Power and cache joint optimization method in space-air-ground fusion vehicle-mounted network scene

Abstract

The invention provides a power and cache joint optimization method in a space-air-ground fusion vehicle-mounted network scene, and relates to the field of edge computing, and the method comprises the steps: S1, building a communication resource library; s2, taking task calculation, caching and power distribution as optimization variables, taking task execution time delay, power consumption and caching efficiency as objective functions, and establishing a general model; s3, solving the general model based on an artificial intelligence algorithm; s4, performing simulation calculation based on a scheme obtained by solving the general model to obtain evaluation data; s5, correcting parameters of the general model according to the evaluation data; and S6, solving the model after parameter optimization, and obtaining a final dynamic optimization power distribution and intelligent cache management scheme. Therefore, a more efficient and more flexible power and cache strategy is realized, the stability and reliability of the space-air-ground fusion vehicle networking system are improved, the energy consumption of the system is reduced, and the resource utilization rate is improved.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to a method for jointly optimizing power and cache in an air-ground-space integrated vehicle network scenario. Background Art

[0002] Air-space-ground integration technology is an important direction for the development of 6G networks. With the rapid development of Internet of Things (IoT) technology, Internet of Vehicles (IoV) is gradually becoming an important part of the intelligent transportation system. Through the information exchange between vehicles and between vehicles and infrastructure, IoV can not only provide an efficient and comfortable driving experience, but also greatly promote the diversified development of transportation services. However, with the continuous emergence of new applications such as in-vehicle multimedia, cruise systems and autonomous driving, vehicles generate a large amount of data during operation, which puts higher requirements on the computing power and processing latency of IoV. In the traditional cloud-edge-end collaborative IoV architecture, the location of the mobile edge computing (MEC) server is usually fixed, which makes it difficult to flexibly respond to the dynamic needs of vehicle users. In recent years, the rapid development of unmanned aerial vehicles (UAV) technology has brought new opportunities for IoV. The integrated air-ground-space vehicle network architecture that combines drones with ground communication networks can not only expand the communication coverage, but also flexibly adjust the deployment of computing resources according to the mobility characteristics of vehicle users, thereby providing more efficient and reliable services; at the same time, the rapid development of low-orbit satellite (LEO) technology has provided a wider coverage capability for the vehicle network. LEO has the advantage of seamless global connection and can solve the coverage blind spot problem of traditional ground networks in remote areas. Under this architecture, task offloading strategies and caching technologies have become the key to improving system performance.

[0003] Task offloading can reduce the computing load and energy consumption of vehicles by reasonably allocating computing tasks to MEC servers or UAVs; while caching technology can reduce network congestion and latency by storing hot data. Dynamic power adjustment and caching strategies are effective ways to reduce system energy consumption and task offloading latency. Domestic and foreign scholars have conducted systematic research on reducing system energy consumption and task offloading latency in air-space-ground vehicle networks. The proposed solutions include: a scheme for collecting task optimization strategies, an algorithm that integrates the gray wolf algorithm and particle swarms, a hybrid task offloading scheme, and an improved soft actor-critic framework, which optimizes the system's collaborative offloading strategy. The above schemes mostly design their own task offloading strategies from a single dimension, such as shortening the system response latency or reducing the energy consumption of mobile terminals. Although these solutions are effective in specific scenarios, they are difficult to fully respond to the dynamic needs of air-space-ground integrated vehicle networks. For example, the traditional power allocation mechanism fails to fully consider the endurance requirements of aerial nodes (such as UAVs and LEOs), resulting in high energy consumption during frequent task offloading, which in turn affects the overall system performance. With the increasing complexity and diversification of vehicle networking application scenarios, existing technologies are unable to meet the demand for more efficient and reliable services. The specific situation is as follows: 1. The traditional static power allocation strategy does not fully consider the dynamic endurance requirements of air nodes, resulting in increased system energy consumption and limited stability of air nodes under high load or frequent task scenarios; 2. Existing cache decision strategies are mostly concentrated in fixed hot spots, which are difficult to adapt to the diverse and random dynamic data requirements in the vehicle network, and the resource utilization rate is low; 3. Existing power and cache optimization solutions rely on a single algorithm, mainly starting from reducing task response delay or energy consumption, lacking comprehensive consideration of multi-dimensional factors in the air-ground fusion environment, and it is difficult to provide flexible and efficient task offloading and cache support, and fail to achieve true resource optimization configuration. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art or related technology, and discloses a method for jointly optimizing power and cache in an air-space-ground-integrated vehicle network scenario, thereby achieving more efficient and flexible power and cache strategies, improving the stability and reliability of the air-space-ground-integrated vehicle network system, reducing system energy consumption, and improving resource utilization.

