• <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 ?

    A Novel Systematic Method of Quality Monitoring and Prediction Based on FDA and Kernel Regression*

    2009-05-14 03:04:44ZHANGXi張曦MASile馬思樂YANWeiwu閻威武ZHAOXu趙旭andSHAOHuihe邵惠鶴
    關(guān)鍵詞:威武

    ZHANG Xi (張曦), MA Sile (馬思樂), YAN Weiwu (閻威武), ZHAO Xu (趙旭) and SHAO Huihe (邵惠鶴)

    ?

    A Novel Systematic Method of Quality Monitoring and Prediction Based on FDA and Kernel Regression*

    ZHANG Xi (張曦)1,2, MA Sile (馬思樂)3,**, YAN Weiwu (閻威武)2, ZHAO Xu (趙旭)2and SHAO Huihe (邵惠鶴)2

    1Guangdong Electric Power Research Institute, Guangzhou 510600, China2Department of Automation, Shanghai Jiao Tong University, Shanghai 200240, China3School of Control Science and Engineering, Shandong University, Jinan 250061, China

    A novel systematic quality monitoring and prediction method based on Fisher discriminant analysis (FDA) and kernel regression is proposed. The FDA method is first used for quality monitoring. If the process is under normal condition, then kernel regression is further used for quality prediction and estimation. If faults have occurred, the contribution plot in the fault feature direction is used for fault diagnosis. The proposed method can effectively detect the fault and has better ability to predict the response variables than principle component regression (PCR) and partial least squares (PLS). Application results to the industrial fluid catalytic cracking unit (FCCU) show the effectiveness of the proposed method.

    quality monitoring, quality prediction, Fisher discriminant analysis, kernel regression, fluid catalytic cracking unit

    1 Introduction

    It is difficult to measure online some important variables in chemical or biological process due to the limitation of the techniques. These variables are normally determined by offline analyses and calculation or online quality analyzer. Mostly, analysis and online analyzer suffer from long measurement delays or high investment and maintenance. Online estimation (soft sensing) techniques, which are developed recently, are now widely used. It predicts the product quality using the online measurable variables, which are correlated to the primary variable.

    Early work on online estimation assumed that a process model was available. Joseph and Brosilow [1]reported an inferential model developed using a Kalmanfilter [2]. In case the process mechanisms were not well understood, empirical models, such as neural network [3, 4], and multivariate statistical methods, such as principal component analysis (PCA) and partial least squares (PLS), were used to derive a regression model [2, 5-7]. In particular, PLS and its variations have been used to solve many practical regression problems in chemical engineering [6, 8]. Other methods include regression based on model [9] and hybrid methods also have been developed [10, 11]. Clustering has also been used for estimating a variable for which there is no online measurement in a distillation column [12].

    However, online process measurements are often contaminated with data points that deviate significantly from the true values due to instrument failure or changes in operating conditions. So, gross error detection is important in quality estimation. Several statistical methods have been presented, such as Global Test (GT) [13], the Nodal Test (NT) [13, 14], the Measurement Test (MT) [15], and the Generalized Likelihood Ratio (GLR) test to detect gross errors. The above-mentioned methods for errors detection must know the mathematical model of process and make necessary statistical assumption. For complex industrial processes, it is difficult to achieve an accurate mathematical model. These limit application of methods based on statistical test. Fisher discriminant analysis (FDA)-related methods are dimensional reduction techniques that have been widely used in the field of pattern classification [16]. They also have been introduced for fault detection and process monitoring in chemical industry [17-22]. Gross error is equal to faultsin systems. So, FDA is introduced for gross error detection in quality estimation and has a good performance.

    In this article, a systematic method of quality monitoring and prediction based on FDA and kernel regression has been proposed. FDA is first used for fault detection. If the process runs under normal condition, kernel regression method is then performed for quality prediction. Otherwise, the contribution plot of weights in fault feature direction is further used for fault diagnosis. Application results of the industrial fluid catalytic cracking unit (FCCU) show that the integrated method can effectively detect the happening faults and perform quality prediction and estimation.

    2 Preliminary principles of FDA and kernel regression

    2.1 Fisher discriminant analysis

    Fisher discriminant analysis (FDA) is a linear dimensional reduction technique widely used in the field of pattern classification [16]. Chiang. [20] used it first to diagnose faults in chemical process. The aim of FDA is to find the optimal Fisher discriminant vector such that the Fisher criterion function is maximized. The data in high-dimensional feature space then can be projected onto the obtained optimal discriminant vectors for constructing a lower-dimensional feature space [21]. The different class data can be separated mostly in the lower-dimensional Fisher space.

    The between-class-scatter matrix is defined as

    wherenis the number of observations in class. It can be concluded that the total-scatter matrix is equal to the sum of the between-class-scatter matrix and within-class-scatter matrix,

    The optimal discriminant vector can be found by maximizing the Fisher criterion function as follows:

    wherebandware the between-class scatter matrix and within-class scatter matrix. It is equipollent to solve the generalized feature equation as follows:

    2.2 Kernel principle component analysis

    Kernel principle component analysis (KPCA) is a nonlinear extension of PCA in kernel feature subspace. Consider a nonlinear mapping [23]:

    then the PC vectors can be denoted as follows.

    where

    2.3 Kernel principle component regression

    Consider the standard regression model in the feature space,

    then,

    Whereis an diagonal matrix of {1,2,···,λ},is an orthogonal matrix. So the linear regression model can be expressed further as,

    where

    is the new regressor in terms of kernel principal component and it satisfies that

    So the least square estimator of coefficientcan be expressed as

    From the above estimator expression, we can see that it is difficult to finddirectly. Below we show how to findusing the kernel function.

