CN110992690B - False data detection method based on space-time outliers in Internet of vehicles - Google Patents

Abstract

The invention proposes a false data detection method based on spatiotemporal outliers in the Internet of Vehicles. Quantify the degree of similarity of data within each group, and detect abnormal data groups with significant data divergence based on Bayesian methods. The area with high correlation of traffic data in the road space is defined as the spatiotemporal neighborhood, and the spatiotemporal neighborhood is established based on the historical data, and the outlier detection algorithm is used to detect the false data that deviates significantly from the whole. The invention utilizes the space-time correlation characteristics of traffic data in the real traffic environment to preliminarily determine the false data, and then combines the space-time outlier detection algorithm to quantify the degree of deviation of each data in the space-time neighborhood from the whole, and find the false data, which can be correct and effective. It can evaluate road traffic conditions and improve traffic management behaviors, so as to realize various upper-layer applications such as traffic safety and intelligent navigation.

Description

False data detection method based on spatiotemporal outliers in the Internet of Vehicles

technical field

The invention relates to the technical field of Internet of Vehicles security, in particular to a false data detection method based on spatiotemporal outliers in the Internet of Vehicles.

Background technique

By installing multiple types of sensors (GPS, radar, acceleration sensors, etc.) on the vehicle, the Internet of Vehicles enables the vehicle to perceive the motion state (position, speed, direction, instantaneous acceleration, etc.) of itself and surrounding vehicles during driving, as well as road environment information. (weather, road conditions, etc.). Use Dedicated Short-Range Communications (DSRC: Dedicated Short-Range Communications) and 5G technology to realize wireless communication between vehicles, between vehicles and roadside infrastructure, or between vehicles and background management centers, and share this information with The surrounding vehicles or the pre-deployed roadside infrastructure (RSU: RoadSide Unit) aggregates the data and provides it to the back-end traffic management center. Based on this information, it can effectively evaluate road traffic conditions, calculate vehicle travel time, or perceive abnormal events such as traffic congestion and traffic accidents, so as to realize various upper-layer applications such as traffic safety and intelligent navigation.

The network architecture of the Internet of Vehicles has the characteristics of distributed and self-organization. Vehicles have equal status, and any vehicle can send traffic data. Due to malicious attacks, user selfishness, sensor failures, etc., vehicle nodes may provide inaccurate or even wrong data. For example, in order to reach their destination faster, selfish vehicle drivers send fake high vehicle density and low driving speed to create the illusion of traffic congestion, and induce surrounding vehicles to leave the current road, so as to achieve the purpose of clearing the road. Vehicles may be misled by wrong data to make wrong responses, such as sudden braking, lane change, etc., which not only interfere with normal traffic order, but may even cause traffic accidents in extreme cases. The background management center will also make wrong judgments on traffic conditions, make wrong traffic management behaviors, and cause traffic chaos.

SUMMARY OF THE INVENTION

In view of the deficiencies in the above background technology, the present invention proposes a false data detection method based on spatiotemporal outliers in the Internet of Vehicles, which solves the technical problem of traffic chaos caused by the injection of false traffic data by internal adversaries in the existing Internet of Vehicles.

The technical scheme of the present invention is realized as follows:

A method for detecting false data based on spatiotemporal outliers in the Internet of Vehicles, the steps of which are as follows:

S1. Data collection and grouping: The RSU deployed on the roadside in the Internet of Vehicles is responsible for collecting the data reported by vehicle nodes within the coverage area, and grouping the data according to the time and space attributes of the data to obtain data groups;

S2. Data anomaly detection: Use RSU to calculate the data consistency factor C of the data group, and judge whether the value of the data consistency factor C has changed significantly on the time axis based on Bayesian decision-making and hypothesis testing methods. If it is marked as an abnormal data group, go to step S3, otherwise, return to step S1;

S3. Establish spatio-temporal neighborhood: use correlation entropy to construct spatio-temporal domain of abnormal data set;

S4, spatiotemporal outlier detection: use an outlier detection algorithm to detect outliers in the spatiotemporal neighborhood of an abnormal data group, and determine the false data injected by the attacker.