[0005] The first aspect of the present invention discloses a power and cache joint optimization method in a space-ground-integrated vehicle network scenario, including: S1, using a unified heterogeneous network resource feature description and expression method to integrate the communication resources of satellite-ground links, air-ground links and ground links into the same vector communication resource library; S2, modeling and analyzing task offloading, task calculation and cache resource allocation according to task type, server computing power and cache rules; taking task calculation, cache and power allocation as optimization variables, taking task execution delay, power consumption and cache efficiency as objective functions, and establishing a dynamic optimization power allocation and intelligent cache system suitable for space-ground-integrated vehicle networks. Management general model; S3, based on the vector communication resource library, use artificial intelligence methods to solve the general model of dynamic optimization power allocation and intelligent cache management, and obtain the optimal dynamic optimization power allocation and intelligent cache management solution; S4, simulate and calculate the obtained dynamic optimization power allocation and intelligent cache management solution to obtain evaluation data; S5, according to the evaluation data, correct the parameters of the general model of dynamic optimization power allocation and intelligent cache management to obtain the dynamic optimization power allocation and intelligent cache management model; S6, solve the dynamic optimization power allocation and intelligent cache management model to obtain the final dynamic optimization power allocation and intelligent cache management solution.

[0006] In this technical solution, the energy consumption characteristics of different nodes in the air-ground-integrated vehicle network are given priority. The power is precisely controlled according to the dynamic endurance requirements of aerial nodes such as UAV and LEO and the mobility of vehicle nodes, and efficient and energy-saving power adjustment is achieved to extend the endurance of aerial nodes and reduce the overall energy consumption of the system. In addition, based on the real-time location and data request characteristics of vehicle nodes, the cache resources in the Internet of Vehicles are dynamically managed, hot spot data is cached first, and the cache content is intelligently adjusted, thereby reducing the response delay of task offloading and network congestion, and significantly improving the overall data transmission efficiency of the system.

[0007] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the unified heterogeneous network resource feature description and expression method specifically includes: calculating the data transmission rate of wireless communication between the drone, edge server and user terminal; calculating the data transmission rate of satellite communication; according to the program size of each task, calculating the task transmission delay: task transmission delay of wireless communication = program size / data transmission rate of wireless communication; task transmission delay of satellite communication = program size / data transmission rate of satellite communication.

[0008] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, wireless communication is applied to air-ground links and ground links, and the method for calculating the data transmission rate of wireless communication specifically includes:

[0009] The data transmission rate of wireless communication between drones, edge servers, and user terminals is calculated using Shannon’s theorem:

[0010]

[0011] Among them, i and j represent the sender and receiver respectively. represents the transmission rate between i and j; B represents the channel bandwidth; P i represents the transmission power of i; w i represents the Gaussian noise power in the channel; H ij represents the channel gain between devices i and j;

[0012] Satellite communication is applied to satellite-to-ground links. The calculation method of the data transmission rate of satellite communication specifically includes:

[0013] Visible light communication is used, and the transmission rate is R space It is expressed as:

[0014] R space =FR×AS×Or

[0015] Among them, FR represents the frame rate, AS represents the antenna array size, and Or represents the modulation order.

[0016] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the steps of modeling and analyzing task offloading, task calculation and cache resource allocation based on task type, server computing capability and cache rules specifically include:

[0017] Assume that the set of all types of subtasks is S ALL ; The task created by terminal k is represented by S k = {SSET,t max}, where t max represents the maximum tolerable delay of the task, SSET represents a task set consisting of M non-cacheable subtasks and N cacheable subtasks, Each subtask S in the offloading task * It is expressed as:

[0018] S * ={TZ, TD, TL}

[0019] Among them, TZ is the amount of computing resources required by the subtask, TD is the size of storage resources required to load the subtask, and TL is the program size of the subtask;

[0020] When the subtask is calculated, the cacheable subtask is placed in the cache, or the subtask is removed from the cache, so that the storage resources of the cached task do not exceed the total storage resources of the server;

[0021] The execution time of a single subtask i on the server is expressed as: ST i =Z i ×f cpu , where f cpu Indicates the computing frequency of the server CPU, Z i The computational load of the task of offloading the vehicle terminal is multiple subtasks S * The sum of the computing resources, namely Z i =∑TZ;

[0022] The computational delay of task offloading to drone nodes UAV Calculate according to the following formula:

[0023]

[0024] in, represents the average power of the drone node during operation, C represents the capacitance of the transistors that make up the CPU, and V dd represents the steady-state voltage within the transistor;

[0025] The computational delay of task offloading to satellite nodes is ST Space Calculate according to the following formula:

[0026]

[0027] in, Represents the average power of the satellite node during operation.

