• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Effect of observation time on source identification of diffusion in complex networks

    2022-08-01 05:58:26ChaoyiShi史朝義QiZhang張琦andTianguangChu楚天廣
    Chinese Physics B 2022年7期
    關(guān)鍵詞:張琦

    Chaoyi Shi(史朝義), Qi Zhang(張琦), and Tianguang Chu(楚天廣),?

    1College of Engineering,Peking University,Beijing 100871,China

    2School of Information Technology and Management,University of International Business and Economics,Beijing 100105,China

    Keywords: complex network,source identification,statistical inference,partial observation

    1. Introduction

    Recent years, researchers have witnessed tremendous growth in studies concerning the diffusion processes in complex networks, such as modeling epidemic spreading,[1–4]identifying influential spreaders,[5–9]and analyzing diffusion dynamics.[10–12]Among a plethora of research thrusts, the problem of identifying the diffusion sources in complex networks based on observed information has gathered increasing attention due to its broad applications in such problems as identifying the origin of a rumor,[13,14]locating the source of epidemic,[15–18]and so on. Clearly, studies in this direction are crucial in preventing or controlling spread of adverse or detrimental influences in a network system consisting of large amount of interconnected individuals or nodes.

    Basically, source identification problem aims at estimating the location of source node that initiates the diffusion in a network by using certain observation information of the diffusion. So far, several approaches have been proposed to solve the source identification problem with observations that involve the states of observed nodes,[13,15–18]or the (infection) time at which the diffusion arrives at the observed nodes.[14,19–22]Early studies often assume complete observation condition, i.e., the state of each node is observed, and make use of the measures of certain centricity to evaluate the source detection probability of the maximum likelihood(ML)estimator.[13,15]However, it is usually hard in practice to obtain complete observations of a network, since the costs for observing all nodes can be expensive and the states of some nodes may be hidden.[3,14]In view of these, considerable efforts have been devoted to the study of source identification problems under partial observation conditions, i.e., the states or infection time of only a fraction of nodes can be available.For instance,in Refs.[16–18]the authors considered a single snapshot of partial observations at a given time and proposed different propagation algorithms to find the source of diffusion in networks.There have been also works addressing source localization by exploiting observed information of propagation delays from the source, under various assumptions on delay distributions, e.g., Gaussian and uniform distributions.[19–21]Another effective approach to locating diffusion source via incomplete observations in a network makes use of statistical inference based on Monte Carlo simulations.[22–24]The basic idea of this approach is to infer the diffusion source by evaluating the similarity between the simulation data and the observed data. Essentially,the approach applies to arbitrary network structures and therefore has wide applications. Notice that most existing studies assume that the initial time of diffusion is known in estimating source location,some of them are also concerned with the impact of the size of observation set on the results.[13,15–18]

    In this paper,we intend to discuss the effect of the observation time on source identification in a diffusive network. So far in literature,little has been available concerning such an issue. Generally speaking,identification of diffusion source in a network essentially relies on the diffusion model,the observation condition,and the underlying network structure. Here we consider the case that the diffusion process is described by a discrete-time SIR model,with the initial condition of all nodes in susceptible states except one in infected state as the source.In literature, this type of models is often used to describe the spread of epidemics[16,17]and computer virus.[15,19]We consider the problem of estimating both the source location and the initial time of diffusion with snapshot of partial nodes,and formulate it as an ML estimator in terms of the marginal probabilities of a node in different states. A method based on Monte Carlo simulation(MCS)is developed to evaluate the marginal probabilities of a candidate source node,whereby to examine the effect of the observation time as well as the fraction of the observers on solutions of the concerned problem with numerical experiments on typical synthetic and real-world networks.

    2. Problem formulation

    We consider the case that the diffusion outbreak occurs at a single node and the states of only a fraction of nodes can be observed with a snapshot. The task of source identification is to determine the location of source and the initial time of diffusion in a network based on the partial observation. In the following, we will introduce the diffusion model and the ML estimator for the source identification problem.

    2.1. The diffusion model

    To describe the discrete-time SIR diffusion process, we model the underlying contact network as an undirected graphG(V,E)withNnodes,whereVis the set of nodes,E ∈V×Vis the set of edges. At timet, each nodei ∈Vhas three possible states represented by a variableXti: susceptible,Xti=S;infected,Xti=I;or recovered,Xti=R.In each time step,an infected nodeiinfects its susceptible neighborjwith probabilityλij,or recovers with probabilityμi,and no longer get infected.Thus,a susceptible nodeibecomes infected with probability,

    where?iis the set of neighbors of nodei.As mentioned above,we assume that at an unknown initial timet0,all nodes are in stateSexcept only one infected nodei0,the source,that initiates diffusion process in the network.

    2.2. Snapshot observation

    Figure 1 shows an example of the diffusion process and snapshot observation on an arbitrary graphG.

    Fig. 1. Visualization of the diffusion process and observed data on graph G. At unknown time t0, the source node i0 (red) initiates the diffusion. The red edges are the links on which the spread occurs. In this example, a snapshot of the observers (green) is taken at time T,which has three types of nodes with the state S, I, or R for each. The goal is to identify the source location and initial time of diffusion from the snapshot observation.

