A reliable unloading optimization method based on self-learning vehicle edge computing and network

Abstract

The invention relates to the technical field of intelligent traffic control, and discloses a reliable unloading optimization method for vehicle edge computing and network based on self-learning, comprising the following steps: constructing a three-layer unloading model of vehicle edge computing and network including a physical network layer, an unloading edge layer and a blockchain network layer; utilizing vehicles in the physical network layer to generate tasks, and designing unloading methods for two types of tasks according to safety-related coefficients in the tasks, and obtaining an unloading matrix for realizing task unloading; defining a new indicator cEDP as an unloading reliability measurement indicator based on total energy consumption, total delay and privacy exposure risk and optimizing it, and obtaining an optimized new indicator cEDP; optimizing the unloading methods for the designed two types of tasks, and constructing a decentralized unloading optimization problem; adopting game theory and a self-learning method of reliable privacy-preserving unloading based on autonomous learning to solve the constructed decentralized unloading optimization problem, and obtaining an optimal unloading matrix; the method ensures and improves unloading reliability.

Description

Technical Field

[0001] The present invention relates to the field of intelligent traffic control technology, and in particular to a reliable unloading optimization method for vehicle edge computing and networks based on self-learning. Background Art

[0002] With the rapid development of Intelligent Transportation Systems (ITSs), the Internet of Vehicles (IoVs) has received extensive attention from academia and industry. The IoV is an emerging transportation service network consisting of roadside units (RSUs) and vehicles that can obtain real-time traffic information and interact with each other. In order to establish the IoV, real-time perception, reliable transmission, and efficient processing of road traffic data are essential. However, due to the large amount of data processed by large sensors, vehicles often cannot receive notifications of processing results in a timely manner. This bottleneck of resource-constrained IoVs seriously affects the driving experience, reduces road operation efficiency, and has an adverse impact on road traffic safety. Therefore, the IoV computing network architecture based on mobile edge computing (MEC), namely the vehicular edge computing and networks (VECONs) paradigm, has emerged. In addition to being processed locally, computationally intensive and latency-critical tasks can also be offloaded to Vehicular Edge Servers (VESs) and Road-side Edge Servers (RESs). In order to provide users with reliable services, VECONs need to consider both offloading efficiency and security during task offloading, which is the key to ensuring reliable task offloading. However, almost no measures have been taken to address security issues such as privacy exposure that may occur when offloading data, which will affect the modeling and optimization of offloading reliability.

[0003] In order to solve the problems associated with the centralized offloading scheme, the existing technology decouples the joint optimization problem of task offloading and resource allocation in the vehicle edge computing and the network into two independent sub-problems, and achieves a near-optimal solution through decentralized optimization; it also solves the computation offloading problem by making decentralized computation offloading decisions by minimizing task delays using the actor-critic algorithm and the deep deterministic policy gradient (DDPG) of the policy gradient; at the same time, a task offloading algorithm based on adaptive learning is designed based on the multi-armed bandit theory with the goal of minimizing the inter-vehicle offloading delay; in addition, a distributed device-to-device offloading system is proposed, considering the mobility of the task generator with the task completion deadline, and other users in the network complete the offloading decision in a decentralized manner based on constrained random games. Although the above decentralized offloading methods are helpful in constructing decentralized offloading problems and solution algorithms, they ignore the semi-trusted premise of decentralized vehicle edge computing and auxiliary computing vehicles and roadside units in the network. Therefore, from the perspective of network security, semi-trusted nodes may try to infer information that they should not obtain, which will lead to the theft and leakage of confidential information from auxiliary computing tasks. In summary, the above decentralized offloading methods only optimize latency and energy consumption, ignoring the risk of privacy exposure during the offloading process, thus undermining the overall reliability of computation offloading. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a reliable unloading optimization method for vehicle edge computing and network based on self-learning. By introducing unloading reliability measurement indicators based on energy consumption, delay and privacy exposure risk, a reliable unloading optimization is constructed, and a reliable privacy protection unloading method based on self-learning is used to solve the optimal unloading problem in a decentralized manner, thereby solving the problem of low reliability of computing offloading in the prior art.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A reliable unloading optimization method based on self-learning vehicle edge computing and network includes the following steps:

[0007] S1. Construct a three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer;

[0008] S2. Based on the constructed three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, the vehicle in the physical network layer is used to generate tasks, and according to the safety-related coefficients in the tasks, two types of task offloading methods are designed to obtain the offloading matrix for realizing task offloading;

[0009] Among them, the two types of tasks are strongly security-related tasks and weakly security-related tasks;

[0010] S3, defining a new indicator cEDP as an offloading reliability measurement indicator based on total energy consumption, total delay, and privacy exposure risk, and optimizing the new indicator cEDP to obtain an optimized new indicator cEDP;

[0011] S4. The new optimization index cEDP is used to optimize the offloading methods of the designed two types of tasks, and a decentralized offloading optimization problem is constructed;

[0012] S5. The game theory and self-learning method of reliable privacy-preserving offloading based on autonomous learning are used to solve the constructed decentralized offloading optimization problem and obtain the optimal offloading matrix.

[0013] The present invention has the following beneficial effects:

[0014] 1. The reliable offloading optimization method for vehicle edge computing and network based on self-learning proposed in the present invention constructs a three-layer offloading model of vehicle edge computing and network including a physical network layer, an offloading edge layer, and a blockchain network layer, and uses fully homomorphic encryption and blockchain to perform authorization in semi-trusted vehicle edge computing and network. At the same time, tasks are classified by security factors, and fully homomorphic encryption is used to pre-process tasks with higher privacy exposure risks before task offloading, thereby ensuring the reliability of offloading;

[0015] 2. A new indicator cEDP is introduced to measure offloading reliability. This new indicator cEDP is defined as an offloading reliability metric based on total energy consumption, total delay, and privacy exposure risk. This fully considers the efficiency and privacy protection in the offloading process, thus improving offloading reliability.

[0016] 3. The new indicator cEDP is optimized, and the optimized new indicator cEDP is used to optimize the offloading methods of the designed two types of tasks: strongly security-related tasks and weakly security-related tasks. By constructing a decentralized offloading optimization problem and using game theory and self-learning methods based on autonomous learning for reliable privacy-preserving offloading to solve the constructed decentralized offloading optimization problem, the vehicle can autonomously and independently learn the optimal offloading response decision, thereby achieving game equilibrium, and at the same time making the convergence speed and convergence results adapt to different scenario requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a reliable unloading optimization method for vehicle edge computing and network based on self-learning proposed in the present invention;

[0018] Figure 2 Schematic diagram of the three-layer offloading model for vehicle edge computing and networking. DETAILED DESCRIPTION

[0019] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0020] like Figure 1 As shown, the reliable unloading optimization method based on self-learning vehicle edge computing and network includes the following steps S1-S5:

[0021] S1. Construct a three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer.

