Service-oriented continuous intelligent vehicle task prediction and scheduling method and device

Abstract

The invention provides an intelligent vehicle task prediction and scheduling method and device oriented to service continuity, and relates to the technical field of network service guarantee and intelligent terminal decision. The method comprises the following steps: acquiring a time sequence data training set; based on the training set, adopting a minimum description length criterion to construct a scoring function of a multivariable fusion prediction model; according to the scoring function, a final multivariable fusion prediction model is constructed based on a dynamic Bayesian network; acquiring an attribute set of the vehicle in a future time period and an attribute set of the current time slot; the attribute set is input into a multivariable fusion prediction model for prediction, and prediction information of the vehicle position is obtained; based on the prediction information, a vehicle unloading strategy guided by the prediction information is optimized through a reinforcement learning method, and a self-adaptive learning intelligent strategy is obtained; based on an intelligent strategy, tasks of vehicle service continuity guarantee are actively scheduled in advance. By adopting the method, the optimal pre-active deployment scheduling of the vehicle task can be realized.

Description

Technical Field

[0001] This invention relates to the fields of network service assurance and intelligent terminal decision-making technology, and in particular to a method and apparatus for intelligent vehicle task prediction and scheduling oriented towards service continuity. Background Technology

[0002] With the development of intelligent vehicle technology, applications such as autonomous driving, cooperative obstacle avoidance, and high-definition visual assessment and analysis are becoming increasingly widespread. These applications generally have strong latency sensitivity and service continuity requirements, meaning that services must be completed seamlessly while the vehicle is in motion, and cannot tolerate task interruptions or cold start issues caused by computing latency or node switching. To meet the real-time and reliability requirements of typical intelligent vehicle applications, edge intelligence is gradually becoming a key supporting technology. During vehicle operation, it can dynamically access roadside units (RSUs) and base station (BS) computing nodes to achieve vehicle-road cooperative perception, intelligent decision-making, and service collaboration.

[0003] However, with the increasing number of intelligent vehicles and the growing complexity of traffic scenarios, traditional methods are gradually revealing their limitations when processing large-scale and dynamically changing traffic data. Current mainstream edge computing scheduling strategies mostly rely on the real-time connection status between vehicles and computing nodes to make offloading decisions; that is, tasks are only distributed and computation begins after a vehicle enters the coverage area of ​​a specific RSU (Roadside Unit). This passive offloading mode struggles to meet the continuous service guarantee requirements in highly dynamic environments, especially in scenarios with high-speed vehicle movement or intensive computing demands, often resulting in task processing delays or service interruptions. Specifically, the traditional passive offloading mode faces two main bottlenecks: first, long task processing delays make it difficult to meet the needs of instantaneous decision-making; second, poor service continuity, as tasks often need to be migrated before completion in high-speed vehicle movement scenarios, leading to service interruptions or duplicate scheduling.

[0004] To overcome the aforementioned bottlenecks, prediction-driven proactive offloading mechanisms have become an important research direction in recent years. In vehicle-to-everything (V2X) networks, tasks typically exhibit phased, decomposable, and prior characteristics, making them suitable for being divided into multiple subtasks and offloaded to multiple computing nodes. This type of mechanism models the predictable states of vehicle trajectory and speed to infer the future access node location of the vehicle in advance, and pre-deploys the task to the target node before access to achieve pre-computation and cache scheduling. Accurate vehicle trajectory prediction can help the system plan the task offloading path in advance and select the optimal offloading target node, thereby achieving the lowest possible latency and energy consumption.

[0005] However, existing trajectory prediction methods still have significant shortcomings in terms of dynamic adaptability, temporal modeling capabilities, and multivariate correlation modeling. Most methods employ simple linear regression, moving averages, or models based on univariate trend prediction, which cannot systematically characterize the complex relationships between the future state of a vehicle and multidimensional influencing factors, including the interactions and causal relationships of road conditions, communication quality, and node load. Especially in large-scale dynamic environments, existing models struggle to model the joint distribution of trajectory, task state, and system resources over time, and lack robustness to sudden traffic disturbances and the non-stationarity of vehicle behavior.

[0006] With the rapid development of vehicle-to-everything (V2X) and intelligent vehicle technologies, higher demands are being placed on real-time scheduling of computing tasks, optimal resource utilization, and service continuity assurance during vehicle movement. To improve system response speed and service quality, numerous prediction-driven task scheduling methods have emerged in recent years. Among these, vehicle trajectory prediction, as a crucial component, is widely applied in traffic flow optimization, task offloading decisions, and collaborative perception and reasoning scenarios. Current mainstream trajectory prediction methods largely rely on reinforcement frameworks to build high-precision models. Existing technologies propose a deep learning trajectory prediction model that integrates physical law constraints. This model combines data-driven and physics-driven modeling methods, intuitively and accurately considering the interactions between vehicles and precisely describing the changes in relevant variables during the prediction process. Existing technologies also propose a vehicle trajectory prediction scheme based on Long Short-Term Memory (LSTM) networks and graph convolutional networks. This scheme effectively improves the accuracy of future trajectory prediction under the influence of surrounding vehicles and the road environment by combining vehicle interaction information and lane information to build a trajectory prediction model. Finally, existing technologies propose a novel network called Sparse Attention Graph Convolutional Network, which aims to comprehensively consider the trajectory interaction details of multiple vehicles to optimize long-sequence time-series predictions for target vehicles. However, the aforementioned methods have numerous problems, including: the prediction methods only focus on static prediction of trajectory or location, lacking a fusion mechanism with computational task scheduling strategies, and thus failing to directly guide task pre-deployment and service continuity assurance; existing solutions have high computational overhead and poor real-time performance, and deep neural networks have high computational complexity during the inference phase, making them difficult to deploy in resource-constrained in-vehicle environments or edge nodes, reducing online decision-making efficiency and system adaptability; and there is a lack of system modeling for multivariate temporal dependencies. Most existing methods model the trajectory itself, failing to uniformly consider the joint evolution of vehicle state, task attributes, and environmental resources, thus limiting the forward-looking judgment of unloading strategies in complex dynamic scenarios. Summary of the Invention

