MEC vehicle-mounted network calculation pre-unloading method based on trajectory prediction

Abstract

The embodiment of the invention relates to the technical field of task unloading, in particular to an MEC vehicle-mounted network calculation pre-unloading method based on trajectory prediction, and the method comprises the steps: building a communication model, an edge server calculation model and a local calculation model for a multi-vehicle multi-MEC scene, and building a target function; acquiring historical trajectory data of each vehicle, preprocessing the historical trajectory data, and inputting the preprocessed historical trajectory data into the LSTM model to obtain a predicted position of each vehicle at the next moment; on the basis of the predicted position of each vehicle at the next moment and the position and service range of each MEC server, determining an available MEC server of each vehicle at the next moment; comprehensively scoring the priority, importance and resource demand of the task of each vehicle, and determining a weighted importance score of each task; and inputting the predicted position of each vehicle at the next moment, the MEC server available at the next moment and the weighted importance score of the task of each vehicle into a DDQN network to obtain a task unloading strategy. According to the method, the resource utilization efficiency can be well improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of task offloading, and in particular to a method for pre-offloading of MEC vehicle network calculations based on trajectory prediction. Background Art

[0002] In recent years, in-vehicle networks, in-vehicle applications and services have gradually diversified, providing users with a better user experience. However, the computing power of vehicle terminals is very limited, and the increase in in-vehicle applications and services has brought huge computing pressure to vehicle terminals. In order to solve this problem, the smart car industry has introduced cloud computing technology, which transfers the Internet of Vehicles computing tasks to the cloud computing center for execution through computing offloading, breaking through the resource limitations of in-vehicle equipment, while reducing the energy consumption of vehicles and ensuring efficient execution of tasks. However, cloud computing technology also has certain problems: the first is the data transmission problem. With the surge in the number of Internet of Vehicles applications, the bandwidth cost of data transmission is getting higher and higher; the second is the timeliness of in-vehicle task processing. The cloud computing center is far away from the terminal, and the data processing and transmission speed delay is large, which seriously affects the real-time and efficiency of in-vehicle applications; finally, there is the problem of data security and privacy. It takes a long time to upload massive data, and there are security issues such as data loss and information leakage during this period.

[0003] Mobile edge computing is considered an effective way to solve this problem. Since the edge server is closer to the vehicle terminal, short-distance computing offloading reduces energy consumption. At the same transmission rate, the transmission delay is significantly reduced. In addition, the edge server is deployed nearby with small scale and distributed characteristics, which provides stronger protection for user information. However, mobile edge computing faces the problem of vehicle mobility, the problem of the constant change of available edge servers in the surrounding area, and the problem that individual tasks have differences in computing requirements, delay sensitivity, data size, etc., so choosing a suitable offloading strategy is very critical.

[0004] At present, many teams at home and abroad have conducted research on task offloading. Based on different optimization objectives, the proposed task offloading algorithms can be roughly divided into three categories: algorithms for delay optimization, algorithms for energy cost optimization, and algorithms for delay and energy cost trade-off optimization. These algorithms perform well in static user environments. However, the increase in target mobility makes the actual situation more complicated and the problem more difficult. Due to the dynamic characteristics of the mobile environment, such decisions often lead to substandard performance. Highly mobile devices such as vehicles bring additional challenges because their connections to edge servers may change rapidly, resulting in increased latency and potential task failures. Current research on task offloading rarely considers the pre-processing of future location task offloading, and the mobility of vehicles and the different requirements of tasks are not considered enough in offloading decisions and resource allocation, resulting in low resource utilization of the system. Summary of the invention

[0005] In order to solve the above technical problems, the embodiment of the present application proposes a MEC vehicle network computing pre-unloading method based on trajectory prediction, which predicts the vehicle's position at the next moment through the vehicle's historical trajectory, and studies the task unloading strategy based on the predicted position. This enables the edge server to prepare for the arrival of the task in advance, thereby reducing the overall task processing delay and improving resource utilization efficiency.

[0006] In the first aspect, an embodiment of the present application proposes a method for pre-unloading calculation of a MEC vehicle network based on trajectory prediction, comprising the following steps: constructing a communication model, an edge server calculation model and a local calculation model for a multi-vehicle multi-MEC scenario, and building an objective function of minimizing delay based on the constructed model; obtaining the historical trajectory data of each vehicle and preprocessing it, inputting the historical trajectory data of each preprocessed vehicle into a pre-trained LSTM model, and obtaining the predicted position of each vehicle at the next moment output by the LSTM model; based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server, determining the MEC server available for each vehicle at the next moment; standardizing the priority, importance and resource requirements of each vehicle's task and then performing a comprehensive score to determine the weighted importance score of each vehicle's task; inputting the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task into a pre-trained DDQN network to obtain a task unloading strategy output by the DDQN network based on the objective function of minimizing delay; executing the task unloading strategy output by the DDQN network to complete the task unloading of each vehicle.

