• <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 ?

    Modeling and Analysis of Production Logistics Spatio-Temporal Graph Network Driven by Digital Twin

    2022-12-09 14:22:22ZHENGLonghui鄭龍輝SUNYicheng孫奕程ZHANGHuihui張輝輝BAOJinsong鮑勁松CHENXiaochuan陳曉川ZHAOZhenhong趙振洪CHENZhonghao陳鐘浩GUANRuifeng管瑞峰
    關鍵詞:勁松

    ZHENG Longhui(鄭龍輝), SUN Yicheng(孫奕程), ZHANG Huihui(張輝輝), BAO Jinsong(鮑勁松), 2*, CHEN Xiaochuan(陳曉川), ZHAO Zhenhong(趙振洪), CHEN Zhonghao(陳鐘浩), GUAN Ruifeng(管瑞峰)

    1 College of Mechanical Engineering, Donghua University, Shanghai 201620, China

    2 Shanghai Frontier Science Research Center for Modern Textiles, Donghua University, Shanghai 201620, China

    3 Guangzhou Zhijing Information Technology Co., Ltd., Guangzhou 510220, China

    Abstract: In the process of logistics distribution of manufacturing enterprises, the automatic scheduling method based on the algorithm model has the advantages of accurate calculation and stable operation, but it excessively relies on the results of data calculation, ignores historical information and empirical data in the solving process, and has the bottleneck of low processing dimension and small processing scale. Therefore, in the digital twin (DT) system based on virtual and real fusion, a modeling and analysis method of production logistics spatio-temporal graph network model is proposed, considering the characteristics of road network topology and time-varying data. In the DT system, the temporal graph network model of the production logistics task is established and combined with the network topology, and the historical scheduling information about logistics elements is stored in the nodes. When the dynamic task arrives, a multi-stage links probability prediction method is adopted to predict the possibility of loading, driving, and other link relationships between task-related entity nodes at each stage. Several experiments are carried out, and the prediction accuracy of the digital twin-based temporal graph network (DTGN) model trained by historical scheduling information reaches 99.2% when the appropriate batch size is selected. Through logistics simulation experiments, the feasibility and the effectiveness of production logistics spatio-temporal graph network analysis methods based on historical scheduling information are verified.

    Key words: digital twin (DT); production logistics; job scheduling; spatio-temporal analysis

    Introduction

    In order to effectively improve the efficiency of production logistics and reduce the cost of production logistics, production logistics distribution as a key part of production operations, task allocation problems, and vehicle routing problems has been widely studied. In order to balance the logistics costs of suppliers and manufacturers, Yang and Wang[1]took the total logistics cost of the two-stage supply chain as the optimization target, and constructed a two-stage collaborative optimization model of delivery time and quantity from multiple suppliers to manufacturers, and from manufacturers’ warehouse to production lines. Yinetal.[2]constructed a double weight directed network model of workshop production logistics with time window, and studied the performance indices of workshop production logistics networks, such as node importance, stability, and self-repair ability. In order to solve the collaborative dynamic scheduling problem of customized workshop production and logistics, Caietal.[3]proposed a complete reactive scheduling information-physical system framework. Zhangetal.[4]studied energy saving oriented path planning for single-load automated guided vehicle (AGV) in discrete manufacturing shop environments. Aiming at the problems of high no-load ratios and low efficiency of stackers and AGVs in drug warehouse, Jiangetal.[5]established an integrated scheduling mixed command sequence operation time model with the goal of the shortest total operation time of orders in and out of warehouses. Chenetal.[6]proposed the multichromo-some genetic algorithm, which unified the solving algorithm of multi-car yards and multi-model problems with the traditional single-car yards and single-model problems. Aiming at the shortcomings of the combination of three-dimensional loading modes and transportation resource sharing modes in vehicle routing problem research, Wangetal.[7]proposed a vehicle routing optimization strategy based on transportation resource sharing under three-dimensional loading constraints. To sum up, in most studies, the process of logistics distribution optimization is as follows: firstly, multi-constraint modeling before task scheduling, then combined optimization of multi-task and multi-objective, and finally, monitoring and real-time control of the distribution process during transportation. However, in the long-term production and operation process, a large number of original scheduling data and manual scheduling experience have been accumulated, and the above research lacks the use of these data and experience.

    With the development of industrial internet, Internet of Things technology, big data and artificial intelligence technology, the emergence of intelligent technology is causing significant changes in today’s logistics and transportation[8]. Chenetal.[9]proposed a novel prediction method, namely the travel time difference model, which exploited the difference between the shortest travel time and the actual travel time to predict next locations. Chenetal.[10]used deep learning to predict the layout of bike parking spaces and dynamic demand for bike scheduling. Zheng[11]summarized the techniques of trajectory data preprocessing, management and spatio-temporal data mining. In order to solve the difficult problem of manufacturing shop scheduling caused by the complexity and changeability of the man-machine material method in the transportation scene, digital twin (DT), as a technical means, has attracted more and more attention along with the proposal of a digital twin workshop(DTW)[12]. DT technology can realize the interconnection and intelligent operation control of the physical world and the information world, and realize the interaction and integration between the physical world and the digital information world of manufacturing. DTW[13-14]consists of physical workshops, virtual workshops, workshop service systems and workshop twin data. Establishment criteria[15], evaluation criteria and evaluation index system[16]of a DT model provide reference for the decision-making process at all stages. Driven by twin data, the DT system realizes the iterative operation of task combination optimization, transportation route planning, and transportation process control of production logistics between the physical world and the upper logistics service system through the real mapping, real-time interaction and closed-loop control between the physical entity and virtual model of production logistics equipment, so as to achieve seamless and intelligent production process logistics of a new production logistics operation mode[12]. In recent years, the use of DT technology to solve intelligent scheduling problems in manufacturing enterprises has been studied by experts at different levels. Since DT is characterized by virtual reality interaction, real-time mapping, and symbiotic evolution, Fangetal.[17]proposed a new job-shop scheduling method based on DT to reduce scheduling bias, and proposed a scheduling resource parameter updating method and a dynamic interactive scheduling strategy to realize real-time accurate scheduling. Zhangetal.[18]designed DT-based control framework to obtain all the information needed for decision making. The DT-enhanced dynamic scheduling method proposed by Zhangetal.[14]performed availability prediction, interference detection and performance evaluation on workshop machines, and generated rescheduling with the support of real-time data of scheduling execution and stored DT data. Liuetal.[19]proposed a real-time data-driven online prediction method for running state of the DTW, and realized the online prediction of the workshop based on continuous transient simulation by integrating real-time data. In order to realize the production line simulation technology based on DT, Shietal.[20]proposed a DT logical model construction method for production line simulation. Many scholars have also introduced DT into enterprise logistics. In order to improve the whole life cycle management of goods distribution and optimize the distribution process, Dengetal.[21]integrated DT into distribution management and studied the distribution management system based on DT. Aiming at the problem of unclear system performance and low optimization degree in distributed scheduling of the multi-robot access system with multi-seeker vehicles, Sun and Zhao[22]proposed a centralized scheduling method based on DT. Aiming at the problem of the asynchronous of assembly execution and material distribution, Zhangetal.[23]proposed a material just-in-time distribution method based on grey theory and multi-model interaction mechanism for a satellite assembly DTW. It is a valuable exploration to apply DT to optimize production logistics distribution. Most of the existing research is from the aspects of system construction, scheduling methods, algorithm decision, and online prediction, but few types of research are using historical scheduling data and manual scheduling experience in DT.