[0007] To address the shortcomings of existing passive-response vehicle-to-everything (V2X) task scheduling methods, such as the lack of a systematic scheduling mechanism oriented towards service continuity, the inability to achieve proactive scheduling and efficient connection of task chains, leading to frequent service interruptions and significant response delays in high-speed vehicle scenarios, and the inability of existing trajectory prediction methods to effectively capture the dynamic dependencies between vehicle state, task attributes, and environmental resources, resulting in coarse modeling granularity and weak temporal expression capabilities, making it difficult to support scheduling decisions in complex time-varying scenarios, this invention provides a service-continuous intelligent vehicle task prediction and scheduling method and apparatus. The technical solution is as follows:

[0008] On the one hand, a service-continuous intelligent vehicle task prediction and scheduling method is provided, which is implemented by a service-continuous intelligent vehicle task prediction and scheduling device, and includes:

[0009] S1. Obtain a training set of time series data of vehicle attributes; based on the training set of time series data of vehicle attributes, construct a scoring function for the multivariate fusion prediction model using the minimum description length criterion; based on the scoring function, construct the final multivariate fusion prediction model using a dynamic Bayesian network.

[0010] S2. Obtain the vehicle's attribute set for the future time period and the attribute set for the current time slot; input the vehicle's attribute set for the future time period and the attribute set for the current time slot into the final multivariate fusion prediction model for prediction to obtain the vehicle's location prediction information;

[0011] S3. Based on the predicted information of vehicle location, the vehicle unloading strategy guided by the predicted information is optimized by reinforcement learning to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is actively pre-scheduled.

[0012] On the other hand, a service-continuous intelligent vehicle task prediction and scheduling apparatus is provided, which is applied to a service-continuous intelligent vehicle task prediction and scheduling method. The apparatus includes:

[0013] A construction unit is used to acquire a training set of time-series data of vehicle attributes; based on the training set of time-series data of vehicle attributes, a scoring function for a multivariate fusion prediction model is constructed using the minimum description length criterion; and based on the scoring function, a final multivariate fusion prediction model is constructed using a dynamic Bayesian network.

[0014] The acquisition unit is used to acquire the attribute set of the vehicle in the future time period and the attribute set in the current time slot; the attribute set of the vehicle in the future time period and the attribute set in the current time slot are input into the final multivariate fusion prediction model for prediction to obtain the predicted information of the vehicle position;

[0015] The scheduling unit is used to optimize the vehicle unloading strategy guided by the predicted information based on vehicle location through reinforcement learning methods to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is proactively pre-scheduled.

[0016] On the other hand, a service-continuous intelligent vehicle task prediction and scheduling device is provided, the service-continuous intelligent vehicle task prediction and scheduling device comprising: a processor; a memory storing computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, any one of the above-described service-continuous intelligent vehicle task prediction and scheduling methods is implemented.

[0017] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement any of the above-described methods of service-oriented intelligent vehicle task prediction and scheduling.

[0018] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0019] This invention first acquires a training set of time-series data on vehicle attributes; based on this training set, a scoring function for a multivariate fusion prediction model is constructed using the minimum description length criterion; according to the scoring function, a final multivariate fusion prediction model is constructed based on a dynamic Bayesian network; the attribute sets for future time periods and current time slots of the vehicle are acquired; these sets are input into the final multivariate fusion prediction model for prediction to obtain vehicle location prediction information; based on the vehicle location prediction information, a reinforcement learning method is used to optimize the vehicle unloading strategy guided by the prediction information to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is proactively pre-scheduled.

[0020] This invention constructs a prediction model based on multivariate fusion. A temporal modeling framework is built upon a dynamic Bayesian network to jointly model multiple variables, including vehicle trajectory, speed, task stage state, and edge node load, forming a multi-time-stack joint probability distribution representation. Through training, state transition probabilities and observation probabilities are obtained, enabling predictive inference of the vehicle's future state and the probability of target node access. This model is suitable for modeling complex temporal dependencies in highly dynamic vehicle-to-everything (V2X) scenarios. Furthermore, the multivariate fusion prediction model utilizes Gibbs sampling to obtain a posterior probability estimate of the vehicle's future trajectory, thus approximating the posterior distribution in a high-dimensional state space under limited data conditions. This significantly reduces reliance on large-scale labeled trajectory data and lowers the cost of data acquisition and model training. The obtained posterior probability distribution serves as part of the input to the reinforcement learning scheduling strategy, providing more accurate prior guidance for subsequent unloading decisions.

[0021] This invention proposes an adaptive service scheduling method based on prediction results. The posterior probability output by the prediction model and the current environmental state are used as the state space input to the adaptive service scheduling method. Through continuous interaction with the environment, this method constructs a reward function based on comprehensive performance indicators such as latency, energy consumption, and offloading success rate. This function guides the policy network to dynamically adjust offloading actions, achieving optimal pre-deployment and scheduling of tasks.

[0022] The proposed proactive pre-deployment architecture for ensuring service continuity in intelligent vehicles, as described in this invention, firstly, accurately characterizes the dynamic uncertainties in vehicle behavior, network state, and task evolution through a multivariate temporal modeling framework, enabling prediction of future states. Secondly, it optimizes the offloading strategy guided by prediction information into an adaptively learnable intelligent strategy through reinforcement learning, thereby achieving proactive pre-scheduling of services, ensuring service continuity, real-time performance, and robustness, and improving the overall system efficiency. Furthermore, the multivariate fusion prediction model constructed in this invention utilizes Gibbs sampling to obtain the posterior probability estimate of the vehicle's future trajectory, approximating the posterior distribution in the high-dimensional state space under limited data conditions, significantly reducing dependence on large-scale labeled trajectory data and lowering the cost of data acquisition and model training. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1This is a flowchart of a service-oriented intelligent vehicle task prediction and scheduling method provided by an embodiment of the present invention;

[0025] Figure 2 This is a diagram of a proactive pre-deployment architecture for ensuring service continuity of intelligent vehicles, provided by an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of a vehicle trajectory prediction network based on a dynamic Bayesian network provided in an embodiment of the present invention.

[0027] Figure 4 This is a flowchart of a multivariate fusion prediction model algorithm for ensuring service continuity of intelligent vehicles, provided by an embodiment of the present invention.