[0007] Optionally, let the MEC server set in the multi-vehicle multi-MEC scenario be M, M = {1, 2, ..., M}, including M MEC servers in total, let the vehicle set in the multi-vehicle multi-MEC scenario be N, N = {1, 2, ..., N}, including N vehicles in total, time is divided into several time slices, each time slice has the same length, and the time slice set is represented by T, T = {t0, t1, ..., t m}, in each time slice, each vehicle randomly generates discrete and inseparable tasks with different attributes. The time slice is also called a moment;

[0008] The task collection is represented as task j,i =(l j,i ,c j,i ,τ j,i ,Z j,i ), where task j,i represents the task generated by the i-th vehicle at the j-th moment, l j,i Indicates the amount of data for the task, c j,i represents the amount of computation required to complete the task, τ j,i represents the deadline of the task, Z j,i represents the weighted importance score of the task;

[0009] Let F j+1,k represents the resources of the kth MEC server at the j+1th time, let ω j+1,i,k represents the resource ratio allocated by the k-th MEC server to the i-th vehicle at the j+1-th time;

[0010] The local computing model is aimed at tasks with low computational complexity, high real-time requirements, small data volume, and high requirements for security and privacy. The latency of the task executed locally is:

[0011]

[0012] in, represents the computing power of the i-th vehicle to perform the task at the j+1th moment, represents the local execution delay of the task generated by the i-th vehicle at the j-th time;

[0013] The communication model is oriented to the case where the vehicle communicates with the MEC server, and the transmission rate is assumed to be:

[0014]

[0015] Among them, B represents bandwidth, N represents the total number of channels, the total number of channels is the same as the total number of vehicles, and the channels correspond to vehicles one by one. i represents the transmission power of the i-th vehicle, g j+1,i,krepresents the channel gain between the i-th vehicle and the k-th MEC server at time j+1, σ 2 Represents noise interference, I j+1,i represents the channel interference of the ith channel at the j+1th time, v j+1,i,k represents the transmission rate between the i-th vehicle and the k-th MEC server at time j+1, h0 is the path loss constant, r is the path loss exponent, r is set to 2, d0 is the preset reference distance, d j+1,i,k represents the distance between the i-th vehicle and the k-th MEC server at the j+1th time;

[0016] The edge server computing model is oriented to the case where tasks are offloaded to the MEC server, including the time it takes for the task to be transmitted to the MEC server, the time it takes for the task to be executed in the MEC server, and the time it takes for the task calculation results to be transmitted back from the MEC server. The time it takes for the task calculation results to be transmitted back from the MEC server is negligible. The total time is assumed to be:

[0017]

[0018] in, represents the time when the i-th task is transmitted to the k-th MEC server at time j+1, It indicates the time when the i-th task is executed in the k-th MEC server at time j+1. Indicates the overall time.

[0019] Optionally, the objective function of minimizing the delay based on the constructed model is expressed as:

[0020]

[0021] Among them, x i ∈{0,1},x i = 0 means that the i-th task is executed locally, x i =1 means that the i-th task is offloaded to the MEC server for execution, y i,k ∈{0,1},y i,k =0 means that the i-th task is not offloaded to the k-th MEC server for execution, y i,k =1 means that the i-th task is offloaded to the k-th MEC server for execution, ω t,k represents the resource ratio of the k-th MEC server to the t-th vehicle, and C represents the objective function of minimizing the delay;

[0022] The constraint condition of the objective function to minimize the delay is: i,1 +y i,2 +y i,3 +...+y i,M=1 and C<τ * , where τ * Indicates the minimum deadline of each task.

[0023] Optionally, based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server, determining the MEC server available to each vehicle at the next moment includes:

[0024] Assume that the predicted position of the i-th vehicle at the next moment is Assume the location of the kth MEC server is Assume that the maximum effective communication distance between the vehicle and the MEC server is R, and the service range of the kth MEC server is

[0025] The distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server is calculated based on the Haversine formula. The calculated distance is expressed as:

[0026] d i,k =R·η i,k

[0027]

[0028] Among them, d i,k represents the distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server;

[0029] Determine d in turn i,k Is it less than R? i,k Is it less than

[0030] If d i,k Less than both R and Then it is determined that the kth MEC server is the MEC server available to the i-th vehicle at the next moment;

[0031] If d i,k Not less than R or d i,k Not less than It is determined that the kth MEC server is not the MEC server available to the i-th vehicle at the next moment.

[0032] Optionally, the priority, importance and resource requirements of each vehicle's task are standardized and then comprehensively scored to determine the weighted importance score of each vehicle's task, which is achieved through the following formula:

[0033] Z j,i =α·z′ j,i,1 +β·z′ j,i,2+λ·z′ j,i,3

[0034]

[0035] f=1,2,3

[0036] Among them, α, β and λ are the preset weight coefficients, z j,i,1 、z j,i,2 and z j,i,3 They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th moment, and Respectively represent the minimum and maximum values ​​of the priority of each task, and Respectively represent the minimum and maximum values ​​of the importance of each task, and Respectively represent the minimum and maximum resource requirements of each task, z′ j,i,1 , z′ j,i,2 and z′ j,i,3 They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th time after normalization, and Z j,i represents the weighted importance score of the task generated by the i-th vehicle at the j-th moment.

[0037] Optionally, obtaining a task offloading strategy of the DDQN network based on an objective function output of minimizing latency includes:

[0038] The state space is constructed based on the set of tasks and the set of available MEC servers. The state space is expressed as:

[0039] S n ={task(μ n ),M(μ n )}

[0040] Among them, task(μ n ) represents the task set, M(μ n ) represents the set of available MEC servers;

[0041] Assume that the action space is a set of task offloading strategies. When there are multiple tasks that select the same MEC server, the resources of the MEC server are allocated according to the weighted importance score of each task. The scheduled end time of each task and the percentage of resource usage of the MEC server are monitored in a timely manner. Before making an offloading decision at the next moment, it is regularly determined whether the resources at the previous moment have been used up by whether the scheduled end time of the task has been reached. If the use has ended, the resources are released.