    2.3. Maximum likelihood estimator

    Therefore,we can formulate the source identification problem as the following ML estimator:

    and clearly,

    We will make use of the marginal probabilities instead of the joint probability in the estimator(1)to infer the source node and initial time of diffusion.

    3. Monte Carlo simulation

    Now we present an MCS approach to estimate the marginal probabilities of a node. The basic idea is that for a candidate source nodei0∈V?OTS,we perform simulations ofmtimes for the diffusion process following the SIR model on time interval [0,tmax], and sample the state of every observeri ∈Oat timet ∈Z∩[0,tmax]. Then, by examining the similarity of the samples and the snapshot to determine the source and initial time of the diffusion.

    To be specific,letYti(i0,l)be the state of nodeiat timetfor thel-th simulation with the candidate source nodei0,it is clear that

    and fort=0,

    forl=1,...,m. In simulations,a node in the network updates its state by the following rule.

    For a susceptible nodeiat timet-1,each infected neighborjwill execute an independent infection attempt with success probabilityλji,triggered by a Bernoulli trial(flipping an independent coin)that generates a random numberr0from the uniform distribution between 0 and 1 as threshold. Namely,the state of nodeiin the next steptis given by

    Similarly, for a nodeiin the stateIat time stept-1, it will execute a recovery attempt with success probabilityμiaccording to a Bernoulli trial numberr0as before, and updates its state at time steptas follows:

    According to the specification of the SIR model described before,all recovered nodes will not be infected anymore and remain the recovered state during the simulation process,i.e.,

    whereδ(·,·)denotes the Kronecker delta function. Similarly,the marginal probabilities of the nodeibeing in statesIandRat timeTare given respectively by

    Algorithm 1 MSC-based source identification algorithm Input: G(V,E);λij;μi;OTS;OTI ;OTR,tmax Output: the source location ?i0;the initial time of diffusion ?t0 for a candidate source node i0 ∈V?OTS do for l=1 to m do Sample the state Yti(i0,l)of each observer i ∈O at time t ∈Z∩[0,tmax]end for for a candidate initial time t0 ∈Z∩[T-tmax,T-1]do Evaluate the marginal probabilities P(XTi =S|i0,t0),P(XTi =I|i0,t0),i ∈O Evaluate the joint probability P(OTS,OTI ,OTR|i0,t0)by Eq.(2)),P(XTi =R|i0,t0 end for end for{?i0,?t0}=argmax )i0∈V?OTS t0∈Z∩[T-tmax,T-1]P(OTS,OTI ,OTR|i0,t0

    We remark that the algorithmic complexity for a single SIR simulation isO(N〈k〉tmax), where〈k〉is the average degree of the graphG, thus runningmsimulations for a candidate source nodei0requiresO(mN〈k〉tmax) time. In experiments we found that it is enough to takem=200 for obtaining a satisfactory identification result, further increasing the simulation times does not improve the accuracy of identification significantly.

    Compared with some related studies, our method can be implemented directly by Monte Carlo simulations,without the need of complicated calculations or additional assumptions.For instance, the methods proposed in Refs. [21,22] require to calculate the multiple integrations of joint probability for partial observations. In Ref. [24], the source likelihood was estimated by using the Gaussian weighting function.

    4. Experiments and discussion

    We give numerical experiments to examine the effect of the observation time as well as the fraction of the observers on source identification for synthetic networks and real-world networks.In all the cases a fractionγ=K/Nof nodes is available in observation at timeTfor the source identification problem, and the sourcei0and the initial timet0are estimated by Algorithm 1. The performance of the method is assessed in terms of the success rate of locating source node for different fractionγand snapshot timeT.

    4.1. Synthetic networks

    We consider three typical synthetic networks,i.e.,Erd¨os–R′enyi(ER)random network,Barab′asi–Albert(BA)scale-free network, and Watts–Strogtz (WS) small world network,[1]all containingN= 1000 nodes with average degree〈k〉 ≈6. For convenience in simulation and without loss of generality,[13,17,24]we consider uniform infection and recovery probabilities by lettingλi j=λandμi=μ. In each experiment, a node is randomly selected to initiate the diffusion at timet=0 and we run SIR process fortmaxsteps,which can be approximated by

    wheredis the diameter of a network,the first term on the left side is the maximum infection time of a node, and the second term is the average recovery time of a node. Then, the state of a fractionγof nodes is observed at timeT,whereγ ∈{0.1,0.2,...,0.9,1.0}and the snapshot timeT ∈{1,...,tmax}.We repeated the experiment 4000 times for each case to get the average success rate of located source.

    Fig.2. Effects of the snapshot time T and the fraction γ of observers on source identification results: (a)–(c)λ =0.4,μ=0.2,(d)–(f)λ =0.8,μ =0.2. The success rate of locating source on ER,BA,and WS networks,with N=1000 nodes and average degree,〈k〉≈6. The diffusion starts at t=0,a fraction γ ∈{0.2,0.4,0.6,0.8}of nodes is observed at time 1 ≤T ≤10.

