• <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 ?

    Machine learning combined with Langmuir probe measurements for diagnosis of dusty plasma of a positive column

    2021-09-10 09:26:34ZheDING丁哲JingfengYAO姚靜鋒YingWANG王瑩ChengxunYUAN袁承勛ZhongxiangZHOU周忠祥AnatolyKUDRYAVTSEVRuilinGAO高瑞林andJieshuJIA賈潔姝
    Plasma Science and Technology 2021年9期
    關(guān)鍵詞:王瑩

    Zhe DING (丁哲),Jingfeng YAO (姚靜鋒),Ying WANG (王瑩),2,Chengxun YUAN (袁承勛),2,?,Zhongxiang ZHOU (周忠祥),2,Anatoly A KUDRYAVTSEV,2,3,Ruilin GAO (高瑞林) and Jieshu JIA (賈潔姝)

    1 School of Physics,Harbin Institute of Technology,Harbin 150001,People’s Republic of China

    2 Heilongjiang Provincial Key Laboratory of Plasma Physics and Application Technology,Harbin 150001,People’s Republic of China

    3 Physics Department,Saint Petersburg State University,St.Petersburg 198504,Russia

    4 Shanghai Electro-Mechanical Engineering Institute,Shanghai 201109,People’s Republic of China

    5 Science and Technology on Electromagnetic Scattering Laboratory,Shanghai 200438,People’s Republic of China

    Abstract This paper reports the use of machine learning to enhance the diagnosis of a dusty plasma.Dust in a plasma has a large impact on the properties of the plasma.According to a probe diagnostic experiment on a dust-free plasma combined with machine learning,an experiment on a dusty plasma is designed and carried out.Using a specific experimental device,dusty plasma with a stable and controllable dust particle density is generated.A Langmuir probe is used to measure the electron density and electron temperature under different pressures,discharge currents,and dust particle densities.The diagnostic result is processed through a machine learning algorithm,and the error of the predicted results under different pressures and discharge currents is analyzed,from which the law of the machine learning results changing with the pressure and discharge current is obtained.Finally,the results are compared with theoretical simulations to further analyze the properties of the electron density and temperature of the dusty plasma.

    Keywords: dusty plasma,machine learning,Langmuir probe

    1.Introduction

    Dusty plasma is a plasma system formed by adding dust particles to a plasma [1].The dust particles become charged due to the nature of the plasma itself [2].Dusty plasmas are often not intentional but sometimes formed intentionally for certain applications.Because the properties of charged dust particles are different from those of electrons and ions,dusty plasma has many unique characteristics [3].It plays an important role in the field of space and engineering applications [4,5].In carrying out laboratory research on dusty plasma,it is important to understand the specific parameters of the dusty plasma [6,7],especially its electron density and electron temperature [8].The diagnostic method is therefore at the heart of experimental research on dusty plasma,so that improving the diagnostic method is of great significance [9].

    The very existence of the dust particles makes the diagnosis of dusty plasma in the laboratory a very challenging problem[10].Commonly used methods include spectrum diagnosis[11],microwave diagnosis[12],and probe diagnosis[13].In spectrum diagnosis,the fiber probe can usually diagnose only outside the plasma area[14].Dust particles in the dusty plasma interfere with the spectrum that the probe can receive,so that the measured spectrum cannot accurately reflect the real situation inside the dusty plasma.For microwave diagnosis,since the dust particles carry electric charges,they will,like electrons,interact with the electromagnetic waves passing through the plasma,causing attenuation of the waves and introducing errors in the measurement of parameters.For probe diagnosis [15],because the diagnostic system applies a voltage to the probe during the diagnostic process,the charged dust particles will adhere to the surface of the probe,preventing the probe from collecting the current in the plasma[10].Because the dust attachment process is relatively random,accurate corrections cannot be made during the diagnostic process and the probe data cannot be calculated correctly [16].So,the dust particles and especially the fact that they are charged prevent the accurate diagnosis of dusty plasma using conventional diagnostic methods.Therefore,improving the diagnostic methods is an important task in the study of dusty plasma [17].

    At the same time,machine learning methods have been used to solve the Boltzmann equation of weakly ionized plasma[18].In plasma diagnosis,Jonathan Chalaturnyk[19]studied the feasibility of machine learning to enhance the diagnosis of dust-free plasma probes.In previous experiments,machine learning has also been successfully implemented to enhance the diagnosis of plasma probes [20].We conclude that improvement in diagnosing dusty plasma is feasible.Since the amount of dust in the dusty plasma has a direct influence on the diagnostic result,the use of machine learning to process the data obtained from the diagnosis and thereby to reduce the error of the result has great research value.At the same time,considering that in common probe diagnostic methods dust particles have relatively little influence,probe diagnosis can be used as the data source for machine learning in order to study parameters such as the electron density and temperature of the dusty plasma.

    Here we present the results of a study in which diagnosing dusty plasma is combined with machine learning and probe diagnosis.The diagnostic result of a smaller dust density is input into the machine learning algorithm; then the result for a larger dust density is predicted,and the predicted result is analyzed.By comparing the results of machine learning with that of fluid model,the reliability of the machine learning algorithm is further verified,and the properties of electron density and temperature of the dust plasma are further analyzed by an improved probe diagnosis method.In section 2,the formulas and principles used in the simulation are introduced,and the machine learning algorithm is applied.In section 3,the experimental device and the process of carrying out the experiment are explained in detail,along with the specific research methods.In section 4,the results of the experiment are analyzed in detail and are compared with the results of a dust-free plasma.Finally,in section 5,conclusions are presented.