[0042] Define the reward function R nFor R n =C;

[0043] In the experience storage stage, the DDQN network initializes an experience playback memory library, obtains the current state, selects an action according to the E-greedy strategy, and executes it in the environment to obtain the reward of the selected action. The environment enters the state of the next moment, and so on. By continuously interacting with the environment, the agent's experience is stored, including the current state, action, and reward brought by the action. The state of the next moment is (s t ,a t ,r t ,s t+1 ) until the capacity of the memory bank reaches a preset threshold and then enters the learning phase, after which when new experience arrives, the oldest experience will be replaced;

[0044] In the learning phase, a batch of experience samples is randomly drawn from the experience replay memory bank, s t As the input of the main Q network, get the Q value and select the action a corresponding to the maximum Q value t Get the next state s t+1 , and the next state s t+1 Output to the target Q network to get Q target , take the Q value as the predicted value of the network, and take r t+1 +γQ target (s t+1 ,a1) as the actual value of the network, use the mean square error loss function to calculate the Q value of the main Q network output and Q target The error between them is calculated by back propagation algorithm, and the parameters of the main Q network are updated to make it close to the Q value output by the main Q network. At the same time, after learning for a fixed number of steps, the parameters of the main Q network are copied to the target Q network to maintain the stability of the target Q network.

[0045] Finally, the task offloading strategy output by the DDQN network is obtained.

[0046] Optionally, when training the LSTM model, use the MAE indicator, MSE indicator, and RMSE indicator to verify the effectiveness and accuracy of the trained LSTM model;

[0047] The MAE index, MSE index and RMSE index are expressed as follows:

[0048]

[0049] Among them, p t+1 It indicates the prediction result of the trained LSTM model for the validation sample. represents the true label on the verification sample, E MAE Represents the MAE indicator, E MSErepresents the MSE index, E RMSE Represents the RMSE indicator.

[0050] Through the above method, communication models, edge server computing models and local computing models were constructed respectively for multi-vehicle and multi-MEC scenarios, and an objective function of minimizing latency was established. The historical trajectory data of each vehicle was used for training and prediction to determine the predicted position of each vehicle at the next moment. Subsequently, the MEC server selection process was modeled as a Markov decision process problem, and the computing resources of multiple servers were reasonably allocated. This solved the delay problem caused by traditional algorithms, reduced the frequency of MEC server switching when switching positions, and improved the accuracy of pre-unloading strategy research, thereby reducing the overall task processing delay and improving resource utilization efficiency.

[0051] In the second aspect, an embodiment of the present application proposes a MEC vehicle network computing pre-unloading system based on trajectory prediction, the system comprising: a scenario construction module, which is used to construct a communication model, an edge server computing model and a local computing model for multi-vehicle and multi-MEC scenarios, and to build an objective function of minimizing latency based on the constructed model; a position prediction module, which is used to obtain and pre-process the historical trajectory data of each vehicle, and input the pre-processed historical trajectory data of each vehicle into a pre-trained LSTM model to obtain the predicted position of each vehicle at the next moment output by the LSTM model; an MEC matching module, which is used to match the predicted position of each vehicle at the next moment, the position of each MEC server, and the position of each The service scope of the MEC server determines the MEC server available to each vehicle at the next moment; the comprehensive scoring module is used to standardize the priority, importance and resource requirements of each vehicle's tasks and then perform a comprehensive score to determine the weighted importance score of each vehicle's tasks; the task offloading strategy generation module is used to input the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's tasks into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing delay; the execution module is used to execute the task offloading strategy output by the DDQN network to complete the task offloading of each vehicle.

[0052] In a third aspect, an embodiment of the present application proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a MEC vehicle network computing pre-unloading method based on trajectory prediction as described in the first aspect above.

[0053] In a fourth aspect, an embodiment of the present application proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement a MEC vehicle network computing pre-unloading method based on trajectory prediction as described in the first aspect above.

[0054] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flowchart of a method for pre-unloading MEC vehicle network calculation based on trajectory prediction provided in one embodiment of the present application;

[0057] Figure 2 is a schematic diagram of a multi-vehicle multi-MEC scenario provided in one embodiment of the present application;

[0058] Figure 3 It is a structural diagram of an LSTM model provided in one embodiment of the present application;

[0059] Figure 4 is a schematic diagram of the structure of a DDQN network provided in one embodiment of the present application;

[0060] Figure 5 is a schematic diagram of a vehicle trajectory provided in one embodiment of the present application;

[0061] Figure 6 is a structural schematic diagram of a MEC vehicle network computing pre-unloading system based on trajectory prediction provided in another embodiment of the present application;

[0062] Figure 7 It is a structural schematic diagram of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. In the various embodiments of the present application, many technical details are proposed in order to make the reader better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is only for the convenience of description, and the specific implementation mode of the present application should not constitute any limitation. The various embodiments can be combined with each other and referenced to each other under the premise of no contradiction.

[0064] An embodiment of the present application proposes a MEC vehicle network computing pre-unloading method based on trajectory prediction, which is applied to an electronic device, wherein the electronic device may be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using a central server as an example. The implementation details of the MEC vehicle network computing pre-unloading method based on trajectory prediction proposed in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0065] The specific process of the MEC vehicle network calculation pre-unloading method based on trajectory prediction proposed in this embodiment can be as follows: Figure 1 As shown, including:

[0066] Step 101, construct a communication model, an edge server computing model, and a local computing model for a multi-vehicle multi-MEC scenario, and build an objective function for minimizing latency based on the constructed models.

[0067] Specifically, the central server first builds a communication model, an edge server computing model, and a local computing model for multi-vehicle and multi-MEC scenarios, and builds an objective function for minimizing latency based on the constructed models.