    The logistics information of the whole process and all elements can be obtained by a DT system. If the traditional optimization algorithm is used, it is difficult to adapt to the huge amount of data, and the solution time is also difficult to control in an effective range. With the continuous growth of spatial data volume, spatial data contains huge value. Traditional data perception storage technology and processing and analysis methods have been unable to fully explore the value of massive spatial data[24]. The actual production logistics system can be regarded as an interactive system that changes dynamically over time, and the dynamic structure contains key information about the system. Graph network is widely used in relational and interactive systems, so we can use the memory module in the spatio-temporal graph network to construct a temporal map of logistics combined with historical experience data to explore the internal temporal and spatial relationships of the logistics system, and then use the model to predict the link relationships of workshops, roads, vehicles, and tasks. Taking full advantage of the characteristics of the DT system, such as accurate virtual-real mapping, real-time decision control and multi-source data fusion, this paper proposes a modeling and analysis method of production logistics spatio-temporal graph network, DT-based temporal graph network (DTGN), which is used to make full use of historical scheduling data and manual experience in the process of production logistics. Different from the previous research work mentioned above, the proposed method uses the memorized logistics data to predict the probability of events occurring in the next stage, so as to carry out reasonable path planning after matching engineering vehicles and tasks, and complete the process of distribution tasks.

    1 DT-Driven Engineering Vehicle Job Scheduling Framework

    The DT aims to digitize physical entities or systems, realize the fusion of physical information world driven by twin data, and simulate the operation process of physical entities in real situations with the help of twin data. At present, a large number of historical data exist in the operation of factory engineering vehicles, which lays a certain foundation for the application of DT technology in the job scheduling of engineering vehicles. The construction vehicle job scheduling infrastructure proposed in this paper based on DT is shown in Fig. 1.

    Fig. 1 DT-driven engineering vehicle job scheduling framework

    1.1 Physical space

    Physical space is a real logistics distribution system, which mainly receives transportation instructions from the model layer through a mobile communication module, performs corresponding workpiece transportation tasks, and collects real-time operation data of vehicles and maps by the data acquisition module. In the DT of logistics vehicle scheduling, the connection between physical space and digital space is established through virtual reality interaction and we use temporal graphs to map vehicle states, road states, workshop states, and task states. The use of virtual reality interaction can make traditional distribution transparent. The real-time monitoring and scheduling of the distribution process can meet the precise distribution requirements of production logistics.

    The sequence diagram is used to represent the collected logistics data. Logistics forecasting is time series analysis on the logistic map sequences of problems, including the addition or deletion of relationships such as distribution and driving, the addition or deletion of task nodes, and the prediction of feature transformation of nodes or relationships. This paper will get logistics data time series {X1,X2, …,Xt, 0

    1.2 Digital space

    Digital space provides a model base for the production logistics DT system to describe the generated logistics process and control the distribution process. The production logistics process was described digitally by the network topology, the temporal graph network, and the prediction model, and the DT model was formed driven by dynamic tasks. Based on the above model, task matching and vehicle route selection are carried out in the logistics process.

    The solution to scheduling problems is often time-consuming and laborious, and the solution obtained by an ordinary optimization algorithm cannot be directly applied to practical engineering problems, and the influencing factors considered are too abstract. Because DT can provide the whole process and the whole factor information of the logistics process, this paper proposes the method of a temporal graph to describe the process, and the core of the problem is how to reveal the spatio-temporal correlation in transportation data.

    1.3 Vehicle scheduling DT data representation

    In the DT of logistics vehicle scheduling, the connection between physical space and digital space is established through virtual reality interaction. The steps of virtual reality interaction are as follows. Firstly, various logistics data in physical space are acquired. Real-time logistics data is the basis to realize the interaction between physical space and digital space. Logistics data can be collected through radio frequency identification (RFID), surveillance cameras, Internet of Things and sensor networks. Secondly, the logistics data between physical space and digital space can be synchronized by establishing data transfer protocol. Finally, to express the spatio-temporal property of logistics-related entities in more detail, this paper presents the logistics data with a graph network. The temporal graph network is used to store and analyze the data in digital space. By analyzing the dynamic events in the logistics process, the scheduling strategy is used to adjust and update the logistics scheme. In the temporal graph, vehicles, roads, workshops, and tasks are mapped as nodes, and relationships between entities are mapped as edges. The node information is shown in Table 1, and the edge information is shown in Table 2.

    Table 1 Information of nodes

    Table 2 Information of edges

    As shown in Fig. 2, the dynamic process of logistics can be represented by a temporal graph.Sis the workshop or warehouse,Ris the internal transport road of the factory,Vis the transport of engineering vehicles, andTis the dynamic generation task. If there is a corresponding relationship between the two entities, namely related events, there will be an edge.

    Fig. 2 Logistic transportation scheduling temporal graph network

    2 Vehicle Job Scheduling Mathematical Model

    2.1 Problem description

    Production logistics is responsible for the integration and transfer of goods in the production base (such as raw materials, fuels, finished products, and waste), and plays a key role in connecting the preceding and the following in the factory base. The operation scheduling of engineering vehicles involves many persons, many vehicles, wide area, and complex distribution tasks. Engineering vehicles need to deliver materials to the workshop in time and transfer the finished products to avoid occupying inventory and affecting the takt time. Therefore, the operation scheduling of engineering vehicles is a link with high risk. To effectively arrange and dispatch all kinds of distribution tasks and transportation resources is the main research point of improving production logistics distribution in factories. Because the number of logistics engineering vehicles in the factory can reach thousands, and the flow momentum and flow distance of resources are relatively large, this paper mainly considers the scheduling strategy research of multiple transportation engineering vehicles in a limited range.

    In conclusion, the production logistics scheduling problem can be described as: in a production cycle, there are multiple transportation tasks dynamically generated, which need to be transported by engineering vehicles. Based on relevant constraints, transport tasks are assigned, and route planning is performed. The transportation vehicle is as far as possible to achieve the transportation optimization goal. In order to solve the optimization problem of engineering vehicle job scheduling, the following assumptions are considered.

    (1)The road planning map in the factory is known, including each distribution application point, distribution destination, and parking area of engineering vehicles.

    (2)Quantities, types, loading capacities, and speed limits of transportation resources in the factory are known.

    (3)Except for the speed control, the operation of construction vehicle drivers shall be subject to subjective operation, and other operations shall conform to the operation specifications of drivers stipulated by the factory.