[0028] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed in the present invention, preferably, the steps of establishing a general model of dynamic optimization power allocation and intelligent cache management applicable to the air-ground-integrated vehicle network specifically include:

[0029] For the unloaded task i, the normalized value of the unloading delay, the normalized value of the energy consumption during task unloading, the inverse of the average cache hit rate during task unloading, and the average cache utilization during task unloading are added and optimized to the minimum, which is expressed as a multi-objective joint optimization equation:

[0030]

[0031] st

[0032] C1:0≤T i ≤T max

[0033] C2:0≤E i ≤E max

[0034] C3:φ∈[0,1]

[0035] C4:

[0036] C5:0≤p≤P k

[0037] C6:TZ i ≥0

[0038] C7:0≤TD i ≤D k

[0039] Among them, T i represents the delay of offloading task i, T max represents the upper limit of the single task offloading delay, T i represents the energy consumption of offloading task i, T max represents the upper limit of energy consumption of a single server, φ is the cache hit rate, is the cache utilization, st means subject to constraints C1, C2, C3, C4, C5, C6 and C7; constraint C5 means that the real-time power consumption p of the server cannot exceed the maximum power P of the server k , k∈{m,a,s}, m represents the computing power of the edge server, a represents the computing power of the drone server, and s represents the computing power of the low-orbit satellite server; constraint C6 represents: the computing resources TZ required for task i i is a non-negative value; constraint C7 indicates that the cache resource TD of task i i Smaller than the storage resource D owned by the computing node.

[0040] …

[0041] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the modeling process of task execution delay specifically includes:

[0042] In the space-ground integrated vehicle network, the unloading delay T of task i i It is expressed as:

[0043]

[0044] in, It is represented as the transmission delay of task i uploaded to the server; represents the waiting delay of task i; represents the return delay of the calculation result of task i;

[0045] After the terminal offloads the computing task to the computing node, if the computing node has an ongoing computing task, the task is added to the queue.

[0046] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the power consumption modeling process specifically includes:

[0047] The energy consumption of a single computing node is expressed as the sum of the energy consumption of the cache and the energy consumption of the CPU when all tasks are executed:

[0048] E=E memory +∑E calculate +∑E transmit

[0049] Among them, E memory It is represented as the energy consumption of cache during operation, E calculate Indicates the energy consumption of the task during calculation, E transmit represents the transmission energy consumption;

[0050] E transmit =P trans ·T

[0051] P trans represents the transmit power of the computing node, and T represents time;

[0052]

[0053] D(t) represents the function of cache occupancy changing over time, D k represents the storage resources owned by the computing node, U represents a drone, S represents a low-orbit satellite, and P represents 0 Indicates the maximum power of the computing node on storage resources;

[0054]

[0055] P(t)=P k L(t),k∈{U,S}

[0056] P k It represents the computing power of the server. L(t) is a function that varies between 0 and 1, which represents the CPU utilization rate. When L(t) = 1, it means that the CPU is running at full load. When L(t) = 0, it means that the CPU is idle.

[0057] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the cache efficiency modeling process specifically includes:

[0058] Define the cache hit rate φ as:

[0059]

[0060] Among them, count() represents the aggregation function of the number of elements in the statistical set, SNN represents the set of cacheable subtasks; SNN * Represents the set of subtasks that are cache hits when executing a task;

[0061] Defining cache utilization for:

[0062]

[0063] in, Represents the average storage usage of the i-th computing node.

[0064] According to the power and cache joint optimization method in the air-ground-integrated vehicle network scenario disclosed by the present invention, preferably, the steps of using an artificial intelligence method to solve the general model of dynamic optimization of power allocation and intelligent cache management specifically include:

[0065] Based on the general model of dynamic optimization of power allocation and intelligent cache management, a deep reinforcement assisted learning algorithm is used to solve the probability of offloading tasks to each server, the probability of various power adjustment operations of the server, the server's cache operations on each task, and the operation sequence of releasing tasks in the cache.

[0066] According to the power and cache joint optimization method in the air-space-ground integrated vehicle network scenario disclosed by the present invention, preferably, the air-space-ground integrated vehicle network includes the following communication execution entities: user terminal, edge server, drone and low-orbit satellite, and the server can be set up in the drone or the low-orbit satellite.