[0068] In the specific implementation, the central server sets the set of MEC servers in the multi-vehicle multi-MEC scenario as M, M = {1, 2, ..., M}, and the scenario contains M MEC servers. At the same time, the set of vehicles in the multi-vehicle multi-MEC scenario is set as N, N = {1, 2, ..., N}, and the scenario contains N vehicles. The time is divided into several time slices (time slices are also called moments), and the time length of each time slice is the same. The set of time slices is represented by T, T = {t0, t1, ..., t m}, in each time slice, each vehicle randomly generates discrete and indivisible tasks with different attributes.

[0069] The task collection is represented as task j,i =(l j,i ,c j,i ,τ j,i ,Zj,i ), where task j,i represents the task generated by the i-th vehicle at the j-th moment, l j,i Indicates the amount of data for the task, c j,i represents the amount of computation required to complete the task, τ j,i represents the deadline of the task, Z j,i Represents the weighted importance score of the task.

[0070] Let F j+1,k represents the resources of the kth MEC server at the j+1th time, let ω j+1,i,k It represents the resource ratio allocated by the k-th MEC server to the i-th vehicle at the j+1-th time.

[0071] Let P t ′={P start ,P1,P2,...P t-1} represents the motion trajectory of the vehicle before time t.

[0072] Assume that the task offloading strategy consists of three parts: x indicating whether the task is offloaded i , indicating which MEC server the y is offloaded to i,k , which represents the corresponding allocated resource ratio ω t,k .

[0073] Therefore, the task allocation problem in this scenario is described as follows: given a set of vehicles N and the corresponding on-board tasks, given a set of MEC servers M and the corresponding resources, given a continuous historical trajectory P of the vehicle t ,The goal is to predict the trajectory of the vehicle and obtain the ,task offloading strategy based on the predicted position to minimize the system ,time cost.

[0074] The central server builds a communication model, an edge server computing model, and a local computing model for the entire system scenario. These models are suitable for different task offloading strategies. The trajectory data set is then preprocessed, and the data is then allocated to training sets and test sets in proportion. After training, trajectory prediction is performed, and evaluation indicators are set to judge the performance. The results are input into the DDQN network, and finally the task offloading strategy is output.

[0075] The local computing model is aimed at tasks with low computational complexity, high real-time requirements, small data volume, and high requirements for security and privacy. In order to ensure the timeliness and reliability of tasks, these tasks will be executed locally instead of being offloaded. The central server sets the delay of local execution of tasks as:

[0076]

[0077] in, represents the computing power of the i-th vehicle to perform the task at the j+1th moment, It represents the time delay of local execution of the task generated by the i-th vehicle at the j-th time.

[0078] The communication model is oriented to the communication between vehicles and MEC servers. It is given by Shannon's theorem. Assume that each vehicle selects a channel to connect to the MEC server. For the convenience of research, we assume that a vehicle will not change after selecting a channel. Based on this, the transmission rate can be obtained as:

[0079]

[0080] Among them, B represents bandwidth, N represents the total number of channels, the total number of channels is the same as the total number of vehicles, and the channels correspond to vehicles one by one. i represents the transmission power of the i-th vehicle, g j+1,i,k represents the channel gain between the i-th vehicle and the k-th MEC server at time j+1, σ 2 Represents noise interference, I j+1,i represents the channel interference of the ith channel at the j+1th time, v j+1,i,k represents the transmission rate between the i-th vehicle and the k-th MEC server at time j+1, h0 is the path loss constant, r is the path loss exponent, r is set to 2, d0 is the preset reference distance, d j+1,i,k represents the distance between the i-th vehicle and the k-th MEC server at the j+1-th time.

[0081] The edge server computing model is oriented to the case where tasks are offloaded to the MEC server, including the time it takes for the task to be transmitted to the MEC server, the time it takes for the task to be executed in the MEC server, and the time it takes for the task calculation results to be transmitted back from the MEC server. The time it takes for the task calculation results to be transmitted back from the MEC server is negligible. The total time is assumed to be:

[0082]

[0083] in, represents the time when the i-th task is transmitted to the k-th MEC server at time j+1, It indicates the time when the i-th task is executed in the k-th MEC server at time j+1. Indicates the overall time.

[0084] In one example, the objective function of minimizing the delay of the model constructed by the central server is expressed as:

[0085]

[0086] Among them, xi ∈{0,1},x i = 0 means that the i-th task is executed locally, x i =1 means that the i-th task is offloaded to the MEC server for execution, y i,k ∈{0,1},y i,k =0 means that the i-th task is not offloaded to the k-th MEC server for execution, y i,k =1 means that the i-th task is offloaded to the k-th MEC server for execution, ω t,k represents the resource ratio of the k-th MEC server to the t-th vehicle, C represents the objective function of minimizing the delay, and the constraint condition of the objective function of minimizing the delay can be expressed as: i,1 +y i,2 +y i,3 +...+y i,M =1 and C<τ * , τ * Indicates the minimum deadline of each task.

[0087] Step 102, obtain the historical trajectory data of each vehicle and preprocess it, input the preprocessed historical trajectory data of each vehicle into the pre-trained LSTM model, and obtain the predicted position of each vehicle at the next moment output by the LSTM model.

[0088] In the specific implementation, after the central server completes the construction of the communication model, edge server computing model, local computing model and objective function, it can obtain the historical trajectory data of each vehicle and preprocess it, and input the preprocessed historical trajectory data of each vehicle into the pre-trained LSTM model to obtain the predicted position of each vehicle at the next moment output by the LSTM model.