    Figure 1.Schematic diagram of discharge device: (a) anode,(b) cathode,(c) dust particle,(d) Langmuir probe.

    2.Setting up the experiment

    In this section,the plasma generator,parameter selection,and some early data processing methods are introduced.

    2.1.Dusty plasma generator and parameter selection

    In order to obtain results of the initial dusty plasma probe diagnosis,the dusty plasma generator used by Ding [21] is adopted,as shown in figure 1.The experimental device can completely trap the dust particles in the curved part of the tube and thereby control the distribution of dust particles in the plasma.Because the density of dust particlesndis difficult to control accurately in the experiment,it is impossible to analyze accurately the influence of dust density on the experimental results.But by injecting a known amount of dust particles in this experimental device,the density of local dust particles can be changed regularly with changing the pressure and voltage,although the density of dust cannot be accurately known.Furthermore,accurate collection of probe data is realized,which provides better data for the prediction of the algorithm.

    Figure 1 gives the overall structure of the plasma generator.It is bent 40 cm glass tube with an inside diameter of 3 cm.The electrode spacing is 25 cm.Some dust particles are placed in the glass tube,and after the discharge starts,they become suspended in the plasma to form a dusty plasma.The dust particles are made of aluminum oxide,there are about 200–300 dust particles in the bend.The dust particles are spherical and have a diameter of 5 μm.The plasma gas is helium,and DC glow discharge is used as the plasma environment in which the dust is suspended.The probe is at the curved part of the glass tube to measure the dusty plasma density and temperature in the positive column in that area.The probe used for measurement is the Impedans commercial probe system.By using high temperature to soften the wall of the glass tube,and passing the probe through the wall,the air tightness of the device is not affected.The plasma powersupply is a CE 1500 005T programed power-supply,which can directly read the power-supply current and voltage and continuously adjust its output parameters.The maximum power-supply voltage is 1500 V,and enough data can be obtained for machine learning training and verification.A 100 kW resistor is connected in series in the circuit.

    In the probe diagnosis of plasma [22],due to the difficulty in measuring discharge current,the pressure and the total voltage of the circuit were selected as the input parameters of the algorithm for analysis.When diagnosing dusty plasma,by improving the accuracy of discharge current diagnosis,more basic discharge current can be selected as the input parameter.Therefore,discharge current and gas pressure are used as input parameters,and electron temperature and electron density are used as output parameters.The diagnostic results at a specific dust particle density are used to train the machine learning algorithms,in order to obtain diagnostic results at other dust particle densities.

    2.2.Implementation details of machine learning

    Because of its scalability and data-fitting ability,the multilayer perceptron (MLP) algorithm,with excellent performance in machine learning,was selected.The number of input nodes of the multilayer perceptron,=d2,corresponds to the pressure and voltage of the plasma.Similarly,the number of nodes in the output layer,=q2,corresponds to the electron temperature and electron density.Through a large number of experimental measurements,the model achieves the best performance when the number of network layers is =l1.The optimal number of nodes in each hidden layer ish1=20,h2=40,h3=40,andh4=20.The code is implemented by Pytorch 1.0 in Python,and the computing device is a desktop computer with Intel(R) Core(TM) i5-6500 CPU @ 3.20 GHz and NVIDIA GeForce GTX 1060 3 GB.To achieve better convergence of the model and avoid gradient disappearance,Gaussian regularization is adopted for both input and output data.We take the electron density data measured by experiments as the true value as the standard to evaluate the accuracy of data prediction.

    The treatment method is as follows:

    wherexis the data vector to be regularized,xregis the result of regularization,andTis the length of the vector.To measure the accuracy of the predicted results,an accuracy calculation method Ac is designed:

    where(R∈0),1 is the maximum acceptable threshold,is the number of predicted data meeting condition equation(2)in the test phase.Ois the predicted result on the test set and is explained in detail in section 3.2.?Ois the true value of test data,andPtotalis the number of test data,while ?OandOare all the original values which are not regularized by equation (1).

    3.Theories of dusty plasma and machine learning

    After choosing the device to be used in the experiment and the selected parameters,the data obtained through the experiment need to be input into the machine learning algorithm for learning and training,and the predicted result needs to be compared with the simulation result of COMSOL Multiphysics software.In this section,the formulas for dusty plasma simulation and the machine learning algorithm that we used are introduced.

    3.1.Model formulation of dusty plasma

    We used Liang’s improved plasma fluid model [23] with the addition of the dust-charging process,and used helium as the gas.Simulations were performed using the finite element method in COMSOL Multiphysics software,which allows relatively simple solutions to complex problems in various fields.Below we briefly introduce the equations of the dusty plasma fluid model system.

    The absorption term of each charged particle by dust particles is added to the plasma particle continuity equation:

    where

    k=(e,i) represents electrons or ions,nkis the density of electrons or ions,Skrepresents the sources and sinks caused by the plasma chemical processes,Ikis the corresponding electron or ion charging current,Γkis the flux density,μkandDkare,respectively,the corresponding mobility and diffusion coefficients,andEis the electric field,zkis the number of charges of the charged particles in the plasma.

