A personalized vehicle-road-cloud collaborative task offloading allocation method and system

Abstract

The application mainly solves the task unloading distribution problem in vehicle-road cloud cooperation by using a recommendation algorithm. Mainly includes: (1) the application constructs an evaluation index which can measure the individual satisfaction degree of users to the task unloading distribution scheme; (2) the application constructs a set of individualized vehicle-road cloud cooperation task unloading distribution method and system based on the recommendation algorithm. First, through the individual satisfaction degree evaluation function, the satisfaction degree of a large number of tasks is calculated, and the optimal task unloading distribution scheme is selected for each task, and these task models and selected service nodes are constructed as the offline data set of the recommendation system; second, the recommendation algorithm model is constructed, and the algorithm model is trained by using the offline data set, so that the model can take the user request as the input and output the recommended task unloading distribution method, effectively meet the individual unloading demand of the user, and obtain the task unloading distribution scheme which is more in line with the user preference and has higher satisfaction degree.

Description

TECHNICAL FIELD

[0001] The present application relates to a task offloading allocation method, more particularly to a personalized vehicle-road cloud collaborative task offloading allocation method and system. BACKGROUND

[0002] The vehicle-road cloud collaborative system effectively solves the problem of insufficient computing resources of vehicles by introducing the multi-access edge computing (MEC) technology into the Internet of Vehicles, and meets the requirements of vehicle-mounted applications for time delay. The MEC sinks cloud services to the edge of the wireless access network to provide computing services near mobile vehicles. Each task has different resource requirements, including computing resources required for task execution and communication resources required for task transmission. By deploying the MEC server on the roadside unit side, the task requesting vehicle can offload the task to the MEC server for calculation through the roadside unit, or offload the task to the adjacent vehicle for calculation. Compared with remote cloud computing, the vehicle-road cloud collaborative system makes up for the time delay jitter and capacity limitation of remote cloud computing, and is more suitable for the low time delay requirement of the Internet of Vehicles.

[0003] In the vehicle-road cloud collaborative system, the commonly used edge computing resource service nodes include: (1) nearby idle vehicles, i.e., vehicle-mounted edge clouds; (2) RSU servers on the roadside unit (RSU) side; and (3) MEC servers on the base station side. However, intelligent networked vehicles have a large number of high-concurrency computing tasks. At the same time, there are many edge computing resource service nodes, and the computing capabilities, resource costs, and processing speeds of the edge computing resource service nodes are quite different. In addition, different user tasks have different requirements, and users have different preferences for different requirements. If an inappropriate service node is selected for task offloading, it will result in low service quality, mutual interference, low computing efficiency, high offloading cost, waste of node resources, and other problems. Therefore, it is of great significance to improve user experience and the service quality of the offloading process to allocate service nodes that meet the personalized preferences of different user tasks.

[0004] The patent with publication number CN112749010A discloses an edge server database storing position information and computing capacity of a plurality of edge servers deployed on roadside facilities which can start and execute tasks according to requirements. When a task request is received, the filtered edge server information is filtered by a filtering algorithm, and then the filtered edge server information is clustered to obtain an executable edge server. The executable edge server is recommended to the cloud, and the cloud caches it and combines it with the edge server which did not participate in the task execution in the previous stage. The selected recommended and cached edge server which can execute the current task is recommended to the task sender. However, the above-mentioned method only considers whether it can execute the current task, and the cached edge server node information recommended to the user can only optimize the time delay, and does not consider other requirements of the user, such as unloading cost, transmission energy consumption, etc.

[0005] The patent with publication number CN114567909A establishes a user pairing model, an energy collection model, a local calculation model and a calculation offloading model to construct a network system minimum energy consumption model, and performs resource optimization allocation and optimal user matching on the minimum energy consumption model to minimize the task offloading energy consumption while meeting the time delay requirement. However, the above-mentioned method only minimizes the energy consumption target, and the scene is fixed and single, which cannot be applied to different requirements of users in actual scenes. SUMMARY

[0006] In view of the deficiencies in the prior art, the purpose of the present application is to provide a personalized vehicle-road cloud cooperative task offloading allocation method and system based on a recommendation algorithm, to obtain a task offloading allocation scheme that better meets the individual preferences of users, and to optimize the quality of service for users and improve user satisfaction.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solution: a personalized vehicle-road cloud cooperative task offloading allocation method, comprising the following steps:

[0008] Step S1: constructing a vehicle-road cloud cooperative task offloading system model;

[0009] Step S2: constructing a vehicle-road cloud cooperative task offloading target model and an individual satisfaction evaluation function;

[0010] Step S3: generating an offline data set of the recommendation system based on the models in step S1 and step S2 and the individual satisfaction evaluation function;

[0011] Step S4: constructing a recommendation algorithm model, so that the recommended allocation scheme can be obtained through the offline data set in step S3;

[0012] Step S5: Discretize time into time slots, the system receives all user task requests in a time slot t as a set of task requests input at time slot t, and then uses the recommended algorithm model in step S4 to recommend the optimal service node for the user.