    (4)The maximum mileage limit for transport vehicles is not considered.

    (5)The loading and unloading time of materials are not considered.

    (6)For the delivery task, the next task can be directly executed after the completion of a single task.

    (7)The distribution task is the smallest unit, which is no longer divided, and only one engineering vehicle is needed for transportation.

    2.2 Mathematical model

    On the premise that the above assumptions of the model are established, relevant definitions and mathematical models of engineering vehicle job scheduling are established as follows.

    Definition1Driving maps.G=(S,R) represents the road network topology in the factory, whereS={1, 2, …,sshop} represents the workshop set, andR={1, 2, …,rroad} represents the road sections between nodes, as shown in Fig. 3.

    Fig. 3 Road network diagram inside plant

    Definition2Transport vehicles. The set of existing transport vehicles isV={1, 2, …,vveshicle}, and each car has a corresponding average speed, an average load, and other attributes.

    Definition3Distribution tasks. The set of the existing distribution task isT= {1, 2, …,ttask}. The set of the request point isA= {1, 2, …,a} ?S. The set of the destination point isB= {1, 2, …,b}?S. The request points and the destination points have mapping relationships according to the distribution request. Each delivery task has corresponding delivery requirements, such as cargo weights and volumes.

    To facilitate description, parameters and symbols are defined in Table 3.

    Table 3 Definition of related parameters and symbols

    In the defined logistic graph sequenceXt, the job scheduling problem of multiple delivery pickup, unloading points and multiple engineering vehicles has multiple targetsfand multiple constraintsg. Therefore, the problem can be described as

    optFt=optff(f1,f2, …,fn)=optfv(St,Rt,Vt,Tt,Et,t),

    (1)

    s.t.

    gcons(St,Rt,Vt,Tt,Et,t)≤0,cons=1, 2, …,l.

    (2)

    Equation (1) represents the optimization objective. In the time range (tstart,tend), the whole logistics processFtneeds to be optimized. {f1,f2, …,fn} represents different optimization objective functions, andffrepresents the relationship function between these optimization objective functions. Each of the optimization objective functions is related to the basic logistics elements or decision variables {St,Rt,Vt,Tt,Et}, andfvrepresents the relationship function between these elements.Strepresents the change of workshop nodes with timet, including the progress of order task completion;Rtrepresents the change of road nodes with timet, including whether the road is usable;Vtrepresents the change of vehicle nodes with timet, including speeds, transportation tasks, and transportation routes;Ttrepresents the change of task nodes with timet, including the change of generation time of the workshop transport task;Etrepresents the change of the connection relationship between nodes with timet. Constraint (2) means that there arelconstraints, and they are all related to {St,Rt,Vt,Tt,Et}.

    To illustrate the problem, the following examples are given. The objective is to minimize the empty driving mileage of the used engineering vehicles and to balance the use of the dispatched engineering vehicles.

    (3)

    (4)

    (5)

    s.t.

    (6)

    (7)

    (8)

    (9)

    (10)

    Equation (3) represents the minimum empty driving mileage of engineering vehicles used in the distribution process. Equation (4) represents the balanced use of all logistics equipment involved in dispatching. Equation (5) represents the cargo mileage of vehiclee. Constraint (6) refers to the load constraint of each vehicle when the vehicle loads goods. Equation (7) indicates that each transport vehicle can only perform one task at a time, and only when the state is idle can the next job scheduling be performed. Equation (8) indicates that each delivery task is assigned to only one vehicle and can only be assigned once. Constraint (9) refers to the constraint of the number of vehicles in the job scheduling, and the number of vehicles in the job should be less than or equal to the total number of vehicles. Equation (10) indicates that engineering vehicles entering nodeiof the road network must leave from this node.

    As this problem is a nonlinear integer programming problem and a time-sequence related dynamic problem, it is difficult to solve by traditional methods when optimization objectives, problem constraints, and data volume increase. Therefore, this paper adopts the temporal graph network method to describe and solve the problem.

    3 Logistics Scheduling Strategy Based on DT

    3.1 Spatio-temporal analysis and multi-stage prediction

    To solve the scheduling problem of multiple vehicles in the dynamic environment, and to be able to adjust the logistics and transportation strategy adaptively in the dynamic environment, it is necessary to monitor the running status and task changes of transport vehicles in real-time, and provide real-time data support for the digital space. Combined with the road network topology, the constraints of task requirements, and the feedback data of the physical space, the logistics process can be adjusted in a timely manner. After getting data from physical space, logistics transportation is analyzed from spatial correlation and temporal correlation.

    3.1.1Spatialcorrelation

    Because the production scope of enterprises is relatively closed and the spatial layout is relatively fixed, the change of logistics vehicle transportation path is mainly affected by the distribution of workshops and the topology of transportation roads. Because of the different locations of the road, the frequency of use is different. The road near the production workshop or the main road is used more frequently, while some peripheral routes are shared by the main road when the task is heavy. When each transport vehicle runs on the public route, there will be the phenomenon of resource competition, which will interfere with each other and jam the traffic condition, so it is necessary to allocate the transport route reasonably in space.

    3.1.2Temporalcorrelation

    The amount of transportation tasks changes dynamically with time. Because workshop production has a production beat and the production cycle of each workshop is relatively fixed, the transportation tasks are also periodic, and the tasks are similar in each production cycle. Each workshop task affects the production status of other workshops through the transfer effect. There is a time window in the transportation process of each task. The arrival time of producing goods in the upstream workshop directly affects the production efficiency of the downstream workshop.

    To realize the dynamic adjustment of job scheduling, there will inevitably be all kinds of emergencies in the process of vehicle scheduling, such as new distribution tasks that need to be processed by inserting orders. Due to various reasons, such as urgent orders and vehicle faults, the scheduled distribution task cannot be executed normally, and the job scheduling of this task needs to be re-arranged. Tasks that are being executed or not executed cannot be executed normally due to vehicle failure. In the process of operation, due to the road condition, a section of road may not be passed normally, so it is necessary to re-plan the path. Due to manual operation, the vehicle fails to drive on the specified route, and the optimal route needs to be re-established after discovery. Given the above problems, this paper divides the time into multiple stages according to the sequence of events, and adopts the method of multi-stage prediction to minimize the impact of disturbances on production logistics.

    3.1.3Multi-stageprediction

    In the production logistics scheduling process, the process can be divided into several interrelated stages according to the sequence of events. Prediction is required at each stage for the entire logistics process to achieve optimal optimization goals. The logistics process is divided intorstages according to the sequence of events, and the statesrrepresents the situation at the beginning of each stage. The specific decision-making process is shown in the pseudo-code of Algorithm 1 in Fig. 4.

