Internet of vehicles task processing based on cooperation of multiple road side units

Abstract

The invention discloses a method for processing vehicle tasks through cooperation of multiple road side units. Comprising the following steps: training a prediction model supporting online learning by using historical data, and regularly updating the model by using new data; vehicle data of a task to be unloaded is predicted through the model, the final position of the vehicle after the task is completed is predicted, and task processing delay is estimated; using a resource transaction algorithm to calculate a trading price according to the task unloading related data; and judging whether the transaction is performed smoothly by using a coordinator, determining an actual transaction price, and selecting a road side unit cooperation scheme. According to the method, the dynamic change and elastic requirements of the urban vehicle load information in time and space are met, the success rate of vehicle task processing is improved, and the task processing delay is reduced at the same time.

Description

Technical Field

[0001] The present invention relates to the field of planning for construction of edge computing infrastructure for Internet of Vehicles, and in particular to a method for multiple roadside units to cooperate and assist in vehicle task processing. Background Art

[0002] In recent years, the Internet of Vehicles has attracted much attention due to the rapid development of intelligent transportation applications such as autonomous driving, traffic prediction, and road safety. These applications usually have strict latency constraints and require a large amount of computing resources. However, due to the lack of resources on the vehicle, only a small part of the data and computing can be processed locally. To address this problem, one strategy is to transfer data to a remote, centralized, and resource-rich cloud server for processing. However, cloud servers are usually too far away from the vehicle, resulting in unstable connections and high latency. In addition, when a large number of vehicles send a large amount of data to a remote data center, the backbone network will be congested and latency will increase. Compared with cloud computing, edge computing can provide computing offload services for vehicles because edge servers are deployed near the vehicle, such as roadside units or base stations. Due to the short distance to the vehicle, edge servers can help vehicles process computing-intensive tasks, reduce vehicle latency and energy consumption, and avoid overloading backbone networks and centralized data centers. To solve the problem of limited computing resources in vehicles, the main method currently proposed by researchers is task offloading; some researchers have proposed solutions for optimizing task scheduling in cloud computing systems, but due to the remote location of the cloud center, the transmission delay may be large, which affects the processing of time-sensitive tasks; to solve the problem of long-distance transmission and computing, a common practice is to place edge servers on roadside units; some researchers currently propose to use the idle computing resources of vehicles to assist roadside units, but the mobility of vehicles makes it difficult for vehicles to fully process tasks; some technical solutions also propose to build cooperation between roadside units. The research results show that the above methods can effectively optimize task offloading and achieve efficient use of computing resources. It can be seen that many existing studies have proposed meaningful solutions for vehicle task offloading. However, many studies consider static environments and are difficult to adapt to vehicle networking scenarios with large or highly dynamic traffic flows. Based on this, a task processing strategy for roadside unit collaboration that is suitable for high-volume or highly dynamic traffic flow scenarios needs to be developed urgently. Summary of the invention