[0013] As a further improvement of the present application, the construction of the vehicle-road cloud cooperative task offloading system model in step S1 specifically includes the construction of the task model and the service node model in the vehicle-road cloud cooperative task offloading system, and the task model is as follows:

[0014] At time slot t, the task request i includes the following detailed task information (t i ,q i ,s i ,d i ,p i );

[0015] Where t i is the delay requirement of task request i, i.e., the task is required to be completed within its deadline t i ; q i is the number of CPU cycles required by task request i, s i is the storage resource requirement of task request i, d i is the data packet size of task i, and p i is the uplink transmission power of the vehicle sending the task request i. At time slot t, the tasks from multiple intelligent connected vehicles are formed into a task set T.

[0016] The service node model is as follows: at time slot t, for the jth service node, the available resource information is represented by ;

[0017] Where f represents the computing frequency of the service node j; b is the bandwidth of the service node j; is the computing resource price, storage resource price, and network resource price of the service node j.

[0018] As a further improvement of the present application, the specific steps of constructing the vehicle-road cloud cooperative task offloading target model and the personalized satisfaction evaluation function in step S2 are as follows:

[0019] Step S2.1: Calculate the transmission delay and the computing delay to construct a delay model;

[0020] Step S2.2: Construct a cost model by calculating the resource cost, network resource cost, and storage resource cost, and the constructed cost model is as follows:

[0021] For task i∈T and edge computing node j∈C, the total cost c i,j of the offloading process is calculated as follows:

[0022]

[0023] wherein q i is the number of CPU cycles required for task i, is the price of a unit CPU cycle of a service node; s i is the amount of storage resources required for task i, is the price of a unit storage space of a service node; b i is the network bandwidth that the service node can provide when the task i and the service node establish communication;

[0024] Step S2.3: During the process of transmitting data from the vehicle end to the service node by the task and receiving the data result of the service node, the energy consumption model of the user equipment is constructed, and the constructed energy consumption model is as follows:

[0025] For task i∈T and edge computing node j∈C, the total energy consumption E i,j of the offloading process is calculated as follows:

[0026]

[0027] wherein p i is the uplink transmission power of the vehicle sending the task request i, is the uplink transmission time.

[0028] Therefore, the total energy consumption E i,j between the task i and the edge computing node j is calculated as follows:

[0029]

[0030] Step S2.4: The personalized dynamic weight coefficient is constructed by the weight coefficient of the delay feature, the weight coefficient of the cost feature and the weight coefficient of the energy consumption feature.

[0031] Step S2.5: The personalized satisfaction evaluation function is constructed by combining the personalized weight coefficient and comprehensively calculating the satisfaction degree of the user to the offloading to the node, and the satisfaction degree formula is defined as follows:

[0032] Q=r c ×c i,j +r t ×t i,j +r E ×E i,j .

[0033] As a further improvement of the present application, the specific steps of constructing the delay model in step S2.1 are as follows:

[0034] Step S2.11, first, the transmission delay is calculated, which refers to the time required for task data to be transmitted from the vehicle end to the service node via the network. For task i∈T and edge computing node j∈C, the transmission delay of the task is The calculation formula is as follows:

[0035]

[0036] where d i is the size of the data packet to be transmitted, b j is the network bandwidth between user i and service node j;

[0037] Step S2.12, second, the calculation delay is calculated, which refers to the time required for the task to be calculated and processed in the service node. For task i∈T and edge computing node j∈C, the calculation delay of the task is The calculation formula is as follows:

[0038]

[0039] where q i is the number of CPU cycles required for the task, and f j is the calculation frequency of the service node j;

[0040] Step S2.13, finally, the total delay is calculated. Since the downlink transmission delay of the result returned from the edge to the vehicle is ignored, the total delay t i,j between task i and edge computing node j is the sum of the transmission delay and the calculation delay , and the calculation formula is as follows:

[0041]

[0042] The total delay after increasing the queuing time is:

[0043]

[0044] As a further improvement of the present application, the calculation formula for constructing the personalized dynamic weight coefficient in step S2.4 is as follows:

[0045]

[0046] where the weight coefficient r t of the delay feature, the weight coefficient r c of the cost feature, and the weight coefficient r E of the energy consumption feature, w c , w t , and w E are the demand degrees of the user for the delay, the cost, and the energy consumption, respectively, which are calculated only by inputting the feature parameters of the user and are irrelevant to the features of the service node.c , w t and w E The calculation process is as follows:

[0047] For task i∈T, the degree of demand of the user task for the delay feature w t The calculation formula is as follows:

[0048]

[0049] Where t Exp is t i , the delay requirement of the task request i; t ExpMin is the minimum value of the delay requirement in the delay interval, t ExpMax is the maximum value of the delay requirement in the delay interval;

[0050] For task i∈T, the degree of demand of the user task for the cost feature w c The calculation formula is as follows:

[0051]