The method for data abnormality detection in the step S2 is:

S21. The calculation method of the data consistency factor C of the data group is:

Among them, G _s (t)={d ₁ , d ₂ ,...d _N } is a data group of traffic data collected by a group of N different vehicles in road segment s and time slot t, t=1, 2,...T , i,j=1,2,...,N;

S22. According to the principle of hypothesis testing, establish a hypothesis model:

Among them, Θ ^T = {C ₁ ,...C _T } is the consistency factor vector of the current data group on time slots 1, 2,...T, H ₀ means that the value of C _t has not changed significantly, H ₁ is the expression C The value of _t changes significantly at time slot t ₀ ;

S23. According to the Bayesian principle, use the posterior probability to determine the probability that the value of the data consistency factor C changes significantly at time t:

Among them, q∈{1, 2}, P(H _q |Θ ^T ) is the likelihood function, and P(H _q ) is the prior probability;

S24. According to the likelihood formula, it is judged that the data consistency factor C has changed significantly at t=t ₀ :

S25 , regard the data group G _s (t ₀ ) as an abnormal data group, and construct a spatiotemporal neighborhood to determine the false data injected by the attacker.

The spatiotemporal neighborhood includes a direct temporal neighborhood, a direct spatial neighborhood, and an indirect spatiotemporal neighborhood; the direct temporal neighborhood includes the data sets G _s (t-1), G _s collected at the same location and adjacent time slots (t+1); the direct spatial neighborhood: including the same time slot, data G _s-1 (t) and G _s+1 (t) of adjacent road segments; the indirect space-time neighborhood: including time, space are all indirectly adjacent datasets.

In the step S3, the method of constructing the spatiotemporal domain of the abnormal data group by using the association entropy is:

S31. Let X be a random variable, representing the traffic data collected by the vehicle, and discretize the X value into Q small sub-intervals;

S32. The data group G _s (t)=(d ₁ , d ₂ ,...d _N ) is the data d ₁ , d ₂ ,... d _N reported by N vehicles, where d _i ∈[0,Q-1] ; Among the N' velocity values, the number of velocity values falling in the jth sub-interval is s _j , where 0≤j≤Q-1, and the entropy of the random variable X is:

S33. Let X _G1 be the random variable of the data group G ₁ (t), and X _G2 be the random variable of the data group G ₂ (t), then the correlation entropy of the data groups G ₁ (t) and G ₂ (t) is:

表示

且

的概率，则随机变量

和

的互信息为：in,

express

and

probability, then the random variable

and

The mutual information is:

S34. Use redundant information R(G ₁ , G ₂ ) between the data groups G ₁ (t) and G ₂ (t) to quantify the association between the data groups G ₁ (t) and G ₂ (t):

或者

但

或者

将l^-或者l⁺定义数据组G_s(t)的时空邻域边界；G_s(t)的空间邻域为

其中，l^-和l⁺均是正整数；S35. Existence of ^sl- or s+l ⁺ is satisfied

or

but

or

Let l- or l ⁺ define the space ^- time neighborhood boundary of the data set G _s (t); the spatial neighborhood of G _s (t) is

Among them ^, l- and l ⁺ are positive integers;

S36. The time neighborhood of a data group includes data groups collected in the same road section but in different time slots, if there is tl ^- or t+l ⁺ satisfying R(G _s (t), G _s (tl ^- ))≥ε or R(G _s (t),G _s (t+l ⁺ ))≥ε, but R(G _s (t),G _s (tl ^- -1))<ε or R(G _s (t),G _s (t+l ⁺ +1))<ε, call l ^- or l ⁺ as the space-time neighborhood boundary.

The method for detecting the outliers in the spatiotemporal neighborhood of the abnormal data group by using the outlier detection algorithm in the step S4 is:

S41. The spatiotemporal neighborhood of an abnormal data group is composed of m data items, denoted as M={o ₁ , o ₂ , ..p _m }, and the mean value of the data in M is calculated:

其中，

S42. Calculate the standard deviation of the data in M as

in,

S43. Check whether each data item o is within the interval [Ak ₀ ·σ, A+k ₀ ·σ], where k ₀ >1 is a predefined constant parameter; data items that fall outside this interval are regarded as As outliers, it is regarded as fake data injected by the adversary, and the data provider is regarded as the attacker.