[0067] The beneficial effects of the present invention include at least: the present invention provides a method for dynamically optimizing power allocation and intelligent cache management in the scenario of space-ground-integrated vehicle network. From the perspective of reducing system energy consumption and latency, it considers the problems of server dynamic power adjustment optimization and cache intelligent management, and provides strong support for the development of green communications; compared with the deep reinforcement learning scheme based on multi-task and multiple neural networks, the present invention is based on the method of deep reinforcement assisted learning, which can more coordinately reduce system latency and system energy consumption. Specifically, the present invention first considers the task type and server computing power, takes these factors as input data, and analyzes and optimizes them through a deep reinforcement learning model; then, based on real-time task requests and network status, an intelligent cache management strategy is used to dynamically adjust the cache content to reduce task execution latency and energy consumption; finally, through simulation verification and correction of model parameters, continuous optimization of system performance is achieved. Through the present invention, the problem of lack of collaborative optimization of power allocation and cache management in the prior art can be effectively solved, and the system's responsiveness and resource utilization efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A schematic diagram of the system framework of a method for joint optimization of power and cache in an air-ground-space integrated vehicle network scenario according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0069] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the limitations of the specific embodiments disclosed below.

[0070] An embodiment of the present invention proposes a power and cache joint optimization method in a space-ground integrated vehicle network scenario. The system framework of this method in an actual application scenario is as follows: Figure 1 As shown: In the air-space-ground fusion vehicle network, the vehicle collection will generate a large number of tasks that need to be processed. These tasks will be unloaded to the server for calculation through wireless communication links or satellite communication links. The power and cache joint optimization method proposed in the present invention includes a power adjustment strategy and a cache management strategy. The power and cache joint optimization model is solved through a multi-task deep reinforcement assisted learning algorithm to obtain a dynamic optimization power allocation and intelligent cache management solution, thereby reducing system energy consumption and improving resource utilization. The power and cache joint optimization method in the air-space-ground fusion vehicle network scenario disclosed in this embodiment includes:

[0071] (1) Using a unified method for describing and expressing the characteristics of heterogeneous network resources, the communication resources of various wireless communication networks can be aggregated into a vector communication resource library. Cognitive radio technology can be used to perceive the communication resource usage status of various wireless communication networks in real time and describe and express the perceived information.

[0072] (2) Model and analyze task offloading, computing, and cache resource allocation based on task type, server computing power, and cache rules. By constructing a mathematical model for dynamic optimization of power allocation and intelligent cache management, the task's computing, caching, and power allocation are used as optimization variables, and the task execution delay, power consumption, and cache efficiency are used as objective functions. Through the above modeling steps, a general model for dynamic optimization of power allocation and intelligent cache management suitable for space-ground integrated vehicle networks is established;

[0073] (3) Use artificial intelligence methods (deep reinforcement assisted learning) to solve the general model of dynamic optimization of power allocation and intelligent cache management to obtain the optimal dynamic optimization of power allocation and intelligent cache management solution;

[0074] (4) in a simulation environment, verify and evaluate the obtained dynamic optimization power allocation and intelligent cache management scheme, obtain evaluation data, modify relevant parameters of the dynamic optimization power allocation and intelligent cache management model, and obtain the dynamic optimization power allocation and intelligent cache management model;

[0075] (5) Solve the dynamic optimization power allocation and intelligent cache management model to obtain the final dynamic optimization power allocation and intelligent cache management solution.

[0076] According to the above embodiment, preferably, a unified heterogeneous network resource feature description and expression method is used in step (1) to integrate the communication resources of satellite-to-ground links, air-to-ground links and ground links, and build a unified vectorized communication resource library. Corresponding mathematical models are established for different link types to describe link characteristics and communication resource distribution. Cognitive radio technology is used to perceive the resource usage of each wireless communication network in real time, and the perceived key parameters are uniformly characterized to provide accurate and real-time resource description for dynamic optimization of power allocation and intelligent cache management. The unified heterogeneous network resource feature description and expression method specifically includes the following steps:

[0077] (1a) Calculate the data transmission rate of the wireless communication link between the drone, edge server, and user terminal using Shannon's Theorem Denoted as:

[0078]

[0079] Where i and j represent the sender and receiver respectively. represents the transmission rate between i and j; B represents the channel bandwidth; P i represents the transmission power of i; w i represents the Gaussian noise power in the channel; H ij represents the channel gain between devices i and j, H ij Expressed as l ij represents the transmission distance between i and j, and α represents the transmission path attenuation factor. ij It can be expressed as:

[0080]

[0081] (1b) In satellite communication links, visible light communication is mainly used, so the transmission rate R in the satellite-to-ground link is space It can be expressed as:

[0082] R space =FR×AS×Or

[0083] Where FR represents the frame rate, AS represents the antenna array size, and Or represents the modulation order.