[0022] In this embodiment, Figure 2 As shown, Figure 2 The schematic diagram of the three-layer offloading model of vehicle edge computing and network, which includes the physical network layer, the offloading edge layer and the blockchain network layer, is shown in Figure 1. Figure 2It can be seen that the physical network layer, the offloading edge layer, and the blockchain network layer constitute the three-layer offloading model of vehicle edge computing and network. Vehicle edge computing and network are used to frequently generate and offload tasks, which is the basis for the operation of the Internet of Vehicles. The physical network layer consists of roadside units (RSU) and vehicles. Vehicles have task offloading vehicles and task generating vehicles, which are used to unload or generate tasks. Roadside infrastructure is usually equipped with roadside units, which can interact with vehicles within its coverage. These roadside units equipped with powerful computing and storage modules are called roadside servers (RES). They provide task offloading services to vehicles through vehicle-to-infrastructure (V2I) communication. The roadside units in the network are semi-trusted nodes, and these roadside units try to infer information that should not be obtained from a network security perspective, but they will not communicate with any party. In addition, according to the actual situation of vehicle computing and storage resources, vehicles can have three behaviors: (1) If resources are insufficient, vehicles offload tasks to other edge nodes through the vehicle network protocol (V2X); (2) If resources are sufficient, vehicles can choose to perform tasks locally by themselves; (3) If resources are sufficient, vehicles can act as vehicle edge servers (VES) to assist other vehicle entities in offloading tasks; The offloading edge layer includes multiple offloading edge networks. In order to facilitate offloading tasks, each roadside server and the vehicle edge servers within its coverage area form an offloading edge network. Using the task offloading method, the optimal offloading method of multiple tasks can be decentralized. The blockchain network layer includes all authorized roadside units, which is a distributed ledger for storing the optimal offloading matrix uploaded by the roadside units. By using the practical Byzantine fault-tolerant consensus method, the blockchain network can prevent the historical best offloading matrix from being tampered with, thereby improving the traceability and reliability of historical storage records. In addition, V2V offloading is vehicle-to-vehicle offloading.

[0023] Specifically, step S1 includes S11-S14:

[0024] S11. The physical network layer includes roadside units and vehicles. Roadside units are used to exchange information with vehicles, and vehicles are used to generate tasks.

[0025] S12. The offloading edge layer includes an offloading edge network, which includes a roadside server and a vehicle edge server. The roadside server is a roadside unit equipped with powerful computing and storage modules, and provides task offloading services for vehicles through vehicle-to-infrastructure communication. At the same time, each vehicle can act as a vehicle edge server to assist other vehicles in achieving task offloading.

[0026] S13. The blockchain network layer includes all authorized roadside units, wherein the blockchain network layer is used to prevent the optimal unloading matrix from being tampered with.

[0027] S14. Based on the physical network layer, the offloading edge layer and the blockchain network layer, a three-layer offloading model of vehicle edge computing and network is constructed, which includes the physical network layer, the offloading edge layer and the blockchain network layer.

[0028] S2. Based on the constructed three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, the vehicle in the physical network layer is used to generate tasks, and according to the security correlation coefficient in the tasks, the offloading methods of two types of tasks are designed to obtain the offloading matrix for realizing task offloading; among them, the two types of tasks are strongly security-related tasks and weakly security-related tasks.

[0029] In this embodiment, two types of tasks, namely, strongly security-related tasks and weakly security-related tasks, are designed to obtain an offloading matrix of tasks, thereby realizing task offloading, so that the user can reduce the offloading reliability metric during the subsequent steps of offloading.

[0030] Specifically, step S2 specifically includes S21-S25:

[0031] S21. Construct the task set generated by the vehicle, namely:

[0032] T={T1,T2,…,T i ,…,T n}

[0033] Where T represents the task set generated by the vehicle, T1 represents the task generated by the first vehicle, T2 represents the task generated by the second vehicle, and T i represents the task generated by the i-th vehicle, T n Represents the task generated by the nth vehicle.

[0034] S22. The tasks generated by the vehicle consist of data information, data size, processing density, generation time, safety-related coefficient, and uniqueness, namely:

[0035]

[0036] in, Denotes the definer, D i represents the data information of the task generated by the i-th vehicle, S i represents the data size of the task generated by the i-th vehicle, C i represents the processing density of the tasks generated by the i-th vehicle, represents the generation time of the task generated by the i-th vehicle, SR i represents the safety-related coefficient of the task generated by the i-th vehicle, ε i represents the uniqueness of the task generated by the i-th vehicle.

[0037] S23. Design a method for offloading tasks that are strongly security-related, specifically:

[0038] If the safety relevance coefficient of the task generated by the i-th vehicle is SR i =1, then the task T generated by the i-th vehicle i For security-related tasks, namely, through vehicle positioning, vehicle navigation and road condition report generation, and based on the three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, three offloading methods for security-related tasks are designed, namely, task offloading through local processing 1 , using fully homomorphic encryption to offload tasks to the vehicle edge server m 2 And use fully homomorphic encryption to offload tasks to roadside servers m 3 .

[0039] In this embodiment, task offloading is performed through local processing. 1 The task is offloaded to the vehicle edge server m using a fully homomorphic encryption method 2 , using fully homomorphic encryption to offload tasks to roadside servers m 3 In comparison, although the local processing capability of the vehicle is the weakest, which increases the computational delay and energy consumption of the task, the task avoids homomorphic encryption transmission, avoiding the corresponding data expansion, additional delay and energy consumption; the task is offloaded to the vehicle edge server m using a fully homomorphic encryption method. 2 Task offloading through local processing 1 Compared with the traditional 3D video streaming, the 3D video streaming can obtain stronger computing power and lower computing latency through data expansion and task transmission; the fully homomorphic encryption method is used to offload tasks to the vehicle edge server m 2 Using fully homomorphic encryption to offload tasks to roadside servers 3 Although the computing power of the vehicle edge server is not as good as that of the roadside server, its advantages are smaller average transmission distance, lower corresponding transmission delay and energy consumption; the fully homomorphic encryption method is used to offload tasks to the roadside server m 3 Task offloading through local processing 1 , using fully homomorphic encryption to offload tasks to the vehicle edge server m 2 Although offloading tasks to roadside servers can utilize the most powerful computing power, it is still more expensive than m 1 Compared with the need for data expansion on the one hand, the average transmission distance of the task is greater than m 2 , which will generate additional delays and energy consumption. In addition, fully homomorphic encryption (FHE) is an encryption technology that conforms to computational complexity theory and is considered to be an encryption technology that balances collaborative computing and privacy protection.