    We examine the effect of the snapshot timeTand the fractionγof observers on source identification results. Existing studies mainly considered the effect ofγon source localization accuracy, regardless of considering the snapshot timeT.Our results as illustrated in Fig.2 reveal an involved influence of the two factors on source identification problem. We can see that, in general, the source identification problem will be hard to attack for observations with large time delay; and the greater the number of observed nodes,the more accurately our methods can infer the true source. Indeed, a large value ofTallows for a full development of diffusion(hence infection)in a network,this would reduce effectiveness of the snapshot information. The cases for different infection probabilityλand recovery probabilityμare similar and hence omitted.

    In more detail,F(xiàn)ig.2 indicates different trends of the success rate of source localization with respect toTandγ. For very large values ofγ(e.g.,not less than 0.8),the success rate decreases monotonically with the increase ofT; whereas for the major portion of values ofγ(e.g., (0,0.8)), the success rate first rises to a highest level and then drops down with the increase ofT. This implies that one could try to choose an optimal timeTfor observation to get more reliable results. As shown in Fig. 2, it is better in general to take a snapshot at early stage of the diffusion process, and the range for such a timeTalso depends on SIR model parametersλandμ. However,for very smallγ(e.g.,0.2)the accuracy of source localization is low in early stages of diffusion as little information about infected nodes is available. Thus, it may be difficult to infer the source at early stages of diffusion from a very small observation set.

    Figure 3 illustrates the impact of the snapshot timeTand the infection probabilityλon the success rate of locating the source, where the value ofλvaries from 0.1 to 1.0. The results are different, depending on the network structures. For the ER network,the observation timeThas little effect on the source localization when the infection probabilityλis small(e.g., (0.1,0.3)). For the other two types of networks, this is true for relatively large values ofλ(e.g., (0.1,0.4) for BA,(0.1,0.7)for WS).On the other hand,for a large value ofλ,a relatively earlier observation is in favor of source localization in ER and BA networks.For the WS network,the effect of different observation timeTis less evident in the case of larger values ofλ.

    Figure 4 depicts the average snapshot timeTof successfully locating the source for different observation fractionγ. It shows that in generalTdecreases monotonically with the increase ofγfor all networks. This implies that a snapshot of a larger observation set facilitates the inference of the source at early stages of diffusion. This is consistent with the results as shown in Fig.2. In particular,given an observation fractionγ,one could try to choose an appropriate timeTfor observation by Fig.4 to get more reliable results.

    Fig. 3. Effects of the snapshot time T and infection probability λ on success rate of locating source for ER, BA, and WS networks, with N =1000 nodes and average degree 〈k〉≈6. The diffusion starts at t =0 with λ =0.1,...,1.0 and μ =0.1, a fraction γ =0.4 of nodes is observed at time 1 ≤T ≤10.

    Fig.4. Observation time T of successfully locating the source for different fraction γ on ER,BA,and WS networks,all containing N=1000 nodes with average degree〈k〉≈6: (a)λ =0.4,μ =0.2,(b)λ =0.3,μ =0.1. The diffusion process starts at t=0 from a randomly selected node,and a snapshot of γN nodes is taken for γ ∈{0.1,0.2,...,0.9,1}.

    Fig.5. Source localization and initial time estimate results on ER,BA,and WS networks,all containing N=1000 nodes with average degree〈k〉≈6. The diffusion process starts at t =0 from a randomly selected node with λ =0.4 and μ =0.2,and a snapshot of γN nodes is taken at time T for γ ∈{0.1,0.2,...,0.9,1}and 1 ≤T ≤10. (a)Success rates of locating source node for different fraction γ. The lower and upper values of the success rate are calculated by using snapshot observation at different time T. (b)MSEs of the initial time estimates for different snapshot time T. The lower and upper values of each MSE are computed with different γ.

    Figure 5 shows the source localization and initial time estimate results for different fractionγof observers and snapshot timeT. From Fig.5(a)we can see that the success rate of locating source increases rapidly asγincreasing from a small value till to certain value (e.g., 0.6), after that, adding more observer nodes does not improve the identification accuracy significantly. Figure 5(b) gives the mean square error (MSE)of the initial time estimates for different snapshot timeT, it indicates that the snapshots taken at early stage of diffusion result in small MSE values for different graph structures.

    4.2. Real-world networks

    Furthermore,we consider two real-world network examples of email[26]and US West-Coast power grid,[27]as detailed in Table 1, which are often cited to test source identification methods in the literature.[17,18]In our experiments, the diffusion process starts att=0 from a randomly selected node,withλ ∈{0.1,0.2,..., 0.8, 0.9}andμ=0.2 respectively in each running, and a snapshot ofγNnodes is taken at timeTforγ ∈{0.05,0.1,...,0.45,0.5}and 1≤T ≤10 respectively in each time.

    Table 1. The network parameters in the experiments,N and|E|denote the network size and the number of edges,respectively.