[0003] The purpose of the present invention is to provide a vehicle task unloading processing method with collaborative cooperation of multiple road side units, which is used to solve the vehicle networking scenario problems of large traffic volume and highly dynamic traffic flow. The technical solution of the present invention is: A method for multi-road side unit cooperation to assist vehicle task processing, comprising: Use historical data to train neural network models and optimize in real time; The model is used to predict the vehicle's location data in real time to obtain the specific information of the vehicle when the task is completed; Using resource trading algorithms, the buying and selling prices are calculated based on task offloading related data; The coordinator is used to determine whether the transaction can be carried out, determine the actual transaction price, and determine the roadside unit cooperation plan. The use of historical data to train the neural network model and optimize in real time includes: The neural network model with online learning and long short-term memory architecture is trained, and data is updated regularly to ensure that the model can adapt to the dynamic computing needs of the Internet of Vehicles. The online learning neural network model using the long short-term memory architecture is trained and data is updated regularly to ensure that the model can adapt to the dynamic Internet of Vehicles computing needs, including: Analyze historical data related to traffic patterns, vehicle locations, and roadside unit availability; Perform data cleaning, set hyperparameters, determine network structure and loss function, and iteratively train the model so that the model can understand the underlying patterns and dependencies in the data; Regularly add a small amount of new data to the training model and update the model weight parameters so that the model can be continuously optimized over time and improve performance. The real-time prediction of the vehicle's position data by the model to obtain the vehicle information when the task is completed includes: Use the model data trained in the experience pool to predict the final position and task processing delay when the task is completed; Determine the roadside units that receive tasks and process tasks respectively. The method of using the model data trained in the experience pool to predict the final position and task processing delay when the task is completed includes: The position data of each vehicle in the experience pool is stored as an array, including the known vehicle initial position, initial speed, initial acceleration, task size, road congestion coefficient and the unknown vehicle final position. The prediction model is used to predict the final position of the vehicle after the task is completed, and then the task processing delay corresponding to each roadside unit is estimated. The roadside unit for respectively determining the receiving task and the processing task comprises: Before predicting the final position of the vehicle when the task is completed, the model looks for the roadside unit that is closest to the position and has sufficient remaining computing power as the roadside unit that receives the task; After predicting the final position of the vehicle when the task is completed, the model searches for a qualified roadside unit again as the roadside unit to process the task, forming an interconnected network with the roadside unit that receives the task. The use of resource trading algorithm to calculate the buying and selling prices according to task offloading related data includes: The bid and ask price for each vehicle is calculated using an online learning-based driven resource trading algorithm. The method of calculating the bid and ask price of each vehicle using the online learning-based resource trading algorithm includes: The bid for each vehicle is calculated based on the vehicle's maximum tolerable delay, total estimated processing delay time, and the amount of tasks to be unloaded. The asking price for each roadside unit is calculated by calculating the cost and difficulty of task execution. The coordinator is used to determine whether the transaction can be carried out, determine the actual transaction price, and determine the roadside unit cooperation plan including: By comparing the bid and the asking price, we determine whether the transaction will be successful, and by coordinating the two, we determine the payment price and select the final solution. The comparison between the bid and the asking price to determine whether the transaction is successful, and the coordination between the two to determine the payment price and select the final solution include: If the RSU lacks the resources to complete the task and the bid is less than the asking price, it will not participate in the auction and the transaction fails; If the RSU has the resources required to complete the task and the bid is greater than the asking price, it can participate in the auction; For each matching scenario, the actual price paid was determined as the average of the bid and ask prices; The solution with the lowest actual payment price is selected as the final solution. The beneficial effects of the present invention include at least: The method described in the present invention first trains an online learning neural network model of a long short-term memory architecture through past data to obtain a model that supports online learning; then uses the model to predict the position data of vehicles in an experience pool in real time to obtain the final position of the vehicle and the task processing delay when the task is completed; then determines the roadside unit that receives the task based on the initial position, and determines the roadside unit that processes the task based on the prediction result; then uses a double auction mechanism to calculate the buying and selling prices based on task unloading related data; determines whether the transaction can be carried out through a coordinator, determines the actual transaction price, and selects the optimal task solution. The method described in the present invention optimizes the task processing process of the Internet of Vehicles by cooperating with multiple road side units to assist in vehicle task processing, and meets the dynamic changes and elasticity requirements of urban vehicle load information in time and space; through tracking-driven simulation experiments conducted on real data sets, it is concluded after analyzing the results that the method described in the present invention can improve the success rate and task delay of task processing, improve user satisfaction and experience, and avoid data overload at the central node and data over-emptiness at the edge node compared to other popular methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a model diagram of the connected vehicle system where a vehicle moves relative to the roadside unit within a certain period of time; Figure 3 It is a schematic diagram of a model in which roadside units are interconnected to assist vehicles in handling tasks; Figure 4 It is a schematic diagram of the deployment of roadside units; Figure 5 is a comparison chart of task processing success rate and relative delay between the comparison method and the method described in this application; Figure 6 It is a comparison chart of the task processing success rate and relative delay of the comparison method and the method described in this application under different congestion conditions; Figure 7 It is a deployment diagram after changing the number of roadside units; Figure 8 It is a comparison chart of the task processing success rate and relative delay of the comparison method and the method described in this application under different numbers of road side units. DETAILED DESCRIPTION