[0040] S24. Design a method for unloading weakly security-related tasks, specifically:

[0041] If the safety relevance coefficient of the task generated by the i-th vehicle is SR i = 0, then the task T generated by the i-th vehicle i For security weakly related tasks, two offloading methods for security weakly related tasks are designed based on the constructed three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, that is, offloading tasks to roadside servers m 4 , after encryption using fully homomorphic encryption, the task is offloaded to the roadside server m 5 .

[0042] In this embodiment, SR i = 0, the task T generated by the i-th vehicle i For tasks that are weakly related to security, the most common ones are entertainment applications, such as audio-visual, games, shopping, etc. Although the data volume is large, the privacy information density is low, so it is more appropriate to offload such tasks to roadside servers. 4 It is not encrypted. When a large amount of data needs to be processed, encryption will increase the delay and energy consumption of the three stages of preprocessing, transmission and processing. Therefore, the task is offloaded to the roadside server m 4 It is worthwhile to trade a certain risk of privacy exposure for smaller delays and energy consumption; in addition, the task is offloaded to the roadside server m after encryption using a fully homomorphic encryption method. 5 Using fully homomorphic encryption to offload tasks to roadside servers 3 Similarly, the task is offloaded to the roadside server m after encryption using a fully homomorphic encryption method. 5 The method of unloading tasks from the roadside server after retaining task encryption makes the task unloading method more flexible.

[0043] S25. According to the designed unloading method of the strongly safety-related tasks and the unloading method of the weakly safety-related tasks, the unloading matrix of the task set T generated by the vehicle is obtained, that is:

[0044]

[0045] Where m represents the unloading matrix of the task set T generated by the vehicle, Indicates that the task T1 generated by the first vehicle adopts the mth 1 Uninstall methods: Indicates that the task T1 generated by the first vehicle adopts the mth 5 Uninstall methods: represents the task T generated by the nth vehicle n Use the mth 1Uninstall methods: represents the task T generated by the nth vehicle n Use the mth 5 Uninstall methods: represents the task T generated by the i-th vehicle i Use the kth uninstall method.

[0046] S3. Define a new indicator cEDP as an offloading reliability measurement indicator based on total energy consumption, total delay and privacy exposure risk, and optimize the new indicator cEDP to obtain an optimized new indicator cEDP.

[0047] In this embodiment, efficiency indicators such as delay and energy consumption are used to evaluate the pros and cons of the offloading strategy. However, when the task is assigned to other entities for calculation, the privacy exposure risk is not taken into account, which will affect the reliability of the offloading method. Therefore, to solve this problem, the present invention introduces an offloading reliability measurement indicator based on total energy consumption, total delay and privacy exposure risk to solve the problem of low reliability of the offloading method.

[0048] Specifically, step S3 specifically includes S31-S34:

[0049] S31. Define a new indicator cEDP as an offloading reliability metric based on total energy consumption, total delay, and privacy exposure risk, namely:

[0050] CED i =f(t i ,E i ,per i )

[0051] Among them, cEDP i represents the new index cEDP of the ith vehicle, f(·) represents the combination function, t i represents the task T generated by the i-th vehicle i The total delay in computing offloading, E i represents the task T generated by the i-th vehicle i In calculating the total energy consumption during unloading, per i represents the task T generated by the i-th vehicle i Privacy exposure risks during computational offloading.

[0052] S32, calculating the total delay and total energy consumption of the task generated by the vehicle when calculating unloading, that is:

[0053]

[0054] in, represents the task T generated by the i-th vehicle i Delays incurred during local processing, represents the task T generated by the i-th vehicle i The delay caused during the transmission phase, represents the task T generated by the i-th vehicle i The delay caused by the task data calculation stage, represents the task T generated by the i-th vehicle i Energy consumption during the local processing stage, represents the task T generated by the i-th vehicle i The energy consumption during the transmission phase, represents the task T generated by the i-th vehicle i Energy consumption generated during the task data calculation phase.

[0055] S33. The privacy exposure risk of the tasks generated by the computing vehicle during computing offloading, namely:

[0056]

[0057] Where log represents the logarithmic function, e represents the exponential function, and Frs i represents the task T generated by the i-th vehicle i The freshness of the task, represents the task T generated by the i-th vehicle i Data safety factor using the kth uninstallation method, S mid Represents the median size of all tasks generated by the vehicle.

[0058] In this embodiment, the data size S i Indicates the T to be processed i The actual number of bits. When the density of privacy data remains unchanged, per i With S i There is a positive correlation because T i The more information contained, the more private information it contains; once T i The more data is leaked, the greater the damage to data privacy. At the same time, as the amount of data increases, the risk of privacy exposure does not increase linearly, but has an upper limit. Therefore, this paper assumes that per i With S i It is reasonable to have a Sigmoid function relationship between them. And per i In S i =S mid The maximum growth rate is reached when i The time interval between the task generation time and the current time reflects the timeliness of the task. i equal in It is T igeneration moment; in the ever-changing vehicle edge computing and network scenarios, the timeliness of tasks will decrease over time, and the risk of task damage will also decrease, so per i With Frs i Negative correlation; therefore, the present invention uses an exponential function to represent per i With Frs i The nonlinear relationship between ε i Indicates T i The uniqueness of is affected by factors such as task type, generation source and time, and owner identity; for example, tasks originating from navigation applications have a higher privacy exposure risk than tasks originating from entertainment applications, and tasks late at night have a higher privacy exposure risk than tasks generated during the day; therefore, the present invention assumes that ε i It is T i A constant that follows a normal distribution between [0, 1], per i With ε i Inversely proportional. Data security factor This reflects the protection of privacy by appropriate task processing methods; when tasks are processed locally, no device other than the task itself is involved, so When the task is offloaded to the vehicle edge server or roadside server for processing, define This reflects the privacy protection of fully homomorphic encryption (FHE); on the contrary, the tasks that are offloaded without completing the encryption process, In addition, the purpose of using the logarithmic function is to reduce the accidental effects that may be caused by points near the two ends of the value space.

