• <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 fuzzy immune algorithm and its application in solvent tower soft sensor modeling

    2015-03-03 08:01:24MENGKeDONGZhaoyangGAOXiaodanWANGHaimingLIXiao
    關(guān)鍵詞:計(jì)算精度測試函數(shù)徑向

    MENG Ke, DONG Zhao-yang, GAO Xiao-dan, WANG Hai-ming, LI Xiao

    (1. Centre for Intelligent Electricity Networks, The University of Newcastle, Callaghan 2308, Australia;2. School of Electrical and Information Engineering, The University of Sydney, Sydney 2006, Australia;3. School of Computer Science and Control Engineering, North University of China, Taiyuan 030051, China)

    ?

    A fuzzy immune algorithm and its application in solvent tower soft sensor modeling

    MENG Ke1,2, DONG Zhao-yang2, GAO Xiao-dan1, WANG Hai-ming1, LI Xiao3

    (1.CentreforIntelligentElectricityNetworks,TheUniversityofNewcastle,Callaghan2308,Australia;2.SchoolofElectricalandInformationEngineering,TheUniversityofSydney,Sydney2006,Australia;3.SchoolofComputerScienceandControlEngineering,NorthUniversityofChina,Taiyuan030051,China)

    An improved immune algorithm is proposed in this paper. The problems, such as convergence speed and optimization precision, existing in the basic immune algorithm are well addressed. Besides, a fuzzy adaptive method is presented by using the fuzzy system to realize the adaptive selection of two key parameters (possibility of crossover and mutation). By comparing and analyzing the results of several benchmark functions, the performance of fuzzy immune algorithm (FIA) is approved. Not only the difficulty of parameters selection is relieved, but also the precision and stability are improved. At last, the FIA is applied to optimization of the structure and parameters in radial basis function neural network (RBFNN) based on an orthogonal sequential method. And the availability of algorithm is proved by applying RBFNN in modeling in soft sensor of solvent tower.

    immune algorithm; fuzzy system; radial basis function neural network (RBFNN); soft sensor

    0 Introduction

    With the development of immunology and its research methods, the mechanism of biologic immune system has attracted increasing attention from researchers in recent years. Due to the powerful ability of information processing and special characteristics such as diversity, adaptive trait, biologic immune system has become a hot spot of artificial intelligence.

    Being the defense system of mammal, immune system plays a significant role in keeping the normal life activities of animals. If it is weakened or destroyed, lives will be endangered. The process that immune system annihilates viruses can be briefly described as follows:

    Once bacteria invade and enter the bloodstream or lymphatic system, they will encounter B cell and the antibodies withheld within B cell’s membrane will detect antigens in the bacteria. Thenceforth, T cells communicate with B cells based on the received information about the antigen from macrophages earlier and by so doing, B cells are inspired to propagate. The propagated B cells are converted into memory cells and antibodies are produced. With the aid of macrophages and other proteins within biologic bodies, antibodies bind to antigens and kill the antigens after they enter into blood system through the heart.

    Being an innovative optimization algorithm based on immune mechanism, the immune algorithm (IA)[1]is employed to address the multi-modal function optimization problem. It imitating the principle of our defense system annihilating foreign disease-causing bacteria or viruses through self-learning and self-adjusting. The capability of somatic theory and network hypothesis of immune system of multi-modal optimization problems has been examined in Ref.[2]. An IA is introduced in Ref.[3] to search for diverse solutions to design problems for electromagnetic devices, where optimal solutions are aggregated in memory cells.

    Differences in the production system for memory and antibodies distinguish IA from genetic algorithm (GA) although they are quite similar. Besides, IA manipulates a population of candidates simultaneously in the search space whereas GA manipulates just one. Compared with GA and other evolution programming, IA promotes the general search ability through the mechanism based on memory pool. At the same time, it realizes the function of self-adjusting by calculating affinity and concentration. To some extent, it avoids premature convergence.

    1 Soft sensor and RBF neural network

    In order to get eligible production, quality control wields an important role in industrial manufacture. Because of the complexity of industrial process especially in the petrochemical industry, it is very difficult to realize the real-time strict control of the quality of some products. Under many circumstances, the qualities of many products are tested off-line by labor because of the high price, difficulty of maintenance, time latency of on-line measure meters.

    The conception of soft sensors, which combines control knowledge and technologic theories together, was firstly brought forward in Ref.[4]. Some variables which can be easily measured are selected to compute real-time reliable estimates data of other ones which can not or is difficult to be measured by designing proper algorithms. Nonlinear modeling techniques are usually utilized to develop soft sensors to handle the peculiar nonlinearities of processes[4]. Not only can soft sensors be operated alone as a valuable, economic replacement of costly hardware sensors, but also work in parallel with real sensors to allow model-based techniques to be adopted in order to develop fault detection functions devoted to the analysis of the sensor’s health status.

    Radial basis function neural networks (RBFNN) is an excellent neural network in performance. In 1990, Girosi and Poggio had proved RBFNN can approach any nonlinear functions by discretionary precision[5]. RBF networks are gaining increasing popularity in many scientific and engineering fields as a result of their strengths compared with other types of artificial neural networks (ANN), e.g. improved approximation capabilities, simpler network structures and faster learning algorithms.