[0052] Where C Exp is the cost expectation value of the task request i; C ExpMin is the minimum value of the cost requirement in the cost interval, C ExpMax is the maximum value of the cost requirement in the cost interval, and the calculation formula is as follows:

[0053] C Exp = (q + s + b ave ) * p ave

[0054] C ExpMin = (q + s + b ave ) * p min

[0055] C ExpMax = (q + s + b ave ) * p max

[0056] For task i∈T, the degree of demand of the user task for the energy consumption feature w E The calculation formula is as follows:

[0057]

[0058] Where p Exp is p i , the uplink transmission power of the vehicle sending the task request i; p ExpMin is the minimum value of the transmission power in the transmission power interval, p ExpMaxThe maximum value of the transmission power in the transmission power interval. As a further improvement of the present application, the specific steps of generating the offline dataset of the recommendation system in step S3 based on the model and the personalized satisfaction evaluation function in steps S1 and S2 are as follows:

[0059] Step S3.1: Based on the target model and the personalized satisfaction evaluation function in step S2, calculate the satisfaction value of each allocation scheme for each task, and take the allocation scheme corresponding to the maximum value as the optimal allocation scheme. Each optimal scheme can be represented as (t i ,q i ,s i ,d i ,p i ,E i ), where t i ,q i ,s i ,d i ,p i is the basic information of the task randomly generated based on the task model in step S1, and E i is the service node allocated by the optimal scheme for the task according to the target model and the personalized satisfaction evaluation function in step S2.

[0060] Step S3.2: Generate n samples according to the scheme of step S3.1 as the offline dataset that can be used by the recommendation system. Summarize the n allocation schemes and their corresponding original task request contents as the offline dataset of the recommendation system.

[0061] As a further improvement of the present application, the selection step of the optimal allocation scheme in step S3.1 is as follows:

[0062] Step S3.11, for the task input generated by a task model, traverse each service node, and calculate the satisfaction value when the task selects the service node through the satisfaction function in step S2.

[0063] Step S3.12, after all service nodes are calculated, select the service node with the maximum satisfaction value as the optimal scheme finally selected by the task.

[0064] As a further improvement of the present application, the specific steps of constructing the recommendation algorithm model in step S4 are as follows:

[0065] Step S4.1: Define the data object of the recommendation algorithm. The data used by the recommendation algorithm generally includes: items, users and transactions. In this scenario, the user is the vehicle issuing the task request, the item is the service node to be matched, and the transaction is the satisfaction of the vehicle user to the matched service node.

[0066] Step S4.2: selecting a suitable recommendation algorithm, the recommendation algorithm mainly includes content-based recommendation, collaborative filtering recommendation, hybrid recommendation, and knowledge-based, community-based, and context-aware, and a recommendation algorithm meeting the high delay requirement of the Internet of Vehicles application is selected as the recommendation method to be adopted;

[0067] Step S4.3: adapting the recommendation algorithm to the model, that is, the input of the algorithm is the task request model issued by the vehicle, and the recommendation result is obtained by training the offline data set obtained in step S3, and the recommendation result is the output of the algorithm.

[0068] As a further improvement of the application, the algorithm selected in step S4.2 is a user-based collaborative filtering method in collaborative filtering, the similarity of the request user and the users in the offline data set is calculated, the user with the maximum similarity to the request user in the offline data set is selected, and the service node selected by the similar user is the recommended service node, wherein the similarity of the user is calculated as follows:

[0069] Step S4.21, normalizing each parameter in its interval to avoid that some parameter values are too large and have too much influence on the similarity result;

[0070] Step S4.22, calculating the similarity of task i and each task in the offline database by using the Pearson correlation coefficient;

[0071]

[0072] Step S4.23, selecting the service node selected by the task with the maximum Pearson correlation coefficient value as the recommended matching service node.

[0073] Another aspect of the application provides a system applying the above method, comprising:

[0074] Task generation module: simulating real task requests, randomly generating multiple groups of tasks in the real task request data interval range as data set task samples;

[0075] Service node information collection module: collecting all service node information in the communication range, and matching the service node information into a service node model and sending the service node model to the offline data set module to provide service node information for the calculation of the offline data set;

[0076] Data set generation module: generating an optimal matching method according to the task request model, the service node model obtained based on the service node information collection module, and the satisfaction evaluation function, and storing multiple calculation results as the offline data set required by the recommendation module;

[0077] Task request collection module: collect all concurrent tasks within the communication range in the actual scene in time slot t, and send a set of task requests to the recommendation module;

[0078] Recommendation module: based on the task request information obtained by the task request collection module and the offline dataset module information, a recommended matching service node is calculated for the request through a recommendation algorithm model, and the result is sent to the user.

[0079] The beneficial effects of the present application are:

[0080] (1) The present application builds an evaluation index that can measure the user's personalized satisfaction degree of the task offloading allocation scheme, and realizes the calculation of the user's satisfaction degree of the allocated service node from the user's perspective. At the same time, since the weight coefficient is dynamically changed with the different characteristics of each user, unlike the fixed weight coefficient which uses the same set of evaluation standards for each user, the personalized satisfaction degree evaluation index provided by the present application is highly adaptable, can reflect the actual needs and personalized preferences of each user, and can more accurately describe the user's satisfaction degree of the allocated service node.