[0059] S34. According to the total delay and total energy consumption of the tasks generated by the vehicle when calculating the unloading, and the privacy exposure risk of the tasks generated by the vehicle when calculating the unloading, the optimized new indicator cEDP is obtained, that is:

[0060]

[0061] Among them, cEDP′ i represents the new optimization index cEDP of the i-th vehicle, λ1, λ2, λ3 represent weighted parameters, represents the task T generated by the i-th vehicle i The minimum total delay when computing offloads, represents the task T generated by the i-th vehicle i The maximum total delay when calculating offload, represents the task T generated by the i-th vehicle i In calculating the minimum total energy consumption during unloading, represents the task T generated by the i-th vehicle i When calculating the maximum total energy consumption during unloading, represents the task T generated by the i-th vehicle i Minimum privacy exposure risk during computation offloading, represents the task T generated by the i-th vehicle i The biggest privacy exposure risk during computational offloading.

[0062] In this embodiment, λ1+λ2+λ3=1, and the sizes of these three parameters can be adjusted according to the actual needs of the network, and can also be obtained according to the multi-criteria decision-making theory.

[0063] S4. The new optimization index cEDP is used to optimize the offloading methods of the two types of tasks designed, and a decentralized offloading optimization problem is constructed.

[0064] In this embodiment, based on the optimized new index cEDP, the offloading reliability of different safety-related tasks when using different offloading methods is measured, and finally a decentralized offloading optimization problem is constructed.

[0065] Specifically, step S4 specifically includes S41-S46:

[0066] S41, based on the new optimized indicator cEDP, obtain the mth 1 New indicators for vehicles when unloading The specific process is:

[0067] First, when using the mth 1 When the unloading method is used, the total delay of the tasks generated by the vehicle during the calculation of unloading is calculated, that is:

[0068]

[0069] Among them, l means local, Both indicate the use of the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i The total delay in computing offload, f l Represents the local processing capability of the vehicle.

[0070] In this embodiment, when T is selected i When processed locally, there are: in represents the task T generated by the i-th vehicle i Use the mth 1 Data security factor of the uninstall method, represents the task T generated by the i-th vehicle i When calculating unloading, the mth 1 The privacy exposure risk of this uninstall method.

[0071] Secondly, when using the mth 1When the unloading method is used, the local power consumption and total energy consumption of the tasks generated by the vehicle during the calculation of unloading are calculated, that is:

[0072]

[0073] in, Indicates the use of the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i Local power consumption during computation offloading, V represents the vehicle edge server, k l,V represents the local power factor, Both indicate that when the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i The total energy consumption when calculating unloading.

[0074] Finally, according to And the optimized new indicator cEDP, get the mth 1 The new indicators of the vehicle when unloading the method are:

[0075]

[0076] in, Indicates the use of the mth 1 The new index of the i-th vehicle when the unloading method is used.

[0077] S42, based on the optimized new indicator cEDP, obtain the mth 2 New indicators for vehicles when unloading The specific process is:

[0078] In this embodiment, if the i-th vehicle selects Then the vehicle edge server calculates the security strongly related task T after homomorphic encryption i , then: in, represents the task T generated by the i-th vehicle i Use the mth 2 Data security factor of the uninstall method, represents the task T generated by the i-th vehicle i When calculating unloading, the mth 2 The privacy exposure risk of this offloading method; therefore, the vehicle edge server can complete task offloading and calculation in three steps, as follows:

[0079] First, assuming that the encryption time and data expansion of the ciphertext are proportional to the data size, the delay incurred by the task generated by the vehicle in the local processing stage is calculated, that is:

[0080]

[0081] Among them, τ represents the time factor consumed by encryption.

[0082] Calculate the energy consumption of the tasks generated by the vehicle in the local processing stage, that is:

[0083]

[0084] in, represents the task T generated by the i-th vehicle i Energy consumption during the local processing stage.

[0085] Calculate the amount of data added to the tasks generated by the vehicle after full homomorphic encryption, that is:

[0086] S′ i =ξS i

[0087] Among them, S′ i represents the task T generated by the i-th vehicle after fully homomorphic encryption i The increased data size, ξ represents the amplification factor of the increased data size after encryption.

[0088] Secondly, in the transmission phase involving the vehicle edge server, the transmission rate of the tasks generated by the vehicle on the IoV channel is calculated, that is:

[0089]

[0090] in, represents the task T generated by the i-th vehicle i The transmission rate on the IoV channel, L represents the total number of IoV communication licensed channels, B represents the bandwidth of each licensed channel, log2 represents the logarithmic function with base 2, P represents the transmission power of the vehicle, h represents the channel attenuation coefficient, ω0 represents the white Gaussian noise power of the vehicle, θ represents the path loss exponent, represents the distance between the i-th vehicle and the vehicle edge server.

[0091] Assuming that the vehicle remains stationary while performing edge computing, the latency and energy consumption of the tasks generated by the vehicle during the transmission phase are calculated, namely:

[0092]

[0093] in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Delays during the transmission phase, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Energy consumption during the transmission phase.

[0094] Then, in the task data calculation phase involving the vehicle edge server, assuming that the number of computing cores of the vehicle edge server is M V , the processing power of each core of the vehicle edge server is f V , calculate the delay of the task generated by the vehicle in the task data calculation stage, that is:

[0095]

[0096] in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Delay in the task data calculation phase.

[0097] Calculate the power consumption and energy consumption of the tasks generated by the vehicle during the task data calculation phase, namely:

[0098]

[0099] in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Power consumption during the task data calculation phase, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Energy consumption during the task data calculation phase.

[0100] Finally, according to Get adopted mth 2 The total delay and total energy consumption of the offloading method are:

[0101]

[0102] in, Indicates the use of the mth 2 Total delay for each uninstallation method, Indicates the use of the mth 2 The total energy consumption of the unloading method.

[0103] according to And the optimized new indicator cEDP, get the mth 2 The new indicators of the vehicle when unloading the method are:

[0104]

[0105] in, Indicates the use of the mth 2 The new index of the i-th vehicle when the unloading method is used.

[0106] S43, based on the optimized new indicator cEDP, obtain the mth 3 New guidance for vehicles when unloading With the mth 5 New indicators for vehicles when unloading The specific process is:

[0107] In this embodiment, if the i-th vehicle selects or Then T i It will be calculated by the roadside server after homomorphic encryption, then: in, represents the task T generated by the i-th vehicle i Use the mth 3 or m 5 Data security factor of the uninstall method, represents the task T generated by the i-th vehicle i When calculating unloading, the mth 5 or m 5 The privacy exposure risk of the uninstall method; In addition, 3 With m 5 The difference between these two offloading methods is only the type of tasks to be processed, so the task offloading and calculation are as follows:

[0108] First, in the transmission phase of the roadside server, the transmission rate, delay, and energy consumption of the tasks generated by the vehicle on the vehicle-to-infrastructure channel are calculated, namely:

[0109]

[0110] in, represents the task T generated by the i-th vehicle i The transmission rate on the vehicle-to-infrastructure channel, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Delay on the vehicle-to-infrastructure channel, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Energy consumption on the vehicle-to-infrastructure channel, Represents the distance between the i-th vehicle and the roadside server.