    Denotethe-th component ofandthe-th component of; from the linear regression model, we obtain

    So,

    where

    From above, we know that

    where

    From the definition of matrix,

    We can see that

    So

    3 Quality monitoring and prediction based on FDA and kernel regression

    3.1 The relationship among FDA, KPCA, and KPCR in quality monitoring and estimation

    Kernel principle component regression (KPCR) is the extension of KPCA. KPCA is used for feature selection. Then the extracted features can be used as preprocessing step for least square regression in the feature space. The framework of quality monitoring and estimation is shown in Fig. 1.

    It can be seen from the figure that data are first normalized using the mean and standard deviation of each variable. Then FDA is performed for gross error and fault detection. If the system runs under normal condition, KPCR is further used for quality estimation and prediction. Otherwise, contribution plot of weights in fault feature direction is used for fault diagnosis.

    FDA-related methods have been used for pattern matching and process monitoring. But it is introduced for the first time for gross error detection in quality estimation.

    3.2 The construction of detecting index and control limit

    Figure 1 Flow chart of quality monitoring and estimation

    For online quality monitoring, the distance of projection vectors in low-dimensional Fisher space,newcompares with the control limit*to determine whether a fault has occurred. The distance between feature vectors is defined as:

    A Quality monitoring based on FDA

    (3) Determine the control limit*from the normal data. The control limit is set to allow 99% of the total distances under the threshold.

    (6) Compare the distancenewwith the control limit*. Ifnewis less than the predefined threshold*, the current process feature is considered to represent the normal operating condition. Then use procedures B and C to perform quality prediction and estimation; if the fault has occurred, use step (7) for fault diagnosis.

    B Develop normal quality prediction and estimation model

    (1) Acquire the training data and normalize it using mean and standard deviation of each variable.

    where

    (4) Calculate the transformed regressor matrixas follows:

    (6) Calculate the predictions of training data as follows:

    C Online estimation and prediction procedure

    (1) Obtain new data and scale it with the mean and variance of each variable.

    (3) Mean centering of the test kernel matrix is as follows

    (4) Calculate the transformed regressor matrixas follows:

    4 Application studies and discussion

    4.1 Application results of the FCCU process

    Fluid catalytic cracking unit (FCCU) is the core unit of the oil secondary operation. Its operation conditions strongly affect the yield of light oil in petroleum refining. In general, FCCU consists of reactor- regenerator subsystem, fractionator subsystem, absorber-stabilizer subsystem and gas sweetening subsystem. The main aim of fractionator subsystem is to split reaction-cracked oil-gas according to a fractional distillation process. Prime products of fractionator subsystem include crude gasoline, light diesel oil, and slurry [7]. To control the product quality, the yield rate of gasoline is calculated offline every 8 hours. Significant delay (often several hours) will incur such that the measured values cannot be used as feedback signals for quality control systems. So it is necessary to estimate it online. However, when using online estimation, the results are often contaminated with data points that deviate significantly from the true values due to instrument failure or changes in operating conditions. So, it is essential to perform quality monitoring and avoiding wrong operations.

    The proposed method is applied to the quality monitoring and prediction of the yield rate of gasoline. The training and testing samples are collected from data of the distributed control system (DCS) and the corresponding daily laboratory analysis of Shijiazhuang Oil Refinery Factory, China. The input variables for quality monitoring and prediction are selected according to the principle that the variables which affected response variable most are selected first. Based on analysis to the process, it is found that the main variables which contribute to the yield rate of gasoline are flow rate of fresh oil, flow rate of reflux oil, temperature of catalytic reaction, overhead temperature of the main fractionating tower, the extraction temperature of light diesel oil, and bottom temperature of stabilizer column. Hence, these six variables are used as inputs for quality monitoring and prediction. Yield rate is used as the output of estimation model.

    We first collect two data sets and each includes 100 normal samples. They are used as training and testing data. The monitoring result using FDA is given in Fig. 2. We can see that no samples exceed the control limit when the system is normal. The monitoring results using PCA is shown in Fig. 3, it can be seen from the figure that there are a few samples exceed the 99% limit although the process is normal. The monitoring performance of FDA is better than PCA.

    Figure 2 Quality monitoring plot of normal data using FDA

    Figure 3 Quality monitoring plot of normal data using PCA

    Figure 4 Estimation results by KPCR (Normal data)

    Figure 6 Quality monitoring plot of fault data using FDA

    When the process is under normal condition, quality estimation can be further performed and appropriate kernel must be first chosen. The radial basis function (RBF) kernel is used as the selected kernel function andis 5.0 in this case.

    The prediction results using kernel regression are shown in Figs. 4 and 5. The upper part of Fig. 4 shows the actual and predicted values of training and testing data and the lower part shows the absolute error between actual and predicted values. Fig. 5 shows the predicted results by another method. In such plots, the data will fall on the diagonal (predicted values equal to actual values) if the model fits the data perfectly. It also shows the residual values of training and testing data. We can see from Fig. 4 that the KPCR model predicts the actual value with a relatively good accuracy. The absolute error is small. The same results can be seen in Fig. 5, in which the predicted values are plotted against observed data and the data shifts to a compact diagonal distribution on the plot. The prediction residuals are close to zero and have no significant outliers.