    Fig. 6. Effects of the snapshot time T and the fraction γ of observers on source identification results for email and US West-Coast power grid. The diffusion process starts at t =0 with λ ∈{0.1,0.2,...,0.8,0.9} and μ =0.2, and a snapshot of γN nodes is taken at time T for γ ∈{0.05,0.1,...,0.45,0.5}and 1 ≤T ≤10. (a)–(d)Success rate of locating source node for different T and γ.

    Figure 6 depicts the average results over 1000 experiments. It shows an evident impact of the observation time as well as the fraction of the observers on the source identification problem,which is similar to the case of the synthetic networks. Particularly, Figs. 6(a) and 6(c) indicate that the success rate of locating source is small at early stage of the diffusion process,and it will be hard to locate the source for large observation time. Figures 6(b)and 6(d)show that,in general,the success rate of locating source increases with the number of observer for different snapshot timeT. For a small value ofT(e.g., 1), the success rate of locating source increases rapidly as the fractionγof observers increasing,whereas for a big value ofT(e.g.,7),adding more observer nodes does not improve the source localization accuracy significantly.

    5. Summary

    We have discussed the effect of the observation time as well as the size of the observation set on source identification of diffusion in a network with discrete-time SIR spreading process, under incomplete observation conditions. The method makes use of the Monte Carlo simulation in evaluation of the marginal probabilities of a node that are involved in ML estimator of the source. Numerical experiments in synthetic networks and real-world networks reveal interesting yet involved nonlinear effects of the observation time and the fraction of observers on source identification. In general,a large value of snapshot time will not be in favor of the source identification for most values of the fraction of observers,since it allows for a full development of diffusion(hence infection)in a network,reducing effectiveness of the snapshot information. In particular,for very large values of the fraction of observers,the success rate of source localization decreases monotonically with the increase of the observation time. On the other hand,however, for a very small observation set, it is usually difficult to infer the source at early stages of diffusion because little information about infected nodes is available. More complicated and interesting cases occur for neither too small nor too large values of the fraction of observers, where the success rate of source localization first rises to a highest level and then drops down with the increase of the snapshot time,indicating possible choice of an optimal observation time to get more reliable results.

    Acknowledgements

    Project supported by the National Natural Science Foundation of China (Grant Nos. 61673027 and 62106047), the Beijing Social Science Foundation (Grant No. 21GLC042),and the Humanity and Social Science Youth foundation of Ministry of Education,China(Grant No.20YJCZH228).

