Vehicle social network caching method based on digital twinning

Abstract

The application relates to a vehicle social network caching method based on digital twinning, and belongs to the vehicle social network field. The method comprises the following steps: S1, extracting a contact link between a vehicle network and a social network; S2, mapping the vehicle network from a physical layer to a twinning network layer based on a digital twinning technology; S3, extracting a relationship feature of the vehicle from the twinning network layer and establishing a vehicle social network layer; S4, establishing a cache cloud in the vehicle social network layer; and S5, determining a vehicle cache cooperation strategy based on the vehicle social network layer and the cache cloud in the vehicle social network layer. The application effectively improves the content distribution utility of the vehicle-mounted network and has certain advantages in the utility and unloading success rate indexes.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of vehicle social networks, and relates to a vehicle social network caching method based on digital twinning. BACKGROUND

[0002] The rapid development of multimedia application services of vehicles leads to exponential growth of traffic in vehicle networking, and the demand of users for high-quality vehicle communication greatly increases, which makes the network burden more severe. In order to alleviate the problem of insufficient computing resources of vehicles, edge computing technology is introduced into vehicle networking to offload the tasks of vehicles. In the process of edge computing offloading, edge caching technology is utilized. Edge caching is to enable vehicles to have the ability of cache storage, which can reduce the communication energy consumption and other problems in the process of edge computing. In the edge caching vehicle networking system, the hit rate is an important evaluation index of data caching technology, and the transmission and placement of caches affect the hit rate of caches. Some studies discuss the influence of content placement in caches, so the placement of caches is also a content worth considering. Edge caching technology is to store content at each node with storage function in the network, so that vehicles can obtain the required content from the surrounding cache nodes. Although this increases the redundancy of content in the network, it improves the cache hit of content in the service process, reduces the load of the network, and shortens the time delay in the service.

[0003] In vehicle cooperative computing in vehicle networking, the traditional method generally directly considers the nearby vehicles for cooperation without considering the use of social attributes of vehicles to form a social network. In the cache strategy part, the cache content of the cache node and the matching of user requests can be combined, and then the movement trajectory of the vehicle is predicted according to historical information. However, due to the high dynamicity of vehicles, the intermittent connection between the content demanders and the content providers of vehicles can affect the propagation efficiency of socially-aware content. SUMMARY

[0004] Therefore, the present application aims to provide a vehicle social network caching method based on digital twinning, which can obtain the social relationship between vehicles and the current vehicle traffic distribution by using its own technology, more specifically allocate storage resources and content placement, combine the social attributes of users with hot content, and form a vehicle social network. Digital twinning can construct a multi-dimensional mapping from a physical system to a virtual network, and form a closed-loop system for data perception, efficient analysis and intelligent decision-making by means of the data analysis and modeling capabilities of digital twinning. In the Internet of Vehicles, digital twinning can help realize adaptive network management from digital simulation to visual evaluation. Using this technology can realize the whole network planning of multiple regions. The introduction of digital twinning computing offloading in the Internet of Vehicles can bring three advantages: first, real-time monitoring of vehicle conditions brings comprehensive and accurate network analysis of the physical network; second, mapping the physical network layer and the application layer to the twin network layer realizes cross-layer interaction and feedback of the three-layer structure; third, it can perceive the dynamic changes of the network topology and vehicle location.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A vehicle social network caching method based on digital twinning, comprising the following steps:

[0007] S1, extracting the contact link between the vehicle network and the social network;

[0008] S2, mapping the vehicle network from the physical layer to the twin network layer based on the digital twinning technology;

[0009] S3, extracting the relationship characteristics of the vehicle from the twin network layer and establishing a vehicle social network layer;

[0010] S4, establishing a cache cloud in the vehicle social network layer;

[0011] S5, determining the vehicle caching cooperation strategy based on the vehicle social network layer and the cache cloud therein.

