CN103761485B - Privacy protection method - Google Patents

Abstract

The invention provides a privacy protection method. The method includes: pre-defining a privacy location set: S＜i≥{s＜i＞1, s＜i＞2, s＜i＞...}, and setting a user's actual location information: ={L1, L2,...,LT}, an uploaded strategy: ={P1, P2,...PT}, and uploaded crowd sensing data of certain time: ={O1, O2,...,OT}. In the uploaded strategy: ={P1, P2,...PT}, expectation maximum of utility( )=|{t|Ot=/empty}|/T is required while Pr[Lt=s＜i＞j| ]-Pr[Lt=s＜i＞j]≤8 is met. The method has the advantages that information quantity of uploaded data is maximally kept and meanwhile the uploaded group sensing data resists powerful attack.

Description

A kind of method for secret protection

Technical field

The present invention relates to field of computer technology, be specifically related to a kind of method for secret protection.

Background technology

In recent years, along with the sensor being equipped with on smart mobile phone gets more and more, intelligent perception technology develops rapidly.Due to group Intelligence cognition technology can obtain large-scale statistical data and carry out the measurement in all scales, intelligent perception technology by It is applied to the every field in people's life, such as publilc health, public safety, traffic monitoring etc..Typical intelligent perception application By great amount of terminals, cellphone subscriber forms, and they spread over each corner in city, by mobile phone be equipped with various kinds of sensors (as Acceleration transducer, optical sensor, gyroscope, GPS etc.) characteristic of record present position, and real-time uploading comprise The data of positional information to intelligent perception server, server obtain the great amount of terminals user sensing data in different location it After, carry out data process and data analysis, thus the respective services required for user is provided.There is substantial amounts of gunz at present The application of perception, although right later user wants to obtain the service that application is provided, is unwilling to provide some privacy informations, as Positional information.Due to such worry, user how to protect when uploading sensing data privacy positional information become one important Problem.

At present, the method for protection privacy positional information is mainly based upon upset, namely in the information that user uploads, Adding independent noise, this kind of technology is proved to be to be hacked, and an assailant can use filtering technique to carry out structure again Build the distribution of initial data, it is hereby achieved that the privacy positional information of user.

Summary of the invention

For the deficiencies in the prior art, the present invention provides a kind of method for secret protection, it is possible to the data making user upload exist While resisting powerful attack, maximize the quantity of information keeping uploading data.

For achieving the above object, the present invention is achieved by the following technical programs:

A kind of method for secret protection, the method includes:

Pre-defined privacy location setsIf the positional information of user's reality isThe strategy uploaded isThe intelligent perception data that certain is uploaded areThe strategy wherein uploadedNeed all intelligent perception data uploadedMeetingOn the premise of so thatExpectation maximum；

Wherein, L_tExpression optional position,Represent any privacy position,Represent and upload data known In the case of successfully speculate the probability of privacy position,It is to represent to speculate in the case of there is no any Given information The probability of privacy position,Value for presetting, δ represents the degree of protection privacy, and it is the least that δ is arranged, and more can protect Protect privacy；

DescribedRepresent the intelligent perception data that user uploadsThe information available quantity comprised, T is user's sum of position when uploading intelligent perception data；

DescribedFor

Wherein,

$\begin{array}{c}\mathrm{Pr}\[\stackrel{\→}{L}|\stackrel{\→}{O}\]=\frac{1}{Z\left(\stackrel{\→}{O}\right)}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}$

Wherein, λ_k、β_k、γ_kAnd θ_kIt is characterized parameter；g_k(L_t,O_t) it is characterized function,Represent the temporal associativity between positional information,Represent positional information and upload between intelligent perception data Spatial correlation,Represent positional information and the feature association uploaded between intelligent perception data, g_k(L_t,O_t) table Showing positional information and the decision-making relatedness uploaded between intelligent perception data, normalization factor is:

$\begin{array}{c}Z\left(\stackrel{\→}{O}\right)=\underset{L}{\Σ}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}.$

It is preferred that described positional information and the decision-making Relating Characteristic function g that uploads between intelligent perception data_k(L_t,O_t) For:

$g_{k1}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=e)$

$g_{k2}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=D_t)$

Wherein, e represents sky empty, O_t=e represents that uploading data in current location is sky, O_t=D_tRepresent in current location Upload the data that data are intelligent perception,Representing the position that some user passes by, δ (w) is true equal to 1, otherwise when w condition Equal to 0.