    Maxwellian electron energy distribution function was used for the calculation of rate constants in this work.The electron energy balance equation is written as

    where

    Hereneis the electron energy density and Γεis the heat flux density,εDis the electron energy diffusion coefficient.The first term on the right side of equation (6) expresses the thermal conductivity and the second term describes heat transfer due to electron drift in the electric fieldE.In equation (5),εSis the electron energy change due to the elastic and inelastic collisions.Energy loss of a single electron to a dust particle is assumed to beeφd.

    The spatial distribution of the electric field is determined by the electric potential,obtained from the modified Poisson equation

    whereε0is the vacuum dielectric constant.

    The dust particle radiusrd=5μm.Assuming that at the center of the dust particle area,the electron density isne=4.01×1015m?3,the ion density isni=4.03×1015m?3,and the concentration of neutral molecules isng=1.56×1022m?3.

    We use the above formula to simulate dusty plasmas with different pressures,different discharge currents,and different dust particle densities,and compare the results obtained with the results obtained by the algorithm to further illustrate the feasibility of the algorithm and analyze the properties of electron density and temperature of the dusty plasma.

    3.2.Theories of machine learning

    MLP,also known as artificial neural network,is a neural network composed of fully connected layers with at least one hidden layer.The output of each hidden layer of the multilayer perceptron is transformed by the activation function so that the neural network can acquire the ability to fit a nonlinear function.Specifically,a small batch of sampleX∈Rn×dis given,with batch sizenand number of inputsd.For a MLP ofllayers,the number of hidden layers is ?l1,and the number of neurons in each hidden layer ishi.We let the output of each hidden layer beHi∈Rn×hi,and the weight and bias parameters of a hidden layer be,respectively,Wi∈Rhi?1×hiandbi∈Rn×hi,so the output of the networkO∈Rn×qcan be written

    whereqis the number of neurons in the output layer andφis the activation function.In our experiment,LeakyRelu [24] is selected as the activation function to avoid the problem of a vanishing gradient.This is expressed as

    whereais hyperparameters,we follow the original value of LeakyRelu[24],which is set to 0.01 in the experiment.Mean square error (MSE) is the most commonly used loss function in regression problems.It is the mean of the sum of squares of the difference between the predicted value and the target value,and can be expressed as:

    whereLis the loss function and ?Ois the true value of training data.In order to avoid falling into a local optimum in the training process,momentum parameters are added into the batch gradient descent,and the parameter update rule of the network is whereη> 0 is the learning rate,μis the hyperparameter of momentum,and ?Lmse(θt)is the gradient at the tunable hyperparameterθt(includingWandB).

    Figure 2.Distribution of probe measurement data.

    4.Result

    In the experiment,the plasma parameters under different dust injection volume,discharge currents,and pressures were measured.Regarding the distribution of pressure and discharge current in the probe data,as shown in figure 2,repeated measurements were made for different dust injection volumes,and 10%of the data were selected as the test set,the rest of the data being the training set.In the end,the machine learning algorithm was used to predict the electron temperature and electron density in a larger parameter range.The specific results are shown in figure 3.By fixing the volume of dust injected,the parameters of voltage and pressure will affect the dust density.Although it is difficult to directly control the dust density,the effect of the continuously changing dust density on the experimental results can clearly be seen and the machine learning algorithm allows the dust density to be simply used as an input parameter in subsequent applications.

    Figure 4 shows the loss rate and accuracy rate of the machine learning algorithm as functions of the number of iterations.In figure 4(a),under the 10% standard,the accuracy rate of the electron density is finally stable at 92.23%,and under the 30% standard,the final accuracy can reach 100%,indicating that all data errors are less than 30%,according to the probe equation,the electron density can be calculated:

    whereAis the surface area of the probe,andIesis the electron saturation current.

    Figure 3.Distribution of predicted data: (a) electron density,(b) electron temperature.

    Figure 4.Various results as functions of the number of iterations.(a)Accuracy of electron density,(b)accuracy of electron temperature,(c)loss rates of electron density and electron temperature.

    In the diagnosis of dust plasma,due to the adhesion of dust particles to the surface of the probe,Awill be reduced,but the originalAis still used in the calculation process,which causes the calculated electron density to be less than the true value.This is the reason why the accuracy rate of the electron density cannot reach 100%.As shown in figure 5,in the dust plasma,the electron saturation current measured by the probe decreases slightly,and the voltage that reaches the saturation current increases.The accuracy of parameters is corresponding to the situation of the probe surface influenced by dust.

    For the electron temperature,according to the equation of the probe,the electron temperature can be calculated

    whereIpis the probe current,VBis the probe potential,and ΦPis the plasma space potential.

    Figure 5.I–V characteristic curve of the probe with free-dust and dust.

    Figure 6.Error of each group of verification data: (a) electron density,(b) electron temperature.

    There is noAterm in this equation,and it can be considered that the electron temperature diagnosis process is less affected by dust particles.Therefore,the machine learning algorithm can get good prediction results.One can see the expected results in figure 4(b).At the same time,for the same reason,the loss rate of electron temperature is also significantly lower than that of the electron density(figure 4(c)).