[0084] (1c) The transmission delay of the task is abstracted as follows:

[0085]

[0086] Where L represents the program size of each task.

[0087] According to the above embodiment, preferably, in step (2), "modeling and analyzing task offloading, computing and cache resource allocation based on task type, server computing capability and cache rules" specifically includes the following steps:

[0088] (2a) Modeling the computing task, assuming that the set of all types of subtasks is represented by S ALL ; The task created by terminal k is represented by S k = {SSET,t max}. Where t max It is represented as the maximum tolerable delay of the task; SSET is represented as a task set consisting of M non-cacheable subtasks and N cacheable subtasks. ALL The relationship is expressed as: Where S NO Indicates non-cacheable subtasks, that is, these tasks cannot be cached in memory, such as real-time calculation results in the program or operations that traverse real-time updated data; S NO,i , i≤M represents the i-th non-cacheable subtask. S j ,j≤N represents the jth cacheable subtask. This part of the task can be cached in the memory of the computing node. When the server executes the task, if the task is cached, it only needs to read the calculation result from the cache. * It can be expressed as:

[0089] S * ={TZ,TD,TL},*∈{NO,1,2......}

[0090] Where TZ is the amount of computing resources required by the subtask (in CPU cycles), TD is the size of the storage resources required to load the subtask, and TL is the program size of the subtask;

[0091] (2b) When the subtask is calculated, if the subtask is a cacheable subtask, the system will decide whether to put the task into the cache; at the same time, the system will also decide whether to remove some tasks from the cache to ensure that the storage resources of the cached tasks do not exceed the total storage resources of the server;

[0092] (2c) Assume that the power of a single transistor is P g =f×C×V dd , where f is the frequency, C is the capacitance, V dd Represents the steady-state voltage in the transistor. The CPU of a computer can be regarded as a component composed of several transistors, so the power P of the CPU is c It can be expressed as:

[0093] P c =∑f×C×V dd

[0094] =f cpu ×∑C×V dd

[0095]

[0096] In the above formula, f cpu represents the computing frequency of the server CPU. The execution time of a single subtask i on the server can be expressed as: ST i =Z i ×f cpu , where Z i The computational effort of the vehicle terminal unloading task is the sum of all subtasks S under this task. * The sum of the computing resources, namely Z i =∑TZ. Therefore, the computing time of the task offloaded to the drone node can be expressed as:

[0097]

[0098]

[0099] in Represents the average power of the drone node during the operation. Since the power changes over time in the air-ground fusion vehicle network model, the calculation delay can be approximately equal to the average power during the operation multiplied by the ratio of the total required computing resources to the total transistor charge;

[0100] Similarly, the computational delay of the space node low-orbit satellite can be expressed as:

[0101]

[0102] in represents the average power of the satellite node during the operation;

[0103] According to the above embodiment, preferably, in step (2), "taking the calculation, cache and power allocation of the task as optimization variables, and taking the task execution delay, power consumption and cache efficiency as the objective function" specifically includes the following steps:

[0104] (2d) The delay model in the space-ground-vehicle network includes task transmission time, task waiting time, computing time, and task return time. These factors together determine the delay performance of the entire task offloading process. Therefore, the offloading delay T of task i i It can be expressed as:

[0105]

[0106] in It is represented as the transmission delay of task i uploaded to the server; represents the waiting delay of task i; represents the return delay of the calculation result of task i. Assuming that the transmission time of upload and download is the same, that is, Therefore, the simplified task offloading delay can be expressed as:

[0107]

[0108] After the terminal unloads the computing task to the computing node, if the computing node has an ongoing computing task, the task is added to the execution queue. Assuming that the position of the task in the execution queue is the mth, then It can be expressed as:

[0109]

[0110] (2e) When modeling energy consumption, multiple factors are taken into account, including the power consumption of the computer CPU and the power consumption of the memory. Therefore, the energy consumption of a single computing node can be expressed as the sum of the energy consumption of the cache and the energy consumption of the CPU when all tasks are executed, as shown below:

[0111] E=E memory +∑E calculate +∑E transmit

[0112] Where E transmit represents the transmission energy consumption, which can be expanded to E transmit =P trans ·T, P trans represents the transmit power of the computing node, T represents time; E memory Expressed as the energy consumption of the cache during operation, it can be expanded as:

[0113]

[0114] Where D(t) represents the function of cache occupancy changing over time, D k Indicates the storage resources owned by the computing node, U represents a drone, and S represents a low-orbit satellite. 0Indicates the maximum power of the computing node on the storage resources. calculate It represents the energy consumption of the task during calculation, which can be expressed as: Where P(t) = P k L(t),k∈{U,S},P k Indicates the computing power of the CPU (server) of the drone or low-orbit satellite. L(t) is a function that varies between 0 and 1, indicating the CPU utilization rate. When L(t) = 1, it means that the CPU is fully loaded, and when L(t) = 0, it means that the CPU is idle; t 0 Indicates the start time of calculation; t 0 +ST indicates the end time of the calculation.