    The estimation performance can also be evaluated in terms of the root-mean-square-error (RMSE) criterion. The RMSE values for training and testing data of KPCR method in this case are 0.5112 and 0.5843, respectively. It also shows relatively good estimation results.

    Remark The RMSE index is defined as [28]:

    Figure 7 Contribution plot of weights in the feature direction

    Figure 8 Quality monitoring plot of fault data using PCA

    Figure 9 Estimation results by KPCR (Fault data)

    We then select a test data set of 100 samples including a fault of 10% decrease in the flow rate of fresh oil. The fault is introduced in sample 50 and persists to the end of the process. The PCA monitoring chart for the fault is shown in Fig. 8. It indicates that the square predicted error (SPE) statistic has a response to the fault happening; only a few samples delay. But it is not very clear. The corresponding FDA monitoring chart is shown in Fig. 6. In contrast to Fig. 8, we can see that the statistic distance increases drastically when the fault occurs at sample 50 and exceeds the 99% control limit. It also has better performance than PCA.

    After a fault has occurred, fault diagnosis is performed to identify the root cause. Fig. 7 is the contribution plot in the fault feature direction. It displays that variable 1 is primarily responsible for the fault deviation. Variable 1 corresponds to the flow rate of fresh oil. So the diagnosis result is right. The early detection and determination of the fault’s root cause will guide the operator to take correct action promptly, which could maintain the final product well.

    The prediction results of yield rate based on kernel regression are shown in Figs. 9 and 10. From Fig. 9, we can see that before the fault occurs, KPCR has good prediction results. But after that, the prediction results become worse. The absolute error is large and the prediction values far derivate from the actual values. The same results can be seen in Fig. 10. Before the fault is introduced, the data distribution is compactly towards the diagonal and the residual is low. But the test data distribution is scattered and the residuals are large after the fault happens. The total RMSE for the test data is 1.9824. If the faulty values are used as feedback signals in the control system, it will lead to wrong operations. So, quality monitoring is very important and necessary to guarantee the prediction results well.

    4.2 Further discussion

    The performance of the estimation method based on KPCR is influenced by many factors. We will use thesamples used in Section 4.1 to discuss them extensively.

    I Performance comparison of different methods

    To demonstrate the predominance of the proposed method, PCR and PLS are further applied, followed by a comparison with the KPCR prediction method. The estimation results using PCR and PLS are shown in Figs. 11-14.

    Figure 11 Estimation results by PCR

    Figure 13 Estimation results by PLS

    Figure 15 Estimation results with linear kernel

    From the figures, we can see that, compared with PCR and PLS, KPCR (shown in Figs. 4 and 5) has better estimation performance.

    Because KPCR is nonlinear modeling method, it has better prediction results. So it has a better ability to deal with nonlinear data. In this study, there are complex nonlinear relationships among input and response variables. So, projecting nonlinear input data onto a linear subspace by PCR or PLS cannot model the nonlinear relationships properly. In contrast, KPCR tries to model such nonlinear relationships preferably by a nonlinear kernel mapping into a feature space.

    II Performance influence of different kernels in KPCR

    When we estimate quality variables by KPCR, the choice of kernel is important. Different kernels have different influence on the performance of prediction. We use the linear kernel, polynomial kernel, and radial basis kernel for comparison, which are listed below.

    The KPCR model estimation results using linear and polynomial kernel (the parameteris chosen as 5.0)are shown in Figs. 15-18. It can be seen from the figures that using linear and polynomial kernels cannot obtain good estimation results compared to using radial basis kernel (shown in Figs. 4 and 5).

    The reason for this is that the input variables of the data set used are nonlinear correlated with each other and with the response variable. When linear and polynomial kernels are used, they cannot model the nonlinear correlation structure properly due to the risk of including noise in the model while trying to account nonlinearity. From our experience, radial basis kernel is more suitable for modeling such nonlinear relationships compared to other kernel functions by selecting appropriate parameter.

    III Performance influence of parameterin the RBF kernel

    Figures 19 and 20 present the estimation results of response variables for training and testing data with parameterequals to 0.01, 0.05, 0.5, 5, and 20. It can be seen from the figures that for training and testing data, whenis larger or smaller over a specific threshold, the estimation performance becomes worse. Therefore, to get good prediction results, the appropriate choice of parameteris important. In this case, whenis equal to 5.0, good estimation performance can be achieved. The same results can be obtained from the RMSE values of different c (shown in Table 1). Whenis equal to 5.0, the RMSE values are 0.5112 and 0.5843. They are the smallest among all the training and testing data.

    Figure 16 Estimation parity plot with linear kernel

    Figure 17 Estimation results with polynomial kernel

    Table 1 RMSE of different parameter c in kernel principal component regression (RBF kernel)

    5 Conclusions

    In this article, a systemic quality monitoring and prediction method based on FDA and kernel regression is proposed. The FDA monitoring method is first performed to detect process condition. If the process is normal, kernel regression is further used to do quality prediction and estimation. If fault has occurred, contribution plot of the weights in the feature direction is used for fault diagnosis. The improvement in prediction performance observed, from PCR, PLS to KPCR, suggests that nonlinear correlation structures should not be modeled using linear approaches due to the risk of including noise in the model while trying to account the nonlinearity. The application results of the industrial fluid catalytic cracking unit (FCCU) show the effectiveness of the proposed method.