    Fig. 4 Algorithm 1: multi-stage prediction process

    3.2 Temporal graph network prediction method based on DT

    Traditional path optimization algorithms only calculate the shortest path distance between the start nodes and end nodes in the graph, and usually do not consider other information about paths, vehicles, and tasks. However, a temporal graph network can be embedded with the information of nodes and edge relationships, and a continuous-time dynamic graph can be represented as a list of events, which may include edge relationship addition or deletion, entity node addition or deletion, and node or edge feature transformation. Therefore, the temporal graph network can be used to carry out deep learning on the temporal graph which takes the time series of event occurrence as node relationships in transportation process.

    Combining the temporal graph network model and DT technology, this paper designs a DTGN prediction method to solve task assignment and path planning problems in DT logistics scheduling environment. The network prediction process of production logistics spatio-temporal graph network in the DT environment is shown in Fig. 5. The first step is to obtain the historical scheduling data and network topology in the DT environment. The second step is to obtain the arrival time of new taskTnewunder real-time mapping. The third step is to predict the probability of new taskTnewgenerating transport link relationship with each transport vehicle at a certain time by using a temporal graph network model. The fourth step is to predict the probability of driving link relationships between the vehicle and each road at a certain time after the vehicle is determined by transport probability, and select the road with the highest probability. The fifth step is to use the output information to predict the scheduling state.

    Fig. 5 Production logistics spatio-temporal graph network prediction based on DT

    Fig. 6 Training and prediction processes of DTGN

    The specific prediction method is that the graph neural network model is regarded as a pair of encoder-decoder, and the temporal graph network generated in the digital space is mapped to nodes through the encoder. The decoder takes the workshop-road-vehicle-task node and the definite moment as the input and predicts the link probability between nodes. In this paper, the temporal graph network proposed in Ref. [25] is used as the encoder. In the actual logistics transportation process, logistics transportation events are defined as a continuous-time dynamic graph, and the method of link prediction is adopted to predict the occurrence time and probability of future interactive events based on the observed process of logistics interaction events in the past in the logistics spatio-temporal graph. Figure 6 shows training and prediction processes of DTGN.

    The following is a detailed description of the DTGN components.

    (1)DT environment. On the one hand, the environment is to provide historical scheduling data, road network topology, and other data, and build the temporal graph network for training after data pretreatment. On the other hand, the link relationship predicted by the spatio-temporal relationship prediction model is transformed into the actual transport control instructions to control vehicle transport operations.

    (2)Store scheduling raw message. The state of all entity nodes (workshops, roads, vehicles, and tasks) in the digital space before timetis represented by vectorn(t).When an event occurs, the nodes corresponding will be updated, and using this module can retain the long-term spatio-temporal relationship of each node.

    (3)Message function. When interactive events (loading, unloading,etc.) occur between source nodeniand target nodenjin the logistics entity at timet, two messagesmiandmjabout the two nodes can be calculated by Eqs. (11) and (12). The information function is trained by multi-layer perceptronMLP.

    mi(t)=MLPs(ni(t-),nj(t-), Δt,E(ni,nj,t)).

    (11)

    mj(t)=MLPd(nj(t-),ni(t-), Δt,E(ni,nj,t)).

    (12)

    (4)Message aggregation. Because the same entity node may involve multiple events in the same period, for example, the same road may carry several cars at the same time, this paper usesmeanfunction to aggregate message in the period oft∈(tstart,tend).

    (13)

    (5)Memory update. Each event involving the node itself will update the memoryni(t) of the node, and the gated recurrent unitGRUis used as the memory update function.

    (14)

    (6)Embedding. The embedding module is used to generate the temporal embeddingzi(t) of nodeiat timet. The main goal of the embedded module is to avoid the so-called memory obsolescence problem and use the graph attention networkGATto embed operations.

    nj(t),E(ni,nj,t)).

    (15)

    After the above steps, the information of the predicted vehicle scheduling state can be obtained, that is, the matching probability between each node (task-vehicle, vehicle-road,etc.) at a certain time. If multiple transport vehicles have matching probability with taskTat timet, the selected vehicle isV=V(index (max(P1,P2,P3)).Thus, the matching relationship between different tasks, vehicles, and roads at different times can be obtained to guide the production logistics activities of manufacturing enterprises.

    4 Case Analysis

    The DT-driven engineering vehicle job scheduling framework described in this paper is applied to a steel factory in Shanghai, China. Iron and steel production process is the ferrous metal smelting and processing of industrial production activities. Raw materials include metal iron, chromium, manganese and other minerals. The production activities also need to transport fuel to provide energy, recycle waste, middleware produced in upstream workshops,etc. Production logistics refers to the transportation of raw materials, fuel, middleware, steel slag, scrap,etc. The process of material transportation in steel production can be described as follows. In the plant, there are multiple workshop units, multiple storage units, and a transportation yard. The engineering vehicle dispatching management center arranges the dispatching of engineering vehicles, and organizes and arranges multiple vehicles to collect goods from multiple distribution application workshops at the same time. The engineering vehicles will send the goods to the corresponding yard or other process workshop within the specified time according to the distribution request and scheduling arrangement. After finishing the delivery request, each engineering vehicle shall continue to perform the task according to the task arrangement or park in the nearby engineering vehicle parking area waiting for the next task deployment. The job scheduling of engineering vehicles is the process of allocating resources reasonably, optimizing distribution routes and completing distribution tasks efficiently. Since the workshops of the factory are scattered, and vehicles are mainly used to distribute materials, the rational use of engineering vehicles is an important part of ensuring the normal production of steel. In this paper, the actual production situation of the enterprise is investigated, combined with the actual production historical data of the factory and the simulation data in the test.

    According to the content mentioned above, the DT management platform as shown in Fig. 7 is constructed, which mainly provides data services for the entire vehicle scheduling process, including map information storage, vehicle information storage, task information release, and scheduling algorithm service integration. Firstly, the temporal graph network of logistics scheduling is established according to the transportation tasks of different workshops. Nodes store resource information and side stores transportation relationship information. Then, logistics scheduling strategy is used for task assignment. Finally, the temporal graph network is updated by feedback of entity node information. In the process of information iteration update, the optimization control of logistics vehicle transportation task is realized.

    Fig. 7 DT management platform

    The partial layout of the factory is shown in Fig.8(a). This area contains 16 workshop-level units and multiple storage units, including wharfs, power plants, blast furnaces, steel works, and warehouses. According to the actual transportation distance between each unit, the virtual simulation model shown in Fig.8 (b) is established to conduct simulation tests on the actual factory and generate virtual data. The experimental platform is implemented by Python language, and the experimental environment is Intel(R) Core(TM) i5 CPU, RAM 8 G, and Windows 10 operating system.

    Fig. 8 Factory layout: (a)physical space; (b)virtual space

    Fig. 9 Graph network representation of two vehicles

    4.1 Experimental settings

    4.1.1Logisticssampledata

    Based on the investigation of historical production data and transportation scheduling experience of enterprises, the following methods are determined to generate workshop-task data sets, task-vehicle matching data sets and vehicle-routing planning data sets. The specific settings of three types of data in this experiment are as follows.

