Online Decision-making Method and Device for Vehicle Service Requests in the S3CE System

Abstract

The present invention discloses an online decision-making method and device for vehicle service requests in an S3CE system. The method includes: obtaining the position of a moving vehicle at any moment in a vehicle-road cooperation scenario according to a vehicle movement trajectory prediction model to obtain the transmission range between the vehicle and the roadside unit; establishing a delay model and a cost model related to vehicle service request scheduling, and forming an optimization problem equation for joint decision-making of request scheduling and container reservation under the conditions of meeting distance restrictions and service container resource restrictions, with the optimization objective being to minimize the long-term cost of the system under the premise of meeting the service request delay deadline; decomposing the problem into a request transfer decision problem and a regional S3CE cost problem, and using the STD algorithm and the ASP algorithm to solve them respectively; based on the solution results, forming an online joint decision-making scheme to realize real-time scheduling of vehicle service requests and service container management. The present invention can perform real-time request scheduling and container reservation decisions under vehicle mobility and changes in vehicle service requirements.

Description

Technical Field

[0001] The present invention relates to an online decision-making method and device for vehicle service requests in an S3CE system, and belongs to the field of vehicle-road collaborative scheduling in edge computing. Background Art

[0002] In the Vehicle-Road Collaboration (VRC) scenario, many new in-vehicle applications are emerging, such as augmented reality, driving behavior detection, license plate recognition, etc. Due to the limited computing and storage capabilities of vehicles themselves, it is difficult to meet the requirements of various intelligent transportation service requests. Cloud computing resources are powerful and can configure various in-vehicle service containers, but the communication is unstable and the latency is high; the roadside edge computing nodes are close to the vehicles and are expected to provide convenient service computing for the vehicles. However, due to resource limitations, the number of service containers that can be configured is limited, and it may not be able to meet the vehicle service requests. That is, the method of using cloud computing or edge computing to provide service computing for connected vehicles cannot guarantee the quality of vehicle service requests. The industry has proposed a vehicle-road collaboration solution based on Serverless cloud-edge collaboration computing (S3CE), which cooperates with edge computing and cloud computing to jointly provide various service request calculations for moving vehicles.

[0003] In the S3CE system, on the one hand, it is necessary to configure service containers for various requests of vehicles. Due to the cold start latency of the containers, the service containers configured on the roadside edge face the decision problem of whether to retain them. If the container is retained, that is, the active state of the container is maintained, the cold start latency caused after the service is scheduled can be avoided, but the S3CE system needs to pay the maintenance cost (resource occupation); if the container is not retained, that is, the container is closed, although it does not occupy server resources, once the relevant service request is distributed, the request cannot be completed. At this time, the S3CE system may need to pay the cost of transferring the request or the cost of restarting the service. On the other hand, the S3CE system also needs to dynamically schedule the service requests of moving vehicles. The request scheduling needs to consider both the mobility of the vehicles and the status of the requested services deployed on the roadside edge servers. In addition, the adjacent edges in different regions in the S3CE system can cooperate with each other, and the vehicle requests that cannot be satisfied in this region can also be transferred to the neighboring edges for processing.

[0004] In the context of vehicle-road cooperation, in the face of various dynamic service requests randomly generated by moving vehicles, the S3CE system needs to perform request scheduling and retention decisions for service containers on each server in real time. While completing a large number of vehicle request service calculations, it should pay as low a cost as possible. In addition, due to the development of machine learning in recent years, advanced artificial intelligence (AI) technologies have been applied to various fields. Generally, the interaction between request scheduling and container retention and the S3CE system environment can be modeled as a Markov Decision Process (MDP) problem, and Deep Reinforcement Learning (DRL) technologies can effectively solve this problem. However, most of the existing decision-making schemes based on DRL technologies are centralized offline decision-making strategies, which require scheduling control based on global information and are not suitable for the dynamically changing vehicle-road cooperation scenarios.

[0005] The existing technical solution most similar to the present invention: Batch et al. proposed a task allocation scheme based on regret matching (RM) learning, aiming to minimize the total delay and cost generated by vehicle task offloading in highway scenarios. This work assumes that when a vehicle leaves the coverage area of its serving Road Side Unit (RSU), it will migrate the task to other suitable RSUs or adjacent base stations according to its movement pattern. First, the task allocation problem is formulated as a binary nonlinear programming (BNLP) problem with specific constraints; then, the BNLP problem is transformed into a standard repeated game problem, and then a distributed RM algorithm is proposed. This algorithm decomposes the centralized agent's observation and action into the observations and actions of multiple agents. This iterative game-based learning algorithm can guarantee an equilibrium solution. In particular, this scheme also proposes a forgetting method to accelerate the convergence speed of the traditional RM algorithm, so that the proposed algorithm can effectively handle the high-level mobility problem of vehicles.

[0006] Disadvantages of the prior art: Most of the existing research on vehicle networking task scheduling does not consider the service request types of the scheduled tasks, assuming that the edge server has already configured the task services to be scheduled. However, in reality, especially in the vehicle-road cooperation scenario, the tasks offloaded by moving vehicles all have service attributes, and the configuration and management of services need to be considered during request scheduling. The service request pattern of vehicles in the vehicle-road cooperation scenario is not prior knowledge, the number of service containers that the roadside edge server can accommodate is limited, and it requires a system cost to keep the containers active. Therefore, considering service container retention while scheduling task requests is a very challenging problem. Summary of the Invention

[0007] Objective of the Invention: The present invention proposes an online decision-making method and device for vehicle service requests in an S3CE system, aiming to solve the problems of vehicle service request scheduling and edge service container reservation in vehicle-road cooperation. The proposed method can make real-time request scheduling and container reservation decisions under the changes of vehicle mobility and vehicle service demands, so as to minimize the cost of the cloud-edge collaborative computing system while meeting the latency requirements of service requests.