    猜你喜歡
    張琦
    Self-screening of the polarized electric field in wurtzite gallium nitride along[0001]direction
    Magnetoresistance effect in vertical NiFe/graphene/NiFe junctions
    全民張琦
    商界評論(2022年12期)2022-03-06 13:02:12
    自相似視角下相對貧困成因分析
    基于TXL的源代碼插樁技術(shù)研究
    張琦:家風敗壞的海南第四“虎”
    “海南虎”張琦:一位闖海者的隕落
    雜文選刊(2020年4期)2020-04-19 10:04:31
    ??谑形瘯洀堢糯蠛蠛D稀笆谆ⅰ?/a>
    曹夢媛、崔琪、張琦、趙承鋮作品
    西部縣域經(jīng)濟發(fā)展模式亟待創(chuàng)新
    一级a爱视频在线免费观看| 蜜桃国产av成人99| 高清av免费在线| 日韩人妻精品一区2区三区| 久久人人爽av亚洲精品天堂| 人人妻人人澡人人看| 中文字幕人妻熟女乱码| 免费看十八禁软件| 精品久久久久久久毛片微露脸| 精品视频人人做人人爽| 丝袜喷水一区| 老汉色∧v一级毛片| 97人妻天天添夜夜摸| 欧美日韩国产mv在线观看视频| av一本久久久久| av又黄又爽大尺度在线免费看| 国产97色在线日韩免费| 大香蕉久久成人网| 免费人妻精品一区二区三区视频| 日本a在线网址| 老汉色av国产亚洲站长工具| 国产不卡一卡二| 久久精品aⅴ一区二区三区四区| 91字幕亚洲| 久久久久久久国产电影| 成年版毛片免费区| 成人永久免费在线观看视频 | 久久精品国产综合久久久| 成人18禁高潮啪啪吃奶动态图| 久9热在线精品视频| 高清黄色对白视频在线免费看| 亚洲人成77777在线视频| 久久人妻福利社区极品人妻图片| 欧美激情 高清一区二区三区| 亚洲精品久久成人aⅴ小说| 一区福利在线观看| 亚洲av成人不卡在线观看播放网| 婷婷成人精品国产| 国产精品一区二区精品视频观看| 成人影院久久| 一本—道久久a久久精品蜜桃钙片| 亚洲熟女精品中文字幕| 国产成人影院久久av| 日韩大码丰满熟妇| 欧美精品一区二区大全| 欧美在线一区亚洲| 亚洲av日韩精品久久久久久密| 亚洲久久久国产精品| 久久久欧美国产精品| 在线永久观看黄色视频| 一区二区日韩欧美中文字幕| 日韩 欧美 亚洲 中文字幕| tube8黄色片| 丝袜喷水一区| 十八禁高潮呻吟视频| 18禁观看日本| 欧美日韩亚洲高清精品| 久久久久久免费高清国产稀缺| 激情视频va一区二区三区| 黄色成人免费大全| 国产精品98久久久久久宅男小说| 怎么达到女性高潮| 久久热在线av| 国产成人免费无遮挡视频| 免费在线观看日本一区| 亚洲欧美日韩高清在线视频 | 国产精品久久久久久精品古装| 日韩一区二区三区影片| 一级黄色大片毛片| 欧美性长视频在线观看| 天天躁日日躁夜夜躁夜夜| 精品国内亚洲2022精品成人 | 女人被躁到高潮嗷嗷叫费观| 麻豆乱淫一区二区| 亚洲自偷自拍图片 自拍| 美国免费a级毛片| 人人澡人人妻人| 一区二区三区乱码不卡18| 最新美女视频免费是黄的| 中亚洲国语对白在线视频| 精品国产乱码久久久久久男人| 90打野战视频偷拍视频| 国产免费现黄频在线看| 啦啦啦 在线观看视频| 性高湖久久久久久久久免费观看| 最新的欧美精品一区二区| 色94色欧美一区二区| 男女下面插进去视频免费观看| 国产精品久久久久久人妻精品电影 | 午夜两性在线视频| 色老头精品视频在线观看| 国产精品99久久99久久久不卡| 久久国产精品影院| 亚洲中文字幕日韩| 国产1区2区3区精品| 亚洲色图综合在线观看| 久久ye,这里只有精品| 久久99热这里只频精品6学生| 国精品久久久久久国模美| 人人妻人人添人人爽欧美一区卜| 国产成人免费无遮挡视频| 亚洲色图综合在线观看| av国产精品久久久久影院| aaaaa片日本免费| 欧美成狂野欧美在线观看| 中亚洲国语对白在线视频| kizo精华| 狠狠精品人妻久久久久久综合| 啦啦啦视频在线资源免费观看| 精品少妇久久久久久888优播| 久久ye,这里只有精品| 国产精品一区二区免费欧美| 久久国产精品人妻蜜桃| 国精品久久久久久国模美| 99精品在免费线老司机午夜| 日本av手机在线免费观看| 成人18禁高潮啪啪吃奶动态图| 国产精品 欧美亚洲| 性少妇av在线| 大片电影免费在线观看免费| 国产精品一区二区在线不卡| 脱女人内裤的视频| 国产1区2区3区精品| 色综合婷婷激情| 欧美性长视频在线观看| 亚洲成人免费电影在线观看| 飞空精品影院首页| 国产日韩欧美视频二区| 午夜两性在线视频| 中文字幕制服av| www.