    RBF networks are composed of three layers, including the input, hidden and output layers, which form an unique neural network architecture. The input layer communicates the entire network to its outside environment. In the hidden layer, all the nodes are connected with centers, and they are vectors with a dimension identical to the number of inputs to the network. A RBF is employed to pass the node activity; the feedback from a hidden node is generated. Lastly, the output serves as a summation unit, which is linear. The structure of a typical RBFNN is presented in Fig.1.

    Fig.1 Typical MISO RBFNN

    But how to decide the number of neurons within the hidden layer has always been the problem counteracting the application of RBFNN. There is a possibility that a small network never converges, however, a large network converges fast but lacks the generalization ability. Besides a suitable network size, there are many other questions that need to be answered to use a network for a particular problem. Learning step, proper training procedure, number of layers, network initialization, value of gain and the number of neurons in each layer are some difficulties which block the wide application of neural network. In this paper an orthogonal sequential method[6]is represented producing RBFNN models based on an improved IA, which is used to auto-configure the structure of the network and obtain the model parameters.

    2 Fuzzy immune algorithm

    2.1 Basic principles of immune algorithm

    For the optimization problem, the antigens and antibodies in the immune system are represented as the objective functions and feasible solutions, respectively.

    The coding method for traditional IA is similar to that for the GA, which is coded in binary. In this paper a new real-coding based evolution IA because of the advantages of real coding algorithm in training neural network[7]is represented, which effectively improves the performance of traditional IA, solving the problems such as premature convergence, low speed of calculation and low precision.

    2.2 Calculation strategy of FIA

    The steps of FIA are illustrated as shown in Fig.2.

    Fig.2 Flow chart of FIA

    Step 1 (Recognize antigen)

    Antigen: objective function (generally minimum value).

    Antibody: feasible solutions.

    Step 2 (Produce initial antibody population and memory pool)

    In this step, the antibodies are generated randomly and then compartmentalized to the given intervals. The memory pool is a zero matrix of given size.

    Step 3 (Calculate the affinity values of all antibodies)

    IA uses affinity value as a discriminator of the quality of solutions represented by the antibodies in a population. Because the final target of the algorithm is searching the minimum value, function values of all the antibodies are calculated and sorted in ascending sequence.

    To calculate the affinity valueaffinity(i) of antibodyi, it is given by

    (1)

    whereris a random number in the interval [0.01,0.3].

    Step 4 (Update memory pool)

    Eminent antibodies from the present population are selected by their affinity values and concentrations in order to update memory pool which can be used to generate the offspring antibodies population.

    Step 5 (Select antibodies)

    1) To calculate the concentrationcon(i) of antibodyi, it is given by

    (2)

    where

    (3)

    2) To calculate the selection probabilityPs(i) of antibodyi, it is given by

    (4)

    3) A roulette selection is implemented based on the computed selection probability for the antibodies. This allocates each antibody a probability of being selected proportional to its relative affinity and concentration. A new antibody generation can therefore be formed by spinning the designed roulette.

    Step 6 (Determine crossover and mutation rates through fuzzy method)

    In IA, many parameters play an important role in determining convergence and convergent rate, such as crossover and mutation rates. Crossover is one key IA operator that promotes the new region exploration ability in the search space. Generally, crossover rate should be chosen comparatively big[8], between 0.7 and 1.0. Mutation is another IA operator which guarantees the diversity of the population. In Ref. [8], the mutation rate should be chosen between thousandths and hundredths.

    According to Ref.[9], statistical method, support vector machine or neural network can be utilized to adjust crossover and mutation rates. However, we have found that fuzzy system approach makes better contributions to the IA in both time consumption and precision when compared with above methodologies.

    Themembershipfunctionsforinputfd(t),andoutputΔPcareshowninFigs.3-6.Inthesamewaythemembershipfunctionsforinputfd(t), Pm,ΔPmandfuzzydecisiontableforΔPmcanbedrawn.

    Fig.3 Membership function of fd(t)

    Fig.4 Membership function of Pc

    Fig.5 Membership function of ΔPc

    Fig.6 Membership function of ΔPm

    According to a great deal of experimental data and expert knowledge, the fuzzy decision for ΔPcis made and presented in Table 1. By virtue of the same theory, the fuzzy decision table for ΔPmcan be generated. In the table, NH, NL, NM, NS, ZE, PS, PM, PL and PH are abbreviated for Negative Huge, Negative Large, Negative Medium, Negative Small, Zero, Positive Small, Positive Medium, Positive Large and Positive Huge, respectively.

    Table 1 Fuzzy decision table for ΔPc

    Step 7 (Crossover implementation)

    The crossover operator represents the mixing of antibiotic material from two selected parent antibodies to produce one or two child offspring antibody population. The amount of antibodies take part in crossover implementation is determined by crossover ratePc, which is adjusted by fuzzy method.

    An improved arithmetic crossover operator is described as

    (5)

    whereb1=0.5+b,b2=0.5-b, andbis a random number in interval [0,1].

    If the offspring antibody exceeds the given intervals, another operator will be selected.

    (6)

    Step 8 (Mutation implementation)

    An uneven mutation method[10-11]is described as follows:

    For one given parent antibody, if its elementxmis randomly selected to mutation, the corresponding element in its offspring is likely to change in two possibilities

    (7)

    (8)

    whereTis maximum generation;tis current generation;ris a fixed uneven parameter, usuallyr=2;bis a random number in the interval [0,1].