[0115] (2f) As a key mechanism to improve system performance and reduce repeated calculations and data transmission overhead, cache plays a vital role in the complex and dynamic environment of the Internet of Vehicles. Especially in the integrated air-ground-space architecture, cache technology has become the key to optimizing the overall system energy consumption and improving user experience. The cache hit rate refers to the execution of all cacheable subtasks in the system. It calculates the proportion of subtasks in the system that are cached, reflecting the impact of cache on task processing efficiency. The cache hit rate φ can be expressed as: Where count() represents the aggregation function of the number of elements in the statistical set; SNN represents the set of cacheable subtasks generated in the system; SNN * Represents the set of subtasks that are cached when executing tasks in the system;

[0116] (2g) Cache utilization is also an important indicator for measuring the performance of the cache system. It indicates the occupancy of cache resources by all computing nodes in the system during task execution, reflecting the effective use of cache resources. Optimizing cache utilization is not only a key strategy to improve the performance of the integrated air-ground-ground vehicle network system, but also an important way to achieve energy consumption optimization. It can be expressed as:

[0117]

[0118] in represents the average storage occupancy value of the i-th computing node;

[0119] (2h) The goal of this model is to maximize the processing capacity of the space-ground-integrated vehicle network, which means minimizing the task processing delay and system energy consumption. Considering the offloading decision, communication delay, server energy consumption, and cache strategy, the vehicle task offloading problem in the space-ground-integrated vehicle network can be modeled as a multi-objective joint optimization problem, expressed as:

[0120]

[0121] st

[0122] C1:0≤T i ≤T max

[0123] C2:0≤E i ≤E max

[0124] C3:φ∈[0,1]

[0125] C4:

[0126] C5:0≤p≤P k

[0127] C6:TZ i ≥0

[0128] C7:0≤TD i ≤D k

[0129] This is a minimization problem optimization model, which means: for the unloaded task i, the normalized value of its unloading delay, the normalized value of the energy consumption generated during task unloading, the inverse of the average cache hit rate during task unloading, and the average cache utilization during task unloading are added together to optimize it to the minimum.

[0130] T max It is a hyperparameter, which indicates the upper limit of the offloading delay of a single task in the system;

[0131] E max It represents the upper limit of energy consumption of a single server within 20 time slots in the system.

[0132] Constraints C1 and C2 indicate that the unloading delay and energy consumption in the model are not negative;

[0133] C3 and C4 represent the cache hit rate φ and cache utilization, respectively. Its value range is between 0 and 1;

[0134] Constraint C5 indicates that the real-time power consumption of the server cannot exceed its maximum power;

[0135] Constraint C6 ensures that the computing resources required by the task are non-negative;

[0136] Constraint C7 stipulates that the cache resources of the task should be smaller than the total cache resources of the server.

[0137] According to the above embodiment, preferably, in step (3), "using an artificial intelligence method (deep reinforcement assisted learning) to solve a general model of dynamic optimization of power allocation and intelligent cache management to obtain an optimal dynamic optimization of power allocation and intelligent cache management solution" specifically includes the following steps:

[0138] (3a) In the model proposed in this scheme, it is assumed that each task consists of at most K subtasks; there are N computing servers in the environment that can be used for task offloading; there are M types of subtasks, then the state space State in the environment can be expressed as:

[0139]

[0140] Among them, task x 、task y 、task z Indicates the location coordinates of the vehicle when the task is created, Z 1 ,…,Z K Indicates the computing resources required for each subtask of the created task, D 1 ,…,D K Indicates the cache resources required when each subtask of the creation task is cached, P 1 ,…,P N Indicates the power of each server at the current moment, from POSX 1 ,POSY 1 ,POSZ 1 to POSX N ,POSY N ,POSZ N Indicates the current location of each server, D k,1 ,…,D k,N represents the maximum cache resource of each server, P k,1 ,…,P k,N Indicates the maximum power of each server, in TZ 1 To TZ M Indicates the computing resources required for each subtask in all subtasks, from TD 1 To TD M Indicates the storage resources required by each subtask in all offloading subtasks.