    However, when the proposed method is used, how to choose kernel function and identify the kernel parameter is still an open problem. At present, we only solve this with our experience. How to settle this using a systematic approach is still a challenging issue. We believe that for the problem to be further solved, the proposed method will give more promising results.

    AcknowledgEments

    ..

    1 Joseph, B., Brosilow, C.B., “Inferential control of processes (1) Steady state analysis and design”,., 124, 485-508 (1978).

    2 Lin, B., Recke, B., Knudsen, J.K.H., J?rgensen, S.B, “A systematic approach for soft sensor development”,..., 31, 419-425 (2007).

    3 Qin, S.J., McAvoy, T.J., “Nonlinear PLS modeling using neural networks”,..., 16, 379-391 (1992).

    4 Radhakrishnan, V.R., Mohamed, A.R., “Neural networks for the identification and control of blast furnace hot metal quality”,., 10, 509-524 (2000).

    5 Kresta, J.V., Marlin, T.E., MacGregor, J.F., “Development of inferential process models using PLS”,..., 18, 597-611 (1994).

    6 Park, S., Han, C., “A nonlinear soft sensor based on multivariate smoothing procedure for quality estimation in distillation columns”,..., 24, 871-877 (2000).

    7 Yan, W., Shao, H., Wang, X., “Soft sensing modeling based on support vector machine and Bayesian model selection”,..., 28, 1489-1498 (2004).

    8 Skagerberg, B., MacGrgor, J.F., Kiprissides, C., “Multivariate data analysis applied to low-density polyethylene reactors”,...., 14, 341-356 (1992).

    9 Chen, S., Billings, S.A., Cowan, C.T.F., Grant, P.M., “Practical identification of NARMAX models using radial basis functions”,.., 52, 1327-1350 (1990).

    10 Ljung, L., System Identification: Theory for the User (Information and System Science Series), Prentice-Hall, New Jersey (1987).

    11 Wang, X., Luo, R., Shao, H., “Designing a soft sensor for a distillation column with the fuzzy distributed radial basis function neural network”, In: Proceedings of the 35th IEEE Conference on Decision and Control, Kobe, Japan (1996).

    12 Espinoza, P. A., Gonzalez, G. D., Casali, A., Ardiles, C., “Design of soft sensors using cluster techniques”, In: Proceedings of International Mineral Processing Congress, San Francisco, USA (1995).

    13 Reilly, P., Carpani, R., “Application of statistical theory of adjustments to material Balances”, In: 13th Canadian Chemical Engineering Conference, Montreal, Canada (1963).

    14 Mah, R.S.H., Stanley, G., Downing, D., “Reconciliation and rectification of process flow and inventory data”,....., 15, 175-183 (1976).

    15 Mah, R.S.H., Tamhane, A.C., “Detection of gross errors in process data”,., 28, 828-830 (1982).

    16 Duda, R.O., Hart, P.E., Stork, D.G., “Pattern classification”, 2nd ed., Wiley, New York (2001).

    17 Cho, H.W., “Identification of contributing variables using kernel-based discriminant modeling and reconstruction”,.., 33, 274-285 (2007).

    18 Jemwa, G.T., Aldrich, C., “Kernel-based fault diagnosis on mineral processing plants”,.., 19, 1149-1162 (2006).

    19 Zhang, X., Zhao, X., Yan, W.W., Shao, H.H., “Nonlinear biological batch process monitoring and fault identification based on kernel fisher discriminant analysis”,., 42, 1200-1210 (2007).

    20 Chiang, L.H., Russell, E.L., Braatz, R.D., “Fault diagnosis in chemical processes using Fisher discriminant analysis, discriminant partial least squares, and principal component analysis”,...., 50, 243-252 (2000).

    21 He, Q.P., Qin, S.J., “A new fault diagnosis method using fault directions in fisher discriminant analysis”,., 51, 555-571 (2005).

    22 Zhao, X., Yan, W., Shao, H., “Monitoring and fault diagnosis for batch process based on feature extract in Fisher subspace”,...., 14, 759-764 (2006).

    23 Sch?lkopf, B., Smola, A., Müller, K.R., “Nonlinear component analysis as a kernel eigenvalue problem”,., 10, 1299-1319 (1998).

    24 Lee, J.M., Yoo, C., Choi, S.W., Vanrolleghem, P.A., Lee, I.B., “Nonlinear process monitoring using kernel principal component analysis”,..., 59, 223-234 (2004).

    25 Rosipal, R., Girolami, M., Trejo, L. J., Cichocki, A., “Kernel PCA for feature extraction and de-noising in nonlinear regression”,.., 10, 231-243 (2001).

    26 Dachapak, C., Kanae, S., Yang, Z.J., Wada, K., “Kernel principal component regression in reproducing Hilbert space”,(.....), 34, 213-218 (2002).

    27 Chiang, L.H., Russell, E.L., Braatz, R.D., Fault Detection and Diagnosis in Industrial Systems, Springer, Hong Kong (2001).

    28 Lee, D.S., Lee, M.W., Woo, S.H., Kim, Y.J., Park, J.M., “Multivariate online monitoring of a full-scale biological anaerobic filter process using kernel-based algorithms”,...., 45, 4335-4344 (2006).

    29 Kim, K., Lee, J. M., Lee, I. B., “A novel multivariate regression approach based on kernel partial least squares with orthogonal signal correction”,...., 79, 22-30 (2005).