    In this paper, an improved mutation method is introduced, and its idea mainly comes from differential algorithm[12].

    (9)

    whereantibodybestis the optimal antibody of the current generation which is stored in memory pool.

    Step 9 (Generate new antibody population and update memory pool)

    Antibodies with high affinity value will evolve into next generation and be added into memory pool. Given number of new antibodies will be added into antibody population replacing antibodies with low affinity value.

    Step 10 (Termination criterion).

    For this step, the search is terminated if the following conditions are satisfied:

    1) The values for minvaluedo not change for several generations.

    2) When the set number of evolutionTis achieved.

    2.3 Test examples

    Several standard test functions are used to examine the ability of FIA and its advantages superior to other algorithms in the same test environment and condition. Except for parameters adaptive selection, the FIA is similar to other algorithms in flow and thought. The standard test functions and test results are shown in Table 2 and Table 3, respectively.

    Table 2 Three standard test functions

    Table 3 Results of test functions

    The above data indicate that FIA can effectively solve the premature problem and is suitable for complex optimization problems. The algorithm is not trapped by the local optimal solution and can promptly and accurately obtain a full set of global optimal solutions, which are incomparable in other similar algorithms.

    3 Configuration of RBFNN using FIA

    Like GA and other evolution algorithms[7,13], IA has three main applications in neural network:

    1) The parameters learning of neural network;

    2) The topology structure selection of neural network;

    3) The parameters and structure optimization of neural network.

    And the standard procedure for RBF networks learning problem can be decomposed into two steps: The first one is obtaining the number and centers of the nodes in hidden layer and the second one is calculation of the connection weights using simple linear regression.

    3.1 General ideas and theories[6]

    For typical RBFNN, ifwidenotes output weights, φi(X,Ci)denotestheoutputofithneuron, X=[x1,x2,…,xm]isinputvector, Cidenotesthehiddennodecenterlocationsofithneuronandydenoteslinearsummationofoutputofhiddenlayerneurons.IftheRBFisGaussfunction,

    (10)

    (11)

    Foronesetoftrainingdata,theequationcanbetransformedinto

    (12)

    Andthen

    (13)

    (14)

    (15)

    Sothegivenequationscanbetransformedinto

    (16)

    (17)

    (18)

    3.2 Two-step learning strategy of RBFNN

    3.2.1 Design of network structure

    Real-coded algorithm is suitable for neural network training because the antibodies are the real values in neural network. The real-coded method forithantibody is that the formern+1 columns are relevantncenters and one warp and the last column is affinity value of the antibody.

    The steps of RBFNN training are depicted as:

    Step 1: Initialization.i=1, E0=Y.

    Step 3: If output satisfies stopping criterion, network training will stop. Otherwise,i=i+1, and another neuron will be added.

    3.2.2 Design of network output layer

    Because of the output layer is linear and it serves as a summation unit, the least square method can be chosen to calculate

    (19)

    3.3 Result of soft sensor

    One pure-terephthalic acid (PTA) solvent tower is chosen as research object in this paper and the ultimately target is to establish the soft sensor model for acid content of the bottom flow of the solvent tower. Solvent dehydration is an important unit in PTA manufacture process. Because of the long delay and slow dynamic response of the rectify process, it is very difficult to realize the real-time control of the production quality. The running situation of the control system largely depends on the operators’ technical levels and habits. Although the set can run smoothly in a short time, it cannot reach the optimal state. Great care was taken in both selecting the appropriate set of training examples, which covered all the operating conditions of the plant. According to technologic flow, three parameters (conductance, temperature and pressure) are selected as inputs to the RBF neural network, whereas the output is the relevant acid content. For 175 metrical data, former 100 are chosen to train neural network and the other 75 are used to determine the availability and generalization ability of the neural network.

    To avoid over-learning phenomena, an early stopping approach is used. The parameters in FIA are set as

    Popsize=50, Memorypool=20,

    And the results of training and estimation are shown in Figs.7 and 8. Parameters comparison between different neural netowkrs are presented in Table 4.

    Fig.7 RBFNN training result

    Fig.8 Comparison between NN estimation and corresponding actual data

    Table 4 Parameters comparison between different neural networks

    NetworksNumberofnodesinhiddenlayerMSEMaxrelativeerrorStandardrelativeerrorFIARBF80.11680.01870.0028OLSRBF90.13360.02240.0031ConventionalRBF120.14190.023540.0033

    4 Conclusion

    The simulation results indicate that the proposed methodology is effective and accurate. The parameters of neural network are optimized by using FIA, not only the number of nodes in hidden layer can be reduced, but also the generalization ability can be improved. As the study of combining FIA and RBFNN in soft sensor modeling is emerging recently, there are many aspects we can borrow from the immune system and fuzzy system, and further research is needed.

    [1] Liao G C, Tsao T P. Application embedded chaos search immune genetic algorithm for short-term unit commitment. Electric Power Systems Research, 2004, 71(2): 135-144.

    [2] Fukuda T, Mori K, Tsukiyama M. Parallel search for multi-modal function optimization with diversity and learning of immune algorithm. Artificial Immune Systems and Their Applications, 1999: 210-220.