[0089] In an example, after obtaining the historical trajectory data of each vehicle, the central server needs to process missing values ​​and outliers and normalize the obtained data to a [0,1] interval. These operations are collectively called preprocessing. The central server can use the Min-Max normalization method to maintain the distribution of the data. The normalized data is expressed as:

[0090]

[0091] Among them, P represents the original historical trajectory data, P min Represents the minimum value in the original historical trajectory data, P max represents the maximum value in the original historical trajectory data, P norm Represents the normalized historical trajectory data.

[0092] In an example, the network architecture of the LSTM model is as follows Figure 3 As shown, Figure 3 The memory cells and gate mechanisms are clearly shown, as well as the structural relationships between layers.

[0093] The memory unit is the core part of the LSTM network. It is similar to a transmission belt that runs through the entire sequence and only undergoes a small amount of linear multiplication and addition operations to maintain and transmit important information. The memory unit allows information to flow over a very long sequence without being lost due to gradient vanishing or exploding. The gate mechanism includes the forget gate, input gate, and output gate. The forget gate determines which information in the memory unit should be forgotten. It uses a Sigmoid function to determine the weight of each piece of information. The output range is between 0 and 1, where 0 means complete forgetfulness and 1 means complete retention. The input gate determines how much of the current input information can be added to the memory cell. It consists of two parts: a Sigmoid function determines which information needs to be updated, and a tanh function generates new candidate values. These new values ​​will be added to the memory cell. The output gate determines which part of the information of the memory unit can be used as output. At the same time, it also generates a new hidden state, which is also achieved through a Sigmoid function and a tanh function. Next, the memory unit is updated according to the output of the forget gate and input gate. Finally, the output gate outputs the During training, the central server uses the Adam optimization algorithm for optimization. By adaptively adjusting the learning rate, Adam effectively accelerates the training process of the LSTM network and improves the training effect.

[0094] In an example, when training an LSTM model, the MAE indicator, MSE indicator, and RMSE indicator are used to verify the effectiveness and accuracy of the trained LSTM model. The MAE indicator, MSE indicator, and RMSE indicator are expressed as follows:

[0095]

[0096] Among them, p t+1 It indicates the prediction result of the trained LSTM model for the validation sample. represents the true label on the verification sample, E MAE Represents the MAE indicator, E MSE represents the MSE index, E RMSE Represents the RMSE indicator.

[0097] Step 103, based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server, determine the MEC server available to each vehicle at the next moment.

[0098] Specifically, after completing the trajectory prediction of each vehicle, the central server needs to determine the MEC server available to each vehicle at the next moment based on the predicted position of each vehicle at the next moment, the location of each MEC server, and the service range of each MEC server.

[0099] In the specific implementation, vehicles perform task offloading while moving at high speed, and frequent MEC server switching will increase the overall latency, increase the signaling overhead and data retransmission overhead during the switching process, and also affect the network efficiency. To avoid this problem, the central server prepares the MEC server at the next location in advance based on the predicted location at the next moment, and gives priority to the MEC server located in the intersection area of ​​the two locations based on the signal strength and service range of the MEC server, to ensure seamless task transfer during switching, reduce interruptions, and improve the efficiency of offloading and the stability of the system.

[0100] In one example, when the central server determines the MEC server available to each vehicle at the next moment based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server, the predicted position of the i-th vehicle at the next moment is Assume the location of the kth MEC server is Assume that the maximum effective communication distance between the vehicle and the MEC server is R, and the service range of the kth MEC server is

[0101] The central server calculates the distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server based on the Haversine formula. The calculated distance is expressed as:

[0102] d i,k =R·η i,k

[0103]

[0104] Among them, d i,k Represents the distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server.

[0105] Then, the central server determines d i,k Is it less than R? i,k Is it less than If d i,k Less than both R and Then it is determined that the kth MEC server is the MEC server available to the i-th vehicle at the next moment, if d i,k Not less than R or d i,k Not less than It is determined that the kth MEC server is not the MEC server available to the i-th vehicle at the next moment.

[0106] Step 104 , the priority, importance and resource requirements of each vehicle's task are standardized and then comprehensively scored to determine a weighted importance score of each vehicle's task.

[0107] Specifically, the central server also needs to standardize the priority, importance, and resource requirements of each vehicle's tasks and then perform a comprehensive score to determine the weighted importance score of each vehicle's tasks.

[0108] In a specific implementation, tasks have different attributes and requirements. For example, safety tasks such as collision detection and emergency braking need to be processed immediately, while background tasks such as software updates are less important and can be processed at an appropriate time. In order to ensure the timeliness of tasks, this embodiment provides a comprehensive scoring model that considers factors such as task priority, importance, and required resources. The task priority, maximum task delay limit, and task resource requirements are incorporated into the scoring model to fully and reasonably allocate resources and improve system reliability.

[0109] In one example, the central server normalizes the priority, importance, and resource requirements of each vehicle's task and then performs a comprehensive score to determine the weighted importance score of each vehicle's task, which is achieved through the following formula:

[0110] Z j,i =α·z′ j,i,1 +β·z′ j,i,2 +λ·z′ j,i,3

[0111]

[0112] f=1,2,3

[0113] Among them, α, β and λ are the preset weight coefficients, z j,i,1 、z j,i,2 and z j,i,3 They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th moment, and Respectively represent the minimum and maximum values ​​of the priority of each task, and Respectively represent the minimum and maximum values ​​of the importance of each task, and Respectively represent the minimum and maximum resource requirements of each task, z′ j,i,1 , z′ j,i,2 and z′ j,i,2They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th time after normalization, and Z j,i represents the weighted importance score of the task generated by the i-th vehicle at the j-th moment.