[0005] In order to cope with the demand for high bandwidth and low latency, edge technology is an important computing architecture that pushes data processing and computing resources to the edge of the network for real-time processing closer to the data source. Edge servers can effectively process data from multiple users while reducing data transmission time and bandwidth consumption. Therefore, edge computing is more flexible and efficient, and also provides new methods for task offloading and resource optimization. However, due to the changes in road traffic conditions in time and space, we need a task offloading solution that adapts to large or highly dynamic traffic flows. Traditionally, the solution to the problem of edge servers changing dynamically with the spatiotemporal changes of traffic is mainly static deployment, that is, installing a certain number of edge servers at fixed locations, and the planning is based on historical traffic data. However, this method has many limitations, such as insufficient flexibility, difficulty in dynamic adjustment, and inability to adapt to traffic flow in real time according to changes in user needs. At the same time, this method has low resource utilization, which may cause insufficient or excessive resources in some areas. In addition, it may also cause excessive time delay. To solve this problem, a common method is to use dynamic traffic data and algorithms to optimize the work of edge servers to adapt to the spatiotemporal changes of traffic. In this context, how to choose the right and appropriate algorithm to allocate computing resources to meet user needs, reduce latency, adapt to different traffic conditions, and improve task success rate has become a key technical issue. This challenge requires us to comprehensively consider the computing resources of edge servers, the size of user tasks, distance, latency, and other issues, as well as the coordination of edge servers caused by dynamic changes in users. Currently, many solutions to this problem are still in the initial stage. The main challenge is that the computing workload of edge servers often changes in time and space due to the dynamic changes in traffic flow, which requires cross-server cooperation. Existing strategies and many solutions lack flexibility in solving dynamic problems. Although they reduce latency and optimize computing resources, they often ignore user mobility and have difficulty adapting to the spatiotemporal changes of traffic. Therefore, in order to meet user needs, reduce latency, adapt to different traffic conditions, and improve task success rate, this embodiment designs a multi-edge server collaborative task offloading strategy based on an auction mechanism. First, through historical data, a neural network model that supports online learning is constructed and trained, and the dynamic computing needs of the vehicle network are met by updating the model in real time; then, according to the trained model, the vehicle to be unloaded is predicted, and the final position of the vehicle and the task processing delay are predicted; then, according to the initial position and the prediction results, the roadside units that receive the task and process the task are selected respectively, and a collaborative Internet network of edge servers is established to meet dynamic traffic conditions; then, the double auction mechanism is used to calculate the bid and the asking price based on the task offloading related data; the coordinator is used to determine the actual transaction price, determine the selection of the roadside unit cooperation plan, and meet the dynamic elasticity requirements of vehicle load information in time and space. The overall idea of ​​the present invention is: First, the traffic flow movement pattern is analyzed to draw the conclusion that the cooperation of roadside units can be used to optimize the task processing of the Internet of Vehicles. A prediction model is established. This embodiment establishes an online learning neural network model with a long short-term memory architecture, and maintains the prediction accuracy by updating the model in real time. Finally, based on the predicted data, this embodiment uses a double auction mechanism to match vehicles and roadside units that handle tasks, and determines the cooperative roadside units, which helps to significantly reduce processing delays while improving the success rate of task processing. Below, the present application is further described in conjunction with the accompanying drawings. Specific embodiment 1: like Figure 1 , a vehicle network task processing optimization method based on roadside unit edge server cooperation, specifically comprising the following steps: Step 1: Initialize the long short-term memory neural network prediction model that supports online learning: Integrate the existing data into experience and put it into the experience pool. For example, the data of vehicle k can be integrated into an experience array containing the vehicle's initial position, vehicle's initial speed, vehicle's initial acceleration, the amount of tasks the vehicle needs to handle, road congestion coefficient, and vehicle's final position, recorded as Ep k = {LS k , Vel k , Acc k , TS k ,Con k , L.E. k Next, a long short-term memory neural network model is established, and the model is initially trained using the existing experience in the experience pool. Step 2: Vehicle task processing strategy: The technical solution described in this embodiment adopts a resource trading algorithm driven by online learning, which aims to select a roadside unit that is more suitable for processing the task by predicting the initial position and final position of the vehicle, thereby improving the success rate of the task and reducing delays. Figure 3 It is a schematic diagram of task transmission and processing. a and R b Responsible for handling the tasks of a certain vehicle, the tasks are transmitted from the vehicle to R a Then R a Transfer tasks to R through high-speed interconnect b , R b After processing the task, the result is transmitted back to the vehicle. This embodiment specifically includes the following steps: 1. Integrate the data of the vehicle k to be unloaded into a state array, including the initial position of the vehicle, the initial speed of the vehicle, the initial acceleration of the vehicle, the amount of tasks that the vehicle needs to handle, and the road congestion coefficient, recorded as Pre k = {LS k , Vel k , Acc k , TS k ,Con k}, as the input of the neural network. 2. According to the initial position LS k , select the nearest roadside unit, and record the roadside unit as R a The vehicle transfers the task to R a . 