    30 Wold, S., “Cross-validatory estimation of components in factor and principal components models”,, 20, 397-405 (1978).

    2008-01-27,

    2008-11-05.

    the National Natural Science Foundation of China (60504033) and the Open Project of State Key Laboratory of Industrial Control Technology in Zhejiang University (0708004).

    ** To whom correspondence should be addressed. E-mail: masile@sdu.edu.cn

    猜你喜歡
    威武
    高傲的大公雞
    威武遼寧艦(一)
    山東艦,威武入列
    威武老槍56半
    威武的獅子
    幼兒畫刊(2018年4期)2018-04-11 03:58:56
    萌寵樂園
    威武大角羊
    威武蝦將軍
    小蝦威武
    威武的甲蟲
    久久久精品区二区三区| 美女主播在线视频| 久久久久久久久免费视频了| 狂野欧美激情性bbbbbb| 亚洲av在线观看美女高潮| 亚洲精品一二三| 亚洲七黄色美女视频| 桃花免费在线播放| 99久久综合免费| 黄色 视频免费看| 亚洲欧美成人精品一区二区| 中文天堂在线官网| 国产成人免费观看mmmm| 9191精品国产免费久久| 亚洲一区中文字幕在线| 中文字幕制服av| 精品视频人人做人人爽| 女人久久www免费人成看片| 男女床上黄色一级片免费看| 纵有疾风起免费观看全集完整版| 亚洲精品视频女| 一级毛片 在线播放| 丝袜美足系列| 97在线人人人人妻| 大陆偷拍与自拍| 欧美日韩一级在线毛片| 黄色毛片三级朝国网站| 亚洲av成人不卡在线观看播放网 | 国产精品麻豆人妻色哟哟久久| 看免费成人av毛片| 人人妻人人澡人人看| 久久精品久久精品一区二区三区| 国产亚洲一区二区精品| 成年av动漫网址| 欧美精品高潮呻吟av久久| 欧美人与善性xxx| 汤姆久久久久久久影院中文字幕| 人人澡人人妻人| 久久人人爽人人片av| 美女中出高潮动态图| 天堂中文最新版在线下载| 国产成人91sexporn| 中文字幕高清在线视频| 国产深夜福利视频在线观看| 欧美日韩亚洲综合一区二区三区_| 人人妻人人澡人人看| 啦啦啦视频在线资源免费观看| 免费看av在线观看网站| 亚洲第一青青草原| 中文欧美无线码| 欧美另类一区| 一级黄片播放器| 青春草视频在线免费观看| 精品免费久久久久久久清纯 | 热re99久久国产66热| 美女视频免费永久观看网站| 久久鲁丝午夜福利片| 国产一区二区三区av在线| 青春草亚洲视频在线观看| 亚洲欧美中文字幕日韩二区| 国产精品麻豆人妻色哟哟久久| 欧美日韩一区二区视频在线观看视频在线| 成人三级做爰电影| 一二三四在线观看免费中文在| avwww免费| 人成视频在线观看免费观看| 国产视频首页在线观看| 亚洲av综合色区一区| 午夜福利,免费看| 国产精品偷伦视频观看了| 搡老乐熟女国产| 麻豆乱淫一区二区| 亚洲国产日韩一区二区| 久久久久精品久久久久真实原创| 超碰成人久久| 久久天堂一区二区三区四区| 哪个播放器可以免费观看大片| 两个人看的免费小视频| 国产一级毛片在线| 啦啦啦 在线观看视频| 婷婷色综合大香蕉| 咕卡用的链子| 乱人伦中国视频| 夫妻性生交免费视频一级片| 成年女人毛片免费观看观看9 | 免费日韩欧美在线观看| 欧美97在线视频| 国产精品av久久久久免费| 精品国产一区二区久久| 午夜91福利影院| 91成人精品电影| 欧美日韩综合久久久久久| 精品一区二区免费观看| 丰满少妇做爰视频| 精品国产一区二区三区四区第35| 又大又爽又粗| 日日爽夜夜爽网站| 乱人伦中国视频| 在线天堂最新版资源| 国产精品人妻久久久影院| 国产片特级美女逼逼视频| 一区福利在线观看| 如日韩欧美国产精品一区二区三区| 综合色丁香网| 极品少妇高潮喷水抽搐| 亚洲国产欧美一区二区综合| 欧美日韩视频高清一区二区三区二| 国产成人欧美| 精品国产乱码久久久久久男人| xxx大片免费视频| av女优亚洲男人天堂| av视频免费观看在线观看| 最新的欧美精品一区二区| 亚洲情色 制服丝袜| 90打野战视频偷拍视频| 丝袜美腿诱惑在线| 日韩av免费高清视频| 欧美精品亚洲一区二区| 宅男免费午夜| 日韩制服骚丝袜av| 国产黄色视频一区二区在线观看| 