[0008] Technical Solution: To achieve the above objective of the invention, the technical solution of the present invention is as follows:

[0009] An online decision-making method for vehicle service requests in an S3CE system includes the following steps:

[0010] S1. According to the vehicle movement trajectory prediction model of constant yaw rate and acceleration CYRA, obtain the position of the moving vehicle at any moment in the vehicle-road cooperation scenario, and obtain the transmission range R between the vehicle and the roadside unit RSU, V2R which is used as the distance limit for vehicle service request scheduling;

[0011] S2. Establish a latency model and a cost model related to vehicle service request scheduling. Under the conditions of meeting the distance limit and the service container resource limit of the S3CE system, form an optimization problem equation for joint decision-making of request scheduling and container reservation, and the optimization objective is to minimize the long-term cost of the S3CE system under the premise of meeting the latency deadline of service requests;

[0012] S3. Analyze the problem equation, decompose the problem into a request transfer decision problem and a regional S3CE cost problem, use the transfer decision STD algorithm to solve the request transfer decision problem, and use the adaptive slope-based piecewise linear function ASP optimization algorithm to solve the regional S3CE cost problem;

[0013] S4. Based on the solution results of the STD and ASP algorithms, form an online joint decision-making scheme to realize the real-time scheduling of vehicle service requests and the management of service containers in the S3CE system, and minimize the long-term cost of the S3CE system.

[0014] Furthermore, the step S1 includes:

[0015] The position of the moving vehicle at time slot t is (x(t), y(t)). According to the longitudinal speed v, acceleration a, yaw angle θ, and yaw speed w of the vehicle, determine the time slot t m The position of the vehicle is:

[0016]

[0017] where Δt = t m - t, c x and c yis a fixed initial value constant;

[0018] According to the real-time position of the vehicle, obtain the distance d between the moving vehicle and the RSU at each time slot t R (t), and determine C based on the log-distance path loss model pl(d) = C + 10n f log 10 d;

[0019] Obtain the communication range between the moving vehicle and the RSU wherein, is the transmission power of the RSU at time slot t, represents the threshold of the vehicle received signal strength, n f is the free space path loss parameter.

[0020] Furthermore, in the step S2, the objective of the established optimization problem is as follows:

[0021]

[0022] M t is the total cost for the S3CE system to process all vehicle requests at time slot t, M i is the total cost generated by the S3CE system to process request r i , is the transmission cost for processing request r i ; is the computing cost for processing request r i ; is the transfer cost for processing request r i ; is the container reservation cost for processing request r i ; is the preparation cost for processing request r i ;

[0023] The problem satisfies the following constraint conditions: C.1, the vehicle service request at each time slot t can be scheduled to and only be scheduled to one RSU for processing; C.2, the distance between the vehicle generating the service request and the scheduled RSU must be within their communication range; C.3, all mirror layers of the required service containers must be available on the scheduled RSU; C.4, the number of containers running simultaneously on each RSU n needs to meet its quantity limit; C.5, the total time delay of the vehicle service request is within its deadline range;

[0024] The purpose of the optimization problem is to determine the request scheduling decision, that is, select an RSU in the area to process the request or transfer the request, and the container reservation decision, that is, whether to keep the service container active;

[0025] Among them, b represents the base station b within the coverage area of the RSU, β represents the set of base stations, n represents the nth RSU, and N b represents the set of RSUs within the coverage area of base station b, k represents the kth service type, represents the set of service types, i represents the ith service request, and R k represents the set of requests with service type k.

[0026] Furthermore, the transmission cost is the time of occupying the bandwidth,

[0027] the computing cost is the time of occupying the CPU resources,

[0028] the transfer cost is the time of occupying the bandwidth between base stations,

[0029] the container retention cost is where is the cost of keeping the service container k active on RSUn, where represents the unit price of retaining the service container k for one time slot on RSUn;

[0030] the preparation cost is the generated preparation time,

[0031] Among them, λ t 、λ c 、λ tr 、λ r and λ p are the transmission cost coefficient, computing cost coefficient, transfer cost coefficient, container retention cost coefficient, and preparation cost coefficient respectively; represents the transmission delay generated by scheduling the request r i to RSUn, and B i,n (t) represents the available bandwidth; represents the computing delay generated by processing the request r i on RSUn; w k represents the workload of service type k; represents the transfer delay generated by transferring the request from the BSb area to area; represents BSb and the link bandwidth between; represents the state of the container of service type k on RSUn at time slot t; represents the preparation delay generated by deploying the container.

[0032] Furthermore, in the step S3, the analysis and decomposition of the problem include:

[0033] Analyzing the problem, it can be known that the situation of service containers on all RSUs in the previous time slot BSb area To a certain extent, it determines the request scheduling decision, whether to process in this area or transfer to the neighboring BS area. Let represent the set of requests transferred to the BSb area. Based on the modeling formula of problem P1, the total cost generated by the regional S3CE system of BSb for processing all vehicle service requests is Therefore, problem P1 can be decomposed into the S3CE system service request transfer decision problem and several regional S3CE system cost sub-problems. Among them, the equation of the S3CE system service request transfer decision problem is expressed as follows:

[0034]

[0035] The equation of the regional S3CE system cost sub-problem is as follows:

[0036]

[0037] C.2 - C.5.

[0038] Furthermore, in step S3, using the transfer decision STD algorithm to solve the request transfer decision problem includes:

[0039] Define the BS that covers the predicted position of the deadline of the service request vehicle in the communication range as PT - BS; define the RSU that is always within the communication range of the service request vehicle within the request deadline as P - RSU;

[0040] At time slot t, for any vehicle service request r in the S3CE system i , obtain its PT - BS set and P - RSU set; traverse the PT - BS set to calculate different transfer costs, select the minimum transfer cost and its corresponding PT - BS, and denote them as PT - BS Traverse the P - RSU set, based on the container image layer sharing relationship, select the minimum preparation cost and its corresponding P - RSU, and denote them as P - RSU n; if then Otherwise, where is the scheduling decision variable, indicates that request r is scheduled to be executed on RSUn at time slot t, and 0 means it is not scheduled to be executed on RSUn; is the transfer decision variable, indicates that request r at time slot t i is transferred to Execute on the area. If it is 0, it will not be transferred to Execute on the area.