[0012] Further, in step S1, the vehicle has social attributes affected by the social activities of the driver, the social relationship, user interest and social activity in the social network are the contact link between the vehicle network and the social network; the driver in the vehicle network is personalized recommended according to the user interest in the social network; the social relationship in the social network corresponds to the network topology in the vehicle network, the network topology in the vehicle network is predicted according to the social relationship, the network topology in the social network is maintained, and the social relationship in the social network is extended according to the network topology; the social activity in the social network is abstracted from the mobile model in the vehicle network, and the mobile model is predicted or discovered according to the group nature of the social activity.

[0013] Further, in step S2, in the physical layer, it is assumed that there are N moving vehicles and M roadside units RSUs on the road, the set of vehicles is defined as N = {1, 2, …, N}, and the set of RSUs is defined as M = {1, 2, …, M};

[0014] The demand content of the vehicle is divided into G types, and each type of demand content is represented as where f g represents the size of the content type, represents the maximum delay of its content, μ g represents the sensitivity coefficient of its delay;

[0015] Each RSU is equipped with an edge cache server, and each edge cache server has a cache capability; the demand content of the vehicle is pre-stored in the vehicle or the edge cache server of the RSU.

[0016] Further, in step S2, the information collection module, the control module, the RSU cooperation module, the instruction issuing module, and the digital twin data storage module are included in the twin network layer; the information collection module obtains the state of the vehicle through communication between the vehicle and the RSU; the control module is used to determine the update period and the interaction frequency and data type in the adjustment process, and the adjustment instruction is transmitted to the corresponding vehicle through the instruction issuing module; the RSU cooperation module is used to exchange data and update the state in time; and the digital twin data storage module is used to store data.

[0017] Further, in step S3, the LSTM network is used to extract the social relationship of the vehicle, which includes at least two relationship features, i.e., content matching of both parties and communication contact rate of the vehicle;

[0018] The content matching element is described by a probability β, and it is assumed that the vehicle needs G types of content, and the probability of the demand is β = {β1, β2, β3, …, β G}, where The probability of the demand content of the vehicle and the matching demand content existing in the communication area is β g ;

[0019] The communication contact rate depends on the distribution density and the driving speed of the vehicle in the area; it is assumed that the vehicle drives at a speed v, the specified speed in the section is v0, and the average distance between two consecutive vehicles is represented as:

[0020]

[0021] where d0 represents the safety distance between the two vehicles, T0 represents the reaction time of the vehicle, and δ represents the speed variation factor of the vehicle; according to the coverage range and the average distance of the communication area, the vehicle distribution density λ of the communication area is obtained;

[0022] Two relationship features of the social relationship of the vehicle are represented as {β, λ}.

[0023] Further, in step S4, each cache cloud is composed of several vehicles in this area, and each cache cloud includes one demand content; each vehicle includes several demand contents, and each demand content corresponds to a cache cloud; thus, a multi-dimensional mapping is formed between the cache cloud and the vehicle.

[0024] When the communication delivery is performed, the number of cached vehicles obtained by the content demander is restricted by two factors, one is the size of the cache content of the provider, and the other is the transmission rate between them.

[0025] The signal-to-noise ratio formula of the mixed deterioration communication V2X communication is:

[0026]

[0027] wherein SINR represents the signal-to-noise ratio, d represents the path length, a represents the path loss factor, p w represents the transmission power of the vehicle, and P v represents the additive noise power of the vehicle.

[0028] Suppose γ min is the minimum signal-to-noise ratio that can be decoded by the receiving end when receiving data, and the maximum distance of the V2X communication in the ideal environment without co-channel interference is represented as:

[0029]

[0030] The average rate of the vehicle V2X communication is defined as:

[0031] R v = Blog2(1+SINR)

[0032] wherein B represents the channel bandwidth.

[0033] Further, in step S5, the cache scheduling considers the balance among the communication cost, the model accuracy and the system utility.