3. Use neural network to predict the final position LE of the vehicle k1 , and then estimate the delay when using different roadside units to process the task, and use the prediction results to conduct auction matching and select the roadside unit R that processes the task b : In this embodiment, the vehicle is the buyer and the roadside unit is the seller. The bid of vehicle k for roadside unit i and the maximum tolerable delay D of the vehicle max , estimate the delay when using roadside unit i to process tasks And the amount of tasks that the vehicle needs to handle TS k related to in Including the transmission delay DI of the task from vehicle k to roadside unit i k , the processing delay P of the task in RSU i k , the waiting time W that roadside unit i waits for vehicle k to arrive in its coverage area k , and the transmission delay DO of the processing result from RSU i to vehicle k k . When the remaining resources of the roadside unit are sufficient to handle vehicle tasks, it can participate in the auction. The price that the roadside unit i asks for vehicle k is the computational cost Cost i and task processing difficulty DF k The trade-off, Price i =α×Cost i +(1-α)×DF k Among them, Cost i It can be regarded as a fixed value, involving the hardware equipment cost, equipment maintenance cost, power consumption, etc. of roadside unit i. k and the task size and the maximum tolerable delay D of the vehicle max Correlation, the calculation formula is DF k =β×TS k -D max . Compare the bid and asking price. i >Price i , then the matching strategy is feasible and generates an estimated payment price For multiple strategies, the vehicle will eventually select the matching strategy corresponding to the minimum payment price and record the matching roadside unit as R b . 4. R a The task is transmitted to the RSU through the high-speed interconnection of the RSU. b Processing is performed and the actual final position LE of the vehicle at the time of task processing completion is recorded k2 . Step 3: Update the prediction model: LE k2 and the initial state array Pre k Integrate into a new experience array Ep k = {LS k , Vel k, Acc k , TS k ,Con k , L.E. k2}, put it into the experience pool. At regular intervals t, the updated experience pool is used to update the weight parameters of the prediction model. Verification: The effectiveness of the method described in this example was verified by two comparative experiments: 1. Random algorithm: The default roadside unit R that receives the task a is the roadside unit closest to the vehicle, and randomly matches the vehicle with the roadside unit R that handles the task b . 2. Greedy algorithm: The default roadside unit R that receives the task a is the roadside unit closest to the vehicle. In order to ensure that the task result will not fail in the process of transmitting the task result back to the vehicle, the algorithm always greedily selects the farthest roadside unit as the roadside unit R to process the task. b . Each comparison method proposes a roadside unit cooperation strategy. The comparison experiment designed in this embodiment is intended to strictly evaluate the performance of these methods relative to the method proposed in this embodiment to prove their effectiveness in optimizing the task processing success rate and reducing the task processing delay. Figure 4 This is a schematic diagram of the deployment of roadside units on a road section. This deployment scheme uses the minimum number of roadside units to cover the entire road section. Figure 5 The figure below is a comparison of the task processing success rate and relative latency of different methods. The task processing success rate refers to the ratio of successfully processed tasks to the total number of tasks, and the relative latency refers to the ratio of the increased latency compared to the ideal situation. Our method significantly outperforms the comparison algorithm in terms of both task processing success rate and relative latency. Figure 6 This is a comparison of the task processing success rate and relative latency of different methods under different congestion levels. In all congestion conditions, our method outperforms other algorithms, with the success rate approaching 100%, and although the latency increases slightly as the congestion worsens, it is still significantly lower than the comparison algorithm. Figure 7 These are diagrams of several different roadside unit deployment plans. Figure 8 This is a comparison of the task processing success rate and relative latency of different methods under different roadside unit deployment schemes. When the number and location of roadside units are different, the two indicators of our proposed method are still optimal. The above-mentioned various comparative experiments show the superiority of the algorithm of this embodiment in optimizing the task processing success rate and reducing the task processing delay. In summary, the present invention is based on urban traffic planning. In this study, this embodiment explores the optimization problem of vehicle task processing in the Internet of Vehicles environment. First, a prediction model that supports online learning is established, and the model is preliminarily trained using historical data; then, the model is used for prediction, and the prediction results and the double auction mechanism are used to optimize the matching strategy between the vehicle and the roadside unit, and select the appropriate roadside unit for collaboration. Subsequently, the collaboration strategy is used to process the task, and the vehicle status data after processing is updated to the experience pool. Finally, the updated experience pool is used to update the prediction model so that the model can better adapt to the dynamic Internet of Vehicles computing needs. In general, the method described in the present invention realizes the effective utilization of edge computing resources, improves the success rate of vehicle task processing and reduces task processing latency, ensuring that the edge computing resources in the Internet of Vehicles environment can better meet service needs. The above disclosures are only several specific implementation scenarios of the present invention, but the present invention is not limited thereto, and any changes that can be thought of by a person skilled in the art should fall within the protection scope of the present invention. The above serial numbers of the present invention are only for description and do not represent the advantages and disadvantages of the implementation scenarios.