[0028] Figure 5 This is a block diagram of a service-oriented intelligent vehicle task prediction and scheduling device provided in an embodiment of the present invention;

[0029] Figure 6 This is a schematic diagram of the structure of an intelligent vehicle task prediction and scheduling device for service continuity provided in an embodiment of the present invention. Detailed Implementation

[0030] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0031] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0032] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0033] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0034] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0035] This invention provides a service-continuous intelligent vehicle task prediction and scheduling method, which can be implemented by a service-continuous intelligent vehicle task prediction and scheduling device, which can be a terminal or a server. Figure 1 The flowchart shown is for a service-oriented, continuous intelligent vehicle task prediction and scheduling method. The processing flow of this method may include the following steps:

[0036] S1. Obtain a training set of time-series data on vehicle attributes; based on the training set of time-series data on vehicle attributes, construct a scoring function for the multivariate fusion prediction model using the minimum description length criterion; based on the scoring function, construct the final multivariate fusion prediction model using a dynamic Bayesian network.

[0037] In one feasible implementation, this invention focuses on the vehicle-to-everything (V2X) scenario in a highway context, and constructs a proactive pre-deployment architecture for ensuring service continuity of intelligent vehicles. For example... Figure 2 This invention provides a proactive pre-deployment architecture diagram for ensuring service continuity of intelligent vehicles, comprising multiple individual roadside units (RSUs) and multiple base stations (BSs). The proactive pre-deployment architecture for ensuring service continuity of intelligent vehicles includes four computing modules: local computing, direct-connected RSUs computing, local base station computing, and cooperative base station computing.

[0038] Among them, let This represents a set of highway regions, where each region represents the coverage area of ​​a single roadside unit (RSU), with a total of M regions. The location of a vehicle is defined as the location of a roadside unit within a predicted region, and the set of vehicle locations is represented as... ,in Since lane changing has little impact on task unloading decisions, the method proposed in this embodiment of the invention ignores lane changing scenarios.

[0039] In this embodiment, since the preceding vehicle imposes constraints on the target vehicle's acceleration and position attributes, while the following vehicle has a negligible impact on the preceding vehicle, the influence of the following vehicle on the preceding vehicle is ignored in the modeling of this invention. At time slot t, the attributes of vehicle n can be represented by a single tuple:

[0040]

[0041] in, This represents the highway region where vehicle n is in time slot t. This represents the horizontal coordinate of vehicle n in time slot t; This represents the speed of vehicle n in time slot t; This represents the velocity-acceleration of vehicle n in time slot t; Indicates vehicle The vehicle ahead is in the time slot The highway area at that time; Indicates vehicle The vehicle ahead is in the time slot The horizontal coordinate at that time; Indicates vehicle The vehicle ahead is in the time slot The speed at that time; Indicates vehicle The vehicle ahead is in the time slot acceleration at time; This indicates the road conditions at time slot t; Indicates time slot The weather conditions at that time.

[0042] In one feasible implementation, this embodiment of the invention constructs a multivariate fusion prediction model for ensuring the continuity of services for intelligent vehicles based on a dynamic Bayesian network. The dependencies between the attributes of vehicle n are modeled as a dynamic Bayesian network, referred to as the Vehicle Trajectory Prediction Network (VTP-DBN). The VTP-DBN structure consists of multiple time slices, each containing a set of attribute variables related to the current time. For example... Figure 3 This is a schematic diagram of a vehicle trajectory prediction network based on a dynamic Bayesian network provided in an embodiment of the present invention.

[0043] The final multivariate fusion prediction model is represented by the following formula (1):

[0044] (1)

[0045] in, This represents the network structure, where each node corresponds to an attribute at a given time step. If there is a direct dependency between two attributes, they are connected by an edge; that is, the relationship between attributes can be represented as a directed acyclic graph. Parameters The dependencies between attributes are quantitatively described, and the joint probability distribution is described using a conditional probability table. If in a time slot... hour, It is an attribute exist The set of parent nodes in the data, then The conditional probability table containing each attribute is represented by the following formula (2):

[0046] (2)

[0047] in, A conditional probability table representing each attribute; The attributes representing vehicle n; Indicates on the network The probability of it happening; Represents attributes The set of parent nodes in G.

[0048] The location and region attributes of a vehicle can be inferred from the observations of multiple attributes.

[0049] In one feasible implementation, embodiments of the present invention employ a minimum description length criterion to construct a scoring function for a multivariate fusion prediction model, used to evaluate the degree of fit between the multivariate fusion prediction model and the training data. The minimum description length criterion treats the construction of the multivariate fusion prediction model as a data compression task, aiming to find a multivariate fusion prediction model that can describe the training data with the shortest possible encoding length, including: the byte length describing the multivariate fusion prediction model itself and the byte length using the multivariate fusion prediction model to describe the training samples.

[0050] Optionally, the scoring function of the multivariate fusion prediction model is expressed by the following formula (3):

[0051] (3)

[0052] in, This represents the scoring function of the multivariate fusion prediction model on the time-series data training set of vehicle attributes; This indicates the number of bytes required to describe and encode a multivariate fusion prediction model; Indicates the number of parameters in a multivariate fusion prediction model; This indicates a description of each Number of bytes required; A conditional probability table representing the attributes of vehicle n; This represents the probability distribution corresponding to the multivariate fusion prediction model. describe Number of bytes required Training set of time-series data representing vehicle attributes;

[0053] in, The calculation formula is expressed by the following formula (4):

[0054] (4)

[0055] in, The calculation formula is expressed by the following formula (5):

[0056] (5)

[0057] in, The attribute representing the time slot t of vehicle n;

[0058] The final expression of the scoring function of the multivariate fusion prediction model is obtained according to formulas (4) and (5), and is expressed by the following formula (6):

[0059] (6)

[0060] in, The size of the training set sample space for time-series data representing vehicle attributes.

[0061] In one feasible implementation, if the network structure G is fixed, then It is a constant; searching for what makes it constant. Minimizing a multivariate fusion prediction model is equivalent to minimizing the parameters. Maximum likelihood estimation requires structural search. Since searching for the optimal VTP-DBN structure in all possible network structure spaces is an NP-hard problem and difficult to solve quickly, this embodiment of the invention uses a greedy algorithm to obtain an approximate solution in a finite amount of time.