The beneficial effects that this technical solution can produce: the present invention utilizes the spatiotemporal correlation characteristics of traffic data in a real traffic environment to preliminarily determine false data, and then combines the spatiotemporal outlier detection algorithm to quantify the degree of deviation between each data in the spatiotemporal neighborhood and the whole , find false data, can correctly and effectively evaluate road traffic conditions, improve traffic management behavior, so as to achieve traffic safety, intelligent navigation and other rich and diverse upper-level applications.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

Fig. 1 is the flow chart of the present invention;

2 is a practical map of an embodiment of the present invention;

Fig. 3 is the detection rate of the present invention and the comparison algorithm under different vehicle densities;

Fig. 4 is the detection rate of the present invention and the comparative algorithm under different proportions of attackers;

FIG. 5 is the false detection rate of the present invention and the comparative algorithm under different proportions of attackers.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The present invention uses discretized time and road space. Time is divided into several time slots of equal length, denoted as t=1, 2, . . . The road is divided into several sections, denoted as s=1,2,…. Vehicles use on-board sensors to collect traffic data such as vehicle density, speed, and flow during driving, and share these data through DSRC (Dedicated Short-Range Communications) wireless communication technology to achieve multiple types of upper-layer applications such as traffic safety and traffic management. The general format of the message is o={t,(lon,lat),d,vid}, where t is the time of data collection, lon and lat indicate the location (latitude and longitude) of data collection, d is the traffic data collected by the vehicle, vid is the vehicle identification identifier. It is assumed that the RSU (Roadside Unit) is pre-deployed along the road, responsible for collecting the data reported by the vehicle, executing the detection method proposed by the present invention, evaluating the authenticity of the data reported by the vehicle, and detecting the false data injected by the attacker.

As shown in FIG. 1 , an embodiment of the present invention provides a method for detecting false data based on spatiotemporal outliers in the Internet of Vehicles. The specific steps are as follows:

S1. Data collection and grouping: The RSU deployed on the roadside in the Internet of Vehicles is responsible for collecting the data reported by the vehicle nodes within the coverage area, and grouping the data according to the time and space attributes of the data to obtain a data group; taking the road segment s as an example, the RSU The received data are grouped into: G _s (1), G _s (2),…G _s (T), which represent the data collected in a series of time slots 1, 2,…T in the same road segment s of the road segment.

S2. Data anomaly detection: Use RSU to calculate the data consistency factor C of the data group, and judge whether the value of the data consistency factor C has changed significantly on the time axis based on Bayesian decision-making and hypothesis testing methods. If it is marked as an abnormal data group, go to step S3, otherwise, return to step S1;

The data anomaly detection method is:

S21. The definition of data consistency is that the data consistency factor quantifies the degree of similarity of the data in the group. The calculation method of the data consistency factor C of the data group is:

Among them, G _s (t)={d ₁ , d ₂ ,...d _N } is a data group of traffic data collected by a group of N different vehicles in road segment s and time slot t, t=1, 2,...T , i,j=1,2,...,N;

S22. According to the principle of hypothesis testing, establish a hypothesis model:

Among them, Θ ^T ={C ₁ ,...C _T } is the vector composed of the consistency factors of the current data set on time slots 1, 2,...T, and H ₀ is the null hypothesis, indicating that the value of C _t has not changed significantly , H ₁ is an alternative hypothesis, indicating that the value of C _t changes significantly at time slot t ₀ ;

S23. According to the Bayesian principle, use the posterior probability to determine the probability that the value of the data consistency factor C changes significantly at time t:

Among them, q∈{1, 2}, P(H _q |Θ ^T ) is the likelihood function, and P(H _q ) is the prior probability;

S24. According to the likelihood formula, it is judged that the data consistency factor C has changed significantly at t=t ₀ :

S25, regard the data group G _s (t ₀ ) as an abnormal data group, and perform a further detection process to determine the false data injected by the attacker.

