A mobile vehicle task offloading method and system based on prediction and game

Abstract

The present invention discloses a method and system for mobile vehicle task offloading based on prediction and game theory, which relates to the technical field of vehicle network communication and computing task offloading. By predicting the future position of the vehicle, channel status and the number of RSU tasks in the queue, a comprehensive delay model is constructed, and the utility value calculation and strategy probability distribution update mechanism are combined to achieve dynamic optimization decision-making for vehicle task offloading. The algorithm is ensured to converge stably through adaptive adjustment of temperature parameters and convergence judgment. Finally, the task offloading decision for each vehicle is obtained by random sampling based on the strategy probability distribution. This method fully considers vehicle mobility, channel changes and RSU load conditions, and improves the efficiency of task offloading and system performance in the vehicle network environment.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle network communication and computing task offloading, and in particular to a mobile vehicle task offloading method and system based on prediction and game. Background Art

[0002] With the rapid development of intelligent transportation systems and vehicle-to-everything (V2X) technologies, vehicles are generating an increasing number of computing tasks while driving, such as automated driving assistance, video stream processing, and real-time path planning. Limited onboard computing resources make it difficult to meet the high-concurrency, low-latency processing requirements. Consequently, there is an urgent need to offload some tasks to roadside units (RSUs) or other edge nodes for collaborative computing.

[0003] In this context, task offloading technology has become a key means of improving the quality of service in connected vehicles. The core issue of task offloading is how to select appropriate computing nodes (such as RSUs) for vehicles in a dynamically changing network environment to minimize overall task processing latency and improve system resource utilization.

[0004] However, traditional task offloading methods often assume static or semi-static scenarios, making them difficult to adapt to complex factors in connected vehicles, such as high-speed vehicle movement, frequently changing channel states, and large fluctuations in RSU load. Furthermore, existing methods often neglect predictions of future vehicle states, causing offloading decisions to lag behind actual environmental changes, impacting system performance.

[0005] In recent years, game theory has been introduced to task offloading decisions, enabling distributed optimization by modeling the strategic interactions between vehicles and RSUs. However, most game models fail to consider the dynamic impact of vehicle mobility and lack effective prediction mechanisms for RSU load changes, limiting their application in real-world connected vehicle environments.

[0006] Publication number CN115134778B, titled "A Method for Computation Offloading in the Internet of Vehicles Based on Multi-User Game and Federated Learning," proposes a binary offloading strategy, combined with a multi-user game algorithm to generate an initial offloading plan. This strategy then utilizes a Deep Q-Network (DQN) for reinforcement learning optimization. Finally, a federated learning mechanism aggregates multiple user model parameters to improve model generalization and privacy protection. However, the method lacks a predictive mechanism for vehicle trajectories, channel changes, and RSU load, making it difficult to address task offloading requirements in high-speed mobile environments.

[0007] Publication number CN112367640B, titled "V2V Multi-Task Offloading Method and System Based on Mobile Edge Computing," proposes a method whereby a vehicle determines whether to offload tasks based on task latency requirements while driving. If so, the task is transmitted to the RSU on the preceding road section via multi-hop V2V communication, where it is processed by the corresponding MEC server. The task results are then downloaded when the vehicle reaches that road section. While this method models transmission, processing, and download times, it does not consider the task queuing of RSUs in detail, potentially leading to task backlogs and impacting response speed.

[0008] Therefore, there is an urgent need for a dynamic task offloading method that can comprehensively consider vehicle position prediction, channel state evaluation and RSU load prediction, and achieve efficient, stable and adaptive task scheduling by constructing a reasonable utility function and policy update mechanism. Summary of the Invention

[0009] To solve the above technical problems, the present invention adopts a technical solution: providing a mobile vehicle task unloading method based on prediction and game theory, the method comprising:

[0010] S1: According to the mobile vehicle scenario, obtain RSU information and vehicle information and initialize the strategy probability distribution;

[0011] S2: Predict the future based on RSU information The number of RSU tasks queued at a moment, based on vehicle information, predicts the future The vehicle position trajectory at each moment is used to calculate the future channel state;

[0012] S3: Build a delay model and calculate the utility value;

[0013] S4: updating the strategy probability distribution according to the utility value and updating the temperature parameter;

[0014] S5: Check the convergence of the probability distribution of the strategy. If the convergence condition is met, execute S6; otherwise, update the state information and execute S3.