[0062] Optionally, based on the scoring function of S1 and a dynamic Bayesian network, a final multivariate fusion prediction model is constructed, including:

[0063] S11. Obtain the time series training set, the set of evidence variables and their corresponding values, the variable to be queried and its corresponding value, the set of all possible dependency edges, the threshold of the scoring function, and the number of samples; wherein, the set of evidence variables is the vehicle n in the time slot. The set of attributes;

[0064] S12. Initialize the multivariate fusion prediction model as follows: ;in Contains only independent nodes; sets the first temporary variable. Initialize to the initial scoring function; set the second temporary variable. Initialize to 0;

[0065] S13. Based on the set of all possible dependency edges, calculate the score function when each edge is added to the multivariate fusion prediction model according to the score function of the multivariate fusion prediction model.

[0066] S13. Traverse the set of all possible dependency edges, and save the minimum score function and the index of the corresponding optimal edge; where, after each round, add the optimal edge to the multivariate fusion prediction model and remove the optimal edge from the set of all possible dependency edges.

[0067] S14. Update the first temporary variable based on the latest multivariate fusion prediction model. Repeatedly traverse the set of all possible dependency edges and add the best edge to the multivariate fusion prediction model until the scoring function is less than the threshold of the scoring function, thus obtaining the final multivariate fusion prediction model.

[0068] In one feasible implementation, during the vehicle location prediction stage, ideally, the precise posterior probability can be calculated based on the joint probability distribution defined by the multivariate fusion prediction model. However, it has been proven that such precise inference suffers from an NP-hard problem. Therefore, the multivariate fusion prediction model employs the Gibbs sampling algorithm for approximate inference, reducing the accuracy requirement and typically obtaining an approximate solution within a finite time. Table 1 shows the execution code for constructing the multivariate fusion prediction model.

[0069] Table 1

[0070]

[0071] S2. Obtain the vehicle's attribute set for the future time period and the attribute set for the current time slot; input the vehicle's attribute set for the future time period and the attribute set for the current time slot into the final multivariate fusion prediction model for prediction to obtain the vehicle's location prediction information.

[0072] Optionally, the specific implementation process of S2 includes S21-S24:

[0073] S21. Obtain the set of vehicle attributes for a future time period and assign initial values ​​to each attribute;

[0074] Among them, the initial value is randomly assigned to vehicles From the future Time The set of attributes, and Initialize to .

[0075] S22. Set temporary variables Initialized as an M-dimensional 0-dimensional vector; where, before each sampling round is updated, each vehicle in the time slot is... The values ​​corresponding to the attributes are initialized; among them, temporary variables... It is an m-dimensional vector.

[0076] S23, From the perspective of time Time Sampling was performed, and the time slot for each vehicle was changed. The attribute values ​​are used as one round of sampling; the current values ​​of the variables other than the object are selected according to the multivariate fusion prediction model;

[0077] in, This represents the current value of all variables except the selected object in the multivariate fusion prediction model.

[0078] S24. Based on the pre-set number of Gibbs samplings, obtain the sampling probability according to the multivariate fusion prediction model and the current values ​​of the remaining variables; according to the sampling probability, sample the i-th attribute of vehicle n at time t, and put the sampled attribute set into the evidence variable set to update the value at the next time; obtain the predicted information of vehicle position through multiple samplings.

[0079] Wherein, the sampling probability is expressed as .

[0080] Among them, temporary variables The The numbers in the dimension represent the random sampling location region. The total number of times samples were taken.

[0081] Optionally, the calculation process for the vehicle position prediction information is represented by the following formulas (7)-(9):

[0082] (7)

[0083] (8)

[0084] (9)

[0085] Among them, the function Convert the predicted position of the vehicle into a one-hot vector representation; Represent a A dimensional vector, where the dimensional vector is... The number of dimensions represents the vehicle. In the location area The posterior probability; The variable to be queried is a vector consisting of highway regions in future time slots; A tuple representing the attributes of vehicle n at the current time.

[0086] This invention addresses the continuity of service requirements for intelligent vehicles by proposing an adaptive service scheduling method based on prediction results. It employs reinforcement learning for fitting, using the posterior probability output by the prediction model and the current environmental state as the input state space for the adaptive service scheduling method. Through continuous interaction with the environment, this method constructs a reward function based on comprehensive performance indicators such as latency, energy consumption, and unloading success rate. This function guides the policy network to dynamically adjust unloading actions, achieving optimal pre-deployment and scheduling of tasks. Table 2 shows the execution code for the vehicle location prediction phase.

[0087] Table 2

[0088]