[0041] Furthermore, in the step S3, using the piecewise linear function ASP optimization algorithm based on the adaptive slope to solve the regional S3CE cost problem includes:

[0042] Convert the objective of problem P3 into minimizing the long-term expected cost value:

[0043]

[0044]

[0045] C.2 - C.5

[0046] where S t is the system state of the regional S3CE at time slot t, including all vehicle information and all RSU information within the coverage area; W t is the random information, including various vehicle service request arrival information and mobile vehicle trajectory information;

[0047] Based on the Bellman optimality principle, convert the multi-step optimization decision problem in P4 into a recursive problem, that is, the objective in P4 becomes:

[0048]

[0049] where V t (S t ) is the system cost value of the regional S3CE at time slot t, and V t+1 (S t+1 |S t ) is the system cost value of the regional S3CE at time slot t + 1 on the premise of the state S t ; ζ ∈ [0, 1] is the future decay factor;

[0050] Introduce the post-decision state function to replace the expected value term E(V t+1 (S t+1 |S t )): where D- represents the pre-decision state, D represents the post-decision state, and P4 is converted to P5:

[0051]

[0052] Use to represent the piecewise linear function, where represents the system average cost value from the initial to time slot t, h t,s represents the slope of the piecewise linear function, and the function is evenly divided into segments Nt A function, and the slope of the function is monotonically increasing. P5 is written as P6:

[0053]

[0054] At t = 1, update the system state according to the latest random information Make a decision X by solving P5 under the constraint conditions t , A t , calculate the system cost value t = t + 1, perform iterative calculation until t > T;

[0055] where the slope is designed to be adaptively updated iteratively:

[0056]

[0057] where, α f-1 is the step size for smoothing the estimated slope; represents the sample observation value of the marginal value, and is defined as follows:

[0058]

[0059] Furthermore, the step S4 includes:

[0060] S41: Obtain the approximate cost function for each time slot t according to the online optimization algorithm of the adaptive approximate cost function; Let the time slot t = 1;

[0061] S42: Input the real-time vehicle information and RSU information;

[0062] S43: Obtain the transfer decision according to the simple transfer decision algorithm in step S3;

[0063] S44: Obtain the request scheduling decision X by solving P6 under the condition of satisfying the constraint conditions t and the container reservation decision A t ;

[0064] S45: t = t + 1, return to S42 until t > T.

[0065] The present invention also provides an online decision-making device for vehicle service requests in an S3CE system, including:

[0066] A transmission range determination module, configured to obtain the position of a moving vehicle at any time in a vehicle-road cooperation scenario according to a vehicle movement trajectory prediction model of a constant yaw rate and acceleration CYRA, and obtain the transmission range R between the vehicle and the roadside unit RSU V2R , as the distance limit for vehicle service request scheduling;

[0067] An optimization problem construction module, configured to establish a delay model and a cost model related to vehicle service request scheduling, and form an optimization problem equation for joint decision-making of request scheduling and container reservation under the conditions of satisfying distance constraints and S3CE system service container resource constraints, with the optimization objective of minimizing the long-term cost of the S3CE system under the premise of meeting the delay deadline of service requests;

[0068] An optimization problem analysis and solution module, configured to analyze the problem equation, decompose the problem into a request transfer decision problem and a regional S3CE cost problem, solve the request transfer decision problem using the transfer decision STD algorithm, and solve the regional S3CE cost problem using the adaptive slope-based piecewise linear function ASP optimization algorithm;

[0069] An online decision-making module, configured to form an online joint decision-making scheme based on the solution results of the STD and ASP algorithms, realize real-time scheduling of vehicle service requests and service container management in the S3CE system, and minimize the long-term cost of the S3CE system.

[0070] The present invention also provides a base station, including: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the online decision-making method for vehicle service requests as described above are implemented.

[0071] The present invention also provides an online decision-making system for vehicle service requests, including: mobile vehicles, a base station BS, roadside units RSU, and a cloud center. The cloud center is connected to the BS in the edge layer through a backbone network, and the BSs are connected through optical cables. One BS covers a certain range of traffic roads and a certain number of RSU equipped with edge servers. The RSU are distributed along both sides of the road in the edge layer and are connected through a local area network. The service requests generated by the mobile vehicles are scheduled to the RSU within the coverage range of the BS for processing, and the BS executes the online decision-making method for vehicle service requests as described above to realize real-time scheduling of vehicle service requests and service container management.

[0072] Beneficial effects: Traditional vehicle networking task scheduling methods do not consider the service attributes of tasks and default that the edge server has configured the requested task services, resulting in the task scheduling algorithm being divorced from the actual application scenario. Therefore, in the S3CE framework, the present invention considers the container management of different vehicle service requests in the edge server and makes decisions on service container retention while scheduling requests. In the vehicle-road cooperation scenario, vehicles have high mobility. The present invention establishes a delay, cost, and problem model based on mobile vehicle trajectory prediction, restricting vehicle requests to be scheduled only to RSUs that can always maintain a communication connection. Additionally, due to the relevance of container images for different vehicle requests for services, the dependence of the image layer of service containers is considered in the design of request scheduling and container retention decision-making, reducing the decision space. Most importantly, to address the dynamics and randomness of vehicle service requests, the present invention proposes an online collaborative decision-making scheme OnCo based on the idea of approximate dynamic programming to achieve real-time scheduling of vehicle service requests and service container management in the S3CE system, thereby minimizing the long-term cost of the S3CE system. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 is a network model structure diagram of an online decision-making system for vehicle service requests in serverless cloud-edge collaborative computing according to the present invention;

[0074] Figure 2 is a schematic diagram showing the influence of the relationship between different service container image layers on scheduling decisions according to the present invention;

[0075] Figure 3 is a schematic diagram showing the mutual influence relationship between request scheduling decisions and container retention decisions in different time slots according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0076] The following further describes the implementation method of the present invention with reference to the accompanying drawings.

[0077] Refer to Figure 1 , an online decision-making system for vehicle service requests in serverless cloud-edge collaborative computing (i.e., the S3CE system). The participating objects include mobile vehicles that generate service requests, several mobile base stations BS, multiple RSUs equipped with edge servers, and a cloud center, constituting a three-layer vehicle-edge-cloud architecture. The cloud center defaults to configure all services requested by vehicles and is connected to the BS in the edge layer through a backbone network. The road is divided into B regions, each region is covered by a BS, and let β = {1, 2,..., B} be the set of BSs. Within the coverage area of each BS, several RSUs equipped with edge servers are distributed on both sides of the road and are connected through a local area network. Containers are deployed on the RSUs to provide computing functions for service requests. The control center located on the BS is responsible for scheduling service requests of all connected vehicles in its area and container management on the RSUs.