    To get the error of the test set,we divide the data into two parts.The detailed error of the data of the test set is shown in figure 6.Due to the influence of dust on the probe,the error of the electron temperature is significantly less than that of the electron density.But as can be seen in figure 7,for large-scale prediction data,the distribution of the prediction results in figures 7(b) and (d) is well summarized and reflects the distribution of training data in figures 7(a)and(c).Especially for the electron temperature,the distribution of exploration data in the training set is relatively insignificant,and machine learning algorithms can also achieve good prediction results.In the experimental results,the law of electron density and temperature changing with pressure and discharge current is not clear.At the same time,due to the random nature of dust pollution,the law is difficult to formulate.But machine learning algorithms can reproduce this law well and make it more obvious.At the same time,figure 6(a) shows the obvious periodic distribution of the error distribution in the test set.

    In order to study the periodic distribution of errors in detail,when the pressure is 120 Pa,discharge current is selected as the variable for studying the prediction data,and compared with the test set.Figure 8 shows that the difference between the predicted value and the measured value of the electron density gradually decreases when the discharge current increases.Combining with the experimental phenomenon shown in figure 8(c),it can be seen that when the discharge current increases,the dust density of the measuring part decreases.From equation (13) it can be concluded that when the dust density decreases,the surface area of the probe is less affected during the measurement,so the deviation between the measured value and the predicted value is also reduced.In the same way,the error of the electron temperature in figure 8(b) should remain stable(equation(14)),and the law of the prediction error changing with the discharge current is in good agreement with the experimental phenomenon and theory.One can compare the results of the modified plasma fluid model with the probe results after the machine learning correction.As shown in figure 9,compared with the predicted data and the simulated data,a better match is achieved.Evidently,machine learning has achieved a good prediction of electron density and electron temperature,and successfully demonstrated the influence of dust particles on the plasma diagnostic results.

    Figure 7.Comparison of predicted data and training data: (a) training data of electron density,(b) predicted data of electron density,(c) training data of electron temperature,(d) predicted data of electron temperature.

    5.Conclusions

    In the traditional probe diagnosis method,when diagnosing dust plasma,dust particles will be adsorbed on the surface of the probe during the diagnostic process,which causes the diagnostic result of the probe to deviate from the true value.The adsorption process of dust particles is random,so this effect cannot be corrected physically.For this reason,we extended the machine learning algorithm applied to the traditional probe theory to the probe diagnosis of dust plasma.

    Our probe diagnosis used dusty plasma and processed the measurements using machine learning algorithms.The results show that for the electron density,under the 10% standard,a high accuracy rate cannot be achieved,while the electron temperature has a better accuracy rate.Through the analysis of the principle of probe diagnosis,we believe that such a result conforms to the rule governing the influence of dust particles on the probe.When comparing the predicted results and the measured results in detail,we found that the error showed a periodic distribution.In order to try to understand the reason,the change of the error was observed in detail,with the discharge current as the independent variable.By comparing with the dust density in the dusty plasma experiment,the law of change of the error and the law of change of the dust density were in good agreement.Finally,the results are compared with the results of the fluid mechanics model,and better results are obtained than for the data measured by the probe.We conclude that the machine learning algorithm shows great advantages in the diagnosis of dusty plasma.While revealing the dust’s influence on plasma,it can also correct this influence to a certain extent,achieving a good correction effect on the probe.

    Figure 8.Differences in data:(a)electron density,(b)electron temperature.Image dust particle of different discharge currents:(c)5.9 mA,(d)8.8 mA.

    Figure 9.Comparisons of predictions and simulations.

    Acknowledgments

    The research has been financially supported by National Natural Science Foundation of China (Nos.11775062,11805130 and 11905125) and the Shanghai Sailing Program (Nos.19YF1420900 and 18YF1422200).