[0015] S6: Based on the final strategy probability distribution, randomly sample a strategy as the final task offloading decision for each vehicle and assign the task to the corresponding RSU.

[0016] Furthermore, the step S1 acquires RSU information and vehicle information according to the mobile vehicle scenario, including:

[0017] Get the number of RSUs ,The number of tasks at the historical moment, computing power, and the number of vehicles obtained are ,historical location trajectory;

[0018] Define temperature parameters and temperature learning rate ;

[0019] Define the convergence threshold of the strategy probability distribution change and utility improvement speed convergence threshold .

[0020] Furthermore, the step S1 initializes the strategy probability distribution, including:

[0021] per vehicle Initialize a policy probability distribution , the calculation formula is:

[0022] ,

[0023] in, is the number of RSUs, Indicates vehicle Select RSUs The probability of task offloading.

[0024] Furthermore, the step S2 predicts the future The vehicle position trajectory at each moment includes:

[0025] Utilizing the vehicle's location ,speed , acceleration Delivering the future The vehicle's position at a moment and speed , the recursive calculation formula is:

[0026] ,

[0027] ,

[0028] in, It's a vehicle At the moment location, It's a vehicle At the moment speed, It's a vehicle At the moment The acceleration of is the time period between adjacent moments, is a modified value that follows a Gaussian distribution.

[0029] Furthermore, the step S2 of calculating the future channel state includes:

[0030] The future predicted The vehicle's position at a moment , for each moment , calculate the future channel state moment by moment, the calculation formula is:

[0031] ,

[0032] in, It's a vehicle with RSU At the moment The future channel state, It's a vehicle At the moment location, RSU location, is the reference gain, is the path loss exponent, is the noise power.

[0033] Furthermore, the step S2 predicts the future The number of RSU tasks queued at a moment, including:

[0034] The number of tasks at historical moments in the RSU information Input into the trained LSTM model to get RSU future The number of tasks queued at a moment is expressed as: ,in, RSU The future moment The number of tasks queued.

[0035] Furthermore, the step S3 of constructing the delay model includes:

[0036] For vehicles Select RSUs The total delay for task offloading is calculated as:

[0037] ,

[0038] in, For vehicles Select RSUs The total delay in performing task offloading, For the moment vehicle Select RSUs The transmission delay, For vehicles Select RSUs Required task queuing delay; For vehicles Select RSUs The required task computation delay.

[0039] Furthermore, the transmission delay includes:

[0040] According to the vehicle Uninstall task size , calculate the transmission delay of the task, the specific steps are:

[0041] Initialize the amount of data to be transmitted, expressed as: ;in, is the amount of data to be transferred for the offload task;

[0042] For every moment , calculate the amount of data that can be transmitted, the transmission delay, and the amount of data to be transmitted at each moment, including:

[0043] (a) Calculate the amount of data that can be transmitted: ,

[0044] (b) Calculate the transmission delay, including:

[0045] ,

[0046] (c) Update the amount of data to be transferred: ,

[0047] in, For vehicles Unload task size, is the bandwidth of the wireless channel, is the size of the moment, For the moment vehicle with RSU The future channel state between For the moment vehicle Select RSUs The amount of data to be transmitted for task offloading, For the moment vehicle Select RSUs The amount of data that can be transferred for task offloading, For the moment vehicle Select RSUs Transmission delay for task offloading, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula.

[0048] Furthermore, the task queuing delay includes:

[0049] According to the time The amount of data to be transferred determines the time it takes for the transfer to complete:

[0050] ,

[0051] in, is the task transmission completion time, For the moment The amount of data to be transmitted, The value range is ;

[0052] Calculate the task queuing delay based on the task transmission completion time:

[0053] ,

[0054] in, RSU computing power, RSU At the moment The number of tasks queued, time For vehicles The task reaches RSU moment, is the average computational requirement of a single task, is the time period between adjacent moments, For vehicles Select RSUs Required task queuing delay.