[0078] Let \(N\) b \(=\{1, 2, \ldots, N\}\) denote the set of heterogeneous RSU (i.e., edge servers) in the coverage area of BS \(b\). Let denote the set of service types requested by vehicles. The workload \(w\) of each service k is expressed as floating point operations per second (Floating point Operation, FLOPs), and the deadline of each service is denoted as \(dl\) k . At \(t\in T\), the set of requests of service type \(k\) in the S3CE system is denoted as \(\vert R\) k \vert\) represents the number of requests of service type \(k\). The situation of service containers deployed on RSUn in the area of BS \(b\) is denoted as where means that a container of service type \(k\) is deployed on RSUn. Conversely, means that a container of service type \(k\) is not deployed on RSUn. In addition, define to represent the state of the container of service type \(k\) on RSUn at time slot \(t\). When , it means that the service container remains active. It means that the service container is closed. When the service container is closed, if a related service request is scheduled for processing, initializing the container will cause cold start delay, affecting the quality of service of the request. The cold start delays of different service containers are denoted as Due to the resource limitation of the edge server, the number of service containers that can be accommodated on RSUn at the same time is denoted as

[0079] Since each service container is built from an image file, the cloud center in the S3CE system hosts the image repository of all containers, and the edge server can pull the relevant image files from the cloud center to build service containers. The image layer of the container of service type \(k\) is denoted as \(L\) k \(=\{l_1, l_2, \ldots\}\). The image layers of different service containers may be the same. The set of all container image layers is denoted as \(L = \{l_1, l_2, \ldots, l\) |L| \}\), and the data size of the image layer \(l\) is denoted as \(d\) l . In addition, use to represent the set of image layers contained on RSUn, where means that RSUn has the image layer \(l\). Conversely, means that RSUn does not have the image layer \(l\).

[0080] The online decision-making method for vehicle service requests in the S3CE system proposed by the present invention is executed on the BS and includes the following steps:

[0081] Step 1: According to the vehicle movement trajectory prediction model of Constant Yaw Rate and Acceleration (CYRA), obtain the position of a moving vehicle at any moment in the vehicle-road collaborative scenario, so as to obtain the transmission range R between any vehicle and the roadside RSU. V2R ;

[0082] Step 2: Establish a delay model and a cost model related to vehicle service request scheduling. Under the constraints of meeting the service request delay deadline and the S3CE system service container resources, etc., form an optimization problem equation for joint decision-making of request scheduling and container reservation. The optimization goal is to minimize the long-term cost of the S3CE system under the condition of meeting the service request delay deadline. Analyze the problem equation and decompose the problem into a request transfer decision problem and a regional S3CE cost problem.

[0083] Step 3: For the request transfer decision problem, based on a Simple Transfer Decision (STD) algorithm, make transfer decisions for all vehicle service requests in the S3CE system.

[0084] Step 4: For the regional S3CE cost problem, according to the approximate dynamic programming idea, obtain an approximate value function based on the Adaptive Slope-based Piecewise Linear Function (ASP) optimization algorithm, so as to obtain an approximate cost function generated by vehicle service request scheduling and service container reservation decisions.

[0085] Step 5: Based on the STD and ASP algorithms, form an Online Co-Decision-Making (OnCo) scheme to realize the real-time scheduling of vehicle service requests and service container management in the S3CE system, and minimize the long-term cost of the S3CE system.

[0086] The following describes the specific implementation methods for each step.

[0087] In Step 1, since in the VRC scenario, the mobility of vehicles poses a huge challenge to service request scheduling, the present invention restricts the scheduling range of service requests by obtaining the communication range between vehicles and RSUs. With the maturity of vehicle trajectory prediction research, the present invention predicts the real-time position of vehicles according to the CYRA vehicle movement trajectory prediction model as the scheduling basis. Let the position of a certain moving vehicle at time slot t be (x(t), y(t)), then at time slot t m The position of the vehicle is:

[0088]

[0089] where v is the longitudinal speed, a is the acceleration, θ is the yaw angle, and w is the yaw speed. c x and c y are fixed initial value constants.

[0090]

[0091] According to the real-time position of the vehicle, the distance d between the moving vehicle and all the RSU within the coverage area of the BS can be obtained for each time slot t R (t). It is assumed that the moving speed of the vehicle satisfies a Gaussian distribution, i.e., v ~ N(u, θ 2 ). Considering the radio environment of the S3CE system, a log-distance path loss model is adopted, i.e., pl(d R (t)) = C + 10n f log 10 d R (t), where C = pl(d0) - 10n f log 10 d0, d0 is the path loss distance reference value, and n f is the free space path loss parameter, and pl(d0) represents the corresponding path loss value.

[0092] Based on the above log-distance path loss model, the communication range R between the moving vehicle and the RSU can be obtained V2R :

[0093]

[0094] where, is the transmission power of the RSU at time slot t, represents the threshold of the received signal strength of the vehicle.

[0095] In step two, the problem is first modeled and then decomposed. Among them, an optimization problem equation for the joint decision of service request and container reservation is established: the delay model of the vehicle service request, the cost model generated by the S3CE system for processing the service request, and the limiting conditions to be satisfied are formed into an optimization problem equation. Among them, the optimization goal is to minimize the cost of the S3CE system under the condition of meeting the service request deadline.

[0096] By analyzing the problem, the complex large-scale combinatorial optimization problem is decomposed into several sub-problems to reduce the problem complexity and seek the optimal solution of the problem.

[0097] An online decision-making method for vehicle service requests in serverless cloud-edge collaborative computing according to this example. First, the problem is modeled and then decomposed. The specific implementation process is as follows:

[0098] For each time slot t, for the vehicle request r of service type k in the area of BSb i , the control center on BSb first checks whether the corresponding service containers are deployed on all RSUs in the area. Define the decision variable indicating that the request r i in time slot t is scheduled to be executed on RSUn, otherwise indicating that the request r i in time slot t is not scheduled to be executed on RSUn.

[0099] Delay model: There are two cases:

[0100] 1) If the corresponding service container k has been deployed on the RSUs in the area, that is the control center on BSb will select one of them to process the request r i ,

[0101] ① Transmission delay: Scheduling the request r i to RSUn, the resulting transmission delay is:

[0102]

[0103] where s i is the data size of the request, V i,n (t) is the transmission rate between the vehicle generating the request and RSUn, modeled as:

[0104]

[0105] where B i,n (t) is the available bandwidth, p i is the transmission rate of the vehicle generating the request, |h i,n (t)| 2 is the channel gain between the vehicle and RSUn, is the received noise power of RSUn. In particular, |h i,n (t)| 2 = pl(d i,n (t)), pl(·) is the logarithmic distance path loss model in step one, and d i,n (t) is the distance between the vehicle generating the request and RSUn.