[0034] Let x g represent the probability of pre-storing g type content on the cache cloud, and y g represent the probability of pre-storing g type content on the RSU server; the optimal edge cache problem of maximizing the utility of the cache system under the constraint conditions of the cache size and the delivery delay is represented as:

[0035]

[0036] wherein N represents the set of vehicles; M represents the set of RSUs; ψ g (εg represents the precision of the vehicle social network layer; ε g is the amount of system information collected in the digital twin for training the LSTM network and obtaining the vehicle social network layer; β g is the probability of encountering a vehicle that needs g-type content; represents the time for the vehicle to obtain g-type content from the cache cloud; represents the time spent by the vehicle to obtain content from the RSUr; y g,r represents the probability of pre-storing g-type content on the RSUr server; ζ v is the transmission cost of the vehicle v; ζ r is the transmission cost of the vehicle and the RSUr; ζ c is the transmission cost of the vehicle to the data center; {Z} is an indicator function, which is 1 when {Z} is true, and 0 when {Z} is false.

[0037] Further, in step S5, the optimal cache strategy is set in an iterative manner based on reinforcement learning to solve the optimal edge caching problem. In each iteration, the cache scheduling strategy is obtained according to the given vehicle social network layer, and the amount of information collected in the construction process of the vehicle social network layer is modified according to the determined strategy, and the iteration is continued until the system converges.

[0038] According to the given vehicle social network layer and the precision function ψ g (ε g ), the utility function of the optimal decision is represented as:

[0039]

[0040] The optimal cache strategy π * is represented as:

[0041]

[0042] where η ∈ (0, 1) represents a discount factor for balancing the utility function.

[0043] Further, in step S5, DDPG is applied to the edge cache cooperation management scenario. For a given cache strategy set as π, the average system utility obtained by the state S i and action A i is represented as:

[0044]

[0045] The DDPG algorithm comprises a main network and a target network; each network has an actor part and a critic part, the target network is regarded as an iteration of the main network, target values for training the main network are generated, and finally the strategy is updated by calculating a loss function; the experience replay pool of the DDPG algorithm stores learning experience of action reward and state change, and is used for training actor and critic parameters.

[0046] The present application has the following advantages:

[0047] Compared with the existing cache placement of the Internet of Vehicles, the social cache mechanism based on digital twinning proposed in the application effectively improves the content distribution utility of the vehicle-mounted network, and has certain advantages in utility and offloading success rate indicators. First, the digital twinning method is used to map the physical edge network to the virtual space, the user will cache the content, form a content cache cloud, and fully utilize the characteristics of digital twinning for real-time updating. Second, in order to build a social network, the social attribute classification will be carried out according to the characteristics of the cache and the behavior characteristics of the vehicle itself. Finally, the edge cache placement is carried out based on the social network. The social relationship model between vehicles is obtained. Further, the model is used to mine the cooperation between resource-constrained vehicles in large-scale content delivery, and a learning-based social relationship construction and cache resource scheduling scheme is proposed.

[0048] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, and will be learned from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to make the purpose, technical scheme and advantages of the present application more clear, the preferred detailed description of the present application will be combined with the drawings as follows, in which:

[0050] Figure 1 The schematic diagram of the contact of the vehicle network and the social network of the present application;

[0051] Figure 2 The schematic diagram of the social network architecture of the Internet of Vehicles based on digital twinning of the present application;

[0052] Figure 3 The schematic diagram of the construction process of the vehicle social network layer of the present application;

[0053] Figure 4 The schematic diagram of the vehicle cache cloud of the present application;

[0054] Figure 5 The schematic diagram of the edge cache strategy structure based on DDPG of the present application. DETAILED DESCRIPTION

[0055] The present application is described and explained with additional specificity and detail through the use of the accompanying drawings in which:

[0056] The accompanying drawings are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the application and, together with the description, serve to explain the principles of the application. In the drawings:

[0057] The same or similar components in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "front", "back" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only for illustrative purposes, and cannot be understood as a limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0058] Reference should be made to Figures 1-5 A vehicle social network caching method based on digital twinning includes the following specific steps:

[0059] S1, extracting the contact link between the vehicle network and the social network;

[0060] S2, mapping the vehicle network from the physical layer to the twin network layer based on the digital twinning technology;

[0061] S3, extracting the relationship features of the vehicle from the twin network layer and establishing the vehicle social network layer

[0062] S4, establishing a cache cloud in the vehicle social network layer;

[0063] S5, determining the vehicle caching cooperation strategy based on the vehicle social network layer and the cache cloud therein.