It is preferred that the temporal associativity characteristic function between described positional informationFor

$f_k\left(\stackrel{\→}{L}\right)=\mathrm{\δ}(L_1=l_{i_1},L_2=l_{i_2},...,L_t=l_{i_t})$

WhereinRepresent t related position.

It is preferred that described positional information and upload the spatial correlation characteristic function between intelligent perception dataFor

h_k(L_t,L_t-1,O_t,O_t-1)=δ (L_t=S, L_t-1=C, O_t=e, O_t-1=e).

It is preferred that described positional information and the feature association characteristic function uploaded between intelligent perception data For

r_k1(L_t,L_t-1,O_t,O_t-1)=δ (dir (L_t,L_t-1)=ahead)

×δ(dis(O_t.rss,O_t-1.rss)≤R)

r_k1(L_t,L_t-1,O_t,O_t-1)=δ (dir (L_t,L_t-1)=turn)

×δ(dis(O_t.rss,O_t-1.rss) ＞ R)

If user is from L_t-1Go to L_tDo not turn, then dir (L_t,L_t-1)=ahead, otherwise dir (L_t,L_t-1)= turn；dis(O_t.rss,O_t-1.rss) for calculating Euler's distance of the received signal strength RSS reading of two location points.

It is preferred that the characteristic parameter θ of described decision-making linked character function_kFor

${\mathrm{\θ}}_{k1}=\log P_j^i$

${\mathrm{\θ}}_{k1}=log(1-P_j^i)$

WhereinRepresent the probability uploading intelligent perception data in j position.

It is preferred that the feature parameter"λ" of characteristic function described association in time_kFor

The maximized object function of method of employing maximal condition possibility predication is:

$l\left(\mathrm{\λ}\right)=\underset{j}{\Σ}\log \mathrm{Pr}\[{\stackrel{\→}{L}}^{i\left(j\right)}|{\stackrel{\→}{O}}^{i\left(j\right)}\]$

To λ_kDo local derviation:

$\begin{array}{c}\frac{\∂l\left(\mathrm{\λ}\right)}{\∂{\mathrm{\λ}}_k}=\underset{j=1}{\overset{N}{\Σ}}\underset{t=1}{\overset{T}{\Σ}}{f}_k({L_t}^{i\left(j\right)},{L_{t-1}}^{i\left(j\right)},{L_{t-2}}^{i\left(j\right)}\\ -\underset{J=1}{\overset{N}{\Σ}}\underset{T=1}{\overset{T}{\Σ}}\underset{l,l^{\′},l^{\′\′}}{\Σ}{f}_k(l,l^{\′},l^{\′\′})\mathrm{Pr}(l,l^{\′},l^{\′\′}|{\stackrel{\→}{O}}^{i\left(j\right)})\end{array}$

Use L-BFGS to carry out regularization, obtain λ_k。

It is preferred that the characteristic parameter β of described space correlation characteristic function_kFor

β_k=logPr (L_t=sⁱ _j|L_t-t'=l_j)

Wherein L_t-t'Represent the position before the t ' moment.

It is preferred that the characteristic parameter γ of described feature association characteristic function_kFor

GivenFirst training threshold value R:

$R=\mathrm{argmax}\underset{j}{\Σ}\underset{t=1}{\Σ}\begin{array}{c}\{r_{k1}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.rss,L_{t-1}^{i\left(j\right)}.rss)\\ +r_{k2}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.rss,L_{t-1}^{i\left(j\right)}.rss\}\end{array}$

After determining threshold value R, obtain characteristic parameter:

γ_k1=log (R-dis (x_t.rss,x_t-1.rss))/R

γ_k2=log (dis (x_t.rss,x_t-1.rss)-R)/R。

The present invention at least has a following beneficial effect:

The present invention provides a kind of method for secret protection, portrays positional information by condition random field and uploads intelligent perception The temporal associativity of data, spatial correlation, feature association and decision-making relatedness, so that based on this method for secret protection The data uploaded can maximize, while resisting powerful attack, the quantity of information uploading data.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is the present invention Some embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to These accompanying drawings obtain other accompanying drawing.