    ORCID iDs

    猜你喜歡
    王瑩
    Valley-dependent topological edge states in plasma photonic crystals
    鋼琴性能對音樂創(chuàng)作風(fēng)格的影響
    音樂探索(2022年2期)2022-05-30 21:01:37
    巧用比較策略,突破學(xué)生的學(xué)習(xí)難點
    王瑩作品
    王瑩作品賞析
    王瑩作品
    王瑩作品賞析
    萊儷青年藝術(shù)獎獲獎?wù)?王瑩:《租賃一平方米》的力量
    Theoretical research on the transport and ionization rate coefficients in glow discharge dusty plasma
    王瑩作品
    www日本在线高清视频| av福利片在线| 欧美激情极品国产一区二区三区| 亚洲 国产 在线| 一级a爱视频在线免费观看| 亚洲av电影在线观看一区二区三区| 婷婷色综合大香蕉| 国产精品一区二区在线观看99| 精品国产一区二区三区四区第35| 亚洲午夜精品一区,二区,三区| av一本久久久久| 成年美女黄网站色视频大全免费| av天堂在线播放| 中文字幕av电影在线播放| 久久久久国产一级毛片高清牌| 欧美亚洲 丝袜 人妻 在线| 欧美日韩综合久久久久久| 精品人妻在线不人妻| 热re99久久精品国产66热6| 久久 成人 亚洲| 国产精品 欧美亚洲| 男人操女人黄网站| 在线看a的网站| 国产精品二区激情视频| 午夜影院在线不卡| 在线观看www视频免费| 免费一级毛片在线播放高清视频 | 亚洲男人天堂网一区| 国产在线一区二区三区精| 日本欧美国产在线视频| av天堂久久9| 国产97色在线日韩免费| 18在线观看网站| 最近最新中文字幕大全免费视频 | 桃花免费在线播放| 日韩人妻精品一区2区三区| 在线天堂中文资源库| 亚洲欧美色中文字幕在线| 中文乱码字字幕精品一区二区三区| 欧美黄色淫秽网站| 久久精品国产亚洲av高清一级| 国产精品一区二区在线观看99| 国产精品二区激情视频| av欧美777| 搡老岳熟女国产| 婷婷成人精品国产| 无限看片的www在线观看| 制服人妻中文乱码| 国产亚洲午夜精品一区二区久久| 别揉我奶头~嗯~啊~动态视频 | 亚洲精品一区蜜桃| 一区二区三区四区激情视频| 午夜老司机福利片| 男女边摸边吃奶| 国产成人影院久久av| 国产精品久久久久久人妻精品电影 | 国产主播在线观看一区二区 | 在线看a的网站| 亚洲男人天堂网一区| 日韩欧美一区视频在线观看| 日韩制服丝袜自拍偷拍| 国产高清视频在线播放一区 | 黄网站色视频无遮挡免费观看| 王馨瑶露胸无遮挡在线观看| 精品一区二区三区av网在线观看 | 一本久久精品| 国产99久久九九免费精品| 50天的宝宝边吃奶边哭怎么回事| 91麻豆av在线| 99久久精品国产亚洲精品| 色网站视频免费| 90打野战视频偷拍视频| 亚洲av电影在线进入| 精品国产一区二区三区四区第35| 精品人妻一区二区三区麻豆| 91成人精品电影| 韩国精品一区二区三区| 国产欧美日韩一区二区三区在线| 精品国产超薄肉色丝袜足j| 老司机亚洲免费影院| 国精品久久久久久国模美| 91老司机精品| 国产日韩一区二区三区精品不卡| 成人18禁高潮啪啪吃奶动态图| 91精品国产国语对白视频| 真人做人爱边吃奶动态| 最近手机中文字幕大全| 中文字幕av电影在线播放| 在线观看免费高清a一片| 搡老乐熟女国产| 黑人欧美特级aaaaaa片| 一级片'在线观看视频| 久久久久视频综合| 亚洲综合色网址| 熟女少妇亚洲综合色aaa.| 大片电影免费在线观看免费| av网站在线播放免费| 少妇人妻 视频| 欧美xxⅹ黑人| 国产免费福利视频在线观看| 色网站视频免费| 夫妻性生交免费视频一级片| 日韩熟女老妇一区二区性免费视频| 男女床上黄色一级片免费看| av在线老鸭窝| 91成人精品电影| 一级黄片播放器| 国产片特级美女逼逼视频| 2021少妇久久久久久久久久久| 夜夜骑夜夜射夜夜干| netflix在线观看网站| 中文字幕av电影在线播放| 看免费av毛片| 亚洲,一卡二卡三卡| 国产av国产精品国产| 日日爽夜夜爽网站| 午夜日韩欧美国产| 午夜福利视频精品| 啦啦啦在线免费观看视频4| 高清不卡的av网站| 亚洲国产精品一区三区| 黄色a级毛片大全视频| 久久国产亚洲av麻豆专区| 精品国产国语对白av| 99国产精品99久久久久| 免费一级毛片在线播放高清视频 | 成年女人毛片免费观看观看9 | 18在线观看网站| 久久天躁狠狠躁夜夜2o2o | 日韩免费高清中文字幕av| 欧美精品啪啪一区二区三区 | 男的添女的下面高潮视频| 美女高潮到喷水免费观看| 国产精品.