[0055] Furthermore, the calculation formula for the task calculation delay is:

[0056] ,

[0057] in, For vehicles Select RSUs the required task computation delay, It's a vehicle The computational requirements of the task, RSU computing power.

[0058] Furthermore, the step S3 of calculating the utility value includes:

[0059] Calculate the utility value based on the total latency of the task offloaded :

[0060] ,

[0061] in, For vehicles Select RSUs The total delay in performing task offloading, For vehicles Select RSUs The utility value of performing task offloading.

[0062] Furthermore, the step S4 updates the strategy probability distribution according to the utility value, and the calculation formula for updating the strategy probability distribution is:

[0063] ,

[0064] in, is the temperature parameter, For vehicles Select RSUs The utility value of task offloading, For vehicles Select RSUs The utility value of task offloading, For updated vehicles Select RSUs The probability distribution of strategies for task offloading, is the number of RSUs.

[0065] Furthermore, the step S4 of updating the temperature parameter includes:

[0066] According to the strategy probability distribution before the update and the current strategy probability distribution, calculate the strategy probability distribution change:

[0067] ,

[0068] in, is the change in strategy probability distribution, is the strategy probability distribution before updating, is the current strategy probability distribution, is the number of RSUs, is the number of vehicles;

[0069] Calculate the utility improvement rate based on the utility value before the update and the current utility value:

[0070] ,

[0071] in, Improve speed for utility, is the utility value before updating, is the current utility value, is the number of RSUs, is the number of vehicles;

[0072] Based on the change in the strategy probability distribution and the utility improvement rate, the temperature parameter is updated:

[0073] ,

[0074] in, , is the temperature parameter before updating, and is a constant, controlling the range of temperature parameters, is the temperature learning rate, and is a constant, a weight of the change in the probability distribution of the control strategy and the speed of utility improvement, , is the change in strategy probability distribution, Improve speed for utility, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula.

[0075] Furthermore, the step S5 checks the convergence of the strategy probability distribution, including:

[0076] The convergence threshold based on the change of the strategy probability distribution and the speed of utility improvement is used to determine whether the strategy probability distribution has converged. The convergence conditions are:

[0077] ,

[0078] in, is the change in strategy probability distribution, Improve speed for utility, is the convergence threshold of the strategy probability distribution change, Improve the speed convergence threshold for utility.

[0079] Furthermore, the step S5 of updating the status information includes:

[0080] For every moment , calculate RSU moment by moment The expected number of task arrivals is calculated as follows:

[0081] ,

[0082] in, For the moment RSUs The expected number of task arrivals, is the indicator function, 1 if yes, 0 otherwise. For vehicles The task reaches RSU moment, For vehicles Select RSUs Probability distribution of strategies for task offloading;

[0083] The current Recorded as , when the moment When each RSU is updated moment by moment Number of tasks queued:

[0084] ,

[0085] in, For the moment The number of tasks queued, Before the update at time The number of tasks queued, For the moment RSU The expected number of task arrivals, RSU computing power, is the average computational requirement of a single task, is the time period between adjacent moments, Indicates taking the maximum value in the formula.

[0086] Furthermore, a mobile vehicle task offloading system based on prediction and game theory includes:

[0087] Initialization module: used to obtain basic information of RSU and vehicle, initialize strategy probability distribution and temperature parameters;

[0088] State calculation module: used to predict the future position trajectory of the vehicle, the number of RSU task queues, and the channel status;

[0089] Utility calculation module: builds a total delay model for task offloading, including transmission delay, queuing delay, and computation delay, and calculates the utility value;

[0090] Strategy update module: updates the strategy probability distribution according to the current utility value, adjusts the temperature parameter, and simulates the game process;

[0091] Convergence judgment module: determines whether the strategy has converged. If not, it updates the status information and continues iteration;

[0092] Decision generation module: Samples the task offloading decision for each vehicle based on the final policy probability distribution.