    [3] Chun J S, Lim, J P, Jung H K, et al. Multisolution optimization of permanent magnet linear synchronous motor for high thrust and acceleration operation. In: Proceedings of International Conference on Electric Machines and Drives (IEMD 99), 1999: 57-59.

    [4] Fortuna L, Rizzo A, Sinatra M, et al. Soft analyzers for a sulfur recovery unit. Control Engineering Practice, 2003, 11(12): 1491-1500.

    [5] Girosi F, Piggio T. Networks and the best approximation property. Biological Cybernetics, 1990, 63(3): 169-179.

    [6] BAO Zhi-jun, WANG Xian-lai. RBF neural networks based on orthogonal sequential genetic algorithm. In: Proceeding of the 22nd Chinese Control Conference, Yichang, China, 2003: 1.

    [7] Michalewicz Z. Genetic algorithms + data structures=evolution program. New York: Springer Verlag, 1994.

    [8] Braberman V A. Verification of real-time design: combining scheduling theory with automatic formal verficaton. Software Engineering Notes, 1999, 24(6): 494-511.

    [9] Shi Y, Eberhart R, Chen Y. Implementation of evolutionary fuzzy system. IEEE Transactions on Fuzzy System, 1999, 7(2): 109-119.

    [10] Thompson J M, Miller S P. Specification-based prototyping for embedded. Software Engineering Notes, 1999, 24(6): 163-180.

    [11] Fierz H. The CIP method: component- and model-based construction of embedded system. Software Engineering Notes, 1999, 24(6): 375-393.

    [12] Lopez-Cruz I L, van Willigenburg L G, van Straten G. Efficient differential evolution algorithms for multimodal optimal control problems. Applied Soft Computing, 2003, 3: 97-122.

    [13] Goldberg D E. Genetic Algorithms in search, optimization and machine learning. MA: Addison-Wesley, 1989.

    模糊免疫算法及其在溶劑脫水塔軟測量建模中的應(yīng)用

    孟 科1, 2, 董朝陽2, 高曉丹1, 王海明1, 李 曉3

    (1. Centre for Intelligent Electricity Networks, The University of Newcastle, Callaghan 2308, Australia;2. School of Electrical and Information Engineering, The University of Sydney, Sydney 2006, Australia;3. 中北大學(xué) 計(jì)算機(jī)與控制工程學(xué)院, 山西 太原 030051)

    本文針對(duì)基本免疫算法收斂速度慢、 計(jì)算精度低等缺點(diǎn), 提出了模糊免疫算法。 該算法引入模糊技術(shù), 對(duì)關(guān)鍵參數(shù)(交叉概率和變異概率)實(shí)現(xiàn)了模糊自適應(yīng)調(diào)整。 通過標(biāo)準(zhǔn)測試函數(shù)實(shí)驗(yàn)結(jié)果的對(duì)比, 其可行性和有效性得到證明, 不僅減輕了原始算法中參數(shù)確定存在的困難, 而且提高了算法的計(jì)算速度和精度。 其次, 本文將模糊免疫算法用于徑向基神經(jīng)網(wǎng)絡(luò)的訓(xùn)練, 并將該神經(jīng)網(wǎng)絡(luò)應(yīng)用于溶劑脫水塔軟測量模型。 仿真實(shí)驗(yàn)證明, 模糊免疫算法優(yōu)化的徑向基函數(shù)神經(jīng)網(wǎng)絡(luò)具有良好的泛化性能。

    免疫算法; 模糊系統(tǒng); 徑向基神經(jīng)網(wǎng)絡(luò); 軟測量

    MENG Ke, DONG Zhao-yang, GAO Xiao-dan, et al. A fuzzy immune algorithm and its application in solvent tower soft sensor modeling. Journal of Measurement Science and Instrumentation, 2015, 6(2): 197-204.

    10.3969/j.issn.1674-8042.2015.02.016

    MENG Ke (ke.meng@newcastle.edu.cn)