久久久| 纵有疾风起免费观看全集完整版| 国产欧美日韩综合在线一区二区| 免费在线观看黄色视频的| 一本色道久久久久久精品综合| 亚洲国产欧美在线一区| 欧美日韩亚洲国产一区二区在线观看 | 国产精品一区二区免费欧美 | 免费久久久久久久精品成人欧美视频| 亚洲精品国产一区二区精华液| √禁漫天堂资源中文www| 亚洲成国产人片在线观看| 成人国语在线视频| 午夜免费鲁丝| 亚洲国产欧美网| 99九九在线精品视频| 人妻人人澡人人爽人人| a级毛片黄视频| 悠悠久久av| 久久精品久久久久久久性| 亚洲五月色婷婷综合| 亚洲五月色婷婷综合| 久久久久久免费高清国产稀缺| 热99国产精品久久久久久7| 欧美乱码精品一区二区三区| 午夜激情av网站| 国产精品久久久久久精品古装| 97在线人人人人妻| 精品一区二区三卡| 宅男免费午夜| 国产伦理片在线播放av一区| 久久国产精品大桥未久av| 国产亚洲精品久久久久5区| 男女之事视频高清在线观看 | 日本一区二区免费在线视频| 欧美激情极品国产一区二区三区| 国产成人精品在线电影| 免费观看人在逋| 亚洲激情五月婷婷啪啪| 日韩制服骚丝袜av| 欧美精品av麻豆av| 国产免费一区二区三区四区乱码| 国产亚洲av高清不卡| 亚洲国产精品国产精品| 精品一品国产午夜福利视频| 成年人免费黄色播放视频| 99国产综合亚洲精品| 久久久国产精品麻豆| 中文字幕人妻丝袜制服| 免费女性裸体啪啪无遮挡网站| 51午夜福利影视在线观看| 男女之事视频高清在线观看 | 日韩人妻精品一区2区三区| 久9热在线精品视频| 18禁黄网站禁片午夜丰满| 亚洲精品国产一区二区精华液| xxxhd国产人妻xxx| www日本在线高清视频| 国产精品国产三级国产专区5o| 日韩制服丝袜自拍偷拍| 日韩欧美一区视频在线观看| 免费少妇av软件| 考比视频在线观看| 欧美日本中文国产一区发布| 夫妻性生交免费视频一级片| 欧美精品一区二区大全| 99热国产这里只有精品6| 欧美日韩亚洲国产一区二区在线观看 | avwww免费| 高清视频免费观看一区二区| 欧美日韩综合久久久久久| 麻豆av在线久日| bbb黄色大片| 欧美少妇被猛烈插入视频| 久久国产亚洲av麻豆专区| 水蜜桃什么品种好| 好男人电影高清在线观看| 久久精品熟女亚洲av麻豆精品| 一二三四社区在线视频社区8| 肉色欧美久久久久久久蜜桃| 久久性视频一级片| 狂野欧美激情性xxxx| 精品一区二区三区av网在线观看 | tube8黄色片| 亚洲国产精品999| 一级毛片黄色毛片免费观看视频| 亚洲av国产av综合av卡| 少妇人妻 视频| 精品一区在线观看国产| 国产成人精品久久二区二区免费| 欧美日韩精品网址| 久久久久久人人人人人| 日本欧美视频一区| 亚洲精品美女久久av网站| 永久免费av网站大全| 亚洲欧洲国产日韩| 男女边吃奶边做爰视频| 真人做人爱边吃奶动态| 久久久久久人人人人人| 亚洲精品美女久久av网站| 超色免费av| 精品国产一区二区久久| 久久ye,这里只有精品| 在线观看免费高清a一片| 秋霞在线观看毛片| 波多野结衣av一区二区av| 国产精品二区激情视频| 男女免费视频国产| 亚洲国产av影院在线观看| 国产99久久九九免费精品| 欧美日韩精品网址| 亚洲国产精品国产精品| 黄色视频不卡| 欧美日韩福利视频一区二区| 50天的宝宝边吃奶边哭怎么回事| 国语自产精品视频在线第100页| av中文乱码字幕在线| 亚洲精品国产精品久久久不卡| 欧美在线一区亚洲| 国产日本99.免费观看| 日日摸夜夜添夜夜添小说| 动漫黄色视频在线观看| 国产一区在线观看成人免费| 日韩三级视频一区二区三区| 一个人观看的视频www高清免费观看 | 免费高清在线观看日韩| 一边摸一边抽搐一进一小说| aaaaa片日本免费| 精品国产乱子伦一区二区三区| 国产黄片美女视频| 最好的美女福利视频网| 女性生殖器流出的白浆| 人人妻人人澡人人看| 91老司机精品| 又紧又爽又黄一区二区| 色在线成人网| 亚洲性夜色夜夜综合| 久久久国产成人免费| 国产人伦9x9x在线观看| 白带黄色成豆腐渣| 日韩成人在线观看一区二区三区| 成人亚洲精品一区在线观看| 婷婷丁香在线五月| 中文资源天堂在线| 身体一侧抽搐| 99热只有精品国产| 18禁裸乳无遮挡免费网站照片 | 亚洲国产欧美网| 老司机靠b影院| 在线免费观看的www视频| 日韩av在线大香蕉| 成年免费大片在线观看| 久久人妻av系列| 少妇 在线观看| 搡老岳熟女国产| 国产精品av久久久久免费| 免费在线观看成人毛片| 麻豆一二三区av精品| 91成人精品电影| 听说在线观看完整版免费高清| 