Fig. 1 is the flow chart of method for secret protection in the embodiment of the present invention 1；

Fig. 2 is the flow chart of method for secret protection in the embodiment of the present invention 2；

Fig. 3 is that the characteristic function of the embodiment of the present invention 2 conditional random field describes；

Fig. 4 is that in the embodiment of the present invention 2, temporal associativity describes；

Fig. 5 is that in the embodiment of the present invention 2, spatial correlation describes；

Fig. 6 is that in the embodiment of the present invention 2, RSS feature association describes.

Detailed description of the invention

For making the purpose of the embodiment of the present invention, technical scheme and advantage clearer, below in conjunction with the embodiment of the present invention In accompanying drawing, the technical scheme in the embodiment of the present invention is carried out clear, complete description, it is clear that described embodiment is The a part of embodiment of the present invention rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art The every other embodiment obtained under not making creative work premise, broadly falls into the scope of protection of the invention.

Embodiment 1

The embodiment of the present invention 1 proposes a kind of method for secret protection, sees Fig. 1, comprises the steps:

Step 101: pre-defined privacy location sets

Step 102: find certain data and upload strategy so that the intelligent perception data uploaded are before protection privacy of user Put, maximize data message amount.

If the positional information of user's reality isThe strategy uploaded isCertain is uploaded Intelligent perception data beThe strategy wherein uploadedNeed to upload number to all According toMeetingOn the premise of so thatExpectation maximum；

In this step, L_tExpression optional position,Represent any privacy position,Represent on known The probability of privacy position is successfully speculated in the case of passing data,It is to represent the situation not having any Given information The probability of lower supposition privacy position,Value for presetting, δ represent protection privacy degree, δ arrange the least, More can protect privacy；

DescribedRepresent the intelligent perception data that user uploadsThe information available quantity comprised, T be user upload data time position sum.

DescribedFor

Wherein,

$\begin{array}{c}\mathrm{Pr}\[\stackrel{\→}{L}|\stackrel{\→}{O}\]=\frac{1}{Z\left(\stackrel{\→}{O}\right)}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}$

Wherein, λ_k、β_k、γ_kAnd θ_kIt is characterized parameter；g_k(L_t,O_t) it is characterized function,Represent the temporal associativity between positional information,Represent positional information and upload between intelligent perception data Spatial correlation,Represent positional information and the feature association uploaded between intelligent perception data, g_k(L_t,O_t) table Showing positional information and the decision-making relatedness uploaded between intelligent perception data, normalization factor is:

$\begin{array}{c}Z\left(\stackrel{\→}{O}\right)=\underset{L}{\Σ}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}.$

Visible, in embodiments of the present invention, portrayed by condition random field and upload the temporal associativity of data, pass, space Connection property, feature association and decision-making relatedness so that the data uploaded based on this method for secret protection can resist strong While big attack, maximize the quantity of information uploading data.

Embodiment 2

Below by a specific example, carry out the realization of a preferred embodiment of the more detailed description present invention Journey.Seeing Fig. 2, this process comprises the steps:

Step 201: pre-defined privacy location sets

Step 202: set up secret protection model.

In this step, if the claimed privacy of intimacy protection system is referred to as δ-privacy.It is defined as follows: for One user, the intelligent perception data uploaded areDefine the set of a privacy position simultaneouslySayProtect δ-privacy, if for any position L_t, for any privacy positionMeet:

$\mathrm{Pr}\[L_t=s_j^i|\stackrel{\→}{O}\]-\mathrm{Pr}\[L_t=s_j^i\]\≤\mathrm{\δ}$

From this formula, it can be seen that δ is used to portray a system and resists the ability of attack.δ in system of defense Value arrange the least, assailant is just more difficult to deduce the privacy information of user.