自偷自拍.com| 欧美+亚洲+日韩+国产| 18禁观看日本| 国产成人欧美| 18禁观看日本| av线在线观看网站| 三上悠亚av全集在线观看| 国产高清视频在线播放一区| 午夜免费成人在线视频| 亚洲成人免费av在线播放| 国产亚洲精品第一综合不卡| 亚洲国产毛片av蜜桃av| 黄片小视频在线播放| 黄色视频不卡| 亚洲av日韩精品久久久久久密| 久久久国产精品麻豆| 在线看a的网站| 亚洲精华国产精华精| 欧美日韩一级在线毛片| 一本—道久久a久久精品蜜桃钙片| 久久ye,这里只有精品| 91老司机精品| 国产精品一区二区在线不卡| √禁漫天堂资源中文www| svipshipincom国产片| 最新美女视频免费是黄的| 淫妇啪啪啪对白视频| 一本久久精品| 欧美精品亚洲一区二区| 老汉色∧v一级毛片| 国产日韩欧美亚洲二区| 欧美精品高潮呻吟av久久| 99re在线观看精品视频| kizo精华| 露出奶头的视频| 自拍欧美九色日韩亚洲蝌蚪91| 久久精品国产99精品国产亚洲性色 | 亚洲欧美一区二区三区久久| 欧美日韩国产mv在线观看视频| 曰老女人黄片| 日韩中文字幕欧美一区二区| 精品福利观看| 侵犯人妻中文字幕一二三四区| 亚洲国产成人一精品久久久| 妹子高潮喷水视频| 999久久久精品免费观看国产| 在线亚洲精品国产二区图片欧美| tube8黄色片| 国产不卡av网站在线观看| 亚洲成av片中文字幕在线观看| 一个人免费在线观看的高清视频| 亚洲精品中文字幕一二三四区 | 一本久久精品| www日本在线高清视频| 亚洲一区中文字幕在线| 少妇 在线观看| 999精品在线视频| 狠狠精品人妻久久久久久综合| 欧美久久黑人一区二区| 一区二区三区国产精品乱码| 手机成人av网站| 丝袜在线中文字幕| 国产精品麻豆人妻色哟哟久久| 欧美黄色片欧美黄色片| 亚洲av电影在线进入| 久久久久久久久久久久大奶| 国产成人av教育| 成年女人毛片免费观看观看9 | 最近最新免费中文字幕在线| 国产精品美女特级片免费视频播放器 | 精品国产乱子伦一区二区三区| 美国免费a级毛片| 黄色成人免费大全| 动漫黄色视频在线观看| videos熟女内射| 亚洲视频免费观看视频| 国产免费现黄频在线看| 国产精品 国内视频| 欧美一级毛片孕妇| 免费久久久久久久精品成人欧美视频| 日韩大片免费观看网站| 日韩一卡2卡3卡4卡2021年| 三级毛片av免费| 国产欧美日韩一区二区精品| 中国美女看黄片| 男女下面插进去视频免费观看| 欧美日本中文国产一区发布| 欧美黑人精品巨大| 啦啦啦免费观看视频1| 久久久国产欧美日韩av| 欧美亚洲日本最大视频资源| 中文亚洲av片在线观看爽 | 91国产中文字幕| 久久久久久久精品吃奶| 日韩欧美免费精品| 91成年电影在线观看| 国产伦理片在线播放av一区| 一区福利在线观看| 色在线成人网| 久久精品国产a三级三级三级| 国产精品麻豆人妻色哟哟久久| 亚洲成av片中文字幕在线观看| 交换朋友夫妻互换小说| 亚洲免费av在线视频| 色视频在线一区二区三区| 亚洲午夜精品一区,二区,三区| 男人舔女人的私密视频| 国产精品免费一区二区三区在线 | 午夜视频精品福利| 蜜桃在线观看..| 国产有黄有色有爽视频| 亚洲国产av新网站| 最新美女视频免费是黄的| av不卡在线播放| 天天躁夜夜躁狠狠躁躁| 在线永久观看黄色视频| 在线看a的网站| 桃花免费在线播放| 50天的宝宝边吃奶边哭怎么回事| av电影中文网址| 亚洲中文字幕日韩| 精品亚洲成a人片在线观看| kizo精华| 亚洲精华国产精华精| 亚洲精品乱久久久久久| 国产成人精品久久二区二区免费| 国产视频一区二区在线看| 精品亚洲乱码少妇综合久久| 精品少妇黑人巨大在线播放| 男女免费视频国产| 菩萨蛮人人尽说江南好唐韦庄| 18禁美女被吸乳视频| 免费高清在线观看日韩| 久久久久国内视频| 亚洲成国产人片在线观看| 色综合婷婷激情| 美女视频免费永久观看网站| 国产高清激情床上av| 久久中文字幕一级| 一本大道久久a久久精品| 亚洲人成伊人成综合网2020| aaaaa片日本免费| 国产熟女午夜一区二区三区| 中文欧美无线码| 在线观看免费日韩欧美大片| av网站免费在线观看视频| 色94色欧美一区二区| 99国产综合亚洲精品| 亚洲国产精品一区二区三区在线| 18禁美女被吸乳视频| 国产伦人伦偷精品视频| 国产精品自产拍在线观看55亚洲 | 亚洲国产欧美日韩在线播放| 每晚都被弄得嗷嗷叫到高潮| 超碰97精品在线观看| 飞空精品影院首页| 中文字幕制服av| 少妇被粗大的猛进出69影院| 男女下面插进去视频免费观看| 亚洲精品乱久久久久久| 亚洲专区字幕在线| 丁香六月天网| 