[0111] Secondly, in the task data calculation stage of the roadside server, assuming that the number of computing cores of the roadside server is M R , the processing power of each core of the roadside server is f R, calculate the delay of the task generated by the vehicle in the task data calculation stage of the roadside server, that is:

[0112]

[0113] in, Indicates the use of the mth 3 The i-th vehicle generates task T when the unloading method i The latency of the task data computation phase at the roadside server.

[0114] Calculate the power consumption and energy consumption of the tasks generated by the vehicle during the task data calculation phase, namely:

[0115]

[0116] in, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i The power consumption in the task data calculation phase, k R represents the power factor of the roadside server, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase.

[0117] according to Get adopted mth 3 The total delay and total energy consumption of the offloading method are:

[0118]

[0119] in, Indicates the use of the mth 3 Total delay for each uninstallation method, Indicates the use of the mth 3 The total energy consumption of the unloading method.

[0120] Finally, according to And the optimized new indicator cEDP, get the mth 3 The new indicators of the vehicle when unloading the method are:

[0121]

[0122] in, Indicates the use of the mth 5 The new index of the i-th vehicle when the unloading method is used.

[0123] Similarly, we get the mth 5 The new indicators of the vehicle when unloading the method are:

[0124]

[0125] in, Indicates the use of the mth 5 The new index of the i-th vehicle when the unloading method is Indicates the use of the mth 5 Total delay for each uninstallation method, Indicates the use of the mth 5 The total energy consumption of the unloading method is Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Delay on the vehicle-to-infrastructure channel, Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Energy consumption on the vehicle-to-infrastructure channel, Indicates the use of the mth 5 The i-th vehicle generates task T when the unloading method i The delay in the task data calculation phase of the roadside server, Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase.

[0126] S44, based on the new optimized indicator cEDP, obtain the mth 4 New indicators for vehicles when unloading The specific process is:

[0127] In this embodiment, if the i-th vehicle selects Then the weak security related tasks are transferred in plain text, and then: in, represents the task T generated by the i-th vehicle i Use the mth 4 The data security factor of this uninstallation method; therefore And S i The calculation unloading process remains unchanged, and the specific details are as follows:

[0128] First, due to the use of the mth 4 The delay and energy consumption of this offloading method only occur in the transmission stage and the task data calculation stage, so the calculation adopts the mth 4 The total delay and total energy consumption of the offloading method are:

[0129]

[0130] in, Indicates the use of the mth4 Total delay for each uninstallation method, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Delays during the transmission phase, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Delays in the task data calculation phase, Indicates the use of the mth 4 The total energy consumption of the unloading method is Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the transmission phase, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase.

[0131] Secondly, the privacy exposure risk of the tasks generated by the vehicle is calculated, namely:

[0132]

[0133] in, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i privacy exposure risk.

[0134] Finally, according to And the optimized new indicator cEDP, get the mth 4 The new indicators of the vehicle when unloading the method are:

[0135]

[0136] in, Indicates the use of the mth 4 The new index of the i-th vehicle when the unloading method is used.

[0137] S45, according to Get the coupling index cEDP of the task generated by the vehicle i (m i ,m -i ),Right now:

[0138]

[0139] Among them, cEDP i (m i ,m -i ) represents the task T generated by the i-th vehicle iCoupling index, m i represents the task T generated by the i-th vehicle i The unloading method matrix, m -i Indicates division by m i The offloading method matrix of the remaining tasks, Indicates intermediate temporary variables, which have no practical meaning. m1, m i-1 、m i+1 、m n Respectively represent the unloading method matrix of the tasks generated by the 1st, i-1, i+1, and nth vehicles, if represents the judgment condition, represents the task T generated by the i-th vehicle i Use the mth 1 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 2 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 3 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 4 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 5 Uninstall method.

[0140] In this embodiment, due to the limited network communication and computing resources, the unloading methods between vehicles are coupled. The optimal unloading method of any specific vehicle will change dynamically with the change of the unloading methods of other vehicles, so the coupling index shown above is established. In addition, due to the different data sizes of tasks, it is often the case that a task completes the transmission before other tasks, or a task completes the calculation in the roadside server before other tasks. In this case, additional channel resources or computing resources will be released. If these resources are used by other tasks, will be reduced, so once the present invention determines the optimal decision matrix m, the communication and computing resources allocated to each task will remain unchanged.

[0141] S46, coupling index cEDP of the task generated by the vehicle i (m i ,m -i ), construct the decentralized unloading optimization problem, namely:

[0142]

[0143] Among them, min means to find the minimum value, st means to satisfy it, C1, C2, C3, and C4 represent different constraints respectively. Indicates any symbol, T max represents the maximum total delay of the tasks generated by the vehicle when computing offloading, per max Represents the maximum privacy exposure risk of tasks generated by vehicles when computing is offloaded.

[0144] In this embodiment, by adjusting T i The processing method can make the coupling indicator cEDP i (m i ,m -i ) is minimized, thereby improving the overall reliability of the network model. Among them, the constraint C1 is used to ensure that each task in T selects m 1 -m 5 The offloading method in is used to handle it; Constraint C2 is used to ensure that latency, energy consumption and privacy protection are all taken into account when making task offloading decisions; Constraint C3 indicates that the task processing time cannot exceed the maximum allowed total latency T max , thereby ensuring the real-time performance of task processing; constraint C4 indicates that the privacy exposure risk caused by task processing needs to be kept at a reasonable upper limit per max This is the primary prerequisite for ensuring reliable unloading.

[0145] S5. The game theory and self-learning method of reliable privacy-preserving offloading based on autonomous learning are used to solve the constructed decentralized offloading optimization problem and obtain the optimal offloading matrix.

[0146] In this embodiment, the decentralized unloading optimization problem is solved by using game theory and self-learning methods for reliable privacy-preserving unloading based on autonomous learning, and the complexity of vehicle decision-making and the overhead caused by frequent interactions are reduced by utilizing the intelligence of individual vehicles. At the same time, the vehicle selects the best unloading method through self-learning, and finally the network model reaches Nash equilibrium, and no vehicle is willing to change its unloading method at this time.