Being defined as follows for information available quantity, the intelligent perception data uploaded as a user areSo this group is uploaded comprised information available quantity and is:

$utility\left(\stackrel{\→}{O}\right)=\left|\right\{t|O_t\≠empty\}|/T$

The problem uploading data, can be modeled as a decision problem, namely for single at position L_tGather group Data D of intelligence perception_t, with probability P_tRetain this data, the most do not upload these data.And with 1-P_tUpload this data. In order to protect the acquisition of the privacy information side of being hacked, one intuitively idea be exactly with bigger probability P_tRetain and privacy information Relevant data, and in order to maximize information available quantity, with smaller probability P_tRetain and the incoherent data of privacy information.

According to the theory of condition random field, and space-time relationship that may be present in actual life, a given output Data setAssailant deduces real positional informationProbability be:

$\begin{array}{c}\mathrm{Pr}\[\stackrel{\→}{L}|\stackrel{\→}{O}\]=\frac{1}{Z\left(\stackrel{\→}{O}\right)}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}$

g_k(L_t,O_t) it is characterized function,Represent the time between positional information Relatedness,Represent positional information and upload the spatial correlation between intelligent perception data,Represent position Information and the feature association uploaded between intelligent perception data, g_k(L_t,O_t) represent positional information and upload intelligent perception data Between decision-making relatedness, and normalization factor is:

$\begin{array}{c}Z\left(\stackrel{\→}{O}\right)=\underset{L}{\Σ}\exp \{\underset{t=1}{\overset{T}{\Σ}}\left(\underset{k=1}{\overset{K_1}{\Σ}}{\mathrm{\λ}}_k{f}_k\right(\stackrel{\→}{L})\\ +\underset{k=1}{\overset{K_2}{\Σ}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\Σ}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\Σ}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}$

To describe the definition of characteristic function below one by one in detail:

For decision-making linked character function g_k(L_t,O_t), its essence is and join uploading strategy in condition random field.This Individual characteristic function has two parts to constitute:

$g_{k1}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=e)$

$g_{k2}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=D_t)$

Characteristic parameter the most directly portray into:

${\mathrm{\θ}}_{k1}=\log P_j^i$

${\mathrm{\θ}}_{k1}=log(1-P_j^i)$

So, just the decision method of intimacy protection system is portrayed the system into condition random field.

For portraying the characteristic function of temporal associativityIt is used for describing transfer that may be present between adjacent position.Ratio As in the diagram, from C, B, A, the transfer to next position F may be portrayed by such a characteristic function:

By that analogy, for the transfer between any t position, characteristic function f can be passed through_k(L₁,L₂,...,L_t) come Portray:

$f_k(L_1,L_2,...,L_t)=\mathrm{\δ}(L_1=l_{i_1},L_2=l_{i_2},...,L_t=l_{i_t})$

WhereinRepresent t related position.

Characteristic functionFor portraying spatial correlation.The most in Figure 5, S is privacy position, and B Yu S is high Degree association.If user goes to B from A, he is likely to go to S, so B should be retained with the highest probability.From Another one aspect is said, if user goes to C from A, he can select go to S or go to D, so need not with higher general Rate retains C.For such a relatedness, portrayed by following characteristics function:

h_k(L_t,L_t-1,O_t,O_t-1)=δ (L_t=S, L_t-1=C, O_t=e, O_t-1=e).

For common scenario, a characteristic function can be set, be used for portraying a privacy information set S' and Individual location sets L' that is mutually related with S'.For convenience of description and calculate simplicity, the most only consider a privacy position And the interdependence between a relevant position:

$h_k(L_t,L_{t-t^{\′}},O_t,O_{t-t^{\′}})=\mathrm{\δ}(L_t=s_j^i,L_{t-t^{\′}}=l_j,O_t=e,O_{t-t^{\′}}=e).$

Corresponding characteristic parameter β_kFor portraying such a correlation degree.The most in Figure 5, the feature letter between S and B Number should have bigger characteristic parameter, and the characteristic parameter between S and C is the least.