[0064] Further, in step S1, the driver of the vehicle is a person, and the original vehicle does not have social attributes, but since the person has social attributes and is affected by the social activities of the driver, the vehicle also has social attributes. Such social attributes make it easier for nodes with similar interests or social attributes to meet in a certain area. Considering the social attributes and social relationships between vehicles, the present application defines this network as a vehicle social network (VSN).

[0065] Specifically, the vehicle social network can share information in real time between vehicles and vehicles, vehicles and roadside units, and vehicles and storage devices. Since the high-speed mobility of vehicles can easily cause the interruption of communication links, the social attributes between vehicles need to establish stable connections. As shown in Figure 1 The social relationship, user interest and social activity are the connecting links between the two architectures. The user interest in the social network can be personalized in the vehicle network. The interests and hobbies of different users make the application personalized to the user himself, which promotes the iteration and update of the application. The social relationship in the social network corresponds to the network topology in the vehicle network. The social relationship can predict the network topology in the vehicle network, maintain the network topology in the social network, and extend the social relationship in the social network. Social activities can be abstracted by the movement model of the vehicle network, and the movement model can be predicted or discovered according to the group nature of social activities.

[0066] Further, in step S2, according to the corresponding relationship between the vehicle network and the social network in step S1, the present application proposes a vehicle networking social network architecture based on digital twinning as shown in Figure 2 The vehicle networking social network architecture based on digital twinning includes at least a physical layer and a twin network layer. In the physical layer, N moving vehicles and M roadside units (RSUs) are set on the road. The set of vehicles is defined as N={1, 2, …, N}, and the set of RSUs is defined as M={1, 2, …, M}. The contents required for the application implementation of the intelligent vehicle are divided into G types, and each type of content can be represented as, where f g The first parameter represents the size of the content type, the second parameter represents the maximum delay of its content, and the third parameter represents the sensitivity coefficient of its delay, that is, the unit time utility obtained by reducing the maximum delay during transmission.

[0067] In the vehicle networking architecture mentioned above, each RSU is equipped with an edge cache server, and each server has cache capability. Therefore, in order to avoid long delay between the data center and the vehicle, the server can store part of the content in the cache. In addition, the content can also be pre-stored in the intelligent vehicle.

[0068] Further, in the vehicle social network architecture proposed in step S2, the digital twin network is a digital mapping of the physical network layer, with control functions such as Figure 3 As shown, in general, the digital twin mainly includes three parts: data storage, twin model mapping, and twin management, the data storage is responsible for collecting and storing various network data, for virtual modeling, optimization and prediction. The twin management is responsible for managing and updating the twin network layer, with functions such as model updating, state synchronization, etc. Figure 3 The vehicle social network construction method based on digital twin is shown, the main functions of the digital twin network (twin network layer) are embodied into five modules: information collection module, control module, RSU cooperation module, instruction issuing and digital twin data storage. The information collection module mainly obtains the state of the vehicle through the communication between the vehicle and the RSU, generally including the position information, service capability, driving state and data demand of the vehicle. The digital twin network here adopts a distributed structure, which is generally established on the RSU in a region, so it is necessary to use the RSU cooperation module to exchange data and update the state in time, to ensure the consistency of the mapped digital twin model. The control module mainly determines the update period and the interaction frequency and data type in the adjustment process, the adjustment command will be transmitted to the intelligent vehicle through the instruction issuing module, so as to change the data type, transmission state and mode in time.