    (1)Workshop-task data sets. Different workshops produce different tasks, which are reflected in goods types, weight, volume, distribution requirements,etc. In this paper, goods types and weight are selected for experimental settings, as shown in Table 4. ST01 to ST07 represent workshop types, and GT01 to GT07 represent goods types. To ensure the standardization of model input, the weight is normalized, and the transportation task is uniformly coded according to the time of task occurrence.

    Table 4 Workshop-task data sets

    (2)Task-vehicle matching data sets. The transportation process of enterprises relies on human experience to direct vehicles. Table 5 shows the heuristic rules to simulate human experience. The earliest task release time rule, the proximity principle rule, and the genetic algorithm (GA) are selected to produce different task-vehicle matching data sets.

    Table 5 Logistics task transport rules

    (3)Vehicle-routing planning data sets. Because the locations of workshops and roads are fixed and the quantity is known, the transportation road selection of goods transshipment between workshops remains basically unchanged, that is, the shortest path is basically determined, so Dijkstra method is used to select the shortest path every time.

    The experimental data sets above are used to simulate the historical production process of the enterprise, and is applied to train and test the temporal graph network model, so as to realize the probability prediction of the occurrence of links between two entity nodes in a given time in the future enterprise transportation graph network. Figure 9 shows the graph network to represent the status of two vehicles in the DT management platform.

    4.1.2Experimentalparameters

    A temporal graph prediction network based on the DT is built by using Python language and neural network algorithm library such as Pytorch. Module parameters are listed in Table 6 below.

    Table 6 Parameters of each module

    4.2 Determination of training batch size

    Graph network algorithm in different problem scenarios, and different hyperparameters, such as training periods and batch sizes, has a great influence on the training speed and experimental results. Experiments are done to verify that the prediction model is fast and accurate. Because the batch size has a great influence on the training speed and the algorithm memory update, and directly affects the convergence speed and the prediction accuracy of the model, this paper determines the batch size through the following experiments.

    Because the large batch size leads to a slow training speed, while the small batch size easily leads to the low prediction accuracy of the model, this experiment selects five batch sizes {50, 100, 150, 200, 250} as test values and selects the best value. Figure 10 shows the curve of the loss of the training set for 100 iterations and the convergence of the average accuracy on the test set with the number of iterations, showing the influence of different batch sizes on the loss and the accuracy.

    According to Fig. 10 and the actual training process, when the batch size is 100, the model training speed is fast and the accuracy on the test set is high, reaching 99.2%.

    Fig. 10 Effect of batch size: (a) loss of training set; (b) average accuracy of test set

    4.3 Validity verification of the method

    According to the actual production and the manufacturing situation, a variety of task data was designed to test the model concerning the task delivery request of the selected factory in a certain period. The detailed task information is shown in Table 7. The spatio-temporal analysis method based on the DT proposed in this paper was used to assign the transport task to five transport vehicles.

    In the constructed virtual experimental environment, the batch size is 100, experimental parameters are shown in Table 6, a temporal graph network model of 200 generations is trained as the decision core, and FIFO and GA algorithms of scheduling rules are selected as the control group. GA is aimed at minimizing the empty driving mileage of engineering vehicles and balancing the use of engineering vehicles participating in dispatching. The validity of the proposed spatio-temporal graph network analysis method for production logistics DT system is verified. Empty driving mileageLee, empty driving rateαand equilibrium standard deviationβof each vehicle are measured.αandβare defined by Eqs. (16) and (17).

    Table 7 Information of tasks

    (16)

    (17)

    As can be seen from Table 8, compared with the rules, DTGN can effectively reduce empty driving mileages and balance vehicle loads. However, the selection of transportation rules and the delivery of transportation instructions are relatively simple and can be completed manually. Compared with GA, DTGN reduces less empty driving mileages and has a poor equalization effect. However, it can complete the training process offline and give transportation instructions quickly in the production process. When the transportation process continues and the amount of data increases, GA becomes slower to solve, which is more conducive to training and prediction of DTGN. In addition, due to the reduction of total travel, the empty driving rate after GA and DTGN optimization increases.

    Table 8 Comparison of DTGN with other scheduling methods

    5 Conclusions

    This paper studies and analyzes the production logistics distribution process based on the DT using the spatio-temporal graph network and solves the problem of link relationship prediction in the logistics process. Its main contributions are as follows.

    (1) A DT-driven engineering vehicle job scheduling framework is proposed to describe the logistics process in the way of a sequence diagram, and a dynamic job scheduling process is established in the digital space.

    (2) Considering various constraints in the process of job scheduling, a mathematical model of the engineering vehicle job scheduling is established with the minimum empty driving mileage and the balanced use of dispatching vehicles as the optimization objective.

    (3) The temporal graph network model is established by using the historical scheduling data in the DT and the network topology structure. The spatio-temporal analysis of the historical data of logistics is carried out, and the DTGN algorithm is used to train the prediction model. When there is task input, the multi-stage prediction of the workshop-road-vehicle-task link relationship in the physical space is carried out. It can cooperate with the field dispatch personnel to issue transport instructions and achieve the reasonable scheduling of roads, vehicles, and tasks.

    The logistics distribution of a steel manufacturing enterprise is taken as an example to verify the feasibility of the method. There are still some deficiencies. The high-fidelity logistics transportation model has not been completely established, and only some constraints and optimization objectives are selected according to the actual situation. The scheduling experience and historical data collection have a great influence on the optimization performance of the training model. When the logistics environment changes, the model cannot adapt quickly.

    Given the above problems, the future research directions are as follows: for the more complex distribution environment, more diversified optimization objective functions are chosen to make the model prediction effect more in line with the needs of enterprises; reinforcement learning related content is introduced to optimize the data of the training model so that the effect of the trained prediction model is closer to the optimization goal; transfer learning related content is introduced to adjust the prediction model with the change of transportation scenarios and enhance generalization performance.