[0114] It is worth noting that the purpose of data standardization is to eliminate the impact of the dimensions among the three indicators, convert the data into the same magnitude, retain the distribution characteristics of the data while avoiding the adverse impact of the magnitude on the results.

[0115] Step 105, the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task are input into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing the delay.

[0116] In the specific implementation, after obtaining the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task, the central server can input the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing delay.

[0117] In an example, the network structure of a DDQN network can be as follows: Figure 4 As shown in Figure 1, the DDQN network constructs a state space based on a set of tasks and a set of available MEC servers. The state space is expressed as:

[0118] S n ={task(μ n ),M(μ n )}

[0119] Among them, task(μ n ) represents the task set, M(μ n ) represents the set of available MEC servers.

[0120] Assume that the action space is a set of task offloading strategies. When there are multiple tasks that select the same MEC server, the resources of the MEC server are allocated according to the weighted importance score of each task. The scheduled end time of each task and the percentage of resource usage of the MEC server are monitored in a timely manner. Before making an offloading decision at the next moment, it is regularly determined whether the resources at the previous moment have been used up by whether the scheduled end time of the task has been reached. If so, the resources are released.

[0121] Define the reward function R n For Rn =C.

[0122] In the experience storage stage, the DDQN network initializes an experience playback memory library, obtains the current state, selects an action according to the E-greedy strategy, and executes it in the environment to obtain the reward of the selected action. The environment enters the state of the next moment, and so on. By continuously interacting with the environment, the agent's experience is stored, including the current state, action, and reward brought by the action. The state of the next moment is (s t ,a t ,r t ,s t+1 ) until the capacity of the memory bank reaches a preset threshold and then enters the learning phase, after which when new experience arrives, the oldest experience will be replaced.

[0123] In the learning phase, a batch of experience samples is randomly drawn from the experience replay memory bank, s t As the input of the main Q network, get the Q value and select the action a corresponding to the maximum Q value t Get the next state s t+1 , and the next state s t+1 Output to the target Q network to get Q target , take the Q value as the predicted value of the network, and take r t+1 +γQ target (s t+1 ,a1) as the actual value of the network, use the mean square error loss function to calculate the Q value of the main Q network output and Q target The error between them is calculated by back propagation algorithm to update the parameters of the main Q network so that it approaches the Q value output by the main Q network. At the same time, after learning for a fixed number of steps, the parameters of the main Q network are copied to the target Q network to maintain the stability of the target Q network.

[0124] Finally, the task offloading strategy output by the DDQN network is obtained.

[0125] In one example, the open base station information on the OpenCelliD website and the public GPS data on the Internet are used as training data sets, from which a trajectory with a moderate amount of data is selected, such as Figure 5 As shown, the location information is updated in time according to the trajectory prediction results, and the unloading decision is made at a fixed time interval.

[0126] Step 106: Execute the task offloading strategy output by the DDQN network to complete the task offloading of each vehicle.

[0127] In a specific implementation, after obtaining the task offloading strategy output by the DDQN network, the central server can execute the task offloading strategy output by the DDQN network at the next moment to complete the task offloading of each vehicle.

[0128] In this embodiment, a communication model, an edge server computing model, and a local computing model are constructed for multi-vehicle and multi-MEC scenarios, and an objective function for minimizing latency is established. The historical trajectory data of each vehicle is used for iterative training and prediction to determine the predicted position of each vehicle at the next moment. The MEC server selection process is then modeled as a Markov decision process problem, and the computing resources of multiple servers are reasonably allocated. This solves the latency problem caused by traditional algorithms, reduces the switching frequency of MEC servers when switching positions, and improves the accuracy of pre-unloading strategy research, thereby reducing the overall task processing delay and improving resource utilization efficiency.

[0129] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0130] Another embodiment of the present application proposes a MEC vehicle network computing pre-unloading system based on trajectory prediction. The implementation details of the MEC vehicle network computing pre-unloading system based on trajectory prediction proposed in this embodiment are specifically described below. The following content is only for the convenience of understanding the implementation details provided, and is not necessary for the implementation of this example. Figure 6 It is a structural diagram of a MEC vehicle network computing pre-unloading system based on trajectory prediction proposed in this embodiment, which includes: a scene construction module 201, a location prediction module 202, a MEC matching module 203, a comprehensive scoring module 204, a task offloading strategy generation module 205 and an execution module 206.

[0131] The scenario construction module 201 is used to construct a communication model, an edge server computing model and a local computing model for a multi-vehicle multi-MEC scenario, and to build an objective function for minimizing latency based on the constructed model.

[0132] The position prediction module 202 is used to obtain and preprocess the historical trajectory data of each vehicle, input the preprocessed historical trajectory data of each vehicle into the pre-trained LSTM model, and obtain the predicted position of each vehicle at the next moment output by the LSTM model.

[0133] The MEC matching module 203 is used to determine the MEC server available to each vehicle at the next moment based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server.

[0134] The comprehensive scoring module 204 is used to perform comprehensive scoring after standardizing the priority, importance and resource requirements of each vehicle's task, and determine the weighted importance score of each vehicle's task.

[0135] The task offloading strategy generation module 205 is used to input the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing delay.

[0136] The execution module 206 is used to execute the task offloading strategy output by the DDQN network to complete the task offloading of each vehicle.

