WO2020177484A1 - Localized difference privacy urban sanitation data report and privacy calculation method - Google Patents

Abstract

The present invention discloses a localized difference privacy urban sanitation data report and a privacy calculation method, the specific steps comprise: setting a location privacy parameter and a garbage amount level privacy parameter, and calculating a privacy processing parameter; obtaining a user's original real data by positioning a sanitation site and determining its garbage amount level; performing localized difference privacy processing on the location and garbage amount level data respectively, and uploading the data to a server; the server receives the data uploaded by the user, performs statistics and denoising processing on the data; the server performs aggregation calculation on the data subjected to the denoising processing, obtains real garbage amount levels of all locations to form a city sanitation distribution map. The present invention can, under the premise of protecting the location and data privacy of all data collectors, achieve the goal of urban sanitation dynamic distribution on the basis of mobile group perception.

Description

A localized differential privacy urban sanitation data report and privacy calculation method Technical field

The present invention relates to the field of information security and privacy protection, in particular to a localized differential privacy urban sanitation data report and privacy calculation method.

Background technique

With the continuous improvement of sensor accuracy and the popularization of mobile smart devices, mobile devices represented by smart phones can perceive people and surrounding environmental information anytime and anywhere, so that people can get the status of their environment in time, and better Provide required services for people and society, and promote the continuous development of the field of mobile group perception. Mobile group perception provides a new way for people to perceive the world, enabling anyone to perform perception tasks through mobile terminals to participate in the perception process, and provide people with various services.

In the application of mobile groups, whether the privacy of perception participating users can be guaranteed is a key factor that relates to whether users are willing to participate in perception tasks. Traditional encryption methods have disadvantages such as high computational overhead and various privacy protection methods based on anonymization cannot resist background knowledge attacks. The centralized differential privacy method needs to be based on the honesty and credibility of third-party data collectors, but will not steal or disclose users The assumption of sensitive information is not realistic in practical applications. The localized differential privacy technology inherits the centralized differential privacy technology to quantitatively define privacy attacks, and allows each user to complete the privacy protection process independently, which can achieve more thorough privacy protection.

Random response technology is the mainstream perturbation mechanism of localized differential privacy protection technology. Random response requires a huge amount of sample data. In most mobile group sensing scenarios, the amount of data for participating users is very limited. Therefore, how to apply random response technology in The scene of small and medium sample mobile group perception needs further research.

Summary of the invention

The purpose of the present invention is to overcome the shortcomings of the prior art and provide a localized differential privacy urban sanitation data report and privacy calculation method based on random response. The invention applies random response technology to report collection in the field of mobile group perception, aggregates and calculates and analyzes privacy data, obtains the joint distribution of privacy attributes, and realizes the monitoring of urban sanitation distribution.

The purpose of the present invention can be achieved through the following technical solutions:

A localized differential privacy urban sanitation data report and privacy calculation method, the specific steps include:

Set location privacy parameters and garbage volume level privacy parameters, and calculate privacy processing parameters;

The user obtains the user's original real data by locating the location of the sanitation point and judging the garbage volume level of the sanitation point;

Perform localized differential privacy processing on location and waste level data and upload them to the server;

The server receives data uploaded by users and performs statistics and denoising processing on the data;

The server aggregates and calculates the data after the denoising process to obtain the real garbage volume level of each location, which constitutes the urban sanitation distribution map.

Specifically, in the steps of setting location privacy parameters and garbage volume level privacy parameters, and calculating privacy processing parameters, set location privacy protection parameters ε ₁ and garbage volume level privacy parameters ε ₂ , and announce the parameters to each user.

Calculate the length l of the Bloom filter and the number of hash functions h through the false positive rate k of the Bloom filter and the set of perception values M. The calculation formula is:

According to the privacy parameters ∈ ₁ , ∈ ₂ , the total number of sanitation points F and the number of hash functions h, the privacy processing parameters are calculated:

r represents the number of position reports, and its calculation formula is:

p represents the probability of selecting the real location, and its calculation formula is:

f represents the bit flip probability, and its calculation formula is:

Specifically, the user obtains the original real data of the user by locating the location of the sanitation point and determining the garbage volume level of the sanitation point, the user locates the location of the sanitation point through the mobile smart device carried, and obtains the sanitation point through the location mapping table. The number i of the point. The location mapping table is obtained by the server for the location number of each sanitation point in advance. Through self-determination, the user obtains the level of the garbage volume of the sanitation point. The number and the level of garbage are the user's original real data <i,level>.