[0106] ② Computation delay: The computation delay generated by processing the request r i on RSUn is:

[0107]

[0108] 2) If the corresponding service container k is not deployed on the RSU within the area, that is then the control center on BSb will decide whether to transfer the request r i . Assume that the BSs in the S3CE system only have cooperation relationships with the BSs in the two adjacent hops. Use to represent the set of two-hop neighbor BSs of BSb.

[0109] The control center of BSb checks whether the corresponding service container k is deployed on the RSU within its neighbor area. If it is deployed, that is then the control center can transfer the request to the corresponding neighbor RSU within the area for processing. Define the decision variable to represent that the request r i at time slot t is transferred to the area for execution. Conversely represents that the request r i at time slot t is not transferred to the area for execution.

[0110] ③ Transfer delay: If the decision is to transfer, there will be a transfer delay:

[0111]

[0112] is the wired or wireless link bandwidth between BSb and , α is the coefficient of the transfer delay, is the number of hops between BSs.

[0113] If the corresponding service container k is not deployed on the RSU within the area of its neighbor BS either, that is then the control center will make a request scheduling decision to select the best RSU within the area to process the request, pull the corresponding service container from the cloud center to deploy the container on this RSU, and then process the request. This will result in a preparation delay.

[0114] ④ Preparation delay: It includes the delay of downloading the container image and the delay of starting the container:

[0115]

[0116] Therefore, the total delay of the service request r i is

[0117] Cost model: Processing vehicle service requests in the S3CE system will incur various costs:

[0118] ① Transmission cost: When the S3CE system processes request r i The transmission cost incurred is defined as the time occupying the bandwidth,

[0119] ② Computational cost: The computational cost is the time occupying the CPU resources,

[0120] ③ Handover cost: The handover cost is the time occupying the bandwidth between base stations,

[0121] ④ Container reservation cost: The cost of keeping the service container k active on RSUn is where represents the unit price of reserving the service container k for one time slot on RSUn. Therefore, for RSUn to process request r i the container reservation cost incurred is

[0122] ⑤ Preparation cost: The preparation cost incurred by the S3CE system is the generated preparation time,

[0123] where λ t 、λ c 、λ tr 、λ r and λ p are the coefficients of the transmission cost, computational cost, handover cost, container reservation cost, and preparation cost respectively;

[0124] Therefore, the total cost generated by the S3CE system for processing request r i is Then the total cost of the S3CE system for processing all vehicle requests in time slot t is

[0125] Problem formulation:

[0126] The S3CE system needs to make two major decisions in real time, request scheduling (selecting an RSU in the area to process the request, determining or transferring the request, determining and container reservation (whether to keep the service container active, determining ), with the goal of minimizing the long-term cost of the S3CE system. Based on the above delay model and cost model, the problem equation is modeled as:

[0127]

[0128]

[0129]

[0130]

[0131]

[0132]

[0133] Constraint C.1 means that the vehicle service request in each time slot t can and can only be scheduled to one RSU for processing; Constraint C.2 means that the distance between the vehicle generating the service request and the scheduled RSU must be within their communication range; Constraint C.3 means that all image layers of the required service containers must be available on the scheduled RSU; Constraint C.4 means that the number of containers running simultaneously on each RSUn needs to meet its quantity limit; Constraint C.5 means that the total delay of the vehicle service request is within its deadline range.

[0134] Problem decomposition: By analyzing the problem, it can be seen that the situation of service containers on all RSUs in the BSb area in the previous time slot to a certain extent determines the request scheduling decision, whether to process in this area or transfer to the neighboring BS area. Let represent the set of requests transferred to the BSb area. Based on the above problem modeling formula, the total cost generated by the regional S3CE system in BSb for processing all vehicle service requests can be obtained as

[0135] Therefore, Problem P1 can be decomposed into the S3CE system service request transfer decision problem and several regional S3CE system cost sub-problems. Among them, the equation of the S3CE system service request transfer decision problem is expressed as follows:

[0136]

[0137] s.t C.1~C.5

[0138] And the equation of the regional S3CE system cost sub-problem is as follows:

[0139]

[0140]

[0141] C.2~C.5

[0142] In step three, according to the online decision-making method for vehicle service requests in the serverless cloud-edge collaborative computing of this instance, first design a simple transfer decision algorithm to solve the S3CE system service request transfer decision problem.

[0143] The BS whose communication range covers the predicted position of the service request vehicle at the deadline is defined as the Potential Transfer BS (PT-BS); the RSU that is always within the communication range of the service request vehicle within the request deadline is defined as the Potential RSU (P-RSU).

[0144] Referring to Figure 2 , when there is no corresponding container for the service request in this area and neighboring areas, and it is necessary to pull the image in this area to build the service container, the S3CE system preparation costs generated by scheduling to different RSUs are different, and the sharing relationship between different service container image layers needs to be considered. The mobile vehicle generates a service request a, and the set of P-RSUs for the request is obtained according to the vehicle trajectory prediction, RSU 2, RSU 3, RSU 4. By observing the relationship between different service container image layers, it can be seen that the images of service container a and service container b share l1-l4, and service container b is deployed on RSU 4. Therefore, the lowest preparation cost for the S3CE system to handle the vehicle request is paid when choosing to process it on RSU 4. Based on the above observations, the present invention designs a simple handover decision algorithm, and the specific steps are as follows:

[0145] S31: At time slot t, for any vehicle service request r in the S3CE system i , first obtain its set of PT-BSs and set of P-RSUs;

[0146] S32: Traverse the set of PT-BSs to calculate different handover costs, select the minimum handover cost and its corresponding PT-BS, and denote them as PT-BS

[0147] S33: Traverse the set of P-RSUs, and based on the sharing relationship of the container image layer, select the minimum preparation cost and its corresponding P-RSU, and denote them as P-RSU n;

