.

8. The vehicle location privacy protection method based on local differential privacy according to claim 6, characterized in that: The performing of unary coding on the position information of the vehicle to obtain a one-dimensional vector includes: Numbering the large area and the small areas in the large area in sequence; Get the vehicle's GPS location information; Determine the large area where the GPS location is located and the corresponding small area; In the large area, the small area where the vehicle is located is assigned a value of 1, and the other small areas are assigned a value of 0, and a one-dimensional vector is obtained.

9. The vehicle location privacy protection method based on local differential privacy according to claim 1, characterized in that: After the vehicle position information is disturbed based on the optimal disturbance flip probability, the method further includes: The position vector obtained after the disturbance is transmitted to the cloud management center or the roadside infrastructure equipment to perform aggregation analysis on the position vector to obtain the estimated number of vehicles corresponding to each small area; The estimated number of vehicles is corrected by using maximum likelihood estimation to obtain an unbiased estimate of the number of vehicles corresponding to each small area, and the vehicle distribution of the entire large area is obtained based on the unbiased estimate of the number of vehicles corresponding to each small area.

10. A vehicle location privacy protection system based on local differential privacy, characterized in that: include: An area division module is used to divide the target area of ​​vehicle operation into at least one large area, and divide each large area into multiple small areas; The variance acquisition module is used to obtain the total variance of the estimated number of vehicles in each small area in the large area based on the perturbation flip probability of ε-local differential privacy, and to sum the total variance of the estimated number of vehicles in all small areas in the large area to obtain the comprehensive total variance; An optimal disturbance flip probability acquisition module is used to obtain an optimization function based on the comprehensive total variance, and solve the optimization function to obtain an optimal disturbance flip probability; The perturbation module is used to perturb the vehicle's location information by using the optimal perturbation flip probability as the perturbation flip probability of ε-local differential privacy, so as to achieve privacy protection of the vehicle's location.

Images