[0089] Among them, such as Figure 4 This is a flowchart of a multivariate fusion prediction model algorithm for ensuring service continuity of intelligent vehicles, provided by an embodiment of the present invention. In one feasible implementation, the input includes a time-series data training set of vehicle attributes, a set of evidence variables and their corresponding values, the variable to be queried (i.e., the region where the vehicle is located in the time interval and its corresponding value), a set of all possible dependency edges, a threshold for the scoring function, and the number of Gibbs samplings. The multivariate fusion prediction model is initialized, the current scoring function value is calculated, and the current scoring function value is assigned to a first temporary variable. A second temporary variable is initialized to 0. It is determined whether the first temporary variable is greater than the threshold of the scoring function; if less, the multivariate fusion prediction model is directly output; if greater, the loop count i is initialized to 1, and it is determined whether the loop count is less than or equal to the number of possible dependency edges in the set. If the number of loops is less than or equal to the number of possible dependency edges in the set, the loop count is determined to be greater than or equal to the number of possible dependency edges in the set. If the condition is otherwise met, add the nth edge to the multivariate fusion prediction model and remove the aforementioned edge from the set of all possible dependency edges. Assign the current score function value to the first temporary variable. If the condition is met, select the i-th edge from the set of all possible dependency edges, calculate the current score function, and add the selected edge to the multivariate fusion prediction model. Check if the current score function value is less than the first temporary variable. If the condition is otherwise met, increment the loop count by 1 and return to the process of re-checking if the loop count is less than or equal to the number of edges in the set of all possible dependency edges. If the condition is met, assign the current score function value to the first temporary variable, assign the current loop count to the second temporary variable, increment the loop count by 1, and return to the process of re-checking if the loop count is less than or equal to the number of edges in the set of all possible dependency edges. Based on the output multivariate fusion prediction model, initialize and randomly assign n future times to the vehicle. Time The set of attributes of vehicle n at the current time; assign the set of attributes of vehicle n at the current time to vehicle n at the time. A collection of attributes; setting temporary variables. Initialize to 0, set the loop count t to 1; check if variable t is less than or equal to the Gibbs sampling count, if yes, otherwise output the M-dimensional probability of vehicle n in the M location regions (i.e., (Divide by the value of the Gibbs sampling count); if the determination is yes, then initialize, and set vehicle n in time. The attribute set is assigned values ​​to the evidence variable set; the loop count t is initialized; it is then determined whether the loop count t is less than or equal to... If the condition is met, the predicted position of the vehicle is converted into an M-dimensional one-hot vector representation. The process involves adding the original value to the M-dimensional one-hot vector, incrementing the loop count by 1, and outputting the M-dimensional posterior probability of the vehicle in the M location regions. If the condition is met, the process continues to determine whether the i-th attribute of vehicle n at time t is included in the total vehicle quantity attribute. If the condition is met but not met, after updating the attribute values ​​of the vehicle at each time point, the set of attributes is included in the evidence variable, the loop count t is incremented by 1, and the process returns. If the condition is met, the sampling probability is obtained through the multivariate fusion prediction model and the current values ​​of the other variables excluding the selected object. Based on the sampling probability, the attribute values ​​of the sampled vehicle at each time point are updated, and the loop count is incremented by 1. The process then returns to the step where the i-th attribute of vehicle n at time t is included in the total vehicle quantity attribute.

[0090] S3. Based on the predicted information of vehicle location, the vehicle unloading strategy guided by the predicted information is optimized by reinforcement learning to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is actively pre-scheduled.

[0091] Optionally, the specific implementation process of S3 includes S31-S33:

[0092] S31. Construct a state space based on vehicle location prediction information; construct a policy space based on vehicle task unloading decisions;

[0093] The state space is represented as ;in, This represents the posterior probability of a vehicle in each location region, as output by the variable fusion prediction model. This indicates the distance between each vehicle and the base station and a single roadside unit; Indicates the remaining computing resources; This indicates the remaining bandwidth resources.

[0094] S32. Based on the state space and policy space, construct the reward function by taking actions;

[0095] Wherein, the policy space is represented as ;in, This indicates the task unloading decision for each vehicle; This indicates the channel bandwidth allocation decision.

[0096] Among them, in state Take action Receive instant rewards ;in, It is related to the system's latency and energy consumption.

[0097] Alternatively, the reward function can be expressed by the following formula (10):

[0098] (10)

[0099] in, Indicates a reward; It is a constant; Indicates the first related indicator; This indicates the second correlation indicator; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total latency; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total energy consumption.

[0100] Among them, when the system reaches the minimum overall latency and energy consumption, an immediate reward is given. It has reached its maximum value.

[0101] S33. Based on the reward function, guide the policy network to dynamically adjust the unloading action to obtain an intelligent policy for adaptive learning.

[0102] This invention first acquires a training set of time-series data on vehicle attributes; based on this training set, a scoring function for a multivariate fusion prediction model is constructed using the minimum description length criterion; according to the scoring function, a final multivariate fusion prediction model is constructed based on a dynamic Bayesian network; the attribute sets for future time periods and current time slots of the vehicle are acquired; these sets are input into the final multivariate fusion prediction model for prediction to obtain vehicle location prediction information; based on the vehicle location prediction information, a reinforcement learning method is used to optimize the vehicle unloading strategy guided by the prediction information to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is proactively pre-scheduled.

[0103] This invention constructs a prediction model based on multivariate fusion. A temporal modeling framework is built upon a dynamic Bayesian network to jointly model multiple variables, including vehicle trajectory, speed, task stage state, and edge node load, forming a multi-time-stack joint probability distribution representation. Through training, state transition probabilities and observation probabilities are obtained, enabling predictive inference of the vehicle's future state and the probability of target node access. This model is suitable for modeling complex temporal dependencies in highly dynamic vehicle-to-everything (V2X) scenarios. Furthermore, the multivariate fusion prediction model utilizes Gibbs sampling to obtain a posterior probability estimate of the vehicle's future trajectory, thus approximating the posterior distribution in a high-dimensional state space under limited data conditions. This significantly reduces reliance on large-scale labeled trajectory data and lowers the cost of data acquisition and model training. The obtained posterior probability distribution serves as part of the input to the reinforcement learning scheduling strategy, providing more accurate prior guidance for subsequent unloading decisions.

[0104] This invention proposes an adaptive service scheduling method based on prediction results. The posterior probability output by the prediction model and the current environmental state are used as the state space input to the adaptive service scheduling method. Through continuous interaction with the environment, this method constructs a reward function based on comprehensive performance indicators such as latency, energy consumption, and offloading success rate. This function guides the policy network to dynamically adjust offloading actions, achieving optimal pre-deployment and scheduling of tasks.

[0105] The proposed proactive pre-deployment architecture for ensuring service continuity in intelligent vehicles, as described in this invention, firstly, accurately characterizes the dynamic uncertainties in vehicle behavior, network state, and task evolution through a multivariate temporal modeling framework, enabling prediction of future states. Secondly, it optimizes the offloading strategy guided by prediction information into an adaptively learnable intelligent strategy through reinforcement learning, thereby achieving proactive pre-scheduling of services, ensuring service continuity, real-time performance, and robustness, and improving the overall system efficiency. Furthermore, the multivariate fusion prediction model constructed in this invention utilizes Gibbs sampling to obtain the posterior probability estimate of the vehicle's future trajectory, approximating the posterior distribution in the high-dimensional state space under limited data conditions, significantly reducing dependence on large-scale labeled trajectory data and lowering the cost of data acquisition and model training.