Specifically, in the step of performing localized differential privacy processing on location and garbage level data respectively, the user performs localized differential privacy processing locally, and the processing steps are:

Initialize location collection

And upload the result set

Generate a random number rand between (0,1), and combine the privacy processing parameters to perform localized differential privacy processing on the location data to meet the location privacy protection parameter ∈ ₁ ;

According to the location set L obtained after the localized differential privacy processing, localized differential privacy processing is performed on the garbage volume level data, and the user's upload result set R is calculated.

Furthermore, in the step of generating a random number rand between (0,1) and combining privacy processing parameters to perform localized differential privacy processing on location data, the specific processing method is:

Judge the relationship between the random number rand and the selection probability p of the real position:

If the random number rand is less than p, then the position set is the union of the real position i and any selected r-1 positions in the position set except i, expressed as:

L={i}∪getRandomElements(F\{i},r-1)

Among them, the function getRandomElements(A,b) means to return the result of randomly selecting b elements from the set A;

If the random number rand is greater than p, the position set is the union of r positions arbitrarily selected from the position set except i, expressed as:

L=getRandomElements(F,r)

In summary, the representation of the relationship between the random number rand and the selection probability p of the real location is as follows:

This processing method can ensure that the user's location data meets ε ₁ -differential privacy.

Furthermore, according to the location set L obtained after the localized differential privacy processing, localized differential privacy processing is performed on the garbage volume level data, and in the step of calculating the user’s upload result set R, if the user’s real location i is included in In L, the corresponding item of the real garbage amount is the real garbage amount level level _i . At this time, the position of the element to the other set of positions t L in which the amount of waste is at level corresponding to the amount of waste entry level value set randomly selected middle-level probability level _'t, then the amount of waste for each processing level, specific processing method for the :

Use h hash functions to map the garbage volume level level _i or level' _t to a Bloom filter of length l to obtain a vector S _{t of} length l;

For each of a S _t S _t [j], a random number generation function operation to obtain a random number between q (0,1), if q <1-f / 2, j-th bit remains unchanged, If q>1-f/2, flip the j-th bit, that is, when the original value is 0, flip to 1; when the original value is 1, flip to 0; get S _t ′;

In summary, the way to process the garbage volume level data is:

This processing method ensures that the user's garbage volume level perception data meets ε ₂ -differential privacy.

Combine each position i and the corresponding processed garbage level value S′ _i into a pair (i, S′ _i ), and add them to the upload result set R to obtain the final upload result set R.

Specifically, in the step of receiving the data uploaded by the user by the server and performing statistics and denoising processing on the data, the statistics and denoising processing method are:

Mix the upload data Z of all users, extract all the location data in the upload report, and calculate the number of real reports E _i for each location. The calculation formula is:

q=(r-p)/(F-1)

E _i ＝w _i

L _i = w _i /(pE _i +q(rN-E _i ))

V=∑ _i L _i

E _i ＝E _i (pL _i +q(VL _i )

E _i ＝E _i /r

Among them, Z represents the user upload data, F represents the total number of sanitation points, p represents the actual location selection probability, r represents the number of location reports, N represents the number of users, w _i represents the number of locations i included in Z, i=1,..., F;

For each upload location i, the server filters out all the report pair data sets Di at the location _i . Assuming a total of N _i D _i of data items, the number of transactions reported as E _i, the noise is reported as the number N _i -E _i, of the amount of waste every level data set statistics as follows S _'i: Calculated per a "1" the total number of _{X i [j], X i} [j] represents a j-th bit of the i-th position, 0≤j <l, consisting of a vector X _i, X _i represents the position number i is received Each digit that arrives is a count of "1";

Sources of X _i consists of four parts, as shown in Table 1.

Table 1

In the real data set, assuming that the original amount of waste after the Bloom filter rating, a total Z _i "1", which is the probability (1-f / 2) is maintained at "1", E _i -Z _i th "0", it has the probability of f/2 to flip to "1"; in the same way, assume that the original garbage level of the noise data has U _i "1"s after the Bloom filter, which has (1-f/2) the probability remains _{"1"; N i -E i} -U i "0", a total of f / 2 flip probability is "1."