国产成人一区二区三区免费视频网站| 啦啦啦观看免费观看视频高清| 在线观看免费视频日本深夜| 亚洲美女黄片视频| 露出奶头的视频| 99精品久久久久人妻精品| 韩国av一区二区三区四区| 午夜免费成人在线视频| 国产欧美日韩精品亚洲av| 午夜免费激情av| 午夜久久久久精精品| 夜夜看夜夜爽夜夜摸| 久久精品国产亚洲av香蕉五月| 亚洲黑人精品在线| 人妻久久中文字幕网| 免费无遮挡裸体视频| 婷婷六月久久综合丁香| 制服丝袜大香蕉在线| 午夜激情福利司机影院| 日韩大尺度精品在线看网址| 一边摸一边做爽爽视频免费| 男人舔女人下体高潮全视频| 少妇裸体淫交视频免费看高清 | 宅男免费午夜| 久久性视频一级片| 精品国内亚洲2022精品成人| 亚洲国产欧美日韩在线播放| 久久久水蜜桃国产精品网| 亚洲一卡2卡3卡4卡5卡精品中文| 欧美中文日本在线观看视频| 99久久99久久久精品蜜桃| 自线自在国产av| 国产一区二区激情短视频| 久9热在线精品视频| 国产精品一区二区免费欧美| 黑人巨大精品欧美一区二区mp4| 久久精品国产亚洲av香蕉五月| 精品久久久久久久末码| 级片在线观看| 在线观看www视频免费| 中文字幕人妻丝袜一区二区| 国产极品粉嫩免费观看在线| 中文资源天堂在线| 巨乳人妻的诱惑在线观看| 99国产精品一区二区蜜桃av| 国内久久婷婷六月综合欲色啪| 成人精品一区二区免费| 一级黄色大片毛片| 欧美av亚洲av综合av国产av| 黄色视频不卡| 久久久久九九精品影院| 色av中文字幕| 美女国产高潮福利片在线看| 国产av一区二区精品久久| 91国产中文字幕| 黑人操中国人逼视频| 国产欧美日韩一区二区三| 国产国语露脸激情在线看| 757午夜福利合集在线观看| 精品人妻1区二区| 天天添夜夜摸| 91大片在线观看| 日本免费一区二区三区高清不卡| 久久人妻福利社区极品人妻图片| 一级作爱视频免费观看| 国产亚洲精品综合一区在线观看 | 欧美日韩精品网址| avwww免费| 视频在线观看一区二区三区| 亚洲人成网站高清观看| 久久久久久免费高清国产稀缺| 国产一区二区激情短视频| 欧美成人一区二区免费高清观看 | 亚洲欧美精品综合一区二区三区| 亚洲久久久国产精品| 日本撒尿小便嘘嘘汇集6| 国产精品1区2区在线观看.| 男女下面进入的视频免费午夜 | e午夜精品久久久久久久| 女人爽到高潮嗷嗷叫在线视频| 伊人久久大香线蕉亚洲五| 亚洲狠狠婷婷综合久久图片| 少妇的丰满在线观看| 久久精品91蜜桃| 国产精品精品国产色婷婷| 欧美中文综合在线视频| 亚洲国产高清在线一区二区三 | 国产熟女午夜一区二区三区| 美国免费a级毛片| 国产欧美日韩一区二区三| 亚洲欧美激情综合另类| 夜夜看夜夜爽夜夜摸| 亚洲九九香蕉| 91大片在线观看| 久久人人精品亚洲av| 日韩av在线大香蕉| 我的亚洲天堂| 久久国产乱子伦精品免费另类| 一边摸一边抽搐一进一小说| 国产精品野战在线观看| 国内揄拍国产精品人妻在线 | 中文字幕av电影在线播放| 欧美+亚洲+日韩+国产| 日韩av在线大香蕉| 99精品在免费线老司机午夜| 中文字幕人成人乱码亚洲影| 麻豆成人午夜福利视频| 欧美激情久久久久久爽电影| 国产一区二区三区在线臀色熟女| 国内精品久久久久久久电影| 欧美国产精品va在线观看不卡| 日韩欧美三级三区| 欧美丝袜亚洲另类 | 欧美大码av| 久久国产精品影院| 亚洲真实伦在线观看| 亚洲七黄色美女视频| 成人18禁高潮啪啪吃奶动态图| 一区二区三区高清视频在线| 极品教师在线免费播放| 可以免费在线观看a视频的电影网站| 在线观看66精品国产| 欧美日韩中文字幕国产精品一区二区三区| 亚洲人成电影免费在线| 又黄又粗又硬又大视频| 中文字幕人妻丝袜一区二区| 一级作爱视频免费观看| 夜夜躁狠狠躁天天躁| 久久久久亚洲av毛片大全| 欧美zozozo另类| 国产精品一区二区免费欧美| 国产av在哪里看| 啪啪无遮挡十八禁网站| 国产在线精品亚洲第一网站| 韩国精品一区二区三区| 欧美性猛交╳xxx乱大交人| 精品久久蜜臀av无| 久久国产精品影院| 国产三级黄色录像| 日韩欧美在线二视频| 丝袜在线中文字幕| 国产亚洲精品久久久久5区| 国产片内射在线| 久久久久久人人人人人| 国内毛片毛片毛片毛片毛片| 国产精品久久久久久人妻精品电影| 成年女人毛片免费观看观看9| 国产精品免费视频内射| 亚洲熟妇熟女久久| 国产精品亚洲av一区麻豆| 激情在线观看视频在线高清| 美女高潮喷水抽搐中文字幕| 亚洲欧美激情综合另类| 国产精品99久久99久久久不卡| 亚洲熟妇中文字幕五十中出| 日韩一卡2卡3卡4卡2021年| 白带黄色成豆腐渣| 黄色视频不卡| 午夜久久久在线观看| 韩国精品一区二区三区| 国产成人影院久久av| netflix在线观看网站| 