    猜你喜歡
    勁松
    顧勁松
    藝術家(2024年2期)2024-04-15 08:19:20
    莊勁松美術作品
    電影文學(2022年8期)2022-05-24 02:13:12
    “貓”忍不住
    故事會(2022年7期)2022-04-03 19:26:31
    貓忍不住
    小小說月刊(2022年1期)2022-02-07 10:44:22
    王勁松:行走在邊緣的一顆恒星
    時代郵刊(2019年20期)2019-07-30 08:05:38
    Simulation and experimental research of digital valve control servo system based on CMAC-PID control method①
    Isolated attosecond pulse generation with few-cycle two-color counter-rotating circularly polarized laser pulses?
    余勁松
    閱讀理解精練
    Gross Error Detection and Identification Based on Parameter Estimation for Dynamic Systems*
    久久久成人免费电影| 成人二区视频| 亚洲三级黄色毛片| 亚洲第一区二区三区不卡| av国产免费在线观看| 伦理电影免费视频| 水蜜桃什么品种好| 欧美成人午夜免费资源| 黄色一级大片看看| 免费观看av网站的网址| 日韩中字成人| 一级毛片电影观看| 国产免费视频播放在线视频| 国产成人aa在线观看| 久久久a久久爽久久v久久| 内射极品少妇av片p| 国产永久视频网站| 欧美少妇被猛烈插入视频| 国产在视频线精品| 国产一区有黄有色的免费视频| 免费大片黄手机在线观看| 中文字幕av成人在线电影| 五月玫瑰六月丁香| 国产精品熟女久久久久浪| 久久久久久久久久久丰满| 国产一区二区三区综合在线观看 | 国产av国产精品国产| 一级黄片播放器| 大又大粗又爽又黄少妇毛片口| 大陆偷拍与自拍| 欧美bdsm另类| 91精品伊人久久大香线蕉| 男人狂女人下面高潮的视频| 日韩国内少妇激情av| av在线观看视频网站免费| 有码 亚洲区| 热99国产精品久久久久久7| 新久久久久国产一级毛片| 国产精品.久久久| 亚洲熟女精品中文字幕| 欧美zozozo另类| 成人二区视频| 成人亚洲精品一区在线观看 | 久久久久精品性色| 久久婷婷青草| 亚洲精品,欧美精品| 精品一区在线观看国产| 国产69精品久久久久777片| 欧美变态另类bdsm刘玥| 亚洲精品,欧美精品| 我要看日韩黄色一级片| 成人二区视频| av天堂中文字幕网| 美女国产视频在线观看| 大香蕉97超碰在线| 秋霞伦理黄片| 久久热精品热| 亚洲国产精品一区三区| 国产精品久久久久久久久免| 性高湖久久久久久久久免费观看| 菩萨蛮人人尽说江南好唐韦庄| av卡一久久| 国产爽快片一区二区三区| 久久鲁丝午夜福利片| 亚洲欧美日韩无卡精品| 18禁在线播放成人免费| 国产精品伦人一区二区| 国产黄片美女视频| 免费不卡的大黄色大毛片视频在线观看| 一级毛片我不卡| 97在线人人人人妻| 波野结衣二区三区在线| 国产精品一及| 国产淫片久久久久久久久| 亚洲第一av免费看| 成人毛片60女人毛片免费| 黑人高潮一二区| 我要看黄色一级片免费的| videos熟女内射| 国产高清不卡午夜福利| 丰满人妻一区二区三区视频av| 能在线免费看毛片的网站| av线在线观看网站| 精品国产一区二区三区久久久樱花 | 2018国产大陆天天弄谢| 亚洲欧美一区二区三区国产| 亚洲精品一二三| 2022亚洲国产成人精品| 国产免费视频播放在线视频| 极品少妇高潮喷水抽搐| 午夜福利网站1000一区二区三区| 搡老乐熟女国产| 韩国av在线不卡| 伦精品一区二区三区| 国产一区二区三区av在线| 久久97久久精品| 特大巨黑吊av在线直播| 国产有黄有色有爽视频| 国产精品三级大全| 亚洲欧美日韩另类电影网站 | 成人国产av品久久久| 国产伦精品一区二区三区视频9| 嫩草影院新地址| 国产精品蜜桃在线观看| 一级毛片黄色毛片免费观看视频| 国产亚洲午夜精品一区二区久久| a级毛色黄片| 国产成人freesex在线| 日韩中字成人| 一区二区三区乱码不卡18| 最近最新中文字幕大全电影3| xxx大片免费视频| 国语对白做爰xxxⅹ性视频网站| 久久久a久久爽久久v久久| 午夜福利网站1000一区二区三区| 人人妻人人添人人爽欧美一区卜 | 午夜福利在线观看免费完整高清在| 日韩中文字幕视频在线看片 | 免费高清在线观看视频在线观看| 亚洲aⅴ乱码一区二区在线播放| 亚洲美女搞黄在线观看| 欧美激情极品国产一区二区三区 | 高清在线视频一区二区三区| 亚洲精品亚洲一区二区| 啦啦啦中文免费视频观看日本| 欧美极品一区二区三区四区| 人妻夜夜爽99麻豆av| 欧美国产精品一级二级三级 | 日日摸夜夜添夜夜添av毛片| 观看av在线不卡| 天天躁日日操中文字幕| 成人一区二区视频在线观看| 哪个播放器可以免费观看大片| 日日摸夜夜添夜夜爱| 亚洲图色成人| 边亲边吃奶的免费视频| 免费观看性生交大片5| 欧美高清性xxxxhd video| 精品国产一区二区三区久久久樱花 | 18禁裸乳无遮挡免费网站照片| 老女人水多毛片| 国产女主播在线喷水免费视频网站| 丰满乱子伦码专区| 亚洲精品456在线播放app| 人人妻人人澡人人爽人人夜夜| 色吧在线观看| 三级国产精品片| 日韩,欧美,国产一区二区三区| 一区二区av电影网| 精品一区二区三区视频在线| 久久久欧美国产精品| 少妇人妻一区二区三区视频| 最近手机中文字幕大全| 亚洲人成网站在线观看播放| 日韩精品有码人妻一区| 少妇人妻精品综合一区二区| 黄色日韩在线| 有码 亚洲区| 婷婷色av中文字幕| 欧美xxⅹ黑人| 久久6这里有精品| av在线app专区| 永久免费av网站大全| 久久综合国产亚洲精品| 亚洲第一av免费看| 久久精品久久久久久噜噜老黄| 亚洲综合精品二区| 亚洲av中文字字幕乱码综合| 国产成人91sexporn| 老司机影院成人| 亚洲欧美成人精品一区二区| 一级毛片 