[0081] (2) The present application builds a set of personalized car-road cloud collaborative task offloading allocation method and system based on recommendation algorithm. The construction of offline dataset and the construction of recommendation algorithm model are realized, which can completely realize the function of task offloading allocation recommendation for user tasks, provide a personalized task offloading allocation method with the goal of maximizing user satisfaction, improve the user's satisfaction degree in the task offloading process, and optimize the service quality. BRIEF DESCRIPTION OF DRAWINGS

[0082] Fig. 1 It is a schematic diagram of car-road cloud collaborative task offloading;

[0083] Fig. 2 It is a flow chart of the task offloading allocation method based on recommendation algorithm provided by the present application;

[0084] Fig. 3 It is a module block diagram of the task offloading allocation system based on recommendation algorithm provided by the present application. DETAILED DESCRIPTION

[0085] The present application will be further described in conjunction with the embodiments given in the accompanying drawings.

[0086] Referring to Figs. 1 to 3 The personalized car-road cloud collaborative task offloading allocation method based on recommendation algorithm of the present embodiment includes the following steps:

[0087] Step S1: build a car-road cloud collaborative task offloading system model.

[0088] Step S2: constructing a task offloading target model of vehicle-road cloud cooperation and a personalized satisfaction evaluation function.

[0089] Step S3: generating an offline dataset of the recommendation system based on the model and the personalized satisfaction evaluation function in steps S1 and S2.

[0090] Step S4: constructing a recommendation algorithm model, which can obtain a recommended allocation scheme through the offline dataset in step S3.

[0091] Step S5: discretizing time into time slots, and the system receives all user task requests in a time slot as a set of task request inputs at the time slot τ. The recommendation algorithm model in step S4 is used to recommend an optimal service node for allocation to a user. In this embodiment, the system further comprises the following steps:

[0092] The task offloading system model in step S1 is not limited to the features, and other task features can be added, deleted or replaced.

[0093] The target model in step S2 is not limited to the three features of time delay, cost and energy consumption as the target features of the user, and other optimization targets can be added, deleted or replaced.

[0094] The process of selecting the optimal scheme in step S3 as the offline dataset is not limited to the exhaustive method in the embodiment, and other calculation methods for obtaining the optimal result can be used.

[0095] The recommendation algorithm model in step S4 is not limited to the user-based collaborative filtering method in the embodiment, and other algorithms in the recommendation system can be used, such as content-based collaborative filtering, item-based collaborative filtering, and hidden semantic model.

[0096] Step S1 includes the construction of a task model and a service node model in the vehicle-road cloud cooperative task offloading system. The specific modeling process includes:

[0097] Step S1.1: constructing a task model. At time t, the task request i includes the following detailed task information (t i ,q i ,s i ,d i ,p i ), where t i is the time delay requirement of the task request i, that is, the task is required to be completed within its deadline t i ; q i is the CPU cycle number required by the task request i, s i is the storage resource requirement of the task request i, d i is the data packet size of the task i, and p i is the uplink transmission power of the vehicle sending the task request i.

[0098] At time slot t, the tasks from multiple intelligent connected vehicles form a task set T.

[0099] Step S1.2: Construct a service node model. In a vehicle-road cloud scenario, an edge computing service node is composed of adjacent vehicles, roadside units (RSUs), and MEC servers, and has certain computing, storage, and network resources. At time slot t, for the jth service node, the available resource information is represented as , where f represents the computing frequency of the service node j; b is the bandwidth of the service node j; is the computing resource price, the storage resource price, and the network resource price of the service node j.

[0100] Step S2: Construct a vehicle-road cloud collaborative task offloading target model and a personalized satisfaction evaluation function, including a time delay model, a cost model, an energy consumption model, and a personalized satisfaction evaluation function. The specific modeling process includes:

[0101] Step S2.1: Construct a time delay model. In task offloading and allocation, the time delay feature of the offloading process is considered, that is, the total time for transmitting data from the vehicle end to the service node, processing, and returning the result to the vehicle end should be less than the maximum completion deadline limit of the task. The time delay mainly includes two parts: transmission time delay and computing time delay. Since the returned result data is very small, the downlink time delay from the service node to the vehicle is negligible compared with the uplink time delay from the vehicle end to the service node. In addition, in the same batch allocation, if node j has been selected, the task of selecting node j again will increase the queuing time.

[0102] The transmission time delay refers to the time required for transmitting task data from the vehicle end to the service node via the network. For task i∈T and edge computing node j∈C, the transmission time delay is The calculation formula is as follows:

[0103]

[0104] where d i is the size of the transmitted data packet, and b j is the network bandwidth between user i and service node j.