Therefore the following equation holds:

Since the added noise is uniformly distributed, U _i in the above formula can be obtained by calculation. Therefore, the above formula can be derived as:

From this, a vector Z _i composed of the number of "1"s in each bit of the original real garbage volume level is obtained.

Since there is a certain error in the frequency estimation in step 1, the data generation and processing process also has a certain randomness, and the value calculated by Z _i also has a certain error. Therefore, a certain analysis method is required to obtain the real garbage volume level at position i.

Specifically, the server aggregates and calculates the data after denoising processing to obtain the real garbage volume level of each location, and in the step of forming the city sanitation distribution map, all w candidate values of the garbage volume level are passed through the bloom filter After the h hash functions of, form a matrix of w·l as the characteristic matrix Q, all Z _i [j] (0≤j＜l) obtained by denoising processing form a vector Z _i of length l, and solve Linear regression problem Z _i =A·Q, the garbage volume level at the position corresponding to the maximum value in the vector A(1·w) is the real garbage volume level at position i.

Through the above steps, the real garbage volume level of each location can be calculated, thereby obtaining the urban garbage volume distribution map.

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

1. Compared with the smart trash can based on the Internet of Things technology, the present invention greatly saves the deployment cost of hardware facilities, the communication overhead of network connection and the human resources of equipment operation and maintenance, and fully saves social resources;

2. The present invention does not rely on a trusted third-party server, and the data is completely processed by the user on the mobile terminal, which more thoroughly guarantees the user's data privacy and security;

3. While ensuring the accuracy of aggregation, the present invention greatly reduces the minimum number of user participants, which is a realistic and feasible solution;

4. The present invention increases the speed while maintaining the security of the scheme.

Description of the drawings

Figure 1 is a flowchart of a localized differential privacy urban sanitation data report and privacy calculation method.

Figure 2 is a structural diagram of a localized differential privacy urban sanitation data report and privacy calculation method.

Fig. 3 is a sample diagram of random disturbance of the garbage volume level in the embodiment of the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the examples and drawings, but the implementation of the present invention is not limited thereto.

Example

Figures 1 and 2 are respectively a flowchart and structure diagram of a localized differential privacy urban sanitation data report and privacy calculation method. The specific steps of the method include:

Set location privacy parameters and garbage volume level privacy parameters, and calculate privacy processing parameters;

The user obtains the user's original real data by locating the location of the sanitation point and judging the garbage volume level of the sanitation point;

Perform localized differential privacy processing on location and waste level data and upload them to the server;

The server receives data uploaded by users and performs statistics and denoising processing on the data;

The server aggregates and calculates the data after the denoising process to obtain the real garbage volume level of each location, which constitutes the urban sanitation distribution map.

In this embodiment, it is assumed that the total number of sanitation points in a certain area of the city is 50, the level of garbage volume has 3 levels, namely High, Middle, and Low. The privacy parameter of the set location is ∈ ₁ = 1.6, and the level of garbage volume The privacy parameter ∈ ₂ = 1.9, the false positive rate of the bloom filter is 0.1, then the optimal bloom filter length is calculated

The number of hash functions is

Privacy processing parameters

Suppose that when a perceiving user passes a sanitation point, he needs to report the amount of garbage at the sanitation point: first locate the current sanitation point location through the GPS sensor of the mobile device, and then match the sanitation point through the location-number table set in advance If the number of the sanitation point is 12, the user sets the corresponding garbage volume level by observing the amount of garbage. If it is determined that the actual garbage volume level of the sanitation point is Middle, the data <12, Middle> is the perception Real perception data of participating users.

According to the above calculated parameters, the user runs a random number generation function locally to generate a random number rand=0.67. Since 0.67>0.486, the user's location set is from the set {1,2,...11,13...,50} Eight locations are randomly selected as upload locations. Since the real perception location is not in the location set, the garbage volume levels corresponding to all eight locations are randomly selected from the garbage volume level set. Assuming that the data set obtained after 8 randomization selections is: {＜1,Middle＞,＜5,High＞,＜9,Low＞,＜14,Middle＞,＜21,Low＞,＜23,High＞ ,<36,Low>,<41,Middle>}.