[0093] The beneficial effects of the present invention are:

[0094] The present invention introduces game theory and a dynamic strategy optimization mechanism, combined with vehicle position prediction, channel state evaluation, and RSU task load prediction, to achieve efficient decision-making for task offloading in the Internet of Vehicles environment. Compared with traditional methods, the present invention can more accurately predict the future state of the vehicle and RSU load changes, thereby improving the response speed and resource utilization of task offloading. In addition, by constructing a comprehensive delay model and using utility value as the evaluation criterion, the relationship between transmission delay, queuing delay, and computational delay is effectively balanced, thereby improving the overall performance of the system. At the same time, a strategy probability distribution update mechanism based on temperature parameters is adopted to enhance the algorithm's adaptability and convergence stability, making it suitable for complex and changeable mobile vehicle environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 A flowchart of a mobile vehicle task offloading method based on prediction and game theory provided by an embodiment of the present invention is shown.

[0096] Figure 2 A calculation flow chart of a mobile vehicle task offloading method based on prediction and game theory provided by an embodiment of the present invention is shown.

[0097] Figure 3 A module diagram of a mobile vehicle task unloading system based on prediction and game theory provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0098] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more precise definition of the protection scope of the present invention.

[0099] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present invention, rather than all the embodiments.

[0100] Figure 1 : is a flowchart of a method for unloading tasks from a mobile vehicle based on prediction and game theory provided by an embodiment of the present invention, the method comprising:

[0101] S1: According to the mobile vehicle scenario, obtain RSU information and vehicle information and initialize the strategy probability distribution;

[0102] S2: Predict the future based on RSU information The number of RSU tasks queued at a moment, based on vehicle information, predicts the future The vehicle position trajectory at each moment is used to calculate the future channel state;

[0103] S3: Build a delay model and calculate the utility value;

[0104] S4: updating the strategy probability distribution according to the utility value and updating the temperature parameter;

[0105] S5: Check the convergence of the probability distribution of the strategy. If the convergence condition is met, execute S6; otherwise, update the state information and execute S3.

[0106] S6: Based on the final strategy probability distribution, randomly sample a strategy as the final task offloading decision for each vehicle and assign the task to the corresponding RSU.

[0107] In this embodiment, first, according to the mobile vehicle scenario, the RSU information and vehicle information are obtained to initialize the strategy probability distribution; the future The number of RSU tasks queued at a moment, based on vehicle information, predicts the future The vehicle position trajectory at each moment is used to calculate the future channel state. Subsequently, a delay model is constructed and the utility value is calculated. Next, the strategy probability distribution is updated based on the utility value, and the temperature parameter is updated. Then, the strategy probability distribution is checked for convergence. If not, the system state information is updated and the utility value is recalculated. Finally, based on the final strategy probability distribution, a strategy is randomly sampled as the final task offloading decision for each vehicle, and the task is assigned to the corresponding RSU.

[0108] RSU (Road Side Unit) refers to an intelligent communication device deployed on both sides of the road. It has computing, storage and communication capabilities and can interact with moving vehicles. In the present invention, RSU is used as the target node for task offloading, responsible for receiving and processing task data uploaded by vehicles. Multiple RSUs form a distributed edge computing network, providing low-latency, high-reliability task processing services for mobile vehicles. By comparing the channel status, task queuing status and computing power of different RSUs, the vehicle selects the optimal RSU for task offloading, thereby achieving efficient operation of the entire system.

[0109] Furthermore, the step S1 acquires RSU information and vehicle information according to the mobile vehicle scenario, including:

[0110] Get the number of RSUs ,The number of tasks at the historical moment, computing power, and the number of vehicles obtained are ,historical location trajectory;

[0111] Define temperature parameters and temperature learning rate ;

[0112] Define the convergence threshold of the strategy probability distribution change and utility improvement speed convergence threshold .

[0113] Furthermore, the step S1 initializes the strategy probability distribution, including:

[0114] per vehicle Initialize a policy probability distribution , the calculation formula is:

[0115] ,

[0116] in, is the number of RSUs, Indicates vehicle Select RSUs The probability of task offloading.

[0117] Furthermore, the step S2 predicts the future The vehicle position trajectory at each moment includes:

[0118] Utilizing the vehicle's location ,speed , acceleration Delivering the future The vehicle's position at a moment and speed , the recursive calculation formula is:

[0119] ,

[0120] ,

[0121] in, It's a vehicle At the moment location, It's a vehicle At the moment speed, It's a vehicle At the moment The acceleration of is the time period between adjacent moments, is a modified value that follows a Gaussian distribution.