在线播放| 欧美xxxx性猛交bbbb| 久久久久久久久久成人| xxx大片免费视频| 久久99热6这里只有精品| 久久鲁丝午夜福利片| 亚州av有码| 99热6这里只有精品| 成人高潮视频无遮挡免费网站| 99热网站在线观看| 国产亚洲一区二区精品| 美女脱内裤让男人舔精品视频| 欧美另类一区| 精品亚洲乱码少妇综合久久| 亚洲精品aⅴ在线观看| 国产高清国产精品国产三级 | 爱豆传媒免费全集在线观看| 国产乱来视频区| 亚洲中文av在线| 日韩中文字幕视频在线看片 | 菩萨蛮人人尽说江南好唐韦庄| 精品一区二区三区视频在线| 女性生殖器流出的白浆| a级毛片免费高清观看在线播放| 国产精品久久久久久久久免| 日韩人妻高清精品专区| 亚洲,一卡二卡三卡| 免费黄频网站在线观看国产| 91aial.com中文字幕在线观看| 久久午夜福利片| 日韩一区二区视频免费看| 亚洲熟女精品中文字幕| 简卡轻食公司| 国产有黄有色有爽视频| 亚洲,一卡二卡三卡| 夜夜爽夜夜爽视频| 男女下面进入的视频免费午夜| 国内精品宾馆在线| 国产成人精品福利久久| 国产高清国产精品国产三级 | 有码 亚洲区| 欧美 日韩 精品 国产| 日本免费在线观看一区| 亚洲国产精品一区三区| 免费人成在线观看视频色| 日本av免费视频播放| 一本久久精品| 亚洲最大成人中文| 男女边吃奶边做爰视频| 精品人妻一区二区三区麻豆| 亚洲欧美日韩无卡精品| 日韩 亚洲 欧美在线| 黄色一级大片看看| 偷拍熟女少妇极品色| 97在线人人人人妻| 久久精品国产自在天天线| 在线 av 中文字幕| h日本视频在线播放| 有码 亚洲区| 高清毛片免费看| 日韩一本色道免费dvd| 久久6这里有精品| 午夜免费观看性视频| 精品国产乱码久久久久久小说| 在线 av 中文字幕| 美女脱内裤让男人舔精品视频| 99九九线精品视频在线观看视频| 国产一区有黄有色的免费视频| 成人黄色视频免费在线看| 99久久精品国产国产毛片| 伦精品一区二区三区| 亚洲av福利一区| 国产v大片淫在线免费观看| 三级国产精品欧美在线观看| 亚洲婷婷狠狠爱综合网| 在线播放无遮挡| 一级毛片 在线播放| 干丝袜人妻中文字幕| 久久人妻熟女aⅴ| 视频中文字幕在线观看| 99久久精品一区二区三区| 又大又黄又爽视频免费| 美女高潮的动态| 高清黄色对白视频在线免费看 | 在线亚洲精品国产二区图片欧美 | av天堂中文字幕网| 汤姆久久久久久久影院中文字幕| 五月天丁香电影| 久久鲁丝午夜福利片| 黑人猛操日本美女一级片| 精品人妻一区二区三区麻豆| 免费播放大片免费观看视频在线观看| 狂野欧美白嫩少妇大欣赏| 欧美最新免费一区二区三区| 最近中文字幕高清免费大全6| 亚洲av在线观看美女高潮| 亚洲精品日韩在线中文字幕| 亚洲综合精品二区| 人人妻人人澡人人爽人人夜夜| 久久影院123| av在线老鸭窝| 亚洲av男天堂| 久久久久久人妻| 免费人妻精品一区二区三区视频| 人妻少妇偷人精品九色| 免费观看在线日韩| 美女内射精品一级片tv| 亚洲国产日韩一区二区| 青春草亚洲视频在线观看| 欧美97在线视频| 亚洲欧洲国产日韩| 美女视频免费永久观看网站| 亚洲欧美精品专区久久| 在线免费观看不下载黄p国产| 99视频精品全部免费 在线| 国产综合精华液| 精品视频人人做人人爽| 久久久成人免费电影| 国产精品久久久久久精品古装| 在线看a的网站| 欧美日韩国产mv在线观看视频 | 日本一二三区视频观看| 亚洲最大成人中文| 国产精品一区二区在线不卡| 国产伦精品一区二区三区四那| 色婷婷av一区二区三区视频| 国产爽快片一区二区三区| 久久99蜜桃精品久久| 国产片特级美女逼逼视频| 三级经典国产精品| 国产综合精华液| 少妇的逼水好多| 18禁裸乳无遮挡动漫免费视频| 国产精品蜜桃在线观看| 日本午夜av视频| 在线免费观看不下载黄p国产| 身体一侧抽搐| 高清欧美精品videossex| 少妇高潮的动态图| 最近中文字幕高清免费大全6| 国产午夜精品一二区理论片| 日本猛色少妇xxxxx猛交久久| 97在线人人人人妻| 久久久久久久久大av| 久热久热在线精品观看| 91aial.com中文字幕在线观看| 51国产日韩欧美| 天堂中文最新版在线下载| 美女福利国产在线 | 日本猛色少妇xxxxx猛交久久| 老女人水多毛片| 亚洲怡红院男人天堂| 久久久久久久大尺度免费视频| 这个男人来自地球电影免费观看 | 噜噜噜噜噜久久久久久91| 免费少妇av软件| 成人国产av品久久久| 国产成人免费观看mmmm| 久久久久久久大尺度免费视频| 日本与韩国留学比较| 久久久久国产精品人妻一区二区| 欧美精品一区二区大全| 成年人午夜在线观看视频| 99精国产麻豆久久婷婷| 免费观看无遮挡的男女| 亚洲中文av在线| av在线播放精品| 尾随美女入室| 丰满迷人的少妇在线观看| 久久这里有精品视频免费| 自拍欧美九色日韩亚洲蝌蚪91 | 又爽又黄a免费视频| 欧美xxⅹ黑人| 精品午夜福利在线看| 日韩欧美精品免费久久| 97超碰精品成人国产| 丰满人妻一区二区三区视频av| 啦啦啦啦在线视频资源| 纵有疾风起免费观看全集完整版| 深爱激情五月婷婷| 国产乱人偷精品视频| 亚洲精品久久久久久婷婷小说| 婷婷色综合www| 久久久久国产网址| 美女国产视频在线观看| 久久精品国产亚洲网站| 欧美最新免费一区二区三区| 狂野欧美激情性xxxx在线观看| 国产深夜福利视频在线观看| 天天躁夜夜躁狠狠久久av| 毛片一级片免费看久久久久| h视频一区二区三区| 国产精品av视频在线免费观看| 丝袜脚勾引网站| 少妇高潮的动态图| 免费av中文字幕在线| 伊人久久国产一区二区| 午夜日本视频在线| 好男人视频免费观看在线| 赤兔流量卡办理| 日韩制服骚丝袜av| 少妇 在线观看| 老司机影院毛片| 丰满迷人的少妇在线观看| 亚洲国产毛片av蜜桃av| 日本午夜av视频| 乱系列少妇在线播放| 夫妻性生交免费视频一级片| 成人18禁高潮啪啪吃奶动态图 | 欧美精品一区二区免费开放| 国产精品久久久久成人av| 99re6热这里在线精品视频| 欧美另类一区| 高清在线视频一区二区三区| 在线看a的网站| 在线观看免费高清a一片| 又黄又爽又刺激的免费视频.