    1674-8042(2015)02-0197-08 doi: 10.3969/j.issn.1674-8042.2015.02.016

    Received date: 2015-02-25

    CLD number: TP273+.4 Document code: A

    猜你喜歡
    計(jì)算精度測試函數(shù)徑向
    淺探徑向連接體的圓周運(yùn)動(dòng)
    RN上一類Kirchhoff型方程徑向?qū)ΨQ正解的存在性
    基于PID+前饋的3MN徑向鍛造機(jī)控制系統(tǒng)的研究
    一類無窮下級(jí)整函數(shù)的Julia集的徑向分布
    基于SHIPFLOW軟件的某集裝箱船的阻力計(jì)算分析
    廣東造船(2018年1期)2018-03-19 15:50:50
    具有收縮因子的自適應(yīng)鴿群算法用于函數(shù)優(yōu)化問題
    帶勢函數(shù)的雙調(diào)和不等式組的整體解的不存在性
    約束二進(jìn)制二次規(guī)劃測試函數(shù)的一個(gè)構(gòu)造方法
    單元類型和尺寸對(duì)拱壩壩體應(yīng)力和計(jì)算精度的影響
    鋼箱計(jì)算失效應(yīng)變的沖擊試驗(yàn)
    欧美乱色亚洲激情| 又紧又爽又黄一区二区| xxx96com| 老司机午夜福利在线观看视频| 欧美另类亚洲清纯唯美| 久久精品亚洲精品国产色婷小说| 十八禁高潮呻吟视频| 欧美激情 高清一区二区三区| 成在线人永久免费视频| 香蕉国产在线看| 日本vs欧美在线观看视频| 老司机午夜十八禁免费视频| 在线观看午夜福利视频| 国产成人精品无人区| 欧美午夜高清在线| 国产精品综合久久久久久久免费 | 午夜视频精品福利| 国产免费男女视频| 国产精品免费大片| 久久精品亚洲av国产电影网| av有码第一页| 高清视频免费观看一区二区| 每晚都被弄得嗷嗷叫到高潮| 欧美久久黑人一区二区| 午夜免费鲁丝| 在线播放国产精品三级| 国产精品亚洲av一区麻豆| 久99久视频精品免费| 这个男人来自地球电影免费观看| 成年动漫av网址| 国产精品九九99| 国产成人欧美| 伦理电影免费视频| 国产午夜精品久久久久久| 午夜福利乱码中文字幕| 一边摸一边抽搐一进一出视频| 热99re8久久精品国产| 亚洲片人在线观看| 国产99久久九九免费精品| 亚洲精品美女久久久久99蜜臀| 精品国产国语对白av| 人妻一区二区av| 精品第一国产精品| 亚洲国产欧美一区二区综合| 一本综合久久免费| 香蕉国产在线看| 十八禁网站免费在线| 老司机影院毛片| 老汉色∧v一级毛片| 最近最新中文字幕大全电影3 | 好看av亚洲va欧美ⅴa在| 精品国内亚洲2022精品成人 | 一级a爱视频在线免费观看| 中文字幕av电影在线播放| 我的亚洲天堂| 国产aⅴ精品一区二区三区波| 在线播放国产精品三级| 色尼玛亚洲综合影院| av天堂久久9| 麻豆成人av在线观看| 久久天堂一区二区三区四区| 亚洲熟女精品中文字幕| 亚洲精品av麻豆狂野| 天天躁夜夜躁狠狠躁躁| 欧美日韩乱码在线| www.精华液| 亚洲自偷自拍图片 自拍| 精品亚洲成国产av| 最近最新中文字幕大全免费视频| 国内久久婷婷六月综合欲色啪| 9热在线视频观看99| 一a级毛片在线观看| 精品福利永久在线观看| 久久久久久久久免费视频了| 纯流量卡能插随身wifi吗| 亚洲久久久国产精品| 香蕉国产在线看| av国产精品久久久久影院| 老汉色∧v一级毛片| 日本精品一区二区三区蜜桃| 欧美精品一区二区免费开放| 久久国产亚洲av麻豆专区| 国产精品综合久久久久久久免费 | 天天躁夜夜躁狠狠躁躁| 精品免费久久久久久久清纯 | 99热只有精品国产| 日本一区二区免费在线视频| 又黄又粗又硬又大视频| 国产免费av片在线观看野外av| 麻豆av在线久日| 伦理电影免费视频| 老熟妇仑乱视频hdxx| 女性生殖器流出的白浆| 精品久久蜜臀av无| 中文亚洲av片在线观看爽 | 91在线观看av| avwww免费| 亚洲五月天丁香| 校园春色视频在线观看| 欧美最黄视频在线播放免费 | 中文亚洲av片在线观看爽 | 久久久久久久国产电影| 免费在线观看黄色视频的| 亚洲精品美女久久av网站| 999久久久精品免费观看国产| 99国产精品一区二区三区| 两性午夜刺激爽爽歪歪视频在线观看 | 免费日韩欧美在线观看| 午夜福利乱码中文字幕| 最新的欧美精品一区二区| 国产精品永久免费网站| 丝袜在线中文字幕| 少妇裸体淫交视频免费看高清 | xxx96com| 精品亚洲成a人片在线观看| 久久人妻熟女aⅴ| 大片电影免费在线观看免费| 免费在线观看影片大全网站| 天堂中文最新版在线下载| 亚洲综合色网址| 