S3. Establish spatiotemporal neighborhood: use association entropy to construct the spatiotemporal domain of abnormal data groups; using step S2, data groups with abnormal data can be found, but it is impossible to determine which data groups are injected by attackers. The present invention uses historical traffic data to construct a spatiotemporal domain of abnormal data groups, compares each data item with data in the spatiotemporal neighborhood, and uses a spatiotemporal outlier algorithm to find data items that deviate significantly. In road scenes, traffic data between two adjacent locations generally has a strong spatial and temporal correlation, but the degree of correlation is affected by various factors such as geographic location and traffic rules. For example, in the highway scene, the two adjacent road sections before and after show strong correlation, while in the urban scene, the front and rear road sections divided by intersections are geographically close, but the correlation between vehicle motions affected by traffic lights is weak. Use spatiotemporal neighborhoods to represent geographic areas in road space where traffic data is highly correlated. Specifically, the data group G _s (t) contains data of time slot s and road segment t, and its spatiotemporal neighborhood includes direct temporal neighborhood, direct spatial neighborhood and indirect spatiotemporal neighborhood; the direct temporal neighborhood: including the same location , the data groups G _s (t-1), G _s (t+1) collected in adjacent time slots; the direct spatial neighborhood: including the same time slot, the data G _s-1 (t), G _s+1 (t); the indirect spatiotemporal neighborhood: including data groups that are indirectly adjacent in time and space.

The method for constructing the spatiotemporal domain of anomalous data sets by using the correlation entropy is as follows:

S31. Let X be a random variable, representing the traffic data collected by the vehicle, and discretize the X value into Q small sub-intervals;

S32. The data group G _s ( _t ) _{= ( d 1} , d ₂ , _. Among the N' velocity values, the number of velocity values falling in the jth sub-interval is s _j , where 0≤j≤Q-1, and the entropy of the random variable X is:

S33. Let X _G1 be the random variable of the data group G ₁ (t), and X _G2 be the random variable of the data group G ₂ (t), then the correlation entropy of the data groups G ₁ (t) and G ₂ (t) is:

表示

且

的概率，则随机变量

和

的互信息为：in,

express

and

probability, then the random variable

and

The mutual information is:

S34. Use redundant information R(G ₁ , G ₂ ) between the data groups G ₁ (t) and G ₂ (t) to quantify the association between the data groups G ₁ (t) and G ₂ (t):

或者

但

或者

其中，l^-和l⁺均是正整数，将l^-或者l⁺定义数据组G_s(t)的时空邻域边界；G_s(t)的空间邻域为

S35. Existence of ^sl- or s+l ⁺ is satisfied

or

but

or

Among them, l ^- and l ⁺ are both positive integers, and l ^- or l ⁺ defines the space-time neighborhood boundary of the data group G _s (t); the spatial neighborhood of G _s (t) is

S36. The time neighborhood of a data group includes data groups collected in the same road section but in different time slots, if there is tl ^- or t+l ⁺ satisfying R(G _s (t), G _s (tl ^- ))≥ε or R(G _s (t),G _s (t+l ⁺ ))≥ε, but R(G _s (t),G _s (tl ^- -1))<ε or R(G _s (t),G _s (t+l ⁺ +1))<ε, call l ^- or l ⁺ as the space-time neighborhood boundary.

S4, spatiotemporal outlier detection: use an outlier detection algorithm to detect outliers in the spatiotemporal neighborhood of an abnormal data group, and determine the false data injected by the attacker.

The method for detecting outliers in the spatiotemporal neighborhood of an abnormal data group by using an outlier detection algorithm is:

S41. The spatiotemporal neighborhood of an abnormal data group is composed of m data items, denoted as M={o ₁ , o ₂ , ..o _m }, and the mean value of the data in M is calculated:

其中，

S42. Calculate the standard deviation of the data in M as

in,

S43. Check whether each data item o is within the interval [Ak ₀ ·σ, A+k ₀ ·σ], where k ₀ >1 is a predefined constant parameter; data items that fall outside this interval are regarded as As outliers, it is regarded as fake data injected by the adversary, and the data provider is regarded as the attacker.

The effectiveness of the present invention is verified based on the simulation method, and the technical advantages of the present invention are evaluated by comparative analysis with the existing methods.