[0122] Furthermore, the step S2 of calculating the future channel state includes:

[0123] The future predicted The vehicle's position at a moment , for each moment , calculate the future channel state moment by moment, the calculation formula is:

[0124] ,

[0125] in, It's a vehicle with RSU At the moment The future channel state, It's a vehicle At the moment location, RSU location, is the reference gain, is the path loss exponent, is the noise power.

[0126] Furthermore, the step S2 predicts the future The number of RSU tasks queued at a moment, including:

[0127] The number of tasks at historical moments in the RSU information Input into the trained LSTM model to get RSU future The number of tasks queued at a moment is expressed as: ,in, RSU The future moment The number of tasks queued.

[0128] The structure of the trained LSTM model is:

[0129] First, the input sequence is normalized to its maximum and minimum values, and then enters the first LSTM layer. The hidden layer size is set to 3, the input size is 1, the output size is 1, and the intermediate results are retained to form a sequence. The input size is , the output size is , used to capture long-term dependencies in time series, using the activation function Tanh nonlinear transformation to enhance the model's expressiveness; the second fully connected layer maps the LSTM output to the target dimension, with an input size of , the output size is , use ReLU to filter the negative prediction values, and finally denormalize the result.

[0130] The training process of the trained LSTM model is as follows:

[0131] The number of tasks queued in each time slot of different RSUs is collected, and the data of the number of tasks queued in consecutive time slots of the same RSU is processed using the sliding window method to form a data set of length The data set consists of data of length is historical data, the length is The data is labeled data, and the data set is divided into training set, validation set and test set with a ratio of 70%:15%:15%. For the LSTM model, the input window is , predicted output , the input is a shape of where N is the batch size and T is the time window length (i.e. the number of time slots in the past).

[0132] Use mean squared error (MSE) as the loss function:

[0133] ,

[0134] in, is the value of the label data. is the predicted value of the output.

[0135] Furthermore, the step S3 of constructing the delay model includes:

[0136] For vehicles Select RSUs The total delay for task offloading is calculated as:

[0137] ,

[0138] in, For vehicles Select RSUs The total delay in performing task offloading, For the moment vehicle Select RSUs The transmission delay, For vehicles Select RSUs Required task queuing delay; For vehicles Select RSUs The required task computation delay.

[0139] Furthermore, the transmission delay includes:

[0140] According to the vehicle Uninstall task size , calculate the transmission delay of the task, the specific steps are:

[0141] Initialize the amount of data to be transmitted, expressed as: ;in, is the amount of data to be transferred for the offload task;

[0142] For every moment , calculate the amount of data that can be transmitted, the transmission delay, and the amount of data to be transmitted at each moment, including:

[0143] (a) Calculate the amount of data that can be transmitted: ,

[0144] (b) Calculate the transmission delay, including:

[0145] ,

[0146] (c) Update the amount of data to be transferred: ,

[0147] in, For vehicles Unload task size, is the bandwidth of the wireless channel, is the size of the moment, For the moment vehicle with RSU The future channel state between For the moment vehicle Select RSUs The amount of data to be transmitted for task offloading, For the moment vehicle Select RSUs The amount of data that can be transferred for task offloading, For the moment vehicle Select RSUs Transmission delay for task offloading, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula.

[0148] Furthermore, the task queuing delay includes:

[0149] According to the time The amount of data to be transferred determines the time it takes for the transfer to complete:

[0150] ,

[0151] in, is the task transmission completion time, For the moment The amount of data to be transmitted, The value range is ;

[0152] Calculate the task queuing delay based on the task transmission completion time:

[0153] ,

[0154] in, RSU computing power, RSU At the moment The number of tasks queued, time For vehicles The task reaches RSU moment, is the average computational requirement of a single task, is the time period between adjacent moments, For vehicles Select RSUs Required task queuing delay.

[0155] Furthermore, the calculation formula for the task calculation delay is:

[0156] ,

[0157] in, For vehicles Select RSUs the required task computation delay, It's a vehicle The computational requirements of the task, RSU computing power.