精品福利观看| 国产欧美亚洲国产| 色综合欧美亚洲国产小说| 老司机在亚洲福利影院| 国产一区二区三区综合在线观看| 亚洲精品乱久久久久久| 午夜两性在线视频| 亚洲av电影在线进入| 女人被狂操c到高潮| 欧美国产精品va在线观看不卡| 国产又色又爽无遮挡免费看| 欧美日韩视频精品一区| 亚洲自偷自拍图片 自拍| 久久精品aⅴ一区二区三区四区| 最新的欧美精品一区二区| 亚洲精品久久午夜乱码| 国产成人系列免费观看| 中文欧美无线码| 一级毛片精品| 在线观看免费视频网站a站| 又大又爽又粗| 免费观看精品视频网站| 在线观看一区二区三区激情| 国产色视频综合| 男男h啪啪无遮挡| 国产亚洲精品第一综合不卡| 国产成人影院久久av| 亚洲成国产人片在线观看| 一级毛片高清免费大全| 欧美精品高潮呻吟av久久| www.999成人在线观看| 在线观看日韩欧美| 成年人午夜在线观看视频| 热99久久久久精品小说推荐| 性少妇av在线| 国产熟女午夜一区二区三区| 91在线观看av| 人妻一区二区av| 久久这里只有精品19| av不卡在线播放| 国产精品久久久人人做人人爽| 成年人黄色毛片网站| 丝袜美足系列| 18禁国产床啪视频网站| 操美女的视频在线观看| 久久久精品区二区三区| 午夜亚洲福利在线播放| 黄色怎么调成土黄色| 国产亚洲精品第一综合不卡| 国产精品偷伦视频观看了| 国产成人精品久久二区二区免费| 国产91精品成人一区二区三区| 在线观看免费午夜福利视频| 国产精品免费视频内射| av有码第一页| 欧美大码av| 嫩草影视91久久| 狠狠狠狠99中文字幕| 欧美大码av| 乱人伦中国视频| 夜夜爽天天搞| 亚洲欧美一区二区三区久久| 国产精品免费一区二区三区在线 | 久久久精品区二区三区| 男女免费视频国产| 成人18禁高潮啪啪吃奶动态图| 日本黄色视频三级网站网址 | 在线观看午夜福利视频| 国产三级黄色录像| 午夜精品久久久久久毛片777| 王馨瑶露胸无遮挡在线观看| 麻豆乱淫一区二区| 国产精品免费一区二区三区在线 | 午夜精品在线福利| av片东京热男人的天堂| 黑人欧美特级aaaaaa片| 一本大道久久a久久精品| 亚洲精品成人av观看孕妇| 精品久久久久久,| 国产精品九九99| 69精品国产乱码久久久| 欧美国产精品一级二级三级| 欧美乱色亚洲激情| www.自偷自拍.com| 十八禁网站免费在线| 最新的欧美精品一区二区| 成人国语在线视频| 91精品国产国语对白视频| 999精品在线视频| 少妇裸体淫交视频免费看高清 | 91在线观看av| 亚洲精品美女久久久久99蜜臀| 可以免费在线观看a视频的电影网站| 亚洲精品在线美女| 日韩人妻精品一区2区三区| 精品高清国产在线一区| 99国产精品免费福利视频| 久久精品国产99精品国产亚洲性色 | 女人久久www免费人成看片| 777米奇影视久久| 不卡一级毛片| 老鸭窝网址在线观看| 啦啦啦视频在线资源免费观看| 久久精品熟女亚洲av麻豆精品| 久久香蕉激情| 9热在线视频观看99| 久久精品人人爽人人爽视色| 50天的宝宝边吃奶边哭怎么回事| 一区福利在线观看| 久久久久久亚洲精品国产蜜桃av| 亚洲全国av大片| 亚洲五月天丁香| 9191精品国产免费久久| 精品人妻熟女毛片av久久网站| 欧美成人免费av一区二区三区 | 99riav亚洲国产免费| 日韩免费高清中文字幕av| 亚洲男人天堂网一区| 亚洲免费av在线视频| 建设人人有责人人尽责人人享有的| 黄色视频,在线免费观看| 国产精品一区二区免费欧美| 国产主播在线观看一区二区| 久久精品国产99精品国产亚洲性色 | tocl精华| 欧美成人午夜精品| 一级片免费观看大全| 亚洲成人手机| 国产精品二区激情视频| av天堂久久9| 国产成人系列免费观看| 99久久精品国产亚洲精品| 精品国产亚洲在线| 亚洲一码二码三码区别大吗| 精品高清国产在线一区| 99在线人妻在线中文字幕 | 岛国毛片在线播放| 我的亚洲天堂| 免费在线观看完整版高清| 99久久综合精品五月天人人| 亚洲少妇的诱惑av| 99热国产这里只有精品6| 国产精品1区2区在线观看. | 精品国产国语对白av| 深夜精品福利| 亚洲精品国产色婷婷电影| 91成人精品电影| 91国产中文字幕| 欧美激情久久久久久爽电影 | 欧美激情 高清一区二区三区| 中出人妻视频一区二区| 午夜福利视频在线观看免费| 如日韩欧美国产精品一区二区三区| 亚洲午夜精品一区,二区,三区| 女人久久www免费人成看片| 久久影院123| 国产精品久久久av美女十八| 精品无人区乱码1区二区| 999久久久精品免费观看国产| 亚洲av欧美aⅴ国产| 美国免费a级毛片| 免费在线观看完整版高清| 欧美另类亚洲清纯唯美| 国产xxxxx性猛交| bbb黄色大片| 国产亚洲一区二区精品| 国产黄色免费在线视频| 