九草在线视频观看| 色综合欧美亚洲国产小说| 国产精品熟女久久久久浪| 精品人妻在线不人妻| 亚洲av国产av综合av卡| 操美女的视频在线观看| 高清欧美精品videossex| 久久婷婷青草| 久久久久网色| 久久天堂一区二区三区四区| 99热网站在线观看| 操出白浆在线播放| 国产淫语在线视频| 国产日韩一区二区三区精品不卡| 久久97久久精品| 两个人免费观看高清视频| 亚洲 欧美一区二区三区| 男女下面插进去视频免费观看| 亚洲熟女毛片儿| 热re99久久国产66热| 啦啦啦视频在线资源免费观看| 亚洲精品国产色婷婷电影| 女人精品久久久久毛片| 亚洲国产av新网站| 中国三级夫妇交换| 欧美日韩福利视频一区二区| 亚洲欧洲国产日韩| 可以免费在线观看a视频的电影网站 | 人人妻人人澡人人看| 国产黄色免费在线视频| 叶爱在线成人免费视频播放| 国产欧美日韩一区二区三区在线| 久久影院123| 日日撸夜夜添| av有码第一页| av线在线观看网站| 看免费av毛片| 国产男人的电影天堂91| 99香蕉大伊视频| 交换朋友夫妻互换小说| 国产免费又黄又爽又色| 激情视频va一区二区三区| 综合色丁香网| av又黄又爽大尺度在线免费看| 国产成人精品无人区| 国产成人精品在线电影| 亚洲国产精品999| 少妇精品久久久久久久| 热re99久久精品国产66热6| 久久精品亚洲熟妇少妇任你| 两个人看的免费小视频| 亚洲国产av影院在线观看| av网站免费在线观看视频| 黄片小视频在线播放| www.自偷自拍.com| xxxhd国产人妻xxx| 亚洲欧洲国产日韩| 精品酒店卫生间| 中文字幕制服av| 国产精品熟女久久久久浪| 伊人亚洲综合成人网| 日韩 欧美 亚洲 中文字幕| 欧美另类一区| 中国三级夫妇交换| 尾随美女入室| 久久久久网色| 日韩不卡一区二区三区视频在线| 亚洲在久久综合| 在线免费观看不下载黄p国产| a级毛片黄视频| 午夜福利一区二区在线看| 亚洲精品日韩在线中文字幕| 亚洲,一卡二卡三卡| 成年av动漫网址| videosex国产| 国产男女内射视频| 纯流量卡能插随身wifi吗| 黑丝袜美女国产一区| 国产成人欧美在线观看 | 久热爱精品视频在线9| 亚洲人成网站在线观看播放| 亚洲精品,欧美精品| 国精品久久久久久国模美| 亚洲伊人色综图| 亚洲第一区二区三区不卡| 男女边吃奶边做爰视频| 国产爽快片一区二区三区| 美女脱内裤让男人舔精品视频| 日韩中文字幕视频在线看片| 丰满饥渴人妻一区二区三| www.av在线官网国产| 少妇人妻久久综合中文| 国产精品久久久久久精品电影小说| 久久毛片免费看一区二区三区| 视频在线观看一区二区三区| 毛片一级片免费看久久久久| 午夜福利乱码中文字幕| 日韩不卡一区二区三区视频在线| 丝袜脚勾引网站| a级片在线免费高清观看视频| 日本午夜av视频| 91精品国产国语对白视频| 精品亚洲成a人片在线观看| 永久免费av网站大全| 精品第一国产精品| tube8黄色片| 这个男人来自地球电影免费观看 | 国产精品.久久久| 亚洲国产毛片av蜜桃av| 91精品国产国语对白视频| 悠悠久久av| 午夜激情久久久久久久| 美女大奶头黄色视频| 新久久久久国产一级毛片| av又黄又爽大尺度在线免费看| 午夜福利一区二区在线看| 激情视频va一区二区三区| 欧美日韩av久久| 国产人伦9x9x在线观看| 国产一区亚洲一区在线观看| 欧美日韩福利视频一区二区| 亚洲精品久久久久久婷婷小说| 啦啦啦 在线观看视频| 日韩制服骚丝袜av| 少妇人妻精品综合一区二区| 久久女婷五月综合色啪小说| 精品一区在线观看国产| 男女午夜视频在线观看| 免费人妻精品一区二区三区视频| 最近手机中文字幕大全| 国产精品二区激情视频| 亚洲国产欧美在线一区| 亚洲专区中文字幕在线 | 好男人视频免费观看在线| 国产成人系列免费观看| 91aial.com中文字幕在线观看| 91精品伊人久久大香线蕉| h视频一区二区三区| 波多野结衣一区麻豆| 又黄又粗又硬又大视频| 十八禁网站网址无遮挡| 国产男人的电影天堂91| 中文字幕制服av| 一级片免费观看大全| 啦啦啦视频在线资源免费观看| 日本欧美视频一区| 操出白浆在线播放| 国产日韩欧美视频二区| 欧美日韩亚洲综合一区二区三区_| 一级片'在线观看视频| 国产无遮挡羞羞视频在线观看| 叶爱在线成人免费视频播放| 日本av手机在线免费观看| 91成人精品电影| 汤姆久久久久久久影院中文字幕| 老司机深夜福利视频在线观看 | 国产精品麻豆人妻色哟哟久久| 亚洲精品日韩在线中文字幕| 亚洲综合精品二区| 啦啦啦 在线观看视频| 一级片'在线观看视频| 18禁动态无遮挡网站| 精品免费久久久久久久清纯 | 超碰97精品在线观看| 你懂的网址亚洲精品在线观看| 制服人妻中文乱码| tube8黄色片| 国产精品偷伦视频观看了| 天天躁日日躁夜夜躁夜夜| 秋霞伦理黄片| av片东京热男人的天堂| 建设人人有责人人尽责人人享有的| 90打野战视频偷拍视频| www.