[0147] Specifically, step S5 includes S51-S53:

[0148] S51. According to the constructed decentralized offloading optimization problem, a decentralized privacy-preserving offloading game model is constructed, namely:

[0149]

[0150] Where G represents the decentralized privacy-preserving offloading game model, cEDP i (m) represents the common cost set of all participants, cEDP1(m1,m -1) represents the coupling index of the task T1 generated by the first vehicle, cEDP2(m2,m -2 ) represents the coupling index of task T2 generated by the second vehicle, cEDP n (m n ,m -n ) represents the task T generated by the nth vehicle n coupling index.

[0151] In this embodiment, in the game stage, each vehicle independently adjusts its decision according to the unloading decision of other vehicles, so that the coupling index cEDP i (m i ,m -i )minimum.

[0152] S52. Introduce the potential function to transform the constructed decentralized privacy protection offloading game model into a potential game model, that is:

[0153]

[0154] Among them, Φ(m i ) represents the potential game model, η、n chnl 、n comp Both represent intermediate temporary variables and have no practical significance. represents the task T generated by the i-th vehicle i The delay of using the kth offloading method, represents the task T generated by the i-th vehicle i Energy consumption using the kth unloading method, represents the task T generated by the qth vehicle i Use the kth uninstall method.

[0155] S53. The potential game model is solved by using game theory and self-learning method based on reliable privacy protection offloading to obtain the optimal offloading matrix. The specific process is as follows:

[0156] First, initialize the record memory M of the i-th vehicle i , where the vehicle's recording memory is used to record the unloading methods and coupling indicators of other tasks.

[0157] Secondly, determine whether the safety relevance coefficient of the task generated by the i-th vehicle is SR i =1, if so, the optimal unloading method for the task generated by the i-th vehicle is initialized as Otherwise, the optimal offloading method for the task generated by the i-th vehicle is initialized as

[0158] Then, according to the unloading matrix m of the task set T generated by the initial vehicle, the new index of the initial vehicle is calculated And enter the self-learning stage to obtain the best unloading matrix, which is:

[0159] Initialize the number of iterations ite to ite = 1, and iterate the loop. Generate a random number when no update request message RTU is broadcast.

[0160] For parallel vehicles, an iterative loop is performed, according to the record memory M of the i-th vehicle i Extract the unloading matrix m of the task set T generated by the vehicle and determine whether the safety correlation coefficient of the task generated by the i-th vehicle is SR i =1, if so, then according to as well as calculate as well as And select the new index of the smallest vehicle as Will The corresponding uninstall method is set to Otherwise, according to calculate , and the new index of the smallest vehicle is selected as Will The corresponding uninstall method is set to

[0161] judge If so, the random back-off competition method, the minimum cEDP decision method, or the hybrid decision method is used to update the opportunity, otherwise the optimal unloading matrix is ​​maintained in the next iteration. constant.

[0162] Specifically, the process of updating the opportunity using the random back-off competition method is:

[0163] First, using the retreat time δ of the i-th vehicle i Initialize the timer.

[0164] Secondly, when there is δ i ←δ i -1, judge δ i = = 0, if yes, the competition is successful and the update request message RTU is broadcast, otherwise according to the update request message RTU received from another vehicle, and the received update request message RTU is used to update Among them, ← indicates assignment, and == indicates equality comparison.

[0165] Specifically, the process of updating opportunities using the minimum cEDP decision method is:

[0166] First, the monitoring timer is initialized with the monitoring time ψ, and an update request message RTU is broadcasted.

[0167] Secondly, before the countdown of the monitoring time ψ ends, the update request messages RTU of other vehicles are monitored and stored.

[0168] Finally, compare the RTU in each update request message And according to The corresponding smallest Update the unloading matrix m of the task set T generated by the vehicle and

[0169] Specifically, the process of updating opportunities using the hybrid decision-making method is:

[0170] First, judge If yes, the random back-off competition method is used to update the chance, otherwise the minimum cEDP decision method is used to update the chance, and Ω represents the judgment threshold.

[0171] The update is performed according to the received update request message RTU until the update request message RTU is no longer broadcast.

[0172] Finally, the unloading matrix of the final vehicle-generated task set obtained in the last iteration is returned and uploaded to the regional chain network after reaching a consensus to obtain the optimal unloading matrix.

[0173] The reliable offloading optimization method for vehicle edge computing and network based on self-learning proposed in the present invention constructs a three-layer offloading model of vehicle edge computing and network including a physical network layer, an offloading edge layer and a blockchain network layer, and uses fully homomorphic encryption and blockchain to perform authorization in semi-trusted vehicle edge computing and network. At the same time, the three-layer offloading model of vehicle edge computing and network is used to classify tasks using security factors, and fully homomorphic encryption is used to pre-process tasks with higher privacy exposure risks before task offloading, thereby ensuring the reliability of offloading. In addition, a new indicator cEDP for measuring offloading reliability is proposed, and the new indicator cEDP is defined as the total energy consumption, The unloading reliability measurement index of total delay and privacy exposure risk fully considers the efficiency and privacy protection in the unloading process, thereby improving the unloading reliability. At the same time, by optimizing the new index cEDP, the optimized new index cEDP is used to optimize the unloading methods of the designed two types of tasks, namely, strongly security-related tasks and weakly security-related tasks. By constructing a decentralized unloading optimization problem, it is proposed to use game theory and self-learning methods of reliable privacy-preserving unloading based on autonomous learning to solve the constructed decentralized unloading optimization problem, so that the vehicle can autonomously and independently learn the optimal unloading response decision, thereby achieving game equilibrium, and at the same time making the convergence speed and convergence results adapt to the requirements of different scenarios.