国产精品乱码一区二三区的特点 | 精品人妻熟女毛片av久久网站| 成人18禁在线播放| 一区二区日韩欧美中文字幕| 亚洲综合色网址| 91大片在线观看| 韩国精品一区二区三区| 欧美日韩一级在线毛片| 老司机靠b影院| 18禁裸乳无遮挡免费网站照片 | 99国产综合亚洲精品| 亚洲 欧美一区二区三区| 日韩欧美一区二区三区在线观看 | av天堂在线播放| 青草久久国产| 亚洲三区欧美一区| 18在线观看网站| 老司机亚洲免费影院| 亚洲av成人av| 一级毛片女人18水好多| 天堂动漫精品| 看免费av毛片| 99国产精品免费福利视频| 久久ye,这里只有精品| 国产真人三级小视频在线观看| x7x7x7水蜜桃| 久久精品国产a三级三级三级| 中文字幕人妻熟女乱码| 每晚都被弄得嗷嗷叫到高潮| 中文字幕人妻熟女乱码| 亚洲专区中文字幕在线| 无人区码免费观看不卡| www.自偷自拍.com| 日韩一卡2卡3卡4卡2021年| 少妇粗大呻吟视频| 欧美人与性动交α欧美精品济南到| 变态另类成人亚洲欧美熟女 | 中文字幕另类日韩欧美亚洲嫩草| 亚洲欧美一区二区三区久久| 国产精品av久久久久免费| 国产成人精品久久二区二区免费| 一个人免费在线观看的高清视频| 亚洲成人手机| 极品教师在线免费播放| 天天躁日日躁夜夜躁夜夜| 成人亚洲精品一区在线观看| 美女福利国产在线| 人人妻人人澡人人爽人人夜夜| 高潮久久久久久久久久久不卡| 亚洲视频免费观看视频| 正在播放国产对白刺激| 1024香蕉在线观看| 久久午夜综合久久蜜桃| 久久中文看片网| 国产精品永久免费网站| 久久精品国产清高在天天线| 久久ye,这里只有精品| av视频免费观看在线观看| 欧美日韩亚洲国产一区二区在线观看 | 人人妻人人爽人人添夜夜欢视频| 高清av免费在线| 国产精品综合久久久久久久免费 | 黑人操中国人逼视频| 啦啦啦免费观看视频1| 一边摸一边抽搐一进一小说 | 国产精品二区激情视频| 成人国语在线视频| 激情视频va一区二区三区| 91av网站免费观看| 91老司机精品| 无遮挡黄片免费观看| 欧美日韩成人在线一区二区| 在线观看舔阴道视频| 午夜免费观看网址| 国产精品秋霞免费鲁丝片| 一级a爱视频在线免费观看| 国产免费现黄频在线看| 国内久久婷婷六月综合欲色啪| 亚洲 国产 在线| 中国美女看黄片| 日韩免费高清中文字幕av| 99精品欧美一区二区三区四区| 少妇的丰满在线观看| 精品一区二区三区视频在线观看免费 | 国产精品久久久av美女十八| 欧美日韩亚洲国产一区二区在线观看 | 亚洲成av片中文字幕在线观看| 国产aⅴ精品一区二区三区波| 久久精品国产清高在天天线| 黄色a级毛片大全视频| 在线免费观看的www视频| 搡老乐熟女国产| 国产91精品成人一区二区三区| 搡老乐熟女国产| 亚洲成人国产一区在线观看| 精品国产乱子伦一区二区三区| www.999成人在线观看| 免费在线观看日本一区| av有码第一页| 国产亚洲精品一区二区www | 午夜激情av网站| 久久久久久久精品吃奶| 国产精品亚洲一级av第二区| 久久午夜亚洲精品久久| 伊人久久大香线蕉亚洲五| 国产极品粉嫩免费观看在线| 99国产极品粉嫩在线观看| 99久久精品国产亚洲精品| 国产亚洲精品第一综合不卡| 人妻久久中文字幕网| 在线观看免费视频网站a站| 黑人猛操日本美女一级片| 一进一出好大好爽视频| 人人妻人人添人人爽欧美一区卜| 久久香蕉国产精品| av电影中文网址| 亚洲精品中文字幕一二三四区| 热99国产精品久久久久久7| 亚洲精品中文字幕一二三四区| 一边摸一边抽搐一进一出视频| 日韩人妻精品一区2区三区| 国产亚洲av高清不卡| 欧美精品一区二区免费开放| 大型黄色视频在线免费观看| 18禁裸乳无遮挡动漫免费视频| 欧美av亚洲av综合av国产av| svipshipincom国产片| 他把我摸到了高潮在线观看| xxxhd国产人妻xxx| 亚洲性夜色夜夜综合| 精品高清国产在线一区| 黄频高清免费视频| 午夜激情av网站| 亚洲人成电影观看| 每晚都被弄得嗷嗷叫到高潮| 日本撒尿小便嘘嘘汇集6| 国产一区有黄有色的免费视频| 自拍欧美九色日韩亚洲蝌蚪91| avwww免费| 国产精品一区二区在线不卡| 成年人免费黄色播放视频| 丝袜美足系列| 亚洲一码二码三码区别大吗| 黄网站色视频无遮挡免费观看| 波多野结衣一区麻豆| 大型黄色视频在线免费观看| 久久草成人影院| 午夜福利乱码中文字幕| 国产成人av激情在线播放| 亚洲中文av在线| 91在线观看av| 久久久久久久久久久久大奶| 国产一区二区三区视频了| 18禁黄网站禁片午夜丰满| 精品亚洲成国产av| 男女高潮啪啪啪动态图| 亚洲熟妇中文字幕五十中出 | 人人妻,人人澡人人爽秒播| 久久久久久亚洲精品国产蜜桃av| 国产主播在线观看一区二区| 怎么达到女性高潮| 