| 国产精品成人在线| 熟女人妻精品中文字幕| h日本视频在线播放| 成人18禁高潮啪啪吃奶动态图 | 高清日韩中文字幕在线| 国产精品伦人一区二区| 天天躁夜夜躁狠狠久久av| 最后的刺客免费高清国语| 亚洲在久久综合| 欧美少妇被猛烈插入视频| 观看免费一级毛片| 男女下面进入的视频免费午夜| 免费在线观看成人毛片| 欧美xxⅹ黑人| 91久久精品国产一区二区成人| 午夜免费观看性视频| 亚洲天堂av无毛| 男人爽女人下面视频在线观看| 久久鲁丝午夜福利片| 一级a做视频免费观看| 内地一区二区视频在线| 2022亚洲国产成人精品| 国产精品久久久久成人av| 人妻制服诱惑在线中文字幕| 久久久久久久久久久丰满| 亚洲激情五月婷婷啪啪| 最近中文字幕高清免费大全6| videos熟女内射| 深夜a级毛片| 午夜福利影视在线免费观看| 久久鲁丝午夜福利片| 亚洲国产精品一区三区| 午夜视频国产福利| 亚洲欧美成人综合另类久久久| 人妻少妇偷人精品九色| 国产成人精品久久久久久| 嫩草影院入口| 久久毛片免费看一区二区三区| av国产免费在线观看| 嫩草影院入口| 国产伦在线观看视频一区| 国产伦精品一区二区三区视频9| 日韩一区二区三区影片| 国产免费一级a男人的天堂| 国产人妻一区二区三区在| 亚洲精品日韩在线中文字幕| 欧美成人精品欧美一级黄| 国产精品秋霞免费鲁丝片| av在线app专区| 国产亚洲91精品色在线| 国产精品国产av在线观看| 一级毛片久久久久久久久女| 亚洲av二区三区四区| 欧美成人一区二区免费高清观看| 天堂中文最新版在线下载| 黄色视频在线播放观看不卡| 亚洲精品aⅴ在线观看| 性色avwww在线观看| 色5月婷婷丁香| 26uuu在线亚洲综合色| 乱系列少妇在线播放| 成人亚洲欧美一区二区av| av在线播放精品| 内射极品少妇av片p| 精品一品国产午夜福利视频| 久久久久视频综合| 国产精品人妻久久久久久| 久久99热这里只频精品6学生| 久久久久久久久久久免费av| 男人和女人高潮做爰伦理| 男女边摸边吃奶| 性色av一级| 3wmmmm亚洲av在线观看| 涩涩av久久男人的天堂| 欧美少妇被猛烈插入视频| 99国产精品免费福利视频| 国产免费视频播放在线视频| 寂寞人妻少妇视频99o| 国产成人aa在线观看| 日本av免费视频播放| 99久久人妻综合| 日韩中文字幕视频在线看片 | 久久人人爽人人爽人人片va| 欧美日本视频| 国产成人免费无遮挡视频| 日本免费在线观看一区| 麻豆乱淫一区二区| 26uuu在线亚洲综合色| 少妇精品久久久久久久| 久久久久人妻精品一区果冻| 大香蕉久久网| 久久国内精品自在自线图片| 亚洲精品自拍成人| 18禁动态无遮挡网站| 18禁裸乳无遮挡动漫免费视频| av视频免费观看在线观看| 在线观看美女被高潮喷水网站| 亚洲经典国产精华液单| 老司机影院成人| 自拍偷自拍亚洲精品老妇| 久久精品国产亚洲av天美| 日韩中文字幕视频在线看片 | 99久久综合免费| 国产白丝娇喘喷水9色精品| 国产高清有码在线观看视频| 国产精品99久久久久久久久| 精品久久久久久久久av| 如何舔出高潮| 国产精品偷伦视频观看了| 熟女人妻精品中文字幕| av网站免费在线观看视频| 国产亚洲精品久久久com| 美女脱内裤让男人舔精品视频| 美女福利国产在线 | 国产av码专区亚洲av| av在线播放精品| kizo精华| 国产亚洲最大av| h视频一区二区三区| 精品国产乱码久久久久久小说| 亚洲av综合色区一区| 91午夜精品亚洲一区二区三区| 日本欧美国产在线视频| 免费大片黄手机在线观看| 成人18禁高潮啪啪吃奶动态图 | a级毛片免费高清观看在线播放| 亚洲最大成人中文| 国产精品嫩草影院av在线观看| 黄片无遮挡物在线观看| tube8黄色片| 18禁在线播放成人免费| 在线天堂最新版资源| 免费在线观看成人毛片| 美女国产视频在线观看| 日韩在线高清观看一区二区三区| 国模一区二区三区四区视频| 中文字幕av成人在线电影| 成年av动漫网址| 色哟哟·www| 久久鲁丝午夜福利片| 日韩亚洲欧美综合| 婷婷色综合大香蕉| 日本欧美视频一区| av国产精品久久久久影院| 男女免费视频国产| 国产精品久久久久久精品古装| 免费不卡的大黄色大毛片视频在线观看| 久久毛片免费看一区二区三区| 天天躁日日操中文字幕| 男女边摸边吃奶| av线在线观看网站| 国产视频首页在线观看| 波野结衣二区三区在线| 韩国av在线不卡| 国产成人免费观看mmmm| 色5月婷婷丁香| 免费观看的影片在线观看| 亚洲真实伦在线观看| 国产亚洲av片在线观看秒播厂| 国产精品精品国产色婷婷| 高清午夜精品一区二区三区| 啦啦啦啦在线视频资源| 大码成人一级视频| 免费大片黄手机在线观看| 在线免费十八禁| 婷婷色综合大香蕉| h视频一区二区三区| 美女福利国产在线 | 日本欧美国产在线视频| 五月伊人婷婷丁香| 99久久精品一区二区三区| 亚洲成人手机| 汤姆久久久久久久影院中文字幕| 精品亚洲成国产av| 国产精品女同一区二区软件| 一级毛片黄色毛片免费观看视频| 成人二区视频| 少妇的逼好多水| h日本视频在线播放| 亚洲精品成人av观看孕妇| av免费在线看不卡| 亚洲无线观看免费| 亚洲四区av| 久久国产乱子免费精品| 精品久久久噜噜| 免费观看无遮挡的男女| 亚洲精品乱码久久久久久按摩| 国产久久久一区二区三区| 一级毛片电影观看| 插逼视频在线观看| 美女高潮的动态| 又粗又硬又长又爽又黄的视频| 免费看光身美女| 国产片特级美女逼逼视频| 亚洲欧美成人综合另类久久久| 国产精品人妻久久久久久| 夜夜骑夜夜射夜夜干| 国产精品无大码| 久久婷婷青草| 欧美日韩国产mv在线观看视频 | 新久久久久国产一级毛片| 超碰av人人做人人爽久久| 亚洲欧美精品自产自拍| 国产91av在线免费观看| 久久久久人妻精品一区果冻| 亚洲欧美一区二区三区黑人 | 看免费成人av毛片| 51国产日韩欧美| 国产精品国产三级国产av玫瑰| 国产在线视频一区二区| 亚洲国产色片| 国产成人91sexporn| 十八禁网站网址无遮挡 | 日韩一区二区视频免费看| 国模一区二区三区四区视频| 国产一区二区三区av在线| 秋霞伦理黄片| 插阴视频在线观看视频| 国产真实伦视频高清在线观看| 日韩一区二区视频免费看| 一区二区三区乱码不卡18| 观看av在线不卡| 人人妻人人看人人澡| 国内精品宾馆在线| 亚洲美女视频黄频| 亚洲,欧美,日韩| 国产淫片久久久久久久久|