精华液| 国产欧美日韩一区二区三区在线| 亚洲一区二区三区欧美精品| 老司机深夜福利视频在线观看 | 亚洲精品久久成人aⅴ小说| 美女主播在线视频| 久热爱精品视频在线9| 午夜福利在线免费观看网站| 欧美日韩视频高清一区二区三区二| 高清在线视频一区二区三区| 国产 一区精品| 国产精品久久久久成人av| 大香蕉久久成人网| av电影中文网址| 亚洲,欧美,日韩| 精品国产一区二区久久| 午夜激情av网站| 国产一区二区三区综合在线观看| 国产97色在线日韩免费| 亚洲精品日本国产第一区| 久久精品国产亚洲av涩爱| 性少妇av在线| 激情视频va一区二区三区| 免费观看av网站的网址| 卡戴珊不雅视频在线播放| 美女中出高潮动态图| 午夜av观看不卡| 超碰成人久久| 国产成人精品福利久久| 国产亚洲最大av| 精品人妻一区二区三区麻豆| 日本av免费视频播放| 国产精品久久久久成人av| 国产在线视频一区二区| 欧美精品av麻豆av| 成年动漫av网址| 国产成人91sexporn| 亚洲国产欧美日韩在线播放| 久久天躁狠狠躁夜夜2o2o | 波多野结衣av一区二区av| 如日韩欧美国产精品一区二区三区| 9色porny在线观看| 国产xxxxx性猛交| 少妇人妻 视频| 国产一级毛片在线| 亚洲人成电影观看| 亚洲欧美一区二区三区国产| 永久免费av网站大全| 夜夜骑夜夜射夜夜干| 国产在线免费精品| 黄色怎么调成土黄色| 国产午夜精品一二区理论片| 欧美日韩亚洲国产一区二区在线观看 | 日本欧美国产在线视频| 波野结衣二区三区在线| 9191精品国产免费久久| 亚洲成色77777| 最近中文字幕高清免费大全6| a级片在线免费高清观看视频| 国产精品 欧美亚洲| 搡老岳熟女国产| 99热网站在线观看| 人人妻,人人澡人人爽秒播 | 久久影院123| 成人午夜精彩视频在线观看| 日本猛色少妇xxxxx猛交久久| 欧美在线黄色| 国产男女内射视频| 久久久精品免费免费高清| av又黄又爽大尺度在线免费看| 新久久久久国产一级毛片| 国产麻豆69| 三上悠亚av全集在线观看| 青草久久国产| 青春草国产在线视频| 视频在线观看一区二区三区| 久热爱精品视频在线9| 永久免费av网站大全| 亚洲自偷自拍图片 自拍| 国产精品.久久久| 看免费av毛片| 欧美黑人欧美精品刺激| 欧美久久黑人一区二区| 另类亚洲欧美激情| 亚洲av电影在线观看一区二区三区| 少妇被粗大的猛进出69影院| 亚洲国产中文字幕在线视频| 人妻 亚洲 视频| xxx大片免费视频| 91成人精品电影| 久久久精品国产亚洲av高清涩受| 日韩制服丝袜自拍偷拍| 少妇人妻久久综合中文| 久久国产精品男人的天堂亚洲| 亚洲成国产人片在线观看| 涩涩av久久男人的天堂| 亚洲免费av在线视频| 国产在线一区二区三区精| 欧美日韩福利视频一区二区| 天天躁夜夜躁狠狠躁躁| 观看av在线不卡| 中文字幕制服av| 一区二区av电影网| 精品亚洲乱码少妇综合久久| 亚洲伊人久久精品综合| 国产在线一区二区三区精| 国产国语露脸激情在线看| 欧美 日韩 精品 国产| 两性夫妻黄色片| 秋霞在线观看毛片| 美女大奶头黄色视频| 国产免费一区二区三区四区乱码| 久久婷婷青草| 国产精品成人在线| 大片免费播放器 马上看| 精品亚洲成a人片在线观看| 久久久亚洲精品成人影院| 亚洲av综合色区一区| 伦理电影大哥的女人| 熟女少妇亚洲综合色aaa.| 精品国产超薄肉色丝袜足j| 欧美激情高清一区二区三区 | 久久久欧美国产精品| 色婷婷av一区二区三区视频| 日韩一区二区三区影片| 欧美日韩一区二区视频在线观看视频在线| 亚洲国产欧美在线一区| 婷婷色综合www| 亚洲视频免费观看视频| 精品久久久久久电影网| 亚洲av在线观看美女高潮| 久久这里只有精品19| 一区二区三区激情视频| 国产成人精品久久二区二区91 | 亚洲在久久综合| 美女扒开内裤让男人捅视频| 国产亚洲av片在线观看秒播厂| 菩萨蛮人人尽说江南好唐韦庄| 中文字幕亚洲精品专区| 日本vs欧美在线观看视频| 日日啪夜夜爽| 免费久久久久久久精品成人欧美视频| 日韩精品免费视频一区二区三区| 免费av中文字幕在线| 蜜桃在线观看..