[0069] Further, in step S3, when the digital twin network is established in the application, some key features of the social relationship of the vehicle need to be extracted to construct the vehicle social network layer (vehicle social model). The application considers using LSTM to extract social features from the collected data. The vehicle social network model is mainly stored in the module of the digital twin data storage, and will be updated regularly according to the obtained data information.

[0070] Specifically, the vehicle with cache function can act as a content carrier and forward the cached data to the requesting vehicle through vehicle-to-vehicle communication (V2V). In cooperative offloading, in order to fully utilize the V2V content interaction, the social relationship of the vehicle is considered. When the content of the request side and the supply side is consistent, and the two parties can establish a transmission channel, the application calls it as having social association. Therefore, the social relationship of the vehicle has two relationship characteristics, one is the content matching of the two parties, and the other characteristic is the communication contact rate of the vehicle.

[0071] In the application, LSTM is used to extract the relationship features of the vehicle social network, such as Figure 3As shown in the figure, the LSTM model includes an input layer, a hidden layer, and an output layer. The output layer obtains vehicle traffic information and content demand information in different areas from the established digital twin, then passes this information to the neurons in the hidden layer, ultimately obtaining elements of the vehicle social network layer. Training is performed based on collected historical data, and the parameters of the neurons are periodically updated.

[0072] The present invention uses probability β to describe content matching elements. It is assumed that the vehicle in the system needs G content, and the probability of its demand is β = {β1, β2, β3, ..., β G},in When in an area with cached content, the probability of encountering a vehicle that needs this content is β g .

[0073] The communication contact rate generally depends on the distribution density and driving speed of vehicles in the area. Assuming that the vehicle is traveling at a certain speed v, the prescribed speed in this section is v0, and the average distance between two consecutive vehicles can be expressed as:

[0074]

[0075] In the formula, d0 represents the safe distance between two vehicles, T0 represents the vehicle's reaction time, and δ represents the vehicle's speed change factor. Based on the coverage and average distance of the communication area, the vehicle distribution density λ in the communication area is calculated. Therefore, the output content of a region in the vehicle social network can be expressed as {β,λ}.

[0076] Further, in step S4, as Figure 4 As shown, each cache cloud consists of multiple vehicles within the area, each responsible for carrying a specific type of content. Because each vehicle has a different content cache, it, like a person, has multiple social circles, resulting in multi-dimensional mapping relationships. The size of content storage is limited by its own storage space. Therefore, during communication, the number of vehicles a content demander can access cached content is constrained by two factors: the size of the provider's cached content and the transmission rate between them.

[0077] The communication between vehicles, between vehicles and roadside units, and between vehicles and storage devices is expressed as hybrid communication V2X. The signal-to-noise ratio formula of V2X communication is:

[0078]

[0079] Where SINR stands for signal-to-noise ratio, d stands for path length, α stands for path loss factor, and p w Represents the transmission power of the vehicle, P vrepresents the additive noise power of the vehicle. Therefore, γ min The maximum distance of V2X communication in an ideal environment without co-channel interference can be represented as:

[0080]

[0081] The average rate of vehicle V2X communication is defined as:

[0082] R v = Blog2(1+SINR)

[0083] where B represents the channel bandwidth.

[0084] Further, in step S5, the cache scheduling will rely on the vehicle social network layer obtained from the digital twin social network. The more information collected by the digital twin network, the higher the accuracy of the vehicle social network model. The collection of information is carried out in a V2X hybrid communication mode, which may generate additional communication costs, so the balance between communication costs, model accuracy and system utility needs to be considered in the scheduling of the cache.