Assuming that the garbage volume level passes through the Bloom filter, the corresponding vectors are respectively {High:1000001001000, Middle:0100000000011, Low:1000001000001}. Each of the corresponding garbage volume levels in the set passes the generated random number and parameters after the Bloom filter

Comparing, less than 0.58 remains unchanged, greater than 0.58 to flip, the final uploaded collection is:

{<1,0101010011011>,<5,1011100100110>,<9,0111000101011>,<14,1101000101011>,<21,1010111101010>,<23,0110101101010>,<36,1101010100111>,<41,0011011011010>}

A sample diagram of random disturbance of the garbage volume level is shown in Figure 3.

When the server receives the upload value of all the perceived users, it will perform probability statistics on all the location items, and can roughly estimate the number of true reports for each location. For example, the server receives data from 2000 users in total, and the location data is shared 2000×8=16000, it is estimated that the number of users whose real position is "1" is 50, and all the reports actually received contain 300 users whose position is "1", so it is inferred that all users are "1" About 250 data items in the data items are noise location data. The garbage volume level of the noise location data items is randomly and uniformly selected. From this, it can be estimated that there are about 250/3 noise data for each garbage volume level. It is Noise=(noise)·3.

Combine the values of the original Bloom filter corresponding to the garbage volume level into a matrix Q, expressed as:

The garbage volume level data of each location is filtered out and analyzed separately. For example, all garbage volume level items with location "1" are expressed in the form:

Bit 0	No. 1	…	L position
1	0	1	0
…	…	…	…
0	1	1	0

Sum in the column as a unit, and get a vector of length l (sum ₁ ,...,sum _l ), denoted as X, and calculate the original noise level of "1" according to the noise estimate calculated in the scheme The number vector U, U=Noise·A, according to the actual estimated value, noise value and calculation formula of statistical calculation, the number of each bit originally "1" is calculated, which is the vector Z of length l.

Solving the linear regression equation problem Z=A·Q, the garbage volume level corresponding to the maximum value in the vector A is the estimated real garbage volume level at the location.

The garbage volume levels of all locations can be obtained sequentially through the above process, so as to obtain the garbage volume level distribution map in this area of the city.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above-mentioned embodiments, and any other changes, modifications, substitutions, combinations, etc. made without departing from the spirit and principle of the present invention Simplified, all should be equivalent replacement methods, and they are all included in the protection scope of the present invention.

Claims (8)

1. A localized differential privacy urban sanitation data report and privacy calculation method, which is characterized in that the specific steps include:

   Set location privacy parameters and garbage volume level privacy parameters, and calculate privacy processing parameters;

   The user obtains the user's original real data by locating the location of the sanitation point and judging the garbage volume level of the sanitation point;

   Perform localized differential privacy processing on location and waste level data and upload them to the server;

   The server receives data uploaded by users and performs statistics and denoising processing on the data;

   The server aggregates and calculates the data after the denoising process to obtain the real garbage volume level of each location, which constitutes the urban sanitation distribution map.
2. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 1, wherein in the step of setting location privacy parameters and garbage level privacy parameters, setting location privacy protection parameters ∈ ₁ and The garbage volume level privacy parameter ∈ ₂ , and the parameters are announced to each user;

   Calculate the length l of the Bloom filter and the number of hash functions h through the false positive rate k of the Bloom filter and the set of perception values M. The calculation formula is:

   

   

   According to the privacy parameters ∈ ₁ , ∈ ₂ , the total number of sanitation points F and the number of hash functions h, the privacy processing parameters are calculated. The calculation formula is:

   

   

   

   Among them, r represents the number of position reports, p represents the probability of real location selection, and f represents the probability of bit flip.
3. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 1, wherein the user obtains the user’s original real data by locating the location of the sanitation point and determining the garbage volume level of the sanitation point. In the step, the user locates the location of the sanitation point through the mobile smart device carried, and obtains the number i of the sanitation point through the mapping table; the user determines the level of the garbage volume of the sanitation point by himself, and the number and the number of garbage levels are the user’s The original real data <i,level>.
4. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 1, characterized in that, in the step of performing localized differential privacy processing on location and garbage level data respectively, the user performs locally Localized differential privacy processing, the processing steps are:

   Initialize location collection

   

   And upload the result set

   

   Generate a random number rand between (0,1), and combine the privacy processing parameters to perform localized differential privacy processing on the location data to meet the location privacy protection parameter ∈ ₁ ;

   According to the location set L obtained after the localized differential privacy processing, localized differential privacy processing is performed on the garbage volume level data, and the user's upload result set R is calculated.
5. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 4, characterized in that said generating a random number rand between (0,1), combined with privacy processing parameters, to position data In the steps of localized differential privacy processing, the specific processing method is:

   Judge the relationship between the random number rand and the selection probability p of the real position:

   If the random number rand is less than p, the position set is the union of the real position i and the arbitrary selection of r-1 positions in the position set except i, expressed as:

   L={i}∪getRandomElements(F\{i},r-1)

   Among them, the function getRandomElements(A,b) means to return the result of randomly selecting b elements from the set A;

   If the random number rand is greater than p, the position set is the union of r positions arbitrarily selected from the position set except i, expressed as:

   L=getRandomElements(F,r)

   In summary, expressed as:

   
6. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 4, characterized in that, according to the location set L obtained after the localized differential privacy processing, the garbage volume level data is localized differentially Privacy processing, in the step of calculating the user's upload result set R, if the user's real location i is included in L, the corresponding item of the real garbage volume is the real garbage volume level _i ; at this time, for other locations in the location set L element t, which is the amount of waste level corresponding item value set in the amount of waste medium probability level select a grade level _'t, then the amount of waste for each processing level, specific processing method is:

   Use h hash functions to map the garbage volume level level _i or level' _t to a Bloom filter of length l to obtain a vector S _{t of} length l;

   For each of a S _t S _t [j], a random number generation function operation to obtain a random number between q (0,1), if q <1-f / 2, j-th bit remains unchanged, If q>1-f/2, flip the j-th bit, that is, when the original value is 0, flip to 1, and when the original value is 1, flip to 0 to obtain S _t ′;

   In summary, expressed as:

   

   Combine each position i and the corresponding processed garbage level value S′ _i into a pair (i, S′ _i ), and add them to the upload result set R to obtain the final upload result set R.
7. A localized differential privacy urban sanitation data report and privacy calculation method according to claim 1, wherein the server receives user uploaded data and performs statistics and denoising processing on the data. The method is:

   Mix the upload data Z of all users, extract all the location data in the upload report, and estimate the actual number of reports E _{i for} each location;

   For each upload location i, the server filters out all the report pair data sets D _{i of} location _i ;

   Assuming a total of N _i D _i of data items, the number of transactions reported as E _i, the noise is reported as the number N _i -E _i, of the amount of waste every level data set statistics as follows S _'i: Calculated per a "1" the total number of _{X i [j], X i} [j] represents a j-th bit of the i-th position, 0≤j <l, consisting of a vector X _i, X _i represents the position number i is received Each digit that arrives is a count of "1";

   In the real data set, assuming that the original amount of waste after the Bloom filter rating, a total Z _i "1", which is the probability (1-f / 2) is maintained at "1", E _i -Z _i th "0", it has the probability of f/2 to flip to "1"; in the same way, assume that the original garbage level of the noise data has U _i "1"s after the Bloom filter, which has (1-f/2) the probability remains _{"1"; N i -E i} -U i "0", a total of f / 2 is inverted to the probability of "1";

   Therefore the following equation holds:

   

   Since the added noise is uniformly distributed, U _i in the above formula can be obtained by calculation. Therefore, the above formula is derived:

   

   From this, a vector Z _i composed of the number of "1"s in each bit of the original real garbage volume level is obtained.
8. The localized differential privacy urban sanitation data report and privacy calculation method according to claim 1, wherein the server aggregates and calculates the data after denoising processing to obtain the real garbage volume level of each location. In the step of constructing the urban sanitation distribution map, all w candidate values of the garbage volume level are processed by the h hash function of the Bloom filter to form a w·l matrix as the feature matrix Q, which is obtained through denoising processing All Z _i [j], compose a vector Z _i of length l, and solve the linear regression problem Z _i =A·Q. The garbage volume level at the position corresponding to the maximum value in the vector A(1·w) is the position i Level of real garbage volume.

Images