[0148] S34: If then Otherwise,

[0149] In step four, according to the problem analysis in step two, the regional S3CE system cost sub-problem P2 needs to make joint decisions on request scheduling and container reservation to minimize the long-term cost of the regional S3CE system. Referring to Figure 3, it can be observed that the container reservation decision in the previous time slot affects the request scheduling decision in the current time slot, and the scheduling decision in the current time slot also has a profound impact on the container reservation decision in the subsequent time slot. For example, in time slot t, a vehicle requests service c, and RSU 2 in its P-RSU set has exactly deployed service container c and can handle the request smoothly. The following decision is made: In time slot t+1, another vehicle also requests service c. Since RSU 2 in its P-RSU set reserved service container c in the previous time slot, it can still handle this request smoothly. Considering that service container a on RSU 1 has not been requested for a long time, it is decided to close this container. The specific decision is as follows: In time slot t+2, a vehicle comes to request service a. Since RSU 1 in its P-RSU set closed service container a in the previous time slot, the system will incur a preparation cost or a transfer cost when processing this request. The arrival of vehicle service requests is not prior knowledge for the regional S3CE system. Problem P2 is a Markov multi-step decision-making process, and the decision at each step depends on the current system state and the subsequent system evolution. To handle the multiple uncertainties of the regional S3CE system, the objective of problem P3 is transformed into minimizing the long-term expected cost value:

[0150]

[0151]

[0152] C.2 - C.5

[0153] where S t is the system state of the regional S3CE at time slot t, including all vehicle information (position, longitudinal speed, acceleration, yaw angle, yaw speed) within the coverage area and all RSU information (container service distribution, server capacity situation); W t is random information, including various vehicle service request arrival information and mobile vehicle trajectory information. Based on the Bellman optimality principle, the multi-step optimization decision problem in P4 can be transformed into a recursive problem, that is, the objective in P4 can become:

[0154]

[0155] where V t (S t ) is the system cost value of the regional S3CE at time slot t, and V t+1 (S t+1 |S t ) is the system cost value of the regional S3CE at time slot t+1 on the premise of state S t ; ζ ∈ [0,1] is the future decay factor.

[0156] According to the basic idea of dynamic programming, the time dependence is decomposed using the Bellman equation, and large-scale problems can be decomposed into small problems for iterative solution. Therefore, in principle, P4 can be solved using the dynamic programming method. However, it is computationally difficult for the dynamic programming algorithm to enumerate the entire solution space. In the problem solved by the present invention, there are a large number of service requests to be scheduled and corresponding service container reservation decisions in each time slot. Due to the huge state space, it is difficult to calculate the exact value of each S t corresponding to V t (S t ). The present invention uses the idea of approximate dynamic programming (ADP) to solve the above problems. ADP makes a trade-off between optimality and computational time to obtain an approximate optimal solution. Let represent the approximate value of V t (S t ). In the objective formula of P4, the expected value term E(V t+1 (S t+1 |S t )) is very difficult to calculate. To avoid calculating this expected value term, the present invention introduces a post-decision state function to replace the expected value term: , where D- represents the pre-decision state and D represents the post-decision state.

[0157] The essence of designing the approximate function is to obtain the optimal policy by iterating the cost value corresponding to each state. The iterative process calculates the optimal decision based on the initial estimate of the cost value function, updates the cost value function using the information obtained from the decision, and then applies the updated cost value to the next iteration to make the approximate function continuously approach the accurate cost value. Therefore, P4 can be transformed into P5:

[0158]

[0159] where f is the number of iterations. represents the state at time t-1 and the (f-1)-th iteration. To solve P5, two challenges need to be addressed. One is the design of the approximate cost value function, and the other is the selection of the iterative method. For this purpose, the present invention proposes a piecewise linear function algorithm with an adaptive slope to obtain the approximate cost value of the joint decision of request scheduling and container reservation. Let represent the piecewise linear function, where represents the system average cost value from the initial time to time slot t, and h t,s represents the slope of the piecewise linear function. The function is evenly divided into N t functions, and the slope of the function is monotonically increasing. Then, P5 can be written as P6:

[0160]

[0161] Since the quality of the solution to the P6 problem depends to a certain extent on the slope of the piecewise function, in order to obtain a good decision-making strategy, the present invention designs the slope to be adaptively updated iteratively:

[0162]

[0163] where α f-1 is the step size for smoothing the estimated slope; represents the sample observation value of the marginal value, and is defined as follows:

[0164]

[0165] Based on the above analysis, an online optimization algorithm for an adaptive approximate cost function is formed, and the specific steps are as follows:

[0166] S41: Initialize the slope of the piecewise linear function, the number of iterations F, and the number of segments N t , f = 1;

[0167] S42: Use the Monte Carlo method to generate F training samples, and each training sample is random information containing various changes of the regional S3CE system lasting for 1 h;

[0168] S43: t = 1, update the system state according to the latest random information Make a decision X by solving P6 under the constraint conditions t , A t , calculate the system cost

[0169] S44: t = t + 1, return to S43 until t > T, then jump to S45;

[0170] S45: t = 1, calculate the sample observation value of the marginal value according to formula (1), and update the slope and the piecewise linear function according to formula (2);

[0171] S46: t = t + 1, return to S45 until t > T, then jump to S47;

[0172] S47: f = f + 1, return to S43 until f > F, then jump to S48;

[0173] S48: Obtain the trained piecewise linear function and obtain the approximate cost function.

[0174] In step five, according to the problem solutions in steps three and four, the present invention forms an online joint decision-making scheme OnCo, realizes the joint decision-making of online request scheduling and container reservation, and minimizes the long-term cost of the S3CE system. The specific steps are as follows:

[0175] S51: Obtain the approximate cost function for each time slot t according to the adaptive approximate cost function online optimization algorithm in step 4; Let time slot t = 1;

[0176] S52: Input real-time vehicle information and RSU information;

[0177] S53: Obtain the transfer decision according to the simple transfer decision algorithm in step 3;

[0178] S54: Obtain the request scheduling decision X t and the container reservation decision A t ;

[0179] S55: t = t + 1, return to S52 until t > T.