国产精品自产拍在线观看55亚洲 | 久久性视频一级片| 欧美精品av麻豆av| 国产精品久久久久久精品电影小说| 日韩免费av在线播放| 亚洲精品美女久久av网站| 一区在线观看完整版| 老司机午夜福利在线观看视频 | 12—13女人毛片做爰片一| 99精品欧美一区二区三区四区| 亚洲五月婷婷丁香| 丰满迷人的少妇在线观看| 天堂俺去俺来也www色官网| 久久天堂一区二区三区四区| 又大又爽又粗| 精品国产乱子伦一区二区三区| 久久久久久免费高清国产稀缺| 亚洲精品一二三| 国产精品偷伦视频观看了| 人妻久久中文字幕网| 国产在视频线精品| 亚洲少妇的诱惑av| 极品教师在线免费播放| 国产精品美女特级片免费视频播放器 | 欧美av亚洲av综合av国产av| 少妇裸体淫交视频免费看高清 | 男女下面插进去视频免费观看| 中文字幕色久视频| www日本在线高清视频| 亚洲欧美色中文字幕在线| 亚洲国产精品一区二区三区在线| 成人国语在线视频| 久久精品亚洲熟妇少妇任你| 国产在线观看jvid| 母亲3免费完整高清在线观看| 免费av中文字幕在线| 超碰97精品在线观看| 亚洲av日韩精品久久久久久密| 午夜精品久久久久久毛片777| 日日爽夜夜爽网站| 老汉色∧v一级毛片| 亚洲精品粉嫩美女一区| 中文字幕人妻熟女乱码| 人人妻人人添人人爽欧美一区卜| 老司机靠b影院| 久久精品国产亚洲av香蕉五月 | 色老头精品视频在线观看| 黄色成人免费大全| 蜜桃国产av成人99| 久久狼人影院| 亚洲精品粉嫩美女一区| 黄频高清免费视频| 国产精品偷伦视频观看了| 亚洲国产av影院在线观看| 国产成人啪精品午夜网站| 日本av手机在线免费观看| 亚洲欧美一区二区三区久久| 麻豆乱淫一区二区| 在线观看人妻少妇| 99国产综合亚洲精品| 国产精品免费大片| 热re99久久精品国产66热6| 欧美激情极品国产一区二区三区| 老司机午夜十八禁免费视频| 亚洲午夜理论影院| 国产一区有黄有色的免费视频| 啪啪无遮挡十八禁网站| 淫妇啪啪啪对白视频| 男女午夜视频在线观看| 亚洲av欧美aⅴ国产| 欧美精品av麻豆av| 久久久久久久国产电影| 狠狠狠狠99中文字幕| 中文字幕最新亚洲高清| 女人被躁到高潮嗷嗷叫费观| 精品一区二区三区av网在线观看 | 亚洲精品久久成人aⅴ小说| 高清黄色对白视频在线免费看| 黄色a级毛片大全视频| 久久人人97超碰香蕉20202| 国内毛片毛片毛片毛片毛片| 中文字幕色久视频| 欧美成人午夜精品| 久久国产精品大桥未久av| 久久久久国产一级毛片高清牌| 97人妻天天添夜夜摸| 精品少妇黑人巨大在线播放| 国产av又大| 男男h啪啪无遮挡| 美女午夜性视频免费| 一区二区三区精品91| 国产免费视频播放在线视频| 国产一区有黄有色的免费视频| 最近最新免费中文字幕在线| 最近最新中文字幕大全电影3 | 在线观看免费日韩欧美大片| 真人做人爱边吃奶动态| 日韩视频在线欧美| a在线观看视频网站| 国产av国产精品国产| 丝袜喷水一区| 亚洲人成电影观看| 男男h啪啪无遮挡| 欧美黑人精品巨大| 国产精品成人在线| 大片免费播放器 马上看| 99re在线观看精品视频| 超色免费av| 亚洲av成人一区二区三| 国产精品麻豆人妻色哟哟久久| 中国美女看黄片| 王馨瑶露胸无遮挡在线观看| 一个人免费看片子| 在线观看免费日韩欧美大片| 午夜福利视频精品| 午夜福利乱码中文字幕| 欧美乱妇无乱码| 午夜福利影视在线免费观看| 久久天堂一区二区三区四区| 中文字幕另类日韩欧美亚洲嫩草| 操出白浆在线播放| 亚洲欧美一区二区三区久久| 国产91精品成人一区二区三区 | 免费在线观看影片大全网站| 久久人妻熟女aⅴ| 多毛熟女@视频| 老汉色∧v一级毛片| 久久中文字幕人妻熟女| 久久精品国产99精品国产亚洲性色 | 中文欧美无线码| 91九色精品人成在线观看| 久热这里只有精品99| 国产精品秋霞免费鲁丝片| 欧美日韩精品网址| 国产午夜精品久久久久久| 中文字幕色久视频| 天堂俺去俺来也www色官网| 久久久国产精品麻豆| 亚洲国产中文字幕在线视频| 亚洲精品国产区一区二| 久9热在线精品视频| 婷婷丁香在线五月| 男男h啪啪无遮挡| 国产主播在线观看一区二区| 久久香蕉激情| 在线 av 中文字幕| 亚洲av电影在线进入| 丰满人妻熟妇乱又伦精品不卡| 女人高潮潮喷娇喘18禁视频| 欧美成狂野欧美在线观看| 亚洲va日本ⅴa欧美va伊人久久| 人人妻人人爽人人添夜夜欢视频| 一进一出抽搐动态| 黄色视频在线播放观看不卡| 成人亚洲精品一区在线观看| 国产精品免费大片| 亚洲伊人色综图| 国产精品偷伦视频观看了| 老司机亚洲免费影院| 欧美日韩中文字幕国产精品一区二区三区 | 天天躁狠狠躁夜夜躁狠狠躁| 国产色视频综合| 