Represent feature association, the most namely relatedness of received signal strength RSS data.Such as Fig. 6 institute Show, when user is toward AP walking when, and the value of RSSI can respond increase, and if turn round in midway, RSSI's Value will drastically decline.The feature of such a association is also possible to the side of being hacked and is utilized, and speculates what user truly passed by Path.In order to portray such a relatedness, just define

r_k1(L_t,L_t-1,O_t,O_t-1)=δ (dir (L_t,L_t-1)=ahead)

×δ(dis(O_t.rss,O_t-1.rss)≤R)

r_k1(L_t,L_t-1,O_t,O_t-1)=δ (dir (L_t,L_t-1)=turn)

×δ(dis(O_t.rss,O_t-1.rss) ＞ R)

If user is from L_t-1Go to L_tDo not turn, then dir (L_t,L_t-1)=ahead, otherwise dir (L_t,L_t-1)= turn。dis(O_t.rss,O_t-1.rss) for calculating Euler's distance of two location point RSS readings.

Step 203: the characteristic parameter in training pattern.

In this step, for spatial correlation, the parameter needing study is β_k, shifting according to probability, training method is such as Under:

${\mathrm{\β}}_k=\log \mathrm{Pr}(L_t=s_j^i|L_{t-t^{\′}}=l_j)$

If it is to say, there is stronger relatedness a positional information and a privacy position, then he just should be with Bigger probability goes to retain.

For the relatedness of RSS feature, first have to train threshold value R, given

$R=\mathrm{argmax}\underset{j}{\Σ}\underset{t=1}{\Σ}\begin{array}{c}\{r_{k1}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.rss,L_{t-1}^{i\left(j\right)}.rss)\\ +r_{k2}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.rss,L_{t-1}^{i\left(j\right)}.rss\}\end{array}$

After determining threshold value R, characteristic parameter then can directly calculate:

γ_k1=log (R-dis (x_t.rss,x_t-1.rss))/R

γ_k2=log (dis (x_t.rss,x_t-1.rss)-R)/R

Finally, the characteristic parameter of space correlation characteristic function to be trained, the most complex.Use maximal condition possibility predication Method, need the maximized object function to be:

$l\left(\mathrm{\λ}\right)=\underset{j}{\Σ}\log \mathrm{Pr}\[{\stackrel{\→}{L}}^{i\left(j\right)}|{\stackrel{\→}{O}}^{i\left(j\right)}\]$

To some specific parameter lambda_kAfter doing local derviation, obtain:

$\frac{\∂l\left(\mathrm{\λ}\right)}{\∂{\mathrm{\λ}}_k}=\underset{j=1}{\overset{N}{\Σ}}\underset{t=1}{\overset{T}{\Σ}}{f}_k({L_t}^{i\left(j\right)},{L_{t-1}}^{i\left(j\right)},{L_{t-2}}^{i\left(j\right)}-\underset{J=1}{\overset{N}{\Σ}}\underset{T=1}{\overset{T}{\Σ}}\underset{l,l^{\′},l^{\′\′}}{\Σ}{f}_k(l,l^{\′},l^{\′\′})\mathrm{Pr}(l,l^{\′},l^{\′\′}|{\stackrel{\→}{O}}^{i\left(j\right)})$

L-BFGS can be used to carry out regularization, solve this optimization problem.

Step 204: the data choosing optimum upload strategy, it is achieved the protection of user privacy information.

In this step, after training a condition random field, remaining be how to choose optimum upload plan Slightly, namely for the intelligent perception data gathered on each position, with what kind of probability retain or upload.This One strategy of sample can directly affect assailant and speculate the probability of privacy information.All possible strategy is uploaded it is therefore desirable to travel through, Find out and meet δ-privacy, simultaneously maximum for utility one.

Such as, in the teaching building of Tsing-Hua University, totally 1600 square metres, include totally 16498 records of 4 users. For each user, first the first half data with him carry out the study of system, test by the data of later half.In training Part, we travel through all possible retention strategy P.For each P, we can train a correspondingly condition random field Model, as shown in Figure 1.Can calculate whether assailant deduces the probability of privacy information based on this conditional random field models Meet δ-privacy.Finally, we are from the retention strategy meeting δ-privacy, choose a utility maximum.