久久精品国产亚洲av香蕉五月| 久久香蕉激情| 国产免费男女视频| 深夜精品福利| www国产在线视频色| 999精品在线视频| 免费看十八禁软件| 变态另类成人亚洲欧美熟女| 一个人观看的视频www高清免费观看 | 两个人看的免费小视频| 久久欧美精品欧美久久欧美| 嫁个100分男人电影在线观看| 欧美乱码精品一区二区三区| 亚洲第一青青草原| 性欧美人与动物交配| 999久久久精品免费观看国产| 女性生殖器流出的白浆| 亚洲avbb在线观看| 亚洲全国av大片| 国产精华一区二区三区| 日韩精品中文字幕看吧| 久久久久久国产a免费观看| 黄片播放在线免费| avwww免费| 性欧美人与动物交配| 神马国产精品三级电影在线观看 | 午夜免费激情av| 天天一区二区日本电影三级| 国产欧美日韩精品亚洲av| 欧美成人免费av一区二区三区| 不卡一级毛片| 51午夜福利影视在线观看| 50天的宝宝边吃奶边哭怎么回事| 精品久久久久久,| 亚洲成av人片免费观看| 国产精品一区二区精品视频观看| 亚洲精品国产精品久久久不卡| 午夜久久久在线观看| 亚洲 国产 在线| 国产成人影院久久av| 亚洲一区中文字幕在线| 亚洲中文av在线| 久久久久精品国产欧美久久久| 女人高潮潮喷娇喘18禁视频| 国产欧美日韩一区二区精品| 亚洲欧美精品综合一区二区三区| 欧美精品啪啪一区二区三区| 国产精品98久久久久久宅男小说| 哪里可以看免费的av片| 特大巨黑吊av在线直播 | 国内毛片毛片毛片毛片毛片| 国产激情欧美一区二区| 老汉色av国产亚洲站长工具| 久久久久国产一级毛片高清牌| 午夜免费成人在线视频| 国产真实乱freesex| 国产视频一区二区在线看| 日日干狠狠操夜夜爽| 一级毛片高清免费大全| 日韩欧美三级三区| svipshipincom国产片| 精品久久久久久,| 亚洲成a人片在线一区二区| 国产精品综合久久久久久久免费| 国产精品久久视频播放| 欧美 亚洲 国产 日韩一| 日韩视频一区二区在线观看| 中文字幕av电影在线播放| 男女午夜视频在线观看| 亚洲人成电影免费在线| 国产亚洲精品久久久久久毛片| 少妇裸体淫交视频免费看高清 | 国产精品久久视频播放| 老司机深夜福利视频在线观看| 淫秽高清视频在线观看| 中文字幕高清在线视频| 欧美日韩亚洲综合一区二区三区_| 超碰成人久久| 日本 欧美在线| 亚洲精品一卡2卡三卡4卡5卡| 桃红色精品国产亚洲av| 国产高清视频在线播放一区| 操出白浆在线播放| 国产一级毛片七仙女欲春2 | 亚洲中文字幕日韩| 别揉我奶头~嗯~啊~动态视频| 每晚都被弄得嗷嗷叫到高潮| 美女 人体艺术 gogo| 少妇粗大呻吟视频| av福利片在线| 91字幕亚洲| 非洲黑人性xxxx精品又粗又长| 国产精品久久久久久亚洲av鲁大| 久久这里只有精品19| 午夜免费成人在线视频| 成年免费大片在线观看| 亚洲国产精品999在线| 精品日产1卡2卡| 国产主播在线观看一区二区| 国产精品永久免费网站| 精品国产美女av久久久久小说| 国产aⅴ精品一区二区三区波| 岛国视频午夜一区免费看| 看免费av毛片| 色综合站精品国产| 久久精品国产综合久久久| 少妇裸体淫交视频免费看高清 | 精品久久久久久久久久免费视频| 国产精品一区二区三区四区久久 | www.自偷自拍.com| 99久久精品国产亚洲精品| 波多野结衣av一区二区av| 成熟少妇高潮喷水视频| 黄色女人牲交| 91老司机精品| 亚洲中文av在线| 国产熟女xx| 国产黄色小视频在线观看| 亚洲色图av天堂| 嫁个100分男人电影在线观看| 亚洲一区二区三区色噜噜| 久99久视频精品免费| 国产精品影院久久| 可以免费在线观看a视频的电影网站| 中文字幕人成人乱码亚洲影| 国产精品永久免费网站| 亚洲精品色激情综合| 国产精品野战在线观看| 极品教师在线免费播放| 色老头精品视频在线观看| 久久久久国内视频| 成人av一区二区三区在线看| 久久久精品国产亚洲av高清涩受| 久久人人精品亚洲av| 校园春色视频在线观看| 国产91精品成人一区二区三区| 免费观看精品视频网站| 嫩草影院精品99| 岛国在线观看网站| 免费看十八禁软件| 12—13女人毛片做爰片一| 97超级碰碰碰精品色视频在线观看| 久久国产精品男人的天堂亚洲| 国产精品 欧美亚洲| 中文字幕人妻丝袜一区二区| 99国产精品99久久久久| 日本免费a在线| 变态另类成人亚洲欧美熟女| 亚洲一卡2卡3卡4卡5卡精品中文| 亚洲自拍偷在线| 黄色丝袜av网址大全| 久久久水蜜桃国产精品网| 国产亚洲精品一区二区www| 色播亚洲综合网| 久久久久久免费高清国产稀缺| 一进一出抽搐动态| 一级a爱片免费观看的视频| 高清在线国产一区| 亚洲精品美女久久av网站| 香蕉久久夜色|