[0105] The computing time delay refers to the time required for computing and processing the task at the service node. For task i∈T and edge computing node j∈C, the computing time delay is The calculation formula is as follows:

[0106]

[0107] where q i is the number of CPU cycles required by the task, and fj The calculation frequency of the service node j.

[0108] The total time delay t between the task i and the edge computing node j, ignoring the downlink transmission delay of returning the result from the edge to the vehicle. i,j The transmission time delay The calculation time delay The sum of the transmission time delay and the calculation time delay, and the calculation formula is as follows:

[0109]

[0110] The total time delay after increasing the queuing time:

[0111]

[0112] Step S2.2: Building a cost model: In the task offloading allocation, the cost characteristics of the offloading process are considered, that is, the fees paid by the user in the process of transmitting data from the vehicle end to the service node and processing. The cost mainly includes three parts: the calculation resource cost, the network resource cost and the storage resource cost. For the task i∈T and the edge computing node j∈C, the total cost c of the offloading process is as follows: i,j The calculation formula is as follows:

[0113]

[0114] Wherein, q i is the number of CPU cycles required by the task i, is the price of the unit CPU cycle of the service node; s i is the storage resource amount required by the task i, is the price of the unit storage space of the service node; b i is the network bandwidth that the node can provide when the task i and the service node establish communication.

[0115] Step S2.3: Building an energy consumption model: In the task offloading allocation, the energy consumption characteristics of the offloading process are considered, that is, the energy consumption of the user equipment in the process of transmitting data from the vehicle end to the service node and receiving the data result of the service node. The energy consumption mainly considers the uplink process of transmitting data from the vehicle end to the service node. Since the returned result data is very small, the backhaul energy consumption is ignored. For the task i∈T and the edge computing node j∈C, the total energy consumption E of the offloading process is as follows: i,j The calculation formula is as follows:

[0116]

[0117] Wherein, p i is the uplink transmission power of the vehicle sending the task request i, is the uplink transmission time length.

[0118] Total energy consumption E between task i and edge computing node j i,j The calculation formula is as follows:

[0119]

[0120] Step S2.4: Construct personalized dynamic weight coefficients. In actual scenarios, since each user has different emphasis on time delay, cost and energy consumption, the user's satisfaction with the offloading decision cannot be measured by a comprehensive evaluation objective function with fixed weight coefficients for each feature. According to the different preferences of each user for each feature, the feature weight coefficient that meets the user's demand is adopted. The personalized dynamic weight coefficient mainly includes three parts: the weight coefficient of the time delay feature, the weight coefficient of the cost feature and the weight coefficient of the energy consumption feature.

[0121] The weight coefficient of the time delay feature r t The weight coefficient of the cost feature r c And the weight coefficient of the energy consumption feature r E The emphasis of the user on time delay, cost and energy consumption, the calculation formula is as follows:

[0122]

[0123] Among them, w c , w t And w E The demand degree of the user on time delay, cost and energy consumption is calculated only by the user input feature parameter, which is irrelevant to the service node feature. The specific calculation process is as follows:

[0124] (1) For task i∈T, the demand degree of the user task on the time delay feature w t The calculation formula is as follows:

[0125]

[0126] Among them, t Exp t i The time delay requirement of task request i; t ExpMin The minimum value of the time delay requirement in the time delay interval, t ExpMax The maximum value of the time delay requirement in the time delay interval.

[0127] (2) For task i∈T, the demand degree of the user task on the cost feature w c The calculation formula is as follows:

[0128]

[0129] Among them, C Exp The cost expectation value of task request i; C ExpMinC is the minimum value of the cost requirement in the cost interval ExpMax C is the maximum value of the cost requirement in the cost interval. The calculation formula is as follows:

[0130] C Exp = (q + s + b ave ) * p ave

[0131] C ExpMin = (q + s + b ave ) * p min

[0132] C ExpMax = (q + s + b ave ) * p max

[0133] (3) For task i∈T, since the specific energy consumption is also related to the selected node, the user's attention to energy consumption is determined by its own uplink transmission power. The greater the uplink transmission power, the higher the energy consumption of the task in the transmission process. Therefore, w E The calculation formula is as follows:

[0134]

[0135] Where p Exp is p i , the uplink transmission power of the vehicle sending task request i. p ExpMin is the minimum value of the transmission power in the transmission power interval, p ExpMax is the maximum value of the transmission power in the transmission power interval.

[0136] Step S2.5: Construct a personalized satisfaction evaluation function

[0137] In the task offloading allocation process, the goal is to maximize user satisfaction. For a single task i, the corresponding delay, cost and energy consumption of the task on the matched node are calculated, and the personalized weight coefficient is combined to calculate the user's satisfaction with offloading to the node. The satisfaction formula is defined as follows:

[0138] Q = r c * c i,j + r t * t i,j + r E * E i,j

[0139] Step S3: Generate the offline data set of the recommendation system based on the model in step S1 and the personalized satisfaction evaluation function in step S2. The specific steps include:

[0140] Step S3.1: Based on the target model and the personalized satisfaction evaluation function in step S2, the satisfaction value of each allocation scheme is calculated for each task, and the allocation scheme corresponding to the maximum value is taken as the optimal allocation scheme. Each optimal scheme can be represented as (t i i i i i i ), where t i i i i i is the basic information of the task randomly generated in the parameters based on the task model in step S1, and E i is the service node allocated by the optimal scheme obtained according to the target model and the personalized satisfaction evaluation function in step S2.