[0106] Figure 5 This is a block diagram illustrating a service-continuous intelligent vehicle task prediction and scheduling apparatus according to an exemplary embodiment. The apparatus is used in a service-continuous intelligent vehicle task prediction and scheduling method. (Refer to...) Figure 5 The device includes a construction unit 510, an acquisition unit 520, and a scheduling unit 530. Wherein:

[0107] Construction unit 510 is used to acquire a training set of time series data of vehicle attributes; based on the training set of time series data of vehicle attributes, a scoring function of the multivariate fusion prediction model is constructed using the minimum description length criterion; and based on the scoring function, the final multivariate fusion prediction model is constructed using a dynamic Bayesian network.

[0108] The acquisition unit 520 is used to acquire the attribute set of the vehicle in the future time period and the attribute set in the current time slot; the attribute set of the vehicle in the future time period and the attribute set in the current time slot are input into the final multivariate fusion prediction model for prediction to obtain the predicted information of the vehicle position;

[0109] The scheduling unit 530 is used to optimize the vehicle unloading strategy guided by the predicted information based on the vehicle location through reinforcement learning methods to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is actively pre-scheduled.

[0110] Optionally, the scoring function of the multivariate fusion prediction model is expressed by the following formula (1):

[0111] (1)

[0112] in, This represents the scoring function of the multivariate fusion prediction model on the time-series data training set of vehicle attributes; This indicates the number of bytes required to describe and encode a multivariate fusion prediction model; Indicates the number of parameters in a multivariate fusion prediction model; This indicates a description of each Number of bytes required; A conditional probability table representing the attributes of vehicle n; This represents the probability distribution corresponding to the multivariate fusion prediction model. describe Number of bytes required Training set of time-series data representing vehicle attributes;

[0113] in, The calculation formula is expressed by the following formula (2):

[0114] (2)

[0115] in, The calculation formula is expressed by the following formula (3):

[0116] (3)

[0117] in, The attribute representing the time slot t of vehicle n;

[0118] The final expression of the scoring function of the multivariate fusion prediction model is obtained according to formulas (2) and (3), and is expressed by the following formula (4):

[0119] (4)

[0120] in, The size of the training set sample space for time-series data representing vehicle attributes.

[0121] Optionally, the building unit is used for:

[0122] Obtain the time series training set, the set of evidence variables and their corresponding values, the variable to be queried and its corresponding value, the set of all possible dependency edges, the threshold of the scoring function, and the number of samples; where the set of evidence variables represents the time slot of vehicle n. The set of attributes;

[0123] Initialize the multivariate fusion prediction model as follows: ;in Contains only independent nodes; sets the first temporary variable. Initialize to the initial scoring function; set the second temporary variable. Initialize to 0;

[0124] Based on the set of all possible dependency edges, calculate the score function when each edge is added to the multivariate fusion prediction model according to the score function of the multivariate fusion prediction model;

[0125] Iterate through the set of all possible dependency edges, and save the minimum score function and the index of the corresponding optimal edge; at the end of each round, add the optimal edge to the multivariate fusion prediction model, and remove the optimal edge from the set of all possible dependency edges.

[0126] Based on the latest multivariate fusion prediction model, update the first temporary variable. Repeatedly traverse the set of all possible dependency edges and add the best edge to the multivariate fusion prediction model until the scoring function is less than the threshold of the scoring function, thus obtaining the final multivariate fusion prediction model.

[0127] Optionally, the acquisition unit is configured to:

[0128] Obtain the set of vehicle attributes for a future time period and assign initial values ​​to each attribute;

[0129] Set temporary variables Initialized as an M-dimensional 0-dimensional vector; where, before each sampling round is updated, each vehicle in the time slot is... Initialize the values ​​corresponding to the attributes;

[0130] From time Time Sampling was performed, and the time slot for each vehicle was changed. The attribute values ​​are used as one round of sampling; the current values ​​of the variables other than the object are selected according to the multivariate fusion prediction model;

[0131] Based on a pre-set number of Gibbs samplings, the sampling probability is obtained according to the multivariate fusion prediction model and the current values ​​of other variables. Based on the sampling probability, the i-th attribute of vehicle n at time t is sampled. The sampled attribute set is then placed into the evidence variable set to update the value at the next time. Through multiple samplings, the predicted information of vehicle position is obtained.

[0132] Optionally, the calculation process of the predicted vehicle position information is represented by the following formulas (5)-(7):

[0133] (5)

[0134] (6)

[0135] (7)

[0136] Among them, the function Convert the predicted position of the vehicle into a one-hot vector representation; Represent a A dimensional vector, where the dimensional vector is... The dimension represents the vehicle. In the location area The posterior probability; The variable to be queried is a vector consisting of highway regions in future time slots; A tuple representing the attributes of vehicle n at the current time.

[0137] Optionally, the scheduling unit is configured to:

[0138] Based on the predicted information of vehicle location, a state space is constructed; based on the vehicle task unloading decision, a policy space is constructed.

[0139] Based on the state space and policy space, a reward function is constructed by taking actions;

[0140] Based on the reward function, the policy network is guided to dynamically adjust the unloading action to obtain an intelligent policy that learns adaptively.

[0141] Optionally, the reward function is expressed by the following formula (8):

[0142] (8)

[0143] in, Indicates a reward; It is a constant; Indicates the first related indicator; Indicates the second correlation indicator; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total latency; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total energy consumption.

[0144] This invention proposes a service-oriented intelligent vehicle task prediction and scheduling method. The method uses a dynamic Bayesian network as its core framework, leveraging its ability to effectively handle probabilistic graphical models of time-series data. It describes the dynamic changes in vehicle trajectories by constructing conditional probability distributions. Specifically, by modeling the uncertainties of multivariate temporal relationships and state transitions, it achieves joint distribution modeling of the evolution of multiple dimensions—vehicle behavior, task state, and network resources—over time. The network structure and parameters are dynamically adjusted based on historical vehicle trajectory data, enabling the network to better adapt to the time-varying characteristics of trajectories and improve prediction accuracy. Furthermore, the dynamic Bayesian network can jointly model explicit and implicit variables, making it suitable for describing the dynamic processes of vehicle motion patterns and network states. It can also express stage dependencies and conditional probability propagation mechanisms, possessing strong reasoning and prediction capabilities. It can infer the system state at multiple future moments based on the current observation state, thus supporting forward-looking scheduling. Building upon this, this invention combines the prediction information provided by the dynamic Bayesian network with reinforcement learning methods. Based on reinforcement learning, the system autonomously learns the optimal unloading strategy in a time-varying environment, effectively solving problems such as high state space dimensionality, complex unloading actions, and strong environmental uncertainty in task chain scheduling. This enables intelligent decision optimization and resource scheduling enhancement with prediction assistance, further improving the system's latency response performance and service continuity.