| 人人澡人人妻人| 国产探花极品一区二区| 国产1区2区3区精品| 国产一区二区激情短视频 | 人人妻人人添人人爽欧美一区卜| 岛国毛片在线播放| av电影中文网址| 国产福利在线免费观看视频| 在线免费观看不下载黄p国产| 黄片无遮挡物在线观看| 如何舔出高潮| 少妇人妻久久综合中文| 国产精品久久久av美女十八| 一区二区三区激情视频| 啦啦啦 在线观看视频| 亚洲av电影在线观看一区二区三区| 在线观看免费视频网站a站| 少妇人妻 视频| 国产精品一区二区精品视频观看| 欧美日韩亚洲综合一区二区三区_| 尾随美女入室| av片东京热男人的天堂| 欧美成人精品欧美一级黄| 男女边摸边吃奶| 无限看片的www在线观看| 亚洲中文av在线| 亚洲精品国产色婷婷电影| 丝袜喷水一区| 国产熟女午夜一区二区三区| av女优亚洲男人天堂| 色94色欧美一区二区| 黑人欧美特级aaaaaa片| 成人午夜精彩视频在线观看| 女人精品久久久久毛片| 国产精品蜜桃在线观看| 男女边摸边吃奶| 国产老妇伦熟女老妇高清| 99国产综合亚洲精品| 丁香六月欧美| 看十八女毛片水多多多| 一本大道久久a久久精品| 成人毛片60女人毛片免费| 2021少妇久久久久久久久久久| 欧美日韩亚洲综合一区二区三区_| 丝袜美足系列| 国产欧美日韩一区二区三区在线| 国产成人精品福利久久| 亚洲av日韩在线播放| 日韩欧美精品免费久久| 亚洲人成77777在线视频| 精品少妇久久久久久888优播| 久久精品aⅴ一区二区三区四区| 午夜激情久久久久久久| 精品视频人人做人人爽| av网站在线播放免费| 秋霞伦理黄片| 国产亚洲午夜精品一区二区久久| av在线app专区| 免费在线观看黄色视频的| 国产亚洲一区二区精品| 天天躁夜夜躁狠狠久久av| 老司机深夜福利视频在线观看 | 国产免费一区二区三区四区乱码| 日韩,欧美,国产一区二区三区| 我的亚洲天堂| 男女床上黄色一级片免费看| 伦理电影免费视频| 国产精品国产av在线观看| 欧美另类一区| 亚洲精品久久成人aⅴ小说| av又黄又爽大尺度在线免费看| 97在线人人人人妻| 欧美日韩一级在线毛片| 晚上一个人看的免费电影| 国产精品成人在线| 天堂俺去俺来也www色官网| 久久久久久久久久久久大奶| 久久精品国产综合久久久| 日本黄色日本黄色录像| 欧美另类一区| 搡老岳熟女国产| 国产成人免费观看mmmm| 韩国精品一区二区三区| 成人黄色视频免费在线看| 少妇人妻精品综合一区二区| 亚洲av日韩在线播放| 国产在视频线精品| 老司机影院毛片| 成年人午夜在线观看视频| 街头女战士在线观看网站| 中文字幕制服av| 爱豆传媒免费全集在线观看| 亚洲欧美成人精品一区二区| 午夜福利视频精品| 亚洲av日韩在线播放| 亚洲欧美激情在线| 少妇人妻 视频| 国产色婷婷99| 人人澡人人妻人| 国产男女超爽视频在线观看| 久久精品亚洲熟妇少妇任你| 亚洲av中文av极速乱| 国产亚洲一区二区精品| 精品久久久久久电影网| 亚洲精品久久成人aⅴ小说| 欧美日韩视频高清一区二区三区二| 久久国产精品大桥未久av| 久久久久网色| 91aial.com中文字幕在线观看| 91成人精品电影| 最近最新中文字幕免费大全7| 极品少妇高潮喷水抽搐| 日本91视频免费播放| 1024视频免费在线观看| 人妻人人澡人人爽人人| 热99久久久久精品小说推荐| 91aial.com中文字幕在线观看| 日日爽夜夜爽网站| 亚洲精品国产av蜜桃| 天天添夜夜摸| 久久ye,这里只有精品| 狠狠精品人妻久久久久久综合| 51午夜福利影视在线观看| 丰满迷人的少妇在线观看| 91成人精品电影| 九草在线视频观看| 大片电影免费在线观看免费| 国产1区2区3区精品| 日本欧美国产在线视频| 欧美最新免费一区二区三区| 中文字幕高清在线视频| 亚洲中文av在线| 久久国产亚洲av麻豆专区| 看免费成人av毛片| 中文字幕最新亚洲高清| 免费看不卡的av| 美女扒开内裤让男人捅视频| avwww免费| 夫妻午夜视频| 欧美变态另类bdsm刘玥| av天堂久久9| 欧美黑人欧美精品刺激| 最近中文字幕高清免费大全6| 久久鲁丝午夜福利片| 黄色怎么调成土黄色| www.av在线官网国产| 美女福利国产在线| 大香蕉久久网| 黄色毛片三级朝国网站| 看十八女毛片水多多多| av电影中文网址| 在线看a的网站| 欧美 亚洲 国产 日韩一| 国产精品av久久久久免费| 咕卡用的链子| 一级毛片黄色毛片免费观看视频| 精品卡一卡二卡四卡免费| 亚洲成人av在线免费| 我要看黄色一级片免费的| 99精国产麻豆久久婷婷| 一边摸一边抽搐一进一出视频| 欧美人与性动交α欧美精品济南到| 国产xxxxx性猛交| 久久97久久精品| 中文天堂在线官网| 九九爱精品视频在线观看| 欧美中文综合在线视频| 岛国毛片在线播放| 国产精品99久久99久久久不卡 | 国产99久久九九免费精品| 成年人免费黄色播放视频| av国产精品久久久久影院| 美女高潮到喷水免费观看| 午夜91福利影院| 欧美在线一区亚洲| 97精品久久久久久久久久精品|