[0141] For example, when the optimal scheme is selected by the exhaustive method, the specific steps are as follows:

[0142] (1) For the task input generated for a task model, each service node is traversed, and the satisfaction value when the task selects the service node is calculated by the satisfaction function in step S2.

[0143] (2) After all service nodes are calculated, the service node with the maximum satisfaction value is selected as the optimal scheme finally selected by the task.

[0144] Step S3.2: n samples are generated according to the scheme in step S3.1 as an offline dataset that can be used by the recommendation system, and the n allocation schemes and the corresponding original task request contents are summarized as the offline dataset of the recommendation system.

[0145] Step S4: Build a recommendation algorithm model. By selecting a suitable recommendation algorithm, an algorithm model is established.

[0146] The specific steps include:

[0147] Step S4.1: Define the data object of the recommendation algorithm. The data used by the recommendation algorithm generally includes: items, users, and transactions (the relationship between users and items), in this scenario, the user is the vehicle issuing the task request, the item is the service node to be matched, and the transaction is the satisfaction of the vehicle user to the matched service node.

[0148] Step S4.2: Select a suitable recommendation algorithm. The recommendation algorithm mainly includes content-based recommendation, collaborative filtering recommendation, hybrid recommendation, and various ways based on knowledge, community, and context awareness, and a recommendation algorithm that meets the high delay requirement of the Internet of Vehicles application is selected as the recommended method to be adopted.​​​​​​​​​

[0149] For example, a user-based collaborative filtering method in collaborative filtering is selected, which selects the user in the offline data set with the highest similarity to the request user by calculating the similarity between the request user and the users in the offline data set, and the selected service node of the similar user is the recommended selected service node. The similarity between users is calculated as follows:

[0150] (1) Normalize each parameter in its interval to avoid some parameter values being too large and affecting the similarity result too much.

[0151] (2) Calculate the similarity of task i and each task in the offline database using the Pearson correlation coefficient.

[0152]

[0153] (3) Select the service node selected by the task with the largest Pearson correlation coefficient value as the recommended matching service node.

[0154] Step S4.3: Adapt the recommendation algorithm to the model, i.e. the input of the algorithm is the task request model issued by the vehicle, and the recommendation result is obtained by training the offline data set obtained in step S3, which is the output of the algorithm.

[0155] Step S5: Discretize time into time slots, and the system receives all user task requests in a time slot as a set of task request inputs at time slot τ. The recommendation algorithm model in step S4 is used to recommend the optimal service node for user allocation.

[0156] On the other hand, a specific embodiment of a system applying the above method is provided, including:

[0157] Task generation module: simulate real task requests and randomly generate multiple sets of tasks within the range of real task request data as data set task samples.

[0158] Service node information collection module: collect all service node information within the communication range and match it as a service node model sent to the offline data set module to provide service node information for offline data set calculation.

[0159] Data set generation module: In this module, the optimal matching method is generated according to the task request model and the service node model obtained based on the service node information collection module and the satisfaction evaluation function, and multiple sets of calculation results are stored as the offline data set required by the recommendation module.

[0160] Task request collection module: In time slot t, collect all concurrent tasks within the communication range in the actual scene, and send a set of task requests to the recommendation module.

[0161] The recommendation module: based on the task request information collected by the task request collection module, and the information of the offline data set module, the recommended matching service node is calculated for the request through the recommendation algorithm model. And send the result to the user.

[0162] In summary, the construction of the personalized satisfaction evaluation function of task offloading allocation in step S2. According to the input parameters of the user task, the user's preference degree for some target characteristics is reflected, and the weight coefficient is dynamically changed with different characteristics of each user, which can reflect the actual demand and personalized preference of each user for the service node. And take it as a coefficient to combine the target value to calculate the user's satisfaction degree to the allocated service node, which realizes the measurement index of service quality from the user's point of view.

[0163] The construction of the recommendation algorithm offline data set in step S3 of task offloading allocation. Since there is a lack of actual data statistics in the field of task offloading, the invention constructs the offline data set required by the recommendation algorithm through task model generation and optimal scheme selection. This data set can be used as a personalized offline data set for the recommendation algorithm by selecting the service node with the optimal satisfaction degree, which can ensure the user's satisfaction during the offloading process.

[0164] The construction of the recommendation system in step S4 of task offloading allocation. Through the construction of algorithm model, combined with the offline data set in step S3, a complete personalized car-road cloud collaborative task offloading allocation method and system based on recommendation algorithm is constructed. It can completely realize the function of task offloading allocation recommendation for user task, and provides a personalized task offloading allocation method with the goal of maximizing user satisfaction.