[0085] Let x g represent the probability of pre-storing g-type content on the vehicle cache cloud, y g represent the probability of pre-storing g-type content on the RSU server. Therefore, according to the proposed optimal edge caching problem, the utility of the cache system is maximized under the constraints of cache size and transmission delay, which can be represented as:

[0086]

[0087] In the formula, N represents the set of vehicles, M represents the set of RSUs, ψ g (ε g ) represents the accuracy of the vehicle social network layer, ε g is the amount of system information collected in the digital twin for training the LSTM network and obtaining the vehicle social network layer, β g is the probability of exactly encountering a vehicle that needs g-type content, represents the time for the vehicle to obtain g-type content from the vehicle cache cloud, represents the time for the vehicle to obtain content from RSUr, y g,r represents the probability of pre-storing g-type content on the RSUr server, ζ v is the transmission cost of vehicle v, ζ r is the transmission cost of vehicle and RSUr, ζ cThe transmission cost requested by the vehicle to the data center.

[0088] To solve this problem, the application proposes an iterative method based on reinforcement learning, in each iteration, first obtain the cache scheduling strategy according to the given vehicle social network layer, then modify the information collected in the model construction according to the determined strategy, continue iteration, until the system converges.

[0089] According to the given vehicle social network layer and the precision function ψ g g , explore the optimal cache strategy under this model, the utility function in this environment is represented as:

[0090]

[0091] In order to maximize the utility of the edge cache system, the application seeks the optimal cache strategy π * , since the optimal strategy is affected by many factors, the application proposes a vehicle cache cloud architecture and edge cache scheme based on DDPG (deep deterministic policy gradient algorithm).

[0092]

[0093] Wherein, η∈(0, 1) represents the discount factor, used to weigh the utility function.

[0094] The architecture of DDPG is shown in Figure 5 The learning system of DDPG mainly includes two parts, namely the main network and the target network. These two networks have similar parts, each network has an actor part and a critic part, and both networks are deep neural networks. In the learning process of DDPG, the actor part of the main network mainly explores the edge cache strategy, and the critic network helps the actor to get better strategy through gradient method. The target network can be regarded as an iteration of the main network, generating the target value for training the main network, and finally updating the strategy by calculating the loss function. The experience replay pool stores the learning experience of action reward and state change, which is used to train the actor and critic parameters.

[0095] DDPG is applied to the scene of edge cache cooperation management, and Markov decision process adopts the form of reinforcement learning problem. The given cache strategy is set as π, the average system utility obtained under the state S i and action A i can be represented as:

[0096] ​

[0097] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A digital-twin-based vehicle social network caching method, characterized in that: The method comprises the following steps: S1, extracting the contact link between the vehicle network and the social network; S2, mapping the vehicle network from the physical layer to the twin network layer based on the digital twin technology; S3, extracting the relationship features of the vehicle from the twin network layer and establishing the vehicle social network layer; S4, establishing the cache cloud in the vehicle social network layer; S5, determining the vehicle cache cooperation strategy based on the vehicle social network layer and the cache cloud therein; In step S1, the vehicle has social attributes affected by the social activities of the driver, the social relationship, user interest and social activity in the social network are the contact link between the vehicle network and the social network; the driver in the vehicle network is personalized recommended according to the user interest in the social network; the social relationship in the social network corresponds to the network topology in the vehicle network, the network topology in the vehicle network is predicted according to the social relationship, the network topology in the social network is maintained, and the social relationship in the social network is extended according to the network topology; the social activity in the social network is abstracted from the mobile model in the vehicle network, and the mobile model is predicted or discovered according to the group nature of the social activity; In step S2, in the physical layer, it is set that there are N moving vehicles and M road side units RSUs on the road, the set of vehicles is defined as , and the set of RSUs is defined as ; The demand contents of the vehicle are divided into G classes, and each type of demand content is expressed as wherein represents the size of the content type, represents the maximum delay of its content, represents the sensitivity coefficient of its delay; Each RSU is provided with an edge cache server, and each edge cache server has cache capability; the demand content of the vehicle is stored in the vehicle or the edge cache server of the RSU in advance; In step S3, the social relationship of the vehicle is extracted by using the LSTM network, and at least two relationship features are included, which are content matching of the two parties and communication contact rate of the vehicle; with a probability The content matching element describes that the vehicle requires G-class content, and the probability of its requirement is wherein The probability of the vehicle's required content matching the presence of the required content in the communication area is ; The communication contact rate depends on the distribution density and the driving speed of vehicles in the area; it is assumed that vehicles drive at a speed of The average distance between two consecutive vehicles is denoted by wherein denotes the safety distance between two vehicles, denotes the reaction time of a vehicle, denotes the speed change factor of a vehicle; the vehicle distribution density of the communication area is determined from the coverage of the communication area and the average distance ; The two relationship features of the social relationship of the vehicle are represented as ; In step S4, each cache cloud is composed of a plurality of vehicles in the region, and each cache cloud includes a demand content; each vehicle includes a plurality of demand contents, and each demand content corresponds to a cache cloud; therefore, a multi-dimensional mapping is formed between the cache cloud and the vehicle; When the communication transmission is performed, the number of cached vehicles obtained by the content demander is restricted by two factors, one is the cache content size of the provider, and the other is the transmission rate between them; In step S5, the balance among the communication cost, model accuracy and system utility is considered in the cache scheduling; Let P (g | C) denotes the probability of pre-storing the g-th content on the cache cloud, P (g | RSU) denotes the probability of pre-storing the g-th content on the RSU server; the optimal edge caching problem that maximizes the utility of the caching system under the constraints of cache size and delivery latency is represented as: where N represents a set of vehicles; M represents a set of RSUs; represents the accuracy of the vehicle social network layer; represents the amount of system information collected in the digital twin for training the LSTM network and obtaining the vehicle social network layer; represents the probability of encountering a vehicle that needs class g content; represents the time for a vehicle to obtain class g content from the cache cloud; represents the time spent by a vehicle to obtain content from RSUr; represents the probability of pre-storing class g content on the RSUr server; represents the transmission cost of vehicle v; represents the transmission cost of vehicle to RSUr; represents the transmission cost of vehicle to the data center;{Z} is an indicator function that represents 1 when{Z} is true and 0 when{Z} is false; In step S5, the optimal cache strategy is set by using the iteration method based on reinforcement learning to solve the optimal edge cache problem, in each iteration, the cache scheduling strategy is obtained according to the given vehicle social network layer, the information amount collected in the vehicle social network layer construction process is modified according to the determined strategy, and the iteration is continued until the system converges; According to the given vehicle social network layer and precision function The utility function representing the optimal decision is explored as follows: Optimal caching strategy is represented as: wherein, denotes a discount factor, used to weigh the utility function; In step S5, DDPG is applied to the edge cache collaboration management scenario, for a given cache policy setting , the state and action , the average system utility obtained is represented as: The DDPG algorithm includes a main network and a target network; each network has an actor part and a critic part, the target network is regarded as an iteration of the main network, generates the target value of the training main network, and finally updates the strategy by calculating the loss function; the experience replay pool of the DDPG algorithm stores the learning experience of action reward and state change, which is used to train the actor and critic parameters.

2. The digital-twin-based vehicle social network caching method of claim 1, wherein: In step S2, the twin network layer comprises an information collection module, a control module, an RSU cooperation module, an instruction issuing module and a digital twin data storage module. The information collection module acquires the state of the vehicle through communication between the vehicle and the RSU; the control module is used for determining an update period and an interaction frequency and data type in an adjustment process, and an adjustment instruction is transmitted to the corresponding vehicle through the instruction issuing module; The RSU cooperation module is used for timely exchanging data and updating states; The digital twin data storage module is used for storing data.

3. The digital-twin-based vehicle social network caching method of claim 1, wherein: The signal-to-noise ratio formula of hybrid communication V2X communication is: where SINR stands for signal-to-noise ratio, stands for path length, stands for path loss factor, stands for the transmit power of the vehicle, stands for the additive noise power of the vehicle; Assume The maximum distance of V2X communication in an ideal environment without co-channel interference is expressed as: The average rate of vehicle V2X communication is defined as: wherein denotes the channel bandwidth.

Images