[0137] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0138] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0139] Another embodiment of the present application provides an electronic device, whose specific structure is as follows: Figure 7 As shown, it includes: at least one processor 301; and a memory 302 that is communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions that can be executed by the at least one processor 301, and the instructions are executed by the at least one processor 301 so that the at least one processor 301 can execute a MEC vehicle-mounted network computing pre-unloading method based on trajectory prediction described in the above embodiments.

[0140] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be further described in this article. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0141] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0142] Another embodiment of the present application proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement a MEC vehicle network computing pre-unloading method based on trajectory prediction described in the above embodiments.

[0143] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0144] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A method for pre-unloading calculation of MEC vehicle network based on trajectory prediction, characterized in that: include: Build communication models, edge server computing models, and local computing models for multi-vehicle and multi-MEC scenarios, and build an objective function to minimize latency based on the built models; Obtain the historical trajectory data of each vehicle and perform preprocessing, input the preprocessed historical trajectory data of each vehicle into the pre-trained LSTM model, and obtain the predicted position of each vehicle at the next moment output by the LSTM model; Determine the MEC server available to each vehicle at the next moment based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server; The priority, importance and resource requirements of each vehicle's tasks are standardized and then comprehensively scored to determine the weighted importance score of each vehicle's tasks; The predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task are input into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing latency. Execute the task offloading strategy output by the DDQN network to complete the task offloading of each vehicle.

2. According to the method of pre-unloading of MEC vehicle network calculation based on trajectory prediction in claim 1, it is characterized in that The MEC server set in the multi-vehicle multi-MEC scenario is M, M = {1, 2, ..., M}, which contains M MEC servers in total. The vehicle set in the multi-vehicle multi-MEC scenario is N, N = {1, 2, ..., N}, which contains N vehicles in total. The time is divided into several time slices, and the time length of each time slice is the same. The time slice set is represented by T, T = {t0, t1, ..., t m }, in each time slice, each vehicle randomly generates discrete and inseparable tasks with different attributes. The time slice is also called a moment; The task collection is represented as task j,i =(l j,i ,c j,i ,τ j,i ,Z j,i ), where task j,i represents the task generated by the i-th vehicle at the j-th moment, l j,i Indicates the amount of data for the task, c j,i represents the amount of computation required to complete the task, τ j,i represents the deadline of the task, Z j,i represents the weighted importance score of the task; Let F j+1,k represents the resources of the kth MEC server at the j+1th time, let ω j+1,i,k represents the resource ratio allocated by the k-th MEC server to the i-th vehicle at the j+1-th time; The local computing model is aimed at tasks with low computational complexity, high real-time requirements, small data volume, and high requirements for security and privacy. The latency of the task executed locally is: in, represents the computing power of the i-th vehicle to perform the task at the j+1th moment, represents the local execution delay of the task generated by the i-th vehicle at the j-th time; The communication model is oriented to the case where the vehicle communicates with the MEC server, and the transmission rate is assumed to be: Among them, B represents bandwidth, N represents the total number of channels, the total number of channels is the same as the total number of vehicles, and the channels correspond to vehicles one by one. i represents the transmission power of the i-th vehicle, g j+1,i,k represents the channel gain between the i-th vehicle and the k-th MEC server at time j+1, σ 2 Represents noise interference, I j+1,i represents the channel interference of the ith channel at the j+1th time, v j+1,i,k represents the transmission rate between the i-th vehicle and the k-th MEC server at time j+1, h0 is the path loss constant, r is the path loss exponent, r is set to 2, d0 is the preset reference distance, d j+1,i,k represents the distance between the i-th vehicle and the k-th MEC server at the j+1th time; The edge server computing model is oriented to the case where tasks are offloaded to the MEC server, including the time it takes for the task to be transmitted to the MEC server, the time it takes for the task to be executed in the MEC server, and the time it takes for the task calculation results to be transmitted back from the MEC server. The time it takes for the task calculation results to be transmitted back from the MEC server is negligible. The total time is assumed to be: in, represents the time when the i-th task is transmitted to the k-th MEC server at time j+1, It indicates the time when the i-th task is executed in the k-th MEC server at time j+1. Indicates the overall time.

3. According to claim 2, a method for pre-unloading calculation of a MEC vehicle network based on trajectory prediction is characterized in that: The objective function of minimizing the delay based on the constructed model is expressed as: Among them, x i ∈{0,1},x i = 0 means that the i-th task is executed locally, x i =1 means that the i-th task is offloaded to the MEC server for execution, y i,k ∈{0,1},y i,j = 0 means that the i-th task is not offloaded to the k-th MEC server for execution, y i,k =1 means that the i-th task is offloaded to the k-th MEC server for execution, ω t,k represents the resource ratio of the k-th MEC server to the t-th vehicle, and C represents the objective function of minimizing the delay; The constraint condition of the objective function to minimize the delay is: i,1 +y i,2 +y i,3 +...+y i,M =1 and C<τ * , where τ * Indicates the minimum deadline of each task.

4. According to claim 3, a method for pre-unloading MEC vehicle network computing based on trajectory prediction is characterized in that: Based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server, the MEC server available to each vehicle at the next moment is determined, including: Assume that the predicted position of the i-th vehicle at the next moment is Assume the location of the kth MEC server is Assume that the maximum effective communication distance between the vehicle and the MEC server is R, and the service range of the kth MEC server is The distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server is calculated based on the Haversine formula. The calculated distance is expressed as: d i，k =R·η i,k Among them, d i,k represents the distance between the predicted position of the i-th vehicle at the next moment and the position of the k-th MEC server; Determine d in turn i,k Is it less than R? i,k Is it less than If d i,k Less than both R and Then it is determined that the kth MEC server is the MEC server available to the i-th vehicle at the next moment; If d i,k Not less than R or d i,k Not less than It is determined that the kth MEC server is not the MEC server available to the i-th vehicle at the next moment.