Above example is merely to illustrate technical scheme, is not intended to limit；Although with reference to previous embodiment The present invention is described in detail, it will be understood by those within the art that: it still can be to aforementioned each enforcement Technical scheme described in example is modified, or wherein portion of techniques feature is carried out equivalent；And these are revised or replace Change, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (9)

1. a method for secret protection, it is characterised in that the method includes:

Pre-defined privacy location setsIf the positional information of user's reality is The strategy uploaded isThe intelligent perception data that certain is uploaded areWherein upload StrategyNeed all intelligent perception data uploadedMeetingOn the premise of so thatExpectation maximum；

Wherein, L_tExpression optional position,Represent any privacy position,Represent in the known situation uploading data The lower probability successfully speculating privacy position,It is to represent supposition privacy position in the case of not having any Given information The probability put,Value for presetting, δ represents the degree of protection privacy, and it is the least that δ is arranged, and more can protect hidden Private；

DescribedRepresent the intelligent perception data that user uploads The information available quantity comprised, T is user's sum of position when uploading intelligent perception data；

Described $\mathrm{Pr}[L_t=s_j^i|\stackrel{\→}{O}]$ For

Wherein,

$\begin{array}{c}\mathrm{Pr}\left[\stackrel{\→}{L}\right|\stackrel{\→}{O}]=\frac{1}{Z\left(\stackrel{\→}{O}\right)}\exp \{\underset{t=1}{\overset{T}{\mathrm{\Σ}}}(\underset{k=1}{\overset{K_1}{\mathrm{\Σ}}}{\mathrm{\λ}}_k{f}_k\left(\stackrel{\→}{L}\right)\\ +\underset{k=1}{\overset{K_2}{\mathrm{\Σ}}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\mathrm{\Σ}}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\mathrm{\Σ}}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}$

Wherein, λ_k、β_k、γ_kAnd θ_kIt is characterized parameter；g_k(L_t, O_t) it is characterized function, Represent the temporal associativity between positional information,Represent positional information and upload the sky between intelligent perception data Between relatedness,Represent positional information and the feature association uploaded between intelligent perception data, g_k(L_t, O_t) represent position Confidence breath and the decision-making relatedness uploaded between intelligent perception data, normalization factor is:

$\begin{array}{c}Z\left(\stackrel{\→}{O}\right)=\underset{L}{\mathrm{\Σ}}\exp \{\underset{t=1}{\overset{T}{\mathrm{\Σ}}}(\underset{k=1}{\overset{K_1}{\mathrm{\Σ}}}{\mathrm{\λ}}_k{f}_k\left(\stackrel{\→}{L}\right)\\ +\underset{k=1}{\overset{K_2}{\mathrm{\Σ}}}{\mathrm{\β}}_k{h}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_3}{\mathrm{\Σ}}}{\mathrm{\γ}}_k{r}_k(\stackrel{\→}{L},\stackrel{\→}{O})+\underset{k=1}{\overset{K_4}{\mathrm{\Σ}}}{\mathrm{\θ}}_k{g}_k(L_t,O_t)\left)\right\}\end{array}.$

Method the most according to claim 1, it is characterised in that described positional information and uploading between intelligent perception data Decision-making Relating Characteristic function g_k(L_t, O_t) it is:

$g_{k1}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=e)$

$g_{k2}(L_t,O_t)=\mathrm{\δ}(L_t=l_j^i,O_t=D_t)$

Wherein, e represents sky empty, O_t=e represents that uploading data in current location is sky, O_t=D_tRepresent and upload in current location Data are the data of intelligent perception,Representing the position that some user passes by, δ (w) is true equal to 1 when w condition, is otherwise equal to 0。

Method the most according to claim 1, it is characterised in that the temporal associativity characteristic function between described positional informationFor

$f_k\left(\stackrel{\→}{L}\right)=\mathrm{\δ}(L_1={\text{l}}_{i_1},L_2=l_{i_2},...,L_t=l_{i_t})$

WhereinRepresent t related position.