[0145] This invention first acquires a training set of time-series data on vehicle attributes; based on this training set, a scoring function for a multivariate fusion prediction model is constructed using the minimum description length criterion; according to the scoring function, a final multivariate fusion prediction model is constructed based on a dynamic Bayesian network; the attribute sets for future time periods and current time slots of the vehicle are acquired; these sets are input into the final multivariate fusion prediction model for prediction to obtain vehicle location prediction information; based on the vehicle location prediction information, a reinforcement learning method is used to optimize the vehicle unloading strategy guided by the prediction information to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is proactively pre-scheduled.

[0146] This invention constructs a prediction model based on multivariate fusion. A temporal modeling framework is built upon a dynamic Bayesian network to jointly model multiple variables, including vehicle trajectory, speed, task stage state, and edge node load, forming a multi-time-stack joint probability distribution representation. Through training, state transition probabilities and observation probabilities are obtained, enabling predictive inference of the vehicle's future state and the probability of target node access. This model is suitable for modeling complex temporal dependencies in highly dynamic vehicle-to-everything (V2X) scenarios. Furthermore, the multivariate fusion prediction model utilizes Gibbs sampling to obtain a posterior probability estimate of the vehicle's future trajectory, thus approximating the posterior distribution in a high-dimensional state space under limited data conditions. This significantly reduces reliance on large-scale labeled trajectory data and lowers the cost of data acquisition and model training. The obtained posterior probability distribution serves as part of the input to the reinforcement learning scheduling strategy, providing more accurate prior guidance for subsequent unloading decisions.

[0147] This invention proposes an adaptive service scheduling method based on prediction results. The posterior probability output by the prediction model and the current environmental state are used as the state space input to the adaptive service scheduling method. Through continuous interaction with the environment, this method constructs a reward function based on comprehensive performance indicators such as latency, energy consumption, and offloading success rate. This function guides the policy network to dynamically adjust offloading actions, achieving optimal pre-deployment and scheduling of tasks.

[0148] The proposed proactive pre-deployment architecture for ensuring service continuity in intelligent vehicles, as described in this invention, firstly, accurately characterizes the dynamic uncertainties in vehicle behavior, network state, and task evolution through a multivariate temporal modeling framework, enabling prediction of future states. Secondly, it optimizes the offloading strategy guided by prediction information into an adaptively learnable intelligent strategy through reinforcement learning, thereby achieving proactive pre-scheduling of services, ensuring service continuity, real-time performance, and robustness, and improving the overall system efficiency. Furthermore, the multivariate fusion prediction model constructed in this invention utilizes Gibbs sampling to obtain the posterior probability estimate of the vehicle's future trajectory, approximating the posterior distribution in the high-dimensional state space under limited data conditions, significantly reducing dependence on large-scale labeled trajectory data and lowering the cost of data acquisition and model training.

[0149] Figure 6 This is a schematic diagram of the structure of an intelligent vehicle task prediction and scheduling device for service continuity provided in an embodiment of the present invention, as shown below. Figure 6 As shown, a service-continuous intelligent vehicle task prediction and scheduling device may include the above-mentioned Figure 5 The illustrated intelligent vehicle task prediction and scheduling device is designed for continuous service. Optionally, the intelligent vehicle task prediction and scheduling device 610 designed for continuous service may include a first processor 2001.

[0150] Optionally, the intelligent vehicle task prediction and scheduling device 610 for continuous service may also include a memory 2002 and a transceiver 2003.

[0151] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0152] The following is combined with Figure 6 The components of the intelligent vehicle task prediction and scheduling device 610 for continuous service are described in detail below:

[0153] The first processor 2001 is the control center of the intelligent vehicle task prediction and scheduling device 610 for continuous service. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0154] Optionally, the first processor 2001 can perform various functions of the service-oriented intelligent vehicle task prediction and scheduling device 610 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0155] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 6 CPU0 and CPU1 are shown in the diagram.

[0156] In a specific implementation, as one example, the intelligent vehicle task prediction and scheduling device 610 for service continuity may also include multiple processors, for example... Figure 6 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0157] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0158] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be accessed via the interface circuit of the intelligent vehicle task prediction and scheduling device 610 for continuous service. Figure 6 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0159] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0160] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 6 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0161] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected to the interface circuit of the intelligent vehicle task prediction and scheduling device 610 for continuous service. Figure 6 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0162] It should be noted that, Figure 6 The structure of the service-oriented intelligent vehicle task prediction and scheduling device 610 shown in the figure does not constitute a limitation on the router. The actual knowledge structure identification device may include more or fewer components than shown, or combine some components, or have different component arrangements.

[0163] Furthermore, the technical effects of the service-continuous intelligent vehicle task prediction and scheduling device 610 can be referred to the technical effects of the service-continuous intelligent vehicle task prediction and scheduling method described in the above method embodiments, and will not be repeated here.

[0164] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0165] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0166] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0167] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0168] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0169] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0170] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0171] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0172] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0173] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0175] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0176] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A service-continuous intelligent vehicle task prediction and scheduling method, characterized in that, The method includes: S1. Obtain a training set of time series data of vehicle attributes; based on the training set of time series data of vehicle attributes, construct a scoring function for the multivariate fusion prediction model using the minimum description length criterion; based on the scoring function, construct the final multivariate fusion prediction model using a dynamic Bayesian network. S2. Obtain the vehicle's attribute set for the future time period and the attribute set for the current time slot; input the vehicle's attribute set for the future time period and the attribute set for the current time slot into the final multivariate fusion prediction model for prediction to obtain the vehicle's location prediction information; S3. Based on the predicted information of vehicle location, the vehicle unloading strategy guided by the predicted information is optimized by reinforcement learning to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is actively pre-scheduled.