[0158] Furthermore, the step S3 of calculating the utility value includes:

[0159] Calculate the utility value based on the total latency of the task offloaded :

[0160] ,

[0161] in, For vehicles Select RSUs The total delay in performing task offloading, For vehicles Select RSUs The utility value of performing task offloading.

[0162] Furthermore, the step S4 updates the strategy probability distribution according to the utility value, and the calculation formula for updating the strategy probability distribution is:

[0163] ,

[0164] in, is the temperature parameter, For vehicles Select RSUs The utility value of task offloading, For vehicles Select RSUs The utility value of task offloading, For updated vehicles Select RSUs The probability distribution of strategies for task offloading, is the number of RSUs.

[0165] Furthermore, the step S4 of updating the temperature parameter includes:

[0166] According to the strategy probability distribution before the update and the current strategy probability distribution, calculate the strategy probability distribution change:

[0167] ,

[0168] in, is the change in strategy probability distribution, is the strategy probability distribution before updating, is the current strategy probability distribution, is the number of RSUs, is the number of vehicles;

[0169] Calculate the utility improvement rate based on the utility value before the update and the current utility value:

[0170] ,

[0171] in, Improve speed for utility, is the utility value before updating, is the current utility value, is the number of RSUs, is the number of vehicles;

[0172] Based on the change in the strategy probability distribution and the utility improvement rate, the temperature parameter is updated:

[0173] ,

[0174] in, is the updated temperature parameter, is the temperature parameter before updating, and is a constant, controlling the range of temperature parameters, is the temperature learning rate, and is a constant, a weight of the change in the probability distribution of the control strategy and the speed of utility improvement, , is the change in strategy probability distribution, Improve speed for utility, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula.

[0175] Furthermore, the step S5 checks the convergence of the strategy probability distribution, including:

[0176] The convergence threshold based on the change of the strategy probability distribution and the speed of utility improvement is used to determine whether the strategy probability distribution has converged. The convergence conditions are:

[0177] ,

[0178] in, is the change in strategy probability distribution, Improve speed for utility, is the convergence threshold of the strategy probability distribution change, Improve the speed convergence threshold for utility.

[0179] Furthermore, the step S5 of updating the status information includes:

[0180] For every moment , calculate RSU moment by moment The expected number of task arrivals is calculated as follows:

[0181] ,

[0182] in, For the moment RSUs The expected number of task arrivals, is the indicator function, 1 if yes, 0 otherwise. For vehicles The task reaches RSU moment, For vehicles Select RSUs Probability distribution of strategies for task offloading;

[0183] The current Recorded as , when the moment When each RSU is updated moment by moment Number of tasks queued:

[0184] ,

[0185] in, For the moment The number of tasks queued, Before the update at time The number of tasks queued, For the moment RSU The expected number of task arrivals, RSU computing power, is the average computational requirement of a single task, is the time period between adjacent moments, Indicates taking the maximum value in the formula.

[0186] The following is an introduction to a mobile vehicle task unloading system based on prediction and game provided by an embodiment of the present invention. The mobile vehicle task unloading system based on prediction and game described below and the mobile vehicle task unloading method based on prediction and game described above can refer to each other.

[0187] refer to Figure 3 , a mobile vehicle task offloading system based on prediction and game theory, comprising:

[0188] Initialization module: used to obtain basic information of RSU and vehicle, initialize strategy probability distribution and temperature parameters;

[0189] State calculation module: used to predict the future position trajectory of the vehicle, the number of RSU task queues, and the channel status;

[0190] Utility calculation module: builds a total delay model for task offloading, including transmission delay, queuing delay, and computation delay, and calculates the utility value;

[0191] Strategy update module: updates the strategy probability distribution according to the current utility value, adjusts the temperature parameter, and simulates the game process;

[0192] Convergence judgment module: determines whether the strategy has converged. If not, it updates the status information and continues iteration;

[0193] Decision generation module: Samples the task offloading decision for each vehicle based on the final policy probability distribution.