久久中文字幕一级| 97人妻天天添夜夜摸| 久久九九热精品免费| 丝瓜视频免费看黄片| 麻豆av在线久日| 麻豆成人av在线观看| 欧美日韩视频精品一区| 男女午夜视频在线观看| 黑人猛操日本美女一级片| 亚洲久久久国产精品| 国产成人精品久久二区二区免费| 高清黄色对白视频在线免费看| 老司机深夜福利视频在线观看| cao死你这个sao货| 国产99白浆流出| 视频区图区小说| 国产精品久久久av美女十八| 久久午夜综合久久蜜桃| 成人亚洲精品一区在线观看| 亚洲午夜精品一区,二区,三区| 一区二区三区精品91| 男人舔女人的私密视频| 欧美成人免费av一区二区三区 | 校园春色视频在线观看| 99re6热这里在线精品视频| 亚洲av片天天在线观看| 精品福利永久在线观看| 老司机靠b影院| 十分钟在线观看高清视频www| 日韩精品免费视频一区二区三区| 精品国产乱子伦一区二区三区| 黑丝袜美女国产一区| 热99国产精品久久久久久7| 国产成人影院久久av| 天天躁夜夜躁狠狠躁躁| 一级片'在线观看视频| 欧美激情 高清一区二区三区| 中文字幕高清在线视频| 久久久久久久久久久久大奶| 女人被躁到高潮嗷嗷叫费观| 国产成人欧美在线观看 | 热re99久久国产66热| 满18在线观看网站| 久久婷婷成人综合色麻豆| 美女 人体艺术 gogo| 国产在线观看jvid| 欧美久久黑人一区二区| 久久中文字幕人妻熟女| 侵犯人妻中文字幕一二三四区| 日韩有码中文字幕| 精品视频人人做人人爽| 亚洲 欧美一区二区三区| 午夜亚洲福利在线播放| 激情视频va一区二区三区| 色婷婷av一区二区三区视频| 日韩欧美一区二区三区在线观看 | 人妻一区二区av| 亚洲欧美一区二区三区久久| 一边摸一边抽搐一进一小说 | 国产激情久久老熟女| 人妻一区二区av| 女人爽到高潮嗷嗷叫在线视频| 老鸭窝网址在线观看| 精品电影一区二区在线| 国产精品免费大片| 国产三级黄色录像| 欧美日韩av久久| 欧美日韩一级在线毛片| avwww免费| 国产精品电影一区二区三区 | 婷婷成人精品国产| 久久久国产成人精品二区 | 国产精品成人在线| 日韩成人在线观看一区二区三区| 两个人看的免费小视频| 岛国毛片在线播放| 黑人操中国人逼视频| 18禁美女被吸乳视频| 欧美不卡视频在线免费观看 | 人人妻人人爽人人添夜夜欢视频| 黑人欧美特级aaaaaa片| 欧美黑人欧美精品刺激| 9色porny在线观看| 天堂√8在线中文| 一级毛片女人18水好多| 久久香蕉国产精品| 午夜日韩欧美国产| 丝袜美腿诱惑在线| 老司机深夜福利视频在线观看| 亚洲国产精品一区二区三区在线| 亚洲一区中文字幕在线| 一级片免费观看大全| 一级黄色大片毛片| 在线观看免费日韩欧美大片| 热re99久久精品国产66热6| 黄片大片在线免费观看| 国产99白浆流出| 国产欧美日韩综合在线一区二区| 国产高清视频在线播放一区| 人人妻人人澡人人看| 日韩制服丝袜自拍偷拍| 亚洲欧美激情综合另类| 亚洲欧美色中文字幕在线| 人妻久久中文字幕网| 国产蜜桃级精品一区二区三区 | 久久亚洲精品不卡| 黑人猛操日本美女一级片| 大香蕉久久网| 一级a爱视频在线免费观看| x7x7x7水蜜桃| 亚洲,欧美精品.| 亚洲自偷自拍图片 自拍| 麻豆国产av国片精品| 啦啦啦免费观看视频1| 啦啦啦在线免费观看视频4| 国产精品 国内视频| 亚洲伊人色综图| 成人影院久久| 亚洲人成77777在线视频| 建设人人有责人人尽责人人享有的| 亚洲精品乱久久久久久| 在线天堂中文资源库| 曰老女人黄片| 精品久久久久久电影网| 日韩欧美三级三区| 热99国产精品久久久久久7| 中文字幕人妻熟女乱码| 少妇被粗大的猛进出69影院| 一本综合久久免费| 欧美人与性动交α欧美精品济南到| 国产精品香港三级国产av潘金莲| 日日摸夜夜添夜夜添小说| 中亚洲国语对白在线视频| 精品亚洲成a人片在线观看| 人人妻人人添人人爽欧美一区卜| 欧美在线一区亚洲| 亚洲精品在线观看二区| 久久人人爽av亚洲精品天堂| cao死你这个sao货| 精品人妻1区二区| 我的亚洲天堂| 久久久久久久久久久久大奶| 极品少妇高潮喷水抽搐| 人人澡人人妻人| 亚洲黑人精品在线| av中文乱码字幕在线| 精品国产乱子伦一区二区三区| av网站免费在线观看视频| 99国产极品粉嫩在线观看| 亚洲在线自拍视频| 母亲3免费完整高清在线观看| 免费看a级黄色片| 午夜免费鲁丝| 中亚洲国语对白在线视频| 精品久久久久久,| 日日摸夜夜添夜夜添小说| 国产成人影院久久av| 啦啦啦免费观看视频1| 在线国产一区二区在线| 久久午夜亚洲精品久久| 久久精品aⅴ一区二区三区四区| 精品久久久精品久久久| 一级作爱视频免费观看| 丰满的人妻完整版| 精品卡一卡二卡四卡免费| 中文欧美无线码| 日韩欧美一区视频在线观看|