2. The service-oriented intelligent vehicle task prediction and scheduling method according to claim 1, characterized in that, The scoring function of the multivariate fusion prediction model is expressed by the following formula (1): （1） in, This represents the scoring function of the multivariate fusion prediction model on the time-series data training set of vehicle attributes; This indicates the number of bytes required to describe and encode a multivariate fusion prediction model; Indicates the number of parameters in a multivariate fusion prediction model; This indicates a description of each Number of bytes required; A conditional probability table representing the attributes of vehicle n; This represents the probability distribution corresponding to the multivariate fusion prediction model. describe Number of bytes required Training set of time-series data representing vehicle attributes; in, The calculation formula is expressed by the following formula (2): （2） in, The calculation formula is expressed by the following formula (3): （3） in, The attribute representing the time slot t of vehicle n; The final expression of the scoring function of the multivariate fusion prediction model is obtained according to formulas (2) and (3), and is expressed by the following formula (4): （4） in, The size of the training set sample space for time-series data representing vehicle attributes.

3. The service-oriented intelligent vehicle task prediction and scheduling method according to claim 1, characterized in that, The step of constructing the final multivariate fusion prediction model based on the scoring function in S1 and a dynamic Bayesian network includes: S11. Obtain the time series training set, the set of evidence variables and their corresponding values, the variable to be queried and its corresponding value, the set of all possible dependency edges, the threshold of the scoring function, and the number of samples; wherein, the set of evidence variables is the vehicle n in the time slot. The set of attributes; S12. Integrating multivariate prediction models Initialize to ;in Contains only independent nodes; sets the first temporary variable. And initialize; set the second temporary variable. Initialize to 0; S13. Based on the set of all possible dependency edges, calculate the score function when each edge is added to the multivariate fusion prediction model according to the score function of the multivariate fusion prediction model. S13. Traverse the set of all possible dependency edges, and save the minimum score function and the index of the corresponding optimal edge; where, after each round, add the optimal edge to the multivariate fusion prediction model and remove the optimal edge from the set of all possible dependency edges. S14. Update the first temporary variable based on the latest multivariate fusion prediction model. Repeatedly traverse the set of all possible dependency edges and add the best edge to the multivariate fusion prediction model until the scoring function is less than the threshold of the scoring function, thus obtaining the final multivariate fusion prediction model.

4. The service-oriented intelligent vehicle task prediction and scheduling method according to claim 1, characterized in that, S2 involves acquiring the vehicle's attribute set for the future time period and the attribute set for the current time slot; inputting these two sets into the final multivariate fusion prediction model for prediction to obtain vehicle location prediction information, including: S21. Obtain the set of vehicle attributes for a future time period and assign initial values ​​to each attribute; S22. Set temporary variables Initialized as an M-dimensional 0-dimensional vector; where, before each sampling round is updated, each vehicle in the time slot is... Initialize the values ​​corresponding to the attributes; S23, From the perspective of time Time Sampling was performed, and the time slot for each vehicle was changed. The attribute values ​​are used as one round of sampling; the current values ​​of the variables other than the object are selected according to the multivariate fusion prediction model; S24. Based on the pre-set number of Gibbs samplings, obtain the sampling probability according to the multivariate fusion prediction model and the current values ​​of the remaining variables; according to the sampling probability, sample the i-th attribute of vehicle n at time t; wherein, the sampled attribute set is put into the evidence variable set to update the value at the next time; through multiple samplings, obtain the predicted information of vehicle position.

5. The service-continuous intelligent vehicle task prediction and scheduling method according to claim 1, characterized in that, The calculation process of the predicted vehicle location information is expressed by the following formulas (5)-(7): （5） （6） （7） Among them, the function Convert the predicted position of the vehicle into a one-hot vector representation; Represent a A dimensional vector, where the dimensional vector is... The number of dimensions represents the vehicle. In the location area The posterior probability; The variable to be queried is a vector consisting of highway regions in future time slots; A tuple representing the attributes of vehicle n at the current time.

6. The service-oriented intelligent vehicle task prediction and scheduling method according to claim 1, characterized in that, The vehicle location-based prediction information in S3 is used to optimize the vehicle unloading strategy guided by the prediction information through reinforcement learning, thereby obtaining an adaptive learning intelligent strategy, including: S31. Construct a state space based on vehicle location prediction information; construct a policy space based on vehicle task unloading decisions; S32. Based on the state space and policy space, construct the reward function by taking actions; S33. Based on the reward function, guide the policy network to dynamically adjust the unloading action to obtain an intelligent policy for adaptive learning.

7. The service-oriented intelligent vehicle task prediction and scheduling method according to claim 6, characterized in that, The reward function is expressed by the following formula (8): （8） in, Indicates a reward; It is a constant; Indicates the first related indicator; Indicates the second correlation indicator; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total latency; Indicates the state Take action Afterwards, the simulated entire vehicle networking system in time slots Total energy consumption.

8. A service-continuous intelligent vehicle task prediction and scheduling device, wherein the service-continuous intelligent vehicle task prediction and scheduling device is used to implement the service-continuous intelligent vehicle task prediction and scheduling method as described in any one of claims 1-7, characterized in that, The device includes: A construction unit is used to acquire a training set of time-series data of vehicle attributes; based on the training set of time-series data of vehicle attributes, a scoring function for a multivariate fusion prediction model is constructed using the minimum description length criterion; and based on the scoring function, a final multivariate fusion prediction model is constructed using a dynamic Bayesian network. The acquisition unit is used to acquire the attribute set of the vehicle in the future time period and the attribute set in the current time slot; the attribute set of the vehicle in the future time period and the attribute set in the current time slot are input into the final multivariate fusion prediction model for prediction to obtain the predicted information of the vehicle position; The scheduling unit is used to optimize the vehicle unloading strategy guided by the predicted information based on vehicle location through reinforcement learning methods to obtain an adaptive learning intelligent strategy; based on the adaptive learning intelligent strategy, the task of ensuring vehicle service continuity is proactively pre-scheduled.

9. A service-continuous intelligent vehicle task prediction and scheduling device, characterized in that, The service-continuous intelligent vehicle task prediction and scheduling device includes: processor; A memory storing computer-readable instructions that, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images