[0194] In this embodiment, the initialization module obtains basic information about the RSU and the vehicle, and initializes initial variables such as the strategy probability distribution and temperature parameters; the state calculation module predicts dynamic information such as the vehicle's future position trajectory, the number of tasks queued by the RSU, and the channel status; the utility calculation module constructs a total delay model for task offloading, including transmission delay, queuing delay, and calculation delay, and calculates the utility value; the strategy update module updates the strategy probability distribution according to the current utility value, adjusts the temperature parameter, and simulates the game process; the convergence judgment module determines whether the strategy has converged. If not, it updates the state information and continues to iterate; and the decision generation module samples the task offloading decision for each vehicle based on the final strategy probability distribution.

[0195] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0196] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A mobile vehicle task offloading method based on prediction and game theory, characterized in that: The steps are: S1: According to the mobile vehicle scenario, obtain RSU information and vehicle information and initialize the strategy probability distribution; S2: Predict the future based on RSU information The number of RSU tasks queued at a moment, based on vehicle information, predicts the future The vehicle position trajectory at each moment is used to calculate the future channel state; S3: Build a delay model and calculate the utility value; S4: updating the strategy probability distribution according to the utility value and updating the temperature parameter; S5: Check the convergence of the probability distribution of the strategy. If the convergence condition is met, execute S6; otherwise, update the state information and execute S3. S6: Based on the final strategy probability distribution, randomly sample a strategy as the final task offloading decision for each vehicle and assign the task to the corresponding RSU; The step S3 of constructing the delay model includes: For vehicles Select RSUs The total delay for task offloading is calculated as: ， in, For vehicles Select RSUs The total delay in performing task offloading, For the moment vehicle Select RSUs The transmission delay, For vehicles Select RSUs Required task queuing delay; For vehicles Select RSUs the required task computation latency; The step S3 of calculating the utility value includes: Calculate the utility value based on the total latency of the task offloaded : ， in, For vehicles Select RSUs The total delay in performing task offloading, For vehicles Select RSUs The utility value of performing task offloading; The step S4 updates the strategy probability distribution according to the utility value, and the calculation formula for updating the strategy probability distribution is: ， in, is the temperature parameter, For vehicles Select RSUs The utility value of task offloading, For vehicles Select RSUs The utility value of task offloading, For updated vehicles Select RSUs The probability distribution of strategies for task offloading, is the number of RSUs; The step S4 updates the temperature parameters, including: According to the strategy probability distribution before the update and the current strategy probability distribution, calculate the strategy probability distribution change: ， in, is the change in strategy probability distribution, is the strategy probability distribution before updating, is the current strategy probability distribution, is the number of RSUs, is the number of vehicles; Calculate the utility improvement rate based on the utility value before the update and the current utility value: ， in, Improve speed for utility, is the utility value before updating, is the current utility value, is the number of RSUs, is the number of vehicles; Based on the change in the strategy probability distribution and the utility improvement rate, the temperature parameter is updated: ， in, is the updated temperature parameter, is the temperature parameter before updating, and is a constant, controlling the range of temperature parameters, is the temperature learning rate, and is a constant, a weight of the change in the probability distribution of the control strategy and the speed of utility improvement, , is the change in strategy probability distribution, Improve speed for utility, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula; The step S5 checks the convergence of the strategy probability distribution, including: The convergence threshold based on the change of the strategy probability distribution and the speed of utility improvement is used to determine whether the strategy probability distribution has converged. The convergence conditions are: ， in, is the change in strategy probability distribution, Improve speed for utility, is the convergence threshold of the strategy probability distribution change, Improve the speed convergence threshold for utility.

2. A mobile vehicle task offloading method based on prediction and game theory as claimed in claim 1, characterized in that: The step S1 obtains RSU information and vehicle information according to the mobile vehicle scenario and initializes the strategy probability distribution, including: Get the number of RSUs ,The number of tasks at the historical moment, computing power, and the number of vehicles obtained are ,historical location trajectory; Define temperature parameters and temperature learning rate ; Define the convergence threshold of the strategy probability distribution change and utility improvement speed convergence threshold ; per vehicle Initialize a policy probability distribution , the calculation formula is: ， in, is the number of RSUs, Indicates vehicle Select RSUs The probability of task offloading.

3. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The step S2 predicts the future based on the vehicle information The vehicle position trajectory at each moment is used to calculate the future channel state, including: Utilizing the vehicle's location ,speed , acceleration Delivering the future The vehicle's position at a moment and speed , the recursive calculation formula is: ， ， in, It's a vehicle At the moment location, It's a vehicle At the moment speed, It's a vehicle At the moment The acceleration of is the time period between adjacent moments, is a modified value that follows a Gaussian distribution; The future predicted The vehicle's position at a moment , for each moment , calculate the future channel state moment by moment, the calculation formula is: ， in, It's a vehicle with RSU At the moment The future channel state, It's a vehicle At the moment location, RSU location, is the reference gain, is the path loss exponent, is the noise power.

4. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The step S2 predicts the future The number of RSU tasks queued at a moment, including: The number of tasks at historical moments in the RSU information Input into the trained LSTM model to get RSU future The number of tasks queued at a moment is expressed as: ,in, RSU The future moment The number of tasks queued.

5. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The calculation formula for the task calculation delay is: ， in, For vehicles Select RSUs the required task computation delay, It's a vehicle The computational requirements of the task, RSU computing power.

6. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The transmission delay includes: According to the vehicle Uninstall task size , calculate the transmission delay of the task, the specific steps are: Initialize the amount of data to be transmitted, expressed as: ;in, is the amount of data to be transferred for the offload task; For every moment , calculate the amount of data that can be transmitted, the transmission delay, and the amount of data to be transmitted at each moment, including: (a) Calculate the amount of data that can be transmitted: , (b) Calculate the transmission delay, including: ， (c) Update the amount of data to be transferred: , in, For vehicles Unload task size, is the bandwidth of the wireless channel, is the size of the moment, For the moment vehicle with RSU The future channel state between For the moment vehicle Select RSUs The amount of data to be transmitted for task offloading, For the moment vehicle Select RSUs The amount of data that can be transferred for task offloading, For the moment vehicle Select RSUs Transmission delay for task offloading, It means taking the maximum value in the formula, Indicates taking the minimum value in the formula.

7. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The task queuing delay includes: According to the time The amount of data to be transferred determines the time it takes for the transfer to complete: ， in, is the task transmission completion time, For the moment The amount of data to be transmitted, The value range is ; Calculate the task queuing delay based on the task transmission completion time: ， in, RSU computing power, RSU At the moment The number of tasks queued, time For vehicles The task reaches RSU moment, is the average computational requirement of a single task, is the time period between adjacent moments, For vehicles Select RSUs The required task queuing delay.

8. The method for unloading mobile vehicle tasks based on prediction and game theory as claimed in claim 1, characterized in that: The step S5 of updating the status information includes: For every moment , calculate RSU moment by moment The expected number of task arrivals is calculated as follows: ， in, For the moment RSUs The expected number of task arrivals, is the indicator function, 1 if yes, 0 otherwise. For vehicles The task reaches RSU moment, For vehicles Select RSUs Probability distribution of strategies for task offloading; The current Recorded as , when the moment When each RSU is updated moment by moment Number of tasks queued: ， in, For the moment The number of tasks queued, Before the update at time The number of tasks queued, For the moment RSU The expected number of task arrivals, RSU computing power, is the average computational requirement of a single task, is the time period between adjacent moments, Indicates taking the maximum value in the formula.

9. A mobile vehicle task unloading system based on prediction and game theory, characterized in that: Implementing a mobile vehicle task offloading method based on prediction and game theory as described in any one of claims 1 to 8, comprising: Initialization module: used to obtain basic information of RSU and vehicle, initialize strategy probability distribution and temperature parameters; State calculation module: used to predict the future position trajectory of the vehicle, the number of RSU task queues, and the channel status; Utility calculation module: builds a total delay model for task offloading, including transmission delay, queuing delay, and computation delay, and calculates the utility value; Strategy update module: updates the strategy probability distribution according to the current utility value, adjusts the temperature parameter, and simulates the game process; Convergence judgment module: determines whether the strategy has converged. If not, it updates the status information and continues iteration; Decision generation module: Samples the task offloading decision for each vehicle based on the final policy probability distribution.

Images