Method the most according to claim 1, it is characterised in that described positional information and uploading between intelligent perception data Spatial correlation characteristic functionFor

h_k(L_t, L_t-1, O_t, O_t-1)=δ (L_t=S, L_t-1=C, O_t=e, O_t-1=e).

Method the most according to claim 1, it is characterised in that described positional information and uploading between intelligent perception data Feature association characteristic functionFor

r_k1(L_t, L_t-1, O_t, O_t-1)=δ (dir (L_t, L_t-1)=ahead)

×δ(dis(O_t.rss, O_t-1.rss)≤R)

r_k1(L_t, L_t-1, O_t, O_t-1)=δ (dir (L_t, L_t-1)=turn)

×δ(dis(O_t.rss, O_t-1.rss) ＞ R)

If user is from L_t-1Go to L_tDo not turn, then dir (L_t, L_t-1)=ahead, otherwise dir (L_t, L_t-1)=turn；dis (O_t.rss, O_t-1.rss) for calculating Euler's distance of the received signal strength RSS reading of two location points.

Method the most according to claim 1, it is characterised in that the characteristic parameter θ of described decision-making linked character function_kFor

${\mathrm{\θ}}_{k1}=\log P_j^i$

${\mathrm{\θ}}_{k1}=\log (1-P_j^i)$

WhereinRepresent the probability uploading intelligent perception data in j position.

Method the most according to claim 1, it is characterised in that described association in time characteristic function feature parameter"λ"_kFor

The maximized object function of method of employing maximal condition possibility predication is:

$l\left(\mathrm{\λ}\right)=\underset{j}{\mathrm{\Σ}}\log \mathrm{Pr}\left[{\stackrel{\→}{L}}^{i\left(j\right)}\right|{\stackrel{\→}{O}}^{i\left(j\right)}]$

To λ_kDo local derviation:

$\begin{array}{c}\frac{\∂l\left(\mathrm{\λ}\right)}{\∂{\mathrm{\λ}}_k}=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\underset{t=1}{\overset{T}{\mathrm{\Σ}}}{f}_k({L_t}^{i\left(j\right)},{L_{t-1}}^{i\left(j\right)},{L_{t-2}}^{i\left(j\right)}\\ -\underset{J=1}{\overset{N}{\mathrm{\Σ}}}\underset{T=1}{\overset{T}{\mathrm{\Σ}}}\underset{l,l^{\′},l^{\′\′}}{\mathrm{\Σ}}{f}_k(l,l^{\′},l^{\′\′})\mathrm{Pr}(l,l^{\′},l^{\′\′}|{\stackrel{\→}{O}}^{i\left(j\right)})\end{array}$

Use L-BFGS to carry out regularization, obtain λ_k。

Method the most according to claim 1, it is characterised in that the characteristic parameter β of described space correlation characteristic function_kFor

${\mathrm{\β}}_k=\log \mathrm{Pr}(L_t={s^i}_j|L_{t-t^{\′}}=l_j)$

Wherein L_t-t′Represent the position before the t ' moment.

Method the most according to claim 1, it is characterised in that the characteristic parameter γ of described feature association characteristic function_kFor

GivenFirst training threshold value R:

$R=\arg \max \underset{j}{\mathrm{\Σ}}\underset{t=1}{\mathrm{\Σ}}\begin{array}{c}\{r_{k1}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.\mathrm{rss},L_{t-1}^{i\left(j\right)}.\mathrm{rss})\\ +r_{k2}(L_t^{i\left(j\right)},L_{t-1}^{i\left(j\right)},L_t^{i\left(j\right)}.\mathrm{rss},L_{t-1}^{i\left(j\right)}.\mathrm{rss}\}\end{array}$

After determining threshold value R, obtain characteristic parameter:

γ_k1=log (R-dis (x_t.rss, x_t-1.rss))/R

γ_k2=log (dis (x_t.rss, x_t-1.rss)-R)/R。