[0180] The present invention also provides an online decision-making device for vehicle service requests in an S3CE system, including:

[0181] A transmission range determination module, configured to obtain the position of a moving vehicle at any time in a vehicle-road cooperation scenario according to a vehicle movement trajectory prediction model of a constant yaw rate and acceleration CYRA, and obtain the transmission range R between the vehicle and the roadside unit RSU V2R , as the distance limit for vehicle service request scheduling;

[0182] An optimization problem construction module, configured to establish a delay model and a cost model related to vehicle service request scheduling, and form an optimization problem equation for joint decision-making of request scheduling and container reservation under the conditions of meeting the distance limit and the service container resource limit of the S3CE system, and the optimization goal is to minimize the long-term cost of the S3CE system under the premise of meeting the delay deadline of the service request;

[0183] An optimization problem analysis and solution module, configured to analyze the problem equation, decompose the problem into a request transfer decision problem and a regional S3CE cost problem, use the transfer decision STD algorithm to solve the request transfer decision problem, and use the adaptive slope-based piecewise linear function ASP optimization algorithm to solve the regional S3CE cost problem;

[0184] An online decision-making module, configured to form an online joint decision-making scheme based on the solution results of the STD and ASP algorithms, realize real-time scheduling of vehicle service requests and service container management in the S3CE system, and minimize the long-term cost of the S3CE system.

[0185] The present invention also provides a base station, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the online decision method for vehicle service requests as described above are implemented.

[0186] Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system, or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0188] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent substitutions, and any modification or equivalent substitution that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. An online decision-making method for vehicle service requests in an S3CE system, characterized in that, It includes the following steps: S1. According to the vehicle movement trajectory prediction model of constant yaw rate and acceleration CYRA, obtain the position of the moving vehicle at any moment in the vehicle-road collaborative scenario, and get the transmission range R between the vehicle and the roadside unit RSU V2R , which is used as the distance limit for vehicle service request scheduling; S2. Establish a delay model and a cost model related to vehicle service request scheduling. Under the conditions of meeting the distance limit and the resource limit of the S3CE system service container, form an optimization problem equation for joint decision-making of request scheduling and container reservation. The optimization goal is to minimize the long-term cost of the S3CE system under the time delay deadline for meeting service requests, as follows: M t The total cost for the S3CE system to process all vehicle requests in time slot t M i The total cost generated by the S3CE system to process request r i The transmission cost for processing request r i The computation cost for processing request r i The transfer cost for processing request r i The container reservation cost for processing request r i The preparation cost for processing request r i ​​​​​​ The problem satisfies the following constraint conditions: C.

1. Each vehicle service request in each time slot t can be scheduled to and processed by only one RSU; C.

2. The distance between the vehicle generating the service request and the scheduled RSU must be within their communication range; C.

3. All image layers of the required service containers must be available on the scheduled RSU; C.

4. The number of containers running simultaneously on each RSUn needs to meet its quantity limit; C.

5. The total time delay of the vehicle service request is within its deadline; The purpose of the optimization problem is to determine the request scheduling decision, that is, to select an RSU in the area to process the request or transfer the request, and the container reservation decision, that is, whether to keep the service container active; where b represents the base station b within the coverage area of the RSU, β represents the set of base stations, n represents the nth RSU, and N b represents the set of RSUs within the coverage area of base station b, k represents the kth type of service, represents the set of service types, i represents the ith service request, R k represents the set of requests with the service type k; S3. Analyze the problem equation, decompose the problem into a request transfer decision problem and a regional S3CE cost problem, use the transfer decision STD algorithm to solve the request transfer decision problem, and use the adaptive slope-based piecewise linear function ASP optimization algorithm to solve the regional S3CE cost problem; S4. Based on the solution results of the STD and ASP algorithms, form an online joint decision-making scheme to realize the real-time scheduling of vehicle service requests and service container management in the S3CE system, and minimize the long-term cost of the S3CE system.

2. The method according to claim 1, wherein The step S1 includes: The position of the moving vehicle at time slot t is (x(t), y(t)), and time slot t is determined based on the vehicle's longitudinal speed v, acceleration a, yaw angle θ, and yaw rate w. m The position of the vehicle is: where Δt = t m - t, c x and c y are fixed initial value constants; According to the real-time position of the vehicle, obtain the distance d between the moving vehicle and the RSU at each time slot t R (t), based on the log-distance path loss model pl(d)=C + 10n f log 10 d to determine C; Obtain the communication range R between the moving vehicle and the RSU V2R : where is the transmission power of the time slot tRSU, represents the threshold of the vehicle received signal strength, n f is the free space path loss parameter.

3. The method according to claim 1, wherein The transmission cost is the time for occupying bandwidth. The computing cost is the time taken to occupy the CPU resources. The transfer cost is the time taken to occupy the bandwidth between base stations. The retention cost of the container is where is the cost of keeping service container k active on RSUn, where represents the unit price of retaining service container k for one time slot on RSUn; The preparation cost is the preparation time spent. Among them, λ t , λ c , λ tr , λ r and λ p are the transmission cost coefficient, the computing cost coefficient, the transfer cost coefficient, the container retention cost coefficient, and the preparation cost coefficient, respectively; represents the transmission delay generated by scheduling the request r i to RSUn, and B i,n (t) represents the available bandwidth; represents the computing delay generated by processing the request r i on RSUn; w k represents the workload of service type k; represents the transfer delay generated by transferring the request from the BSb area to area; represents the link bandwidth between BSb and ; represents the state of the container of service type k on the time slot tRSUn; represents the preparation delay generated by deploying the container.

4. The method according to claim 1, characterized in that In the step S3, the analysis and decomposition of the problem include: Analyze the problem, and it can be known that the situation of service containers on all RSUs in the previous time slot BSb area To a certain extent, it determines the request scheduling decision, whether to process in this area or transfer to the neighboring BS area. Use to represent the set of requests transferred to the BSb area. Based on the modeling formula of problem P1, the total cost generated by the regional S3CE system of BSb to process all vehicle service requests is Therefore, problem P1 can be decomposed into the S3CE system service request transfer decision problem and several regional S3CE system cost sub-problems. Among them, the equation of the S3CE system service request transfer decision problem is expressed as follows: s.t C.1~C.5 The equation of the sub-problem of the regional S3CE system cost is as follows: C.2~C.5 5. The method according to claim 4, characterized in that, In the step S3, using the transfer decision STD algorithm to solve the request transfer decision problem includes: Define the BS whose communication range covers the predicted position of the deadline of the service request vehicle as PT-BS; define the RSU that is always within the communication range of the service request vehicle within the request deadline as P-RSU; At time slot t, for any vehicle service request r within the S3CE system i , obtain its PT-BS set and P-RSU set; traverse the PT-BS set to calculate different transfer costs, select the minimum transfer cost and its corresponding PT-BS, and denote them as PT-BS Traverse the P-RSU set, and based on the container image layer sharing relationship, select the minimum preparation cost and its corresponding P-RSU, and denote them as P-RSU n; if then Otherwise, where is the scheduling decision variable, indicates that request r at time slot t is scheduled to be executed on RSUn, and 0 means it is not scheduled to be executed on RSUn; is the transfer decision variable, indicates that request r at time slot t i is transferred to area for execution, and 0 means it is not transferred to area for execution.