(1) Simulation method

In the simulation experiment, the traffic simulation software SUMO is used to generate a highly realistic traffic scene, and the electronic map provided by the OpenStreetMap project is used. The size of the simulation area is 2000 meters × 2000 meters, as shown in Figure 2. All roads are two-way, three lanes in each direction, and the position and direction of the vehicle entering the simulation area are randomly selected. The average density of vehicles in the simulation area is 75 vehicles/km, and the real-time driving speed is reported to the RSU during the driving process. Simulations were run 100 times and all simulation data are the average of 100 runs.

In order to verify the accuracy of the detection method, a collusion attack scenario is set up, assuming that there are multiple colluding attackers, sending fake low vehicle speeds to create the illusion of traffic congestion. The attack location and time are randomly selected, the fake vehicle speed is randomly selected between [0,10]km/h, each vehicle sends the vehicle speed once in a time slot, and the proportion of attackers in all vehicles is denoted as γ .

The performance of the detection method was evaluated using the following metrics. The detection rate (DR) is defined as the proportion of all false data that are correctly detected, and the false positive rate (FPR) is defined as the probability that the data is detected incorrectly, including real data being detected as false data and false data. detected as real data.

(2) Results and analysis

The effects of vehicle density and attacker ratio on the performance of the algorithm are investigated. The simulation results are presented in Figures 3 to 5. Figure 3 shows the performance comparison evaluation of the method of the present invention and the LEAVE (Local eviction of attackers and voting evaluators) protocol under different vehicle densities. The detection rate of the method of the present invention is higher than that of LEAVE under all vehicle densities. When the vehicle density is 80 vehicles/km, the detection rate of this method is 98.3%, while the detection rate of the LEAVE protocol is 91.5%. In addition, when the vehicle density is low, the detection rate of the method of the present invention is lower than that of the high vehicle density scene, and when the vehicle density is 20 vehicles/km, the detection rate is 92.0%. The detection rate is 99.9% when the vehicle density is 200 vehicles/km. This is mainly because the distance between vehicles is large when the vehicle density is low, the space restriction between vehicles is not obvious, and the vehicles can travel at the ideal speed. Since different types of vehicles have different ideal speeds, the speed difference between vehicles is relatively small. If the data is large, the data collected by the vehicle presents a large variance, which increases the difficulty of correct identification.

Figures 4 and 5 evaluate the anti-collision attack capability of the method of the present invention and the LEAVE protocol. Figure 3 presents the detection rates under different proportions of attackers. It can be seen from the figure that the detection rate of the method of the present invention is similar to that of LEAVE under the low proportion of attackers. However, when the proportion of attackers increases, the detection rate of the LEAVE protocol decreases rapidly, when the proportion of attackers reaches 30% and 40%, respectively, the detection rate is 65.5% and 40%. The method of the invention shows strong anti-collusion attack capability. When the proportion of attackers reaches 25%, it is still able to achieve a detection rate of 96.5%. At the same time, the false detection rate is also significantly lower than that of LEAVE, with a false detection rate of 5.9% under a 25% attacker ratio.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the present invention. within the scope of protection.

Claims (4)

1. a false data detection method based on spatiotemporal outliers in the Internet of Vehicles, is characterized in that, its steps are as follows: S1. Data collection and grouping: The RSU deployed on the roadside in the Internet of Vehicles is responsible for collecting the data reported by vehicle nodes within the coverage area, and grouping the data according to the time and space attributes of the data to obtain data groups; S2. Data anomaly detection: Use RSU to calculate the data consistency factor C of the data group, and judge whether the value of the data consistency factor C has changed significantly on the time axis based on Bayesian decision-making and hypothesis testing methods. If it is marked as an abnormal data group, go to step S3, otherwise, return to step S1; The method for data abnormality detection in the step S2 is: S21. The calculation method of the data consistency factor C of the data group is:

Among them, G _s (t)={d ₁ , d ₂ ,...d _N } is a data group of traffic data collected by a group of N different vehicles in road segment s and time slot t, t=1, 2,...T , i,j=1,2,...,N; S22. According to the principle of hypothesis testing, establish a hypothesis model:

Among them, Θ ^T = {C ₁ ,...C _T } is the consistency factor vector of the current data group on time slots 1, 2,...T, H ₀ means that the value of C _t has not changed significantly, H ₁ is the expression C The value of _t changes significantly at time slot t ₀ ; S23. According to the Bayesian principle, use the posterior probability to determine the probability that the value of the data consistency factor C changes significantly at time t:

Among them, q∈{1, 2}, P(H _q |Θ ^T ) is the likelihood function, and P(H _q ) is the prior probability; S24. According to the likelihood formula, it is judged that the data consistency factor C has changed significantly at t=t ₀ :

S25, regard the data group G _s (t ₀ ) as an abnormal data group, and construct a space-time neighborhood to determine the false data injected by the attacker; S3. Establish spatio-temporal neighborhood: use correlation entropy to construct spatio-temporal domain of abnormal data set; S4, spatiotemporal outlier detection: use an outlier detection algorithm to detect outliers in the spatiotemporal neighborhood of an abnormal data group, and determine the false data injected by the attacker. 2. The method for detecting false data based on spatiotemporal outliers in the Internet of Vehicles according to claim 1, wherein the spatiotemporal neighborhood comprises a direct temporal neighborhood, a direct spatial neighborhood and an indirect spatiotemporal neighborhood; the Direct temporal neighborhood: including the data groups G _s (t-1) and G _s (t+1) collected at the same location and adjacent time slots; the direct spatial neighborhood: including the data of the same time slot and adjacent road sections G _s-1 (t), G _s+1 (t); the indirect space-time neighborhood: including data groups that are indirectly adjacent in time and space. 3. the false data detection method based on spatiotemporal outliers in the Internet of Vehicles according to claim 1, is characterized in that, utilizes association entropy in described step S3 to construct the method for the spatiotemporal domain of abnormal data group as: S31. Let X be a random variable, representing the traffic data collected by the vehicle, and discretize the X value into Q small sub-intervals; S32. The data group G _s (t)=(d ₁ , d ₂ ,...d _N ) is the data d ₁ , d ₂ ,... d _N reported by N vehicles, where d _i ∈[0,Q-1] ; Among the N' velocity values, the number of velocity values falling in the jth sub-interval is s _j , where 0≤j≤Q-1, and the entropy of the random variable X is:

S33. Let X _G1 be the random variable of the data group G ₁ (t), and X _G2 be the random variable of the data group G ₂ (t), then the correlation entropy of the data groups G ₁ (t) and G ₂ (t) is:

in,

express

and

probability, then the random variable

and

The mutual information is:

S34. Use redundant information R(G ₁ , G ₂ ) between the data groups G ₁ (t) and G ₂ (t) to quantify the association between the data groups G ₁ (t) and G ₂ (t):

S35. Existence of ^sl- or s+l ⁺ is satisfied

or

but

or

Let l- or l ⁺ define the space ^- time neighborhood boundary of the data set G _s (t); the spatial neighborhood of G _s (t) is

Among them ^, l- and l ⁺ are positive integers; S36. The time neighborhood of a data group includes data groups collected in the same road section but in different time slots, if there is tl ^- or t+l ⁺ satisfying R(G _s (t), G _s (tl ^- ))≥ε or R(G _s (t),G _s (t+l ⁺ ))≥ε, but R(G _s (t),G _s (tl ^- -1))<ε or R(G _s (t),G _s (t+l ⁺ +1))<ε, call l ^- or l ⁺ as the space-time neighborhood boundary. 4. The method for detecting false data based on spatiotemporal outliers in the Internet of Vehicles according to claim 1, wherein in the step S4, an outlier detection algorithm is used to detect outliers in the spatiotemporal neighborhood of abnormal data groups The point method is: S41. The spatiotemporal neighborhood of an abnormal data group is composed of m data items, denoted as M={o ₁ , o ₂ , ..o _m }, and the mean value of the data in M is calculated:

S42. Calculate the standard deviation of the data in M as

in,

S43. Check whether each data item o is within the interval [Ak ₀ ·σ, A+k ₀ ·σ], where k ₀ >1 is a predefined constant parameter; data items that fall outside this interval are regarded as As outliers, it is regarded as fake data injected by the adversary, and the data provider is regarded as the attacker.

Images