[0141] (3b) In the model proposed in this scheme, sub-goal 1 is responsible for the task offloading decision, and its action space can be expressed as A 1 ={a 1 ,a 2 ,...,a N}, where a 1 ,a 2 ,...,a Nrepresents the probability of offloading to each server; sub-goal 2 is responsible for the power adjustment of the server, and its action space can be expressed as A 2 ={β -3 ,β -2 ,β -1 ,β 0 ,β 1 ,β 2 ,β 3}, where β -3 ,β -2 ,β -1 ,β 0 ,β 1 ,β 2 ,β 3 They represent the probability values ​​of seven operations: significantly reducing power, reducing power by a moderate amount, reducing power by a small amount, maintaining the original power, increasing power by a small amount, increasing power by a moderate amount, and significantly increasing power. Sub-goal three is responsible for caching decisions, and its action space can be expressed as A 3 ={ 1 ,x 2 ,...,χ M}, where χ 1 ,x 2 ,...,χ M represents a sequence of operations, if χ i , i∈[1,M] is equal to 1, then cache task i, if χ i ,i∈[1,M] is equal to 0, then cache task i is released.

[0142] To summarize, the method for dynamically optimizing power allocation and intelligent cache management in the scenario of air-space-ground integrated vehicle network provided by the present invention takes into account the problems of dynamic power adjustment optimization of servers and intelligent cache management from the perspective of reducing system energy consumption and latency, and provides strong support for the development of green communications. Compared with the deep reinforcement learning scheme based on multi-task and multiple neural networks, the method for solving the problem based on deep reinforcement assisted learning can more coordinately reduce system latency and system energy consumption.

[0143] All or part of the steps in the various methods of the above embodiments can be completed by controlling the relevant hardware through a program, and the program can be stored in a readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other readable medium that can be used to carry or store data.

[0144] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power and cache joint optimization method in the air-ground fusion vehicle network scenario, characterized in that: include: S1, using a unified heterogeneous network resource feature description and expression method to integrate the communication resources of satellite-to-ground links, air-to-ground links, and ground links into the same vector communication resource library; S2, model and analyze task offloading, task calculation and cache resource allocation according to task type, server computing power and cache rules; take task calculation, cache and power allocation as optimization variables, and task execution delay, power consumption and cache efficiency as objective functions, to establish a general model of dynamic optimization power allocation and intelligent cache management suitable for space-ground integrated vehicle network; S3, based on the vector communication resource library, using artificial intelligence methods to solve the general model of dynamic optimization power allocation and intelligent cache management, to obtain the optimal dynamic optimization power allocation and intelligent cache management solution; S4, performing simulation calculation on the obtained dynamic optimization power allocation and intelligent cache management scheme to obtain evaluation data; S5, modifying the parameters of the general model of dynamic optimization of power allocation and intelligent cache management according to the evaluation data to obtain a dynamic optimization of power allocation and intelligent cache management model; S6, solving the dynamic optimization power allocation and intelligent cache management model to obtain a final dynamic optimization power allocation and intelligent cache management solution.

2. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 1 is characterized in that: The unified heterogeneous network resource feature description and expression method specifically includes: Calculate the data transmission rate of wireless communication between the drone, edge server and user terminal; Calculate data transmission rates for satellite communications; According to the program size of each task, calculate the task transmission delay: The task transmission delay of wireless communication = program size / data transmission rate of wireless communication; The mission transmission delay of satellite communication = program size / data transmission rate of satellite communication.

3. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 2 is characterized in that: Wireless communication is applied to the air-to-ground link and the ground link, and the method for calculating the data transmission rate of the wireless communication specifically includes: The data transmission rate of wireless communication between drones, edge servers, and user terminals is calculated using Shannon’s theorem: Among them, i and j represent the sender and receiver respectively. represents the transmission rate between i and j; B represents the channel bandwidth; P i represents the transmission power of i; w i represents the Gaussian noise power in the channel; H ij represents the channel gain between devices i and j; Satellite communication is applied to the satellite-to-ground link, and the method for calculating the data transmission rate of satellite communication specifically includes: Visible light communication is used, and the transmission rate is R space It is expressed as: R space =FR×AS×Or Among them, FR represents the frame rate, AS represents the antenna array size, and Or represents the modulation order.

4. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 1 is characterized in that: The step of modeling and analyzing task offloading, task computing, and cache resource allocation based on task type, server computing capability, and cache rules specifically includes: Assume that the set of all types of subtasks is S ALL ; The task created by terminal k is represented by S k = {SSET,t max }, where t max represents the maximum tolerable delay of the task, SSET represents a task set consisting of M non-cacheable subtasks and N cacheable subtasks, Each subtask S in the offloading task * It is expressed as: S * ={TZ,TD,TL} Among them, TZ is the amount of computing resources required by the subtask, TD is the size of storage resources required to load the subtask, and TL is the program size of the subtask; When the subtask is calculated, the cacheable subtask is placed in the cache, or the subtask is removed from the cache, so that the storage resources of the cached task do not exceed the total storage resources of the server; The execution time of a single subtask i on the server is expressed as: ST i =Z i ×f cpu , where f cpu Indicates the computing frequency of the server CPU, Z i The computational load of the task of offloading the vehicle terminal is multiple subtasks S * The sum of the computing resources, namely Z i =∑TZ; The computational delay of task offloading to drone nodes UAV Calculate according to the following formula: in, represents the average power of the drone node during operation, C represents the capacitance of the transistors that make up the CPU, and V dd represents the steady-state voltage within the transistor; The computational delay of task offloading to satellite nodes is ST Space Calculate according to the following formula: in, Represents the average power of the satellite node during operation.

5. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 4 is characterized in that: The steps of establishing a general model for dynamic optimization power allocation and intelligent cache management applicable to the air-ground-integrated vehicle network specifically include: For the unloaded task i, the normalized value of the unloading delay, the normalized value of the energy consumption during task unloading, the inverse of the average cache hit rate during task unloading, and the average cache utilization during task unloading are added and optimized to the minimum, which is expressed as a multi-objective joint optimization equation: st C1:0≤T i ≤T max C2:0≤E i ≤E max C3:φ∈[0,1] C4: C5:0≤p≤P k C6:TZ i ≥0 C7:0≤TD i ≤D k Among them, T i represents the delay of offloading task i, T max represents the upper limit of the single task offloading delay, T i represents the energy consumption of offloading task i, T max represents the upper limit of energy consumption of a single server, φ is the cache hit rate, is the cache utilization, st means subject to constraints C1, C2, C3, C4, C5, C6 and C7; constraint C5 means that the real-time power consumption p of the server cannot exceed the maximum power P of the server k , k∈{m,a,s}, m represents the computing power of the edge server, a represents the computing power of the drone server, and s represents the computing power of the low-orbit satellite server; constraint C6 represents: the computing resources TZ required for task i i is a non-negative value; constraint C7 indicates that the cache resource TD of task i i Smaller than the storage resource D owned by the computing node.

6. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 5 is characterized in that: The modeling process of the task execution delay specifically includes: In the space-ground integrated vehicle network, the unloading delay T of task i i It is expressed as: in, It is represented as the transmission delay of task i uploaded to the server; represents the waiting delay of task i; represents the return delay of the calculation result of task i; After the terminal offloads the computing task to the computing node, if the computing node has an ongoing computing task, the task is added to the queue.

7. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 5 is characterized in that: The power consumption modeling process specifically includes: The energy consumption of a single computing node is expressed as the sum of the energy consumption of the cache and the energy consumption of the CPU when all tasks are executed: E=E memory +∑E calculate +∑E transmit Among them, E memory It is represented as the energy consumption of cache during operation, E calculate Indicates the energy consumption of the task during calculation, E transmit represents the transmission energy consumption; E transmit =P trans ·T P trans represents the transmit power of the computing node, and T represents time; D(t) represents the function of cache occupancy changing over time, D k It represents the storage resources owned by the computing node, U represents the UAV, S represents the low-orbit satellite, and P0 represents the maximum power of the computing node on the storage resources; P(t)=P k ·L(t),k∈{U,S} P k It represents computing power. L(t) is a function that varies between 0 and 1, which represents CPU utilization. When L(t) = 1, it indicates that the CPU is running at full load. When L(t) = 0, it indicates that the CPU is idle.

8. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 5 is characterized in that: The cache efficiency modeling process specifically includes: Define the cache hit rate φ as: Among them, count() represents the aggregation function of the number of elements in the statistical set, SNN represents the set of cacheable subtasks; SNN * Represents the set of subtasks that are cache hits when executing a task; Defining cache utilization for: in, Represents the average storage usage of the i-th computing node.

9. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to claim 1 is characterized in that: The step of using an artificial intelligence method to solve the general model of dynamic optimization of power allocation and intelligent cache management specifically includes: Based on the general model of dynamic optimization of power allocation and intelligent cache management, a deep reinforcement assisted learning algorithm is used to solve the probability of offloading tasks to each server, solve the probability of various power adjustment operations of the server, solve the server's cache operations on each task and the operation sequence of releasing tasks in the cache.

10. The power and cache joint optimization method in the air-ground-integrated vehicle network scenario according to any one of claims 1 to 9, characterized in that: The air-space-ground integrated vehicle network includes the following communication execution entities: user terminal, edge server, drone and low-orbit satellite. The server can be set up in the drone or the low-orbit satellite.

Images