[0165] The above is only the preferred embodiment of the present application, the protection scope of the present application is not limited to the above-mentioned examples, all technical solutions belonging to the idea of the present application are within the protection scope of the present application. It should be noted that for ordinary skilled in the art, some improvements and decorations without departing from the principles of the present application, these improvements and decorations should also be considered as the protection scope of the present application.

Claims

1. A personalized vehicle-road-cloud collaborative task offloading and allocation method, characterized in that: Includes the following steps: Step S1: Construct a vehicle-road-cloud collaborative task offloading system model; Step S2: Construct the vehicle-road-cloud collaborative task offloading target model and personalized satisfaction evaluation function; Step S3: Generate the offline dataset for the recommender system based on the model and personalized satisfaction evaluation function from steps S1 and S2; Step S4: Construct a recommendation algorithm model so that it can obtain the recommendation allocation scheme from the offline dataset in step S3; Step S5: Discretize time into time slot combinations. At time slot τ, the system receives all user task requests within that time slot as a set of task request inputs. Then, it uses the recommendation algorithm model from step S4 to recommend the optimal service node assigned to the user. The construction of the vehicle-road-cloud collaborative task offloading system model in step S1 specifically includes the construction of the task model and service node model in the vehicle-road-cloud collaborative task offloading system. The task model is as follows: At time slot t, task request i includes the following detailed task information (t i ,q i ,s i ,d i ,p i ); Among them, t i The time delay requirement for task request i is that the task must meet its deadline t. i completed within; q i s is the number of CPU cycles required for task request i. i For the storage resource requirements of task request i, d i For the size of the data packet for task i, p i To determine the uplink transmission power of the vehicle sending task request i, at time slot t, tasks from multiple intelligent connected vehicles are combined into a task set T. The service node model is as follows: At time slot t, for the j-th service node, its available resource information is obtained using... express; Where f represents the computation frequency of service node j; b is the bandwidth of service node j; The computing resource price, storage resource price, and network resource price for service node j; the specific steps in step S2 for constructing the vehicle-road-cloud collaborative task offloading target model and personalized satisfaction evaluation function are as follows: Step S2.1: Calculate transmission delay and construct a delay model; Step S2.2: Construct a cost model by calculating resource costs, network resource costs, and storage resource costs. The constructed cost model is as follows: For task i∈T and edge computing node j∈C, the total cost of its unloading process is c. i,j The calculation formula is as follows: Where, q i Let i be the number of CPU cycles required for task i. The price per CPU cycle for a service node; s i The amount of storage resources required for task i. The price per unit of storage space for a service node; b i When task i establishes communication with the service node, the network bandwidth that the node can provide; Step S2.3: During the process of transmitting data from the vehicle to the service node and receiving data results from the service node through the task, the user equipment's energy consumption is used to construct an energy consumption model. The constructed energy consumption model is as follows: For task i∈T and edge computing node j∈C, the total energy consumption of its unloading process is E. i,j The calculation formula is as follows: Where, p i The uplink transmission power of the vehicle sending task request i, This refers to the uplink transmission duration; Therefore, the total energy consumption E between task i and edge computing node j i,j The calculation formula is as follows: Step S2.4: Construct personalized dynamic weighting coefficients by using the weighting coefficients of latency characteristics, cost characteristics, and energy consumption characteristics; Step S2.5: Combine the personalized weighting coefficients to calculate the user's satisfaction with uninstalling to this node and construct a personalized satisfaction evaluation function. The satisfaction formula is defined as follows: Q=r c ×c i,j +r t ×t i,j +r E ×E i,j 。 2. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 1, characterized in that: The specific steps for constructing the delay model in step S2.1 are as follows: Step S2.11: First, calculate the transmission delay. The transmission delay refers to the time required for task data to be transmitted from the vehicle to the service node via the network. For task i∈T and edge computing node j∈C, the transmission delay is... The calculation formula is as follows: Where, d i b is the size of the transmitted data packet. j The network bandwidth between user i and service node j; Step S2.12, next calculate the computation latency. Computation latency refers to the time required for a task to be computed and processed on the service node. For task i∈T and edge computing node j∈C, its computation latency is... The calculation formula is as follows: Where, q i f is the number of CPU cycles required for the task. j The computation frequency of service node j; Step S2.13: Finally, calculate the total latency. Since the downlink transmission delay from the edge return result to the vehicle is ignored, the total latency t between task i and edge computing node j is... i,j For transmission delay Computation delay The sum is calculated using the following formula: Total delay after adding queuing time:

3. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 2, characterized in that: The calculation formula for constructing the personalized dynamic weight coefficient in step S2.4 is as follows: Among them, the weighting coefficient r of the time delay feature t Weighting coefficient r of cost characteristics c The weighting coefficient r of energy consumption characteristics E w c w t and w E These represent the user's requirements for latency, cost, and energy consumption, calculated solely from user-input feature parameters and independent of service node characteristics. c w t and w E The calculation process is as follows: For task i∈T, the degree of user task requirement for latency features w t The calculation formula is as follows: Among them, t Exp That is, t i t represents the latency requirement for task request i; ExpMin t represents the minimum delay requirement within the delay interval. ExpMax This represents the maximum delay requirement within the delay interval; For task i∈T, the degree of user task demand for cost features w c The calculation formula is as follows: Among them, C Exp C represents the expected cost of task request i; ExpMin C represents the minimum cost requirement within the cost range. ExpMax The maximum cost requirement within the cost range is calculated using the following formula: C Exp =(q+s+b ave )*p ave C ExpMin =(q+s+b ave )*p min C ExpMax =(q+s+b ave )*p max For task i∈T, w represents the degree of energy consumption requirement of the user task during transmission. E The calculation formula is as follows: Where, p Exp That is, p i , p is the uplink transmission power of the vehicle sending task request i. ExpMin p represents the minimum transmit power within the transmit power range. ExpMax This represents the maximum transmit power within the transmit power range.

4. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 1 or 2, characterized in that: The specific steps in step S3 for generating the offline dataset of the recommender system based on the model and personalized satisfaction evaluation function in steps S1 and S2 are as follows: Step S3.1: Based on the target model and personalized satisfaction evaluation function in Step S2, calculate the satisfaction value for each allocation scheme for each task, and take the allocation scheme corresponding to the largest value as the optimal allocation scheme. Each optimal scheme can be expressed as (t i ,q i ,s i ,d i ,p i E i ), where t i ,q i ,s i ,d i ,p i Based on the basic information of the task randomly generated within the parameters of the task model in step S1, E i The service node is assigned to the optimal solution obtained from the target model and personalized satisfaction evaluation function in step S2 for this task; Step S3.2: Based on the solution in step S3.1, n samples are generated as an offline dataset that the recommendation system can use. These n allocation schemes and their corresponding original task request content are summarized as the offline dataset of the recommendation system.

5. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 4, characterized in that: The steps for selecting the optimal allocation scheme in step S3.1 are as follows: Step S3.11: For a task input generated by a task model, traverse each service node and calculate the satisfaction value when the task selects the service node using the satisfaction function in step S2. Step S3.12: After all service nodes have been calculated, select the service node with the highest satisfaction value as the optimal solution for the task.

6. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 1 or 2, characterized in that: The specific steps for constructing the recommendation algorithm model in step S4 are as follows: Step S4.1: Define the data object for the recommendation algorithm. The data used by the recommendation algorithm generally includes: items, users and transactions. In this scenario, the user is the vehicle that issues the task request, the item is the service node to be matched, and the transaction is the satisfaction of the vehicle user with the matched service node. Step S4.2: Select a suitable recommendation algorithm. The recommendation algorithms mainly include content-based recommendation, collaborative filtering recommendation, hybrid recommendation, as well as knowledge-based, community-based, and context-aware recommendation. Select the recommendation algorithm that meets the latency requirements of the Internet of Vehicles application as the recommendation method to be adopted. Step S4.3: Adapt the recommendation algorithm into the model. That is, the input of the algorithm is the task request model issued by the vehicle. The recommendation result is obtained by training the model based on the offline dataset obtained in step S3. The recommendation result is the output of the algorithm.

7. The personalized vehicle-road-cloud collaborative task offloading and allocation method according to claim 6, characterized in that: The algorithm selected in step S4.2 is a user-based collaborative filtering method. By calculating the similarity between the requesting user and users in the offline dataset, the user with the highest similarity to the requesting user in the offline dataset is selected. The service node selected by the similar user is the recommended service node. The specific steps for calculating the user similarity are as follows: Step S4.21: Normalize each parameter within its interval to avoid some parameters having large values ​​and having too much impact on the similarity results; Step S4.22: Calculate the similarity between task i and each task in the offline database using the Pearson correlation coefficient; Step S4.23: Select the service node chosen by the task with the largest Pearson correlation coefficient as the recommended matching service node.

8. A system applying the personalized vehicle-road-cloud collaborative task offloading and allocation method according to any one of claims 1 to 7, characterized in that: include: Task generation module: Simulates real task requests and randomly generates multiple sets of tasks as dataset task samples within the data range of real task requests; Service Node Information Collection Module: Collects information on all service nodes within the communication range and matches them into a service node model, which is then sent to the offline dataset module to provide service node information for the calculation of the offline dataset. Dataset generation module: This module generates the optimal matching method based on the task request model, the service node model obtained from the service node information collection module, and the satisfaction evaluation function, and stores multiple sets of calculation results as the offline dataset required by the recommendation module. Task request collection module: Within time slot t, collect all concurrent tasks within the communication range of the actual scenario, and send a set of task requests to the recommendation module; Recommendation module: Based on the task request information obtained by the task request collection module and the information from the offline dataset module, the recommendation algorithm model calculates the recommended matching service nodes for the request and sends the results to the user.

Images