国产精品熟女久久久久浪| 日本黄色日本黄色录像| 欧美中文综合在线视频| 女人高潮潮喷娇喘18禁视频| 欧美精品一区二区免费开放| 欧美国产精品va在线观看不卡| 国产一区二区三区综合在线观看| 大码成人一级视频| 老司机靠b影院| 免费看十八禁软件| 亚洲午夜理论影院| 免费在线观看完整版高清| 欧美日韩亚洲国产一区二区在线观看 | 男人操女人黄网站| 757午夜福利合集在线观看| 99re6热这里在线精品视频| 可以免费在线观看a视频的电影网站| 国产高清视频在线播放一区| 国产精品久久久久久人妻精品电影 | 丰满饥渴人妻一区二区三| 久久人妻熟女aⅴ| 国产成人免费无遮挡视频| 18禁国产床啪视频网站| 精品久久蜜臀av无| 久久久久久久大尺度免费视频| 国产精品1区2区在线观看. | 国产精品98久久久久久宅男小说| 美女高潮到喷水免费观看| 最新美女视频免费是黄的| 99re在线观看精品视频| 亚洲国产av新网站| 蜜桃国产av成人99| 嫁个100分男人电影在线观看| 91精品国产国语对白视频| 蜜桃在线观看..| 婷婷丁香在线五月| 嫩草影视91久久| 国产av又大| av免费在线观看网站| 精品国产一区二区三区久久久樱花| 久久99热这里只频精品6学生| 少妇被粗大的猛进出69影院| 国产成人欧美在线观看 | 午夜视频精品福利| 啪啪无遮挡十八禁网站| av电影中文网址| 亚洲精品粉嫩美女一区| 亚洲精品国产色婷婷电影| 97在线人人人人妻| 国产熟女午夜一区二区三区| 国产精品一区二区精品视频观看| 国产亚洲精品久久久久5区| 久久国产亚洲av麻豆专区| 亚洲成av片中文字幕在线观看| 在线播放国产精品三级| 国产精品影院久久| 69精品国产乱码久久久| 亚洲精品一卡2卡三卡4卡5卡| 精品国产乱码久久久久久男人| 日日夜夜操网爽| 久久av网站| 欧美另类亚洲清纯唯美| 午夜福利视频在线观看免费| 人人妻人人澡人人看| 超碰97精品在线观看| 老鸭窝网址在线观看| 一本大道久久a久久精品| 色94色欧美一区二区| 天堂8中文在线网| 欧美午夜高清在线| 午夜久久久在线观看| 免费少妇av软件| 亚洲伊人久久精品综合| 老司机靠b影院| 久久性视频一级片| 国产视频一区二区在线看| 男女免费视频国产| 香蕉久久夜色| 午夜福利在线免费观看网站| 亚洲精品av麻豆狂野| 欧美黑人欧美精品刺激| av一本久久久久| avwww免费| 欧美另类亚洲清纯唯美| 久久国产精品男人的天堂亚洲| 18禁裸乳无遮挡动漫免费视频| 自线自在国产av| 一区在线观看完整版| 日韩中文字幕欧美一区二区| 久久久久久久久久久久大奶| 曰老女人黄片| 纵有疾风起免费观看全集完整版| 国产99久久九九免费精品| 亚洲精品美女久久久久99蜜臀| 欧美久久黑人一区二区| kizo精华| 丰满迷人的少妇在线观看| 蜜桃在线观看..| 欧美精品av麻豆av| 久久精品国产亚洲av香蕉五月 | 自拍欧美九色日韩亚洲蝌蚪91| 波多野结衣av一区二区av| 老司机午夜福利在线观看视频 | 极品教师在线免费播放| 久久 成人 亚洲| 啦啦啦 在线观看视频| 久久99热这里只频精品6学生| 亚洲成av片中文字幕在线观看| 成人18禁高潮啪啪吃奶动态图| 久久精品国产亚洲av香蕉五月 | 黑人欧美特级aaaaaa片| 午夜福利乱码中文字幕| √禁漫天堂资源中文www| 两个人免费观看高清视频| 婷婷丁香在线五月| 无遮挡黄片免费观看| 免费不卡黄色视频| 涩涩av久久男人的天堂| 日韩一卡2卡3卡4卡2021年| 大型黄色视频在线免费观看| 精品熟女少妇八av免费久了| 午夜视频精品福利| 久久精品亚洲av国产电影网| 国产无遮挡羞羞视频在线观看| 免费人妻精品一区二区三区视频| 中文亚洲av片在线观看爽 | 天堂中文最新版在线下载| 亚洲va日本ⅴa欧美va伊人久久| 免费高清在线观看日韩| 欧美乱码精品一区二区三区| 香蕉丝袜av| 一本久久精品| 性色av乱码一区二区三区2| 中文字幕精品免费在线观看视频| 一夜夜www| 欧美激情久久久久久爽电影 | 别揉我奶头~嗯~啊~动态视频| 美女视频免费永久观看网站| 一区在线观看完整版| 大型黄色视频在线免费观看| 一区二区av电影网| 天天躁夜夜躁狠狠躁躁| 三级毛片av免费| 十八禁高潮呻吟视频| 首页视频小说图片口味搜索| tube8黄色片| 亚洲avbb在线观看| 久久久欧美国产精品| 欧美大码av| 91精品三级在线观看| 亚洲 欧美一区二区三区| 国产一区二区激情短视频| 啦啦啦视频在线资源免费观看| 91成年电影在线观看| 在线观看舔阴道视频| 少妇裸体淫交视频免费看高清 | 成年人午夜在线观看视频| 亚洲欧美一区二区三区黑人| 欧美日韩精品网址| 18在线观看网站| 欧美亚洲日本最大视频资源|