6. The method according to claim 5, wherein In the step S3, using the adaptive slope-based piecewise linear function ASP optimization algorithm to solve the regional S3CE cost problem includes: Convert the objective of problem P3 into minimizing the long-term expected cost value: C.2~C.5 Among which S t is the system state of the time slot t regional S3CE, including all vehicle information and all RSU information within the coverage area; W t is random information, including various vehicle service request arrival information and mobile vehicle trajectory information; Based on the Bellman optimality principle, convert the multi-step optimization decision problem in P4 into a recursive problem, that is, the objective in P4 becomes: where V t (S t ) is the system cost value of the regional S3CE in time slot t, and V t+1 (S t+1 |S t ) is the system cost value of the regional S3CE in time slot t + 1 on the premise of state S t ; ζ ∈ [0, 1] is the future decay factor; Introduce the post - decision state function to replace the expected value term E(V t+1 (S t+1 |S t )): where D - represents the pre - decision state, D represents the post - decision state, and P4 is transformed into P5: Use to represent a piecewise linear function, where represents the system average cost value from the initial to time slot t, h t,s represents the slope of the piecewise linear function, and the function is evenly divided into N segments t functions, and the slope of the function is monotonically increasing. P5 is written as P6: At t = 1, update the system state according to the latest random information Make decision X by solving P6 under the constraints t ,A t , calculate the system cost value t = t + 1, perform iterative calculations until t > T; where the slope is designed to be adaptively updated iteratively: where α f-1 is the step size for smoothing the estimated slope; denotes the sample observation value of the marginal value, which is defined as follows:

7. The method according to claim 6, wherein The step S4 includes: S41: Obtain the approximate cost function for each time slot t according to the online optimization algorithm of the adaptive approximate cost function; let the time slot t = 1; S42: Input real-time vehicle information and RSU information; S43: Obtain the transfer decision according to the simple transfer decision algorithm in step S3; S44: Obtain the request scheduling decision X by solving P6 under the condition of satisfying the restriction condition t and the container reservation decision A t ; S45: t = t + 1, return to S42 until t > T.

8. An online decision-making device for vehicle service requests in an S3CE system, characterized in that, It includes: A transmission range determination module, configured to obtain the position of a moving vehicle at any time in a vehicle-road cooperation scenario according to a vehicle movement trajectory prediction model of a constant yaw rate and acceleration CYRA, and obtain a transmission range R between the vehicle and a roadside unit RSU V2R , as a distance limit for vehicle service request scheduling; An optimization problem construction module, configured to establish a latency model and a cost model related to vehicle service request scheduling, and form an optimization problem equation for joint decision-making of request scheduling and container reservation under the conditions of satisfying distance constraints and S3CE system service container resource constraints. The optimization objective is to minimize the long-term cost of the S3CE system under the latency deadline for satisfying service requests, as follows: M t The total cost for the S3CE system to process all vehicle requests in time slot t M i The total cost generated by the S3CE system to process request r i The transmission cost for processing request r i The computation cost for processing request r i The transfer cost for processing request r i The container reservation cost for processing request r i The preparation cost for processing request r i ​​​​​​ The problem satisfies the following constraint conditions: C.1, each vehicle service request in each time slot t can be scheduled to and processed by only one RSU; C.2, the distance between the vehicle generating the service request and the scheduled RSU must be within their communication range; C.3, all image layers of the required service containers must be available on the scheduled RSU; C.4, the number of containers running simultaneously on each RSUn needs to satisfy its quantity limit; C.5, the total latency of the vehicle service request is within its deadline range; The purpose of the optimization problem is to determine the request scheduling decision, that is, to select an RSU in the area to process the request or transfer the request, and the container reservation decision, that is, whether to keep the service container active; where \(b\) represents the base station \(b\) within the coverage area of the RSU, \(\beta\) represents the set of base stations, \(n\) represents the \(n\)th RSU, and \(N\) b represents the set of RSUs within the coverage area of base station \(b\), \(k\) represents the \(k\)th service type, represents the set of service types, \(i\) represents the \(i\)th service request, and \(R\) k represents the set of requests with service type \(k\); An optimization problem analysis and solution module, configured to analyze the problem equation, decompose the problem into a request transfer decision problem and a regional S3CE cost problem, use the transfer decision STD algorithm to solve the request transfer decision problem, and use the adaptive slope-based piecewise linear function ASP optimization algorithm to solve the regional S3CE cost problem; An online decision-making module, configured to form an online joint decision-making scheme based on the solution results of the STD and ASP algorithms, realize real-time scheduling of vehicle service requests and service container management in the S3CE system, and minimize the long-term cost of the S3CE system.

9. A base station, characterized in that, Including: One or more processors; A memory; And One or more programs, where the one or more programs are stored in the memory and configured to be executed by the one or more processors. When the programs are executed by the processors, the steps of the online decision-making method for vehicle service requests described in any one of claims 1-7 are implemented.

10. An online decision-making system for vehicle service requests, characterized in that, Including: Moving vehicles, a base station BS, roadside units RSU, and a cloud center. The cloud center is connected to the BS in the edge layer through a backbone network, and the BSs are connected through optical cables. One BS covers a certain range of traffic roads and a number of RSUs equipped with edge servers. The RSUs are distributed along both sides of the road in the edge layer and are connected through a local area network. The service requests generated by the moving vehicles are scheduled to the RSUs within the coverage range of the BS for processing. The BS executes the online decision-making method for vehicle service requests described in any one of claims 1-7 to realize real-time scheduling of vehicle service requests and service container management.

Images