5. According to claim 4, a method for pre-unloading calculation of a MEC vehicle network based on trajectory prediction is characterized in that: After standardizing the priority, importance, and resource requirements of each vehicle's tasks, a comprehensive score is performed to determine the weighted importance score of each vehicle's tasks, which is achieved through the following formula: WITH j，i =α z′ j，i，1 +β z′ j，i，2 +λ z′ j，i，3 f＝1,2,3 Among them, α, β and λ are the preset weight coefficients, z j,i,1 、z j,i,2 and z j,i,3 They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th moment, and Respectively represent the minimum and maximum values ​​of the priority of each task, and Respectively represent the minimum and maximum values ​​of the importance of each task, and Respectively represent the minimum and maximum resource requirements of each task, z′ j,i,1 , z′ j,i,2 and z′ j,i,3 They represent the priority, importance and resource requirements of the task generated by the i-th vehicle at the j-th time after normalization, and Z j,i represents the weighted importance score of the task generated by the i-th vehicle at the j-th moment.

6. According to claim 5, a method for pre-unloading calculation of a MEC vehicle network based on trajectory prediction is characterized in that: Obtain the task offloading strategy of the DDQN network based on the objective function output of minimizing latency, including: The state space is constructed based on the set of tasks and the set of available MEC servers. The state space is expressed as: S n ={task(μ n ),M(μ n )} Among them, task(μ n ) represents the task set, M(μ n ) represents the set of available MEC servers; Assume that the action space is a set of task offloading strategies. When there are multiple tasks that select the same MEC server, the resources of the MEC server are allocated according to the weighted importance score of each task. The scheduled end time of each task and the percentage of resource usage of the MEC server are monitored in a timely manner. Before making an offloading decision at the next moment, it is regularly determined whether the resources at the previous moment have been used up by whether the scheduled end time of the task has been reached. If the use has ended, the resources are released. Define the reward function R n For R n =C; In the experience storage stage, the DDQN network initializes an experience playback memory library, obtains the current state, selects an action according to the E-greedy strategy, and executes it in the environment to obtain the reward of the selected action. The environment enters the state of the next moment, and so on. By continuously interacting with the environment, the agent's experience is stored, including the current state, action, and reward brought by the action. The state of the next moment is (s t ,a t ,r t ,s t+1 ) until the capacity of the memory bank reaches a preset threshold and then enters the learning phase, after which when new experience arrives, the oldest experience will be replaced; In the learning phase, a batch of experience samples is randomly drawn from the experience replay memory bank, s t As the input of the main Q network, get the Q value and select the action a corresponding to the maximum Q value t Get the next state s t+1 , and the next state s t+1 Output to the target Q network to get Q target , take the Q value as the predicted value of the network, and take r t+1 +γQ target (s t+1 ,a1) as the actual value of the network, use the mean square error loss function to calculate the Q value of the main Q network output and Q target The error between them is calculated by back propagation algorithm, and the parameters of the main Q network are updated to make it close to the Q value output by the main Q network. At the same time, after learning for a fixed number of steps, the parameters of the main Q network are copied to the target Q network to maintain the stability of the target Q network. Finally, the task offloading strategy output by the DDQN network is obtained.

7. A method for pre-unloading MEC vehicle network computing based on trajectory prediction according to any one of claims 1 to 6, characterized in that: When training the LSTM model, the MAE, MSE, and RMSE indicators are used to verify the effectiveness and accuracy of the trained LSTM model; The MAE index, MSE index and RMSE index are expressed as follows: Among them, p t+1 It indicates the prediction result of the trained LSTM model for the validation sample. represents the true label on the verification sample, E MAE Represents the MAE indicator, E MSE represents the MSE index, E RMSE Represents the RMSE indicator.

8. A MEC vehicle network computing pre-unloading system based on trajectory prediction, characterized in that: include: The scenario construction module is used to build communication models, edge server computing models, and local computing models for multi-vehicle and multi-MEC scenarios, and to build an objective function to minimize latency based on the constructed models; The position prediction module is used to obtain the historical trajectory data of each vehicle and perform preprocessing, input the preprocessed historical trajectory data of each vehicle into the pre-trained LSTM model, and obtain the predicted position of each vehicle at the next moment output by the LSTM model; The MEC matching module is used to determine the MEC server available to each vehicle at the next moment based on the predicted position of each vehicle at the next moment, the position of each MEC server, and the service range of each MEC server; The comprehensive scoring module is used to standardize the priority, importance and resource requirements of each vehicle's tasks and then conduct a comprehensive scoring to determine the weighted importance score of each vehicle's tasks; The task offloading strategy generation module is used to input the predicted position of each vehicle at the next moment, the MEC server available to each vehicle at the next moment, and the weighted importance score of each vehicle's task into the pre-trained DDQN network to obtain the task offloading strategy output by the DDQN network based on the objective function of minimizing the delay; The execution module is used to execute the task offloading strategy output by the DDQN network and complete the task offloading of each vehicle.

9. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; In which, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the MEC vehicle network computing pre-unloading method based on trajectory prediction as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it is capable of implementing the MEC vehicle network computing pre-unloading method based on trajectory prediction as described in any one of claims 1 to 7.

Images