Claims

1. A method for processing vehicle tasks by cooperating with multiple roadside units, characterized in that: include: Use historical data to train neural network models and optimize in real time; The model is used to predict the vehicle's location data in real time to obtain the specific information of the vehicle when the task is completed; Using resource trading algorithms, the buying and selling prices are calculated based on task offloading related data; The coordinator is used to determine whether the transaction can be carried out, determine the actual transaction price, and determine the roadside unit cooperation plan.

2. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 1, characterized in that: Use historical data to train neural network models and optimize in real time, including: The neural network model with online learning and long short-term memory architecture is trained, and data is updated regularly to ensure that the model can adapt to the dynamic computing needs of the Internet of Vehicles.

3. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 2, characterized in that: The neural network model with online learning and long short-term memory architecture is used for training, and data is updated regularly to ensure that the model can adapt to the dynamic computing needs of the Internet of Vehicles: Analyze historical data related to traffic patterns, vehicle locations, and roadside unit availability; Perform data cleaning, set hyperparameters, determine network structure and loss function, and iteratively train the model so that the model can understand the underlying patterns and dependencies in the data; Regularly add a small amount of new data to the training model and update the model weight parameters so that the model can be continuously optimized over time and improve performance.

4. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 1, characterized in that: The model is used to predict the vehicle's location data in real time to obtain the vehicle's information when the task is completed, including: Use the model data trained in the experience pool to predict the final position and task processing delay when the task is completed; Determine the roadside units that receive tasks and process tasks respectively.

5. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 4, characterized in that: Use the model data trained in the experience pool to predict the final position and task processing delay when the task is completed, including: The position data of each vehicle in the experience pool is stored as an array, including the known vehicle initial position, initial speed, initial acceleration, task size, road congestion coefficient and the unknown vehicle final position. The prediction model is used to predict the final position of the vehicle after the task is completed, and then the total task processing delay corresponding to each roadside unit is estimated.

6. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 4, characterized in that: Determine the roadside units that receive tasks and process tasks, including: Before predicting the final position of the vehicle when the task is completed, the model looks for the roadside unit that is closest to the position and has sufficient remaining computing power as the roadside unit that receives the task; After predicting the final position of the vehicle when the task is completed, the model searches for a qualified roadside unit again as the roadside unit to process the task, forming an interconnected network with the roadside unit that receives the task.

7. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 1, characterized in that: Using the resource trading algorithm, the buying and selling prices are calculated based on the task offloading related data, including: The bid and ask price for each vehicle is calculated using an online learning-based driven resource trading algorithm.

8. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 7, characterized in that: The bid and ask price for each vehicle is calculated using an online learning-based resource trading algorithm, including: The bid for each vehicle is calculated based on the vehicle's maximum tolerable delay, total estimated processing delay time, and the amount of tasks to be unloaded. The asking price for each roadside unit is calculated by calculating the cost and difficulty of task execution.

9. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 1, characterized in that: The coordinator is used to determine whether the transaction can be carried out, determine the actual transaction price, and determine the roadside unit cooperation plan, including: By comparing the bid and the asking price, we determine whether the transaction will be successful, and by coordinating the two, we determine the payment price and select the final solution.

10. The method for processing vehicle tasks by cooperating with multiple road side units according to claim 9, characterized in that: By comparing the bid and the asking price, we determine whether the transaction is successful, and by coordinating the two, we determine the payment price and select the final solution, including: If the RSU lacks the resources to complete the task and the bid is less than the asking price, it will not participate in the auction and the transaction fails; If the RSU has the resources required to complete the task and the bid is greater than the asking price, it can participate in the auction; For each matching scenario, the actual price paid was determined as the average of the bid and ask prices; The solution with the lowest actual payment price is selected as the final solution.

Images