[0174] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0175] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A reliable unloading optimization method based on self-learning vehicle edge computing and network, characterized in that: The following steps are involved: S1. Construct a three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer; S2. Based on the constructed three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, the vehicle in the physical network layer is used to generate tasks, and according to the security correlation coefficient in the tasks, two types of task offloading methods are designed to obtain the offloading matrix for task offloading; the two types of tasks are security-strongly related tasks and security-weakly related tasks; Among them, the task set generated by the vehicle is: T={T1,T2,…,T i ,…,T n } Where T represents the task set generated by the vehicle, T1 represents the task generated by the first vehicle, T2 represents the task generated by the second vehicle, and T i represents the task generated by the i-th vehicle, T n represents the task generated by the nth vehicle; Among them, the tasks generated by the vehicle are composed of data information, data size, processing density, generation time, safety correlation coefficient and uniqueness, namely: in, Denotes the definer, D i represents the data information of the task generated by the i-th vehicle, S i represents the data size of the task generated by the i-th vehicle, C i represents the processing density of the tasks generated by the i-th vehicle, represents the generation time of the task generated by the i-th vehicle, SR i represents the safety-related coefficient of the task generated by the i-th vehicle, ε i represents the uniqueness of the task generated by the i-th vehicle; Among them, the uniqueness of the task is affected by the task type, the source and time of generation, and the identity of the owner; S3. Define a new indicator cEDP as an offloading reliability measurement indicator based on total energy consumption, total delay, and privacy exposure risk, and optimize the new indicator cEDP to obtain an optimized new indicator cEDP, specifically: S31. Define a new indicator cEDP, namely: cEDP i =f(t i ,AND i ,For i ) Among them, cEDP i represents the new index cEDP of the ith vehicle, f(·) represents the combination function, t i represents the task T generated by the i-th vehicle i The total delay in computing offloading, E i represents the task T generated by the i-th vehicle i In calculating the total energy consumption during unloading, per i represents the task T generated by the i-th vehicle i Privacy exposure risks during computational offloading; S32, calculating the total delay and total energy consumption of the task generated by the vehicle when calculating unloading, that is: in, represents the task T generated by the i-th vehicle i Delays incurred during local processing, represents the task T generated by the i-th vehicle i The delay caused during the transmission phase, represents the task T generated by the i-th vehicle i The delay caused by the task data calculation stage, represents the task T generated by the i-th vehicle i Energy consumption during the local processing stage, represents the task T generated by the i-th vehicle i The energy consumption during the transmission phase, represents the task T generated by the i-th vehicle i Energy consumption generated during the task data calculation phase; S33. The privacy exposure risk of the tasks generated by the computing vehicle during computing offloading, namely: Where log represents the logarithmic function, e represents the exponential function, and Frs i represents the task T generated by the i-th vehicle i The freshness of the task, represents the task T generated by the i-th vehicle i Data safety factor using the kth uninstallation method, S mid represents the median size of all tasks generated by the vehicle; S34, obtaining an optimized new indicator cEDP according to the calculated total delay and total energy consumption of the task generated by the vehicle when the task is calculated and unloaded, and the privacy exposure risk of the task generated by the vehicle when the task is calculated and unloaded; S4. The new optimization index cEDP is used to optimize the offloading methods of the designed two types of tasks, and a decentralized offloading optimization problem is constructed; S5. The game theory and self-learning method of reliable privacy-preserving offloading based on autonomous learning are used to solve the constructed decentralized offloading optimization problem and obtain the optimal offloading matrix.

2. The reliable unloading optimization method based on vehicle edge computing and network based on self-learning according to claim 1 is characterized in that: Step S1 specifically includes: S11, the physical network layer includes roadside units and vehicles, using roadside units to exchange information with vehicles and using vehicles to generate tasks; S12, the offloading edge layer includes an offloading edge network, and the offloading edge network includes a roadside server and a vehicle edge server, wherein the roadside server is a roadside unit equipped with a powerful computing and storage module, and provides task offloading services for vehicles through vehicle-to-infrastructure communication, and each vehicle can act as a vehicle edge server to assist other vehicles in achieving task offloading; S13, the blockchain network layer includes all authorized roadside units, wherein the blockchain network layer is used to prevent the optimal unloading matrix from being tampered with; S14. Based on the physical network layer, the offloading edge layer and the blockchain network layer, a three-layer offloading model of vehicle edge computing and network is constructed, which includes the physical network layer, the offloading edge layer and the blockchain network layer.

3. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 2 is characterized in that: In step S2, the vehicle in the physical network layer is used to generate tasks, and according to the safety-related coefficients in the tasks, two types of task offloading methods are designed to obtain an offloading matrix. The specific process for realizing task offloading is as follows: Design a method for offloading tasks that are strongly related to security, specifically: If the safety relevance coefficient of the task generated by the i-th vehicle is SR i =1, then the task T generated by the i-th vehicle i For security-related tasks, namely, through vehicle positioning, vehicle navigation and road condition report generation, and based on the three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, three offloading methods for security-related tasks are designed, namely, task offloading through local processing 1 , using fully homomorphic encryption to offload tasks to the vehicle edge server m 2 And use fully homomorphic encryption to offload tasks to roadside servers m 3 ; Design a method for unloading weakly security-related tasks, specifically: If the safety relevance coefficient of the task generated by the i-th vehicle is SR i = 0, then the task T generated by the i-th vehicle i For security weakly related tasks, two offloading methods for security weakly related tasks are designed based on the constructed three-layer offloading model of vehicle edge computing and network including physical network layer, offloading edge layer and blockchain network layer, that is, offloading tasks to roadside servers m 4 , after encryption using fully homomorphic encryption, the task is offloaded to the roadside server m 5 ; According to the designed unloading method of strongly safety-related tasks and the unloading method of weak safety-related tasks, the unloading matrix of the task set T generated by the vehicle is obtained, namely: Where m represents the unloading matrix of the task set T generated by the vehicle, Indicates that the task T1 generated by the first vehicle adopts the mth 1 Uninstall methods: Indicates that the task T1 generated by the first vehicle adopts the mth 5 Uninstall methods: represents the task T generated by the nth vehicle n Use the mth 1 Uninstall methods: represents the task T generated by the nth vehicle n Use the mth 5 Uninstall methods: represents the task T generated by the i-th vehicle i Use the kth uninstall method.

4. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 3 is characterized in that: The new indicator cEDP optimized in step S3 is: Among them, cEDP′ i represents the new optimization index cEDP of the i-th vehicle, λ1, λ2, λ3 represent weighted parameters, represents the task T generated by the i-th vehicle i The minimum total delay when computing offloads, represents the task T generated by the i-th vehicle i The maximum total delay when calculating offload, represents the task T generated by the i-th vehicle i In calculating the minimum total energy consumption during unloading, represents the task T generated by the i-th vehicle i When calculating the maximum total energy consumption during unloading, represents the task T generated by the i-th vehicle i Minimum privacy exposure risk during computation offloading, represents the task T generated by the i-th vehicle i The biggest privacy exposure risk during computational offloading.

5. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 4 is characterized in that: Step S4 specifically includes: S41, based on the new optimized indicator cEDP, obtain the mth 1 New indicators for vehicles when unloading The specific process is: First, when using the mth 1 When the unloading method is used, the total delay of the tasks generated by the vehicle during the calculation of unloading is calculated, that is: Among them, l means local, Both indicate the use of the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i The total delay in computing offload, f l Indicates the local processing capability of the vehicle; Secondly, when using the mth 1 When the unloading method is used, the local power consumption and total energy consumption of the tasks generated by the vehicle during the calculation of unloading are calculated, that is: in, Indicates the use of the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i Local power consumption during computation offloading, V represents the vehicle edge server, k l,v represents the local power factor, Both indicate that when the mth 1 When the unloading method is used, the task T generated by the i-th vehicle i When calculating the total energy consumption during unloading; Finally, according to And the optimized new indicator cEDP, get the mth 1 The new indicators of the vehicle when unloading the method are: in, Indicates the use of the mth 1 The new index of the i-th vehicle when the unloading method is used; S42, based on the optimized new indicator cEDP, obtain the mth 2 New indicators for vehicles when unloading The specific process is: First, assuming that the encryption time and data expansion of the ciphertext are proportional to the data size, the delay incurred by the task generated by the vehicle in the local processing stage is calculated, that is: Among them, τ represents the time factor consumed by encryption; Calculate the energy consumption of the tasks generated by the vehicle in the local processing stage, that is: in, represents the task T generated by the i-th vehicle i Energy consumption at the local processing stage; Calculate the amount of data added to the tasks generated by the vehicle after full homomorphic encryption, that is: S′ i =ξS i Among them, S′ i represents the task T generated by the i-th vehicle after fully homomorphic encryption i The increased data size, ξ represents the amplification factor of the increased data size after encryption; Secondly, in the transmission phase involving the vehicle edge server, the transmission rate of the tasks generated by the vehicle on the IoV channel is calculated, that is: in, represents the task T generated by the i-th vehicle i The transmission rate on the IoV channel, L represents the total number of IoV communication licensed channels, B represents the bandwidth of each licensed channel, log2 represents the logarithmic function with base 2, P represents the transmission power of the vehicle, h represents the channel attenuation coefficient, ω0 represents the white Gaussian noise power of the vehicle, θ represents the path loss exponent, represents the distance between the i-th vehicle and the vehicle edge server; Assuming that the vehicle remains stationary while performing edge computing, the latency and energy consumption of the tasks generated by the vehicle during the transmission phase are calculated, namely: in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Delays during the transmission phase, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Energy consumption during the transmission phase; Then, in the task data calculation phase involving the vehicle edge server, assuming that the number of computing cores of the vehicle edge server is M V , the processing power of each core of the vehicle edge server is f V , calculate the delay of the task generated by the vehicle in the task data calculation stage, that is: in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Delays in the task data calculation phase; Calculate the power consumption and energy consumption of the tasks generated by the vehicle during the task data calculation phase, namely: in, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Power consumption during the task data calculation phase, Indicates the use of the mth 2 When the unloading method is used, the task T generated by the i-th vehicle i Energy consumption during the task data calculation phase; Finally, according to Get adopted mth 2 The total delay and total energy consumption of the offloading method are: in, Indicates the use of the mth 2 Total delay for each uninstallation method, Indicates the use of the mth 2 Total energy consumption when using the unloading method; according to And the optimized new indicator cEDP, get the mth 2 The new indicators of the vehicle when unloading the method are: in, Indicates the use of the mth 2 The new index of the i-th vehicle when the unloading method is used; S43, based on the optimized new indicator cEDP, obtain the mth 3 New guidance for vehicles when unloading With the mth 5 New indicators for vehicles when unloading The specific process is: First, in the transmission phase of the roadside server, the transmission rate, delay, and energy consumption of the tasks generated by the vehicle on the vehicle-to-infrastructure channel are calculated, namely: in, represents the task T generated by the i-th vehicle i The transmission rate on the vehicle-to-infrastructure channel, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Delay on the vehicle-to-infrastructure channel, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Energy consumption on the vehicle-to-infrastructure channel, represents the distance between the i-th vehicle and the roadside server; Secondly, in the task data calculation stage of the roadside server, assuming that the number of computing cores of the roadside server is M R , the processing power of each core of the roadside server is f R , calculate the delay of the task generated by the vehicle in the task data calculation stage of the roadside server, that is: in, Indicates the use of the mth 3 The i-th vehicle generates task T when the unloading method i Delay in the task data calculation phase at the roadside server; Calculate the power consumption and energy consumption of the tasks generated by the vehicle during the task data calculation phase, namely: in, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i The power consumption in the task data calculation phase, k R represents the power factor of the roadside server, Indicates the use of the mth 3 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase; according to Get adopted mth 3 The total delay and total energy consumption of the offloading method are: in, Indicates the use of the mth 3 Total delay for each uninstallation method, Indicates the use of the mth 3 Total energy consumption when using the unloading method; Finally, according to And the optimized new indicator cEDP, get the mth 3 The new indicators of the vehicle when unloading the method are: in, Indicates the use of the mth 5 The new index of the i-th vehicle when the unloading method is used; Similarly, we get the mth 5 The new indicators of the vehicle when unloading the method are: in, Indicates the use of the mth 5 The new index of the i-th vehicle when the unloading method is Indicates the use of the mth 5 Total delay for each uninstallation method, Indicates the use of the mth 5 The total energy consumption of the unloading method is Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Delay on the vehicle-to-infrastructure channel, Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Energy consumption on the vehicle-to-infrastructure channel, Indicates the use of the mth 5 The i-th vehicle generates task T when the unloading method i The delay in the task data calculation phase of the roadside server, Indicates the use of the mth 5 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase; S44, based on the new optimized indicator cEDP, obtain the mth 4 New indicators for vehicles when unloading The specific process is: First, due to the use of the mth 4 The delay and energy consumption of this offloading method only occur in the transmission stage and the task data calculation stage, so the calculation adopts the mth 4 The total delay and total energy consumption of the offloading method are: in, Indicates the use of the mth 4 Total delay for each uninstallation method, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Delays during the transmission phase, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Delays in the task data calculation phase, Indicates the use of the mth 4 The total energy consumption of the unloading method is Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the transmission phase, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i Energy consumption during the task data calculation phase; Secondly, the privacy exposure risk of the tasks generated by the vehicle is calculated, namely: in, Indicates the use of the mth 4 The task T generated by the i-th vehicle when the unloading method is i privacy exposure risks; Finally, according to And the optimized new indicator cEDP, get the mth 4 The new indicators of the vehicle when unloading the method are: in, Indicates the use of the mth 4 The new index of the i-th vehicle when the unloading method is used; S45, according to Get the coupling index cEDP of the task generated by the vehicle i (m i ,m -i ),Right now: Among them, cEDP i (m i ,m -i ) represents the task T generated by the i-th vehicle i Coupling index, m i represents the task T generated by the i-th vehicle i The unloading method matrix, m -i Indicates division by m i The offloading method matrix of the remaining tasks, Indicates intermediate temporary variables, which have no practical meaning. m1, m i-1 、m i+1 、m n Respectively represent the unloading method matrix of the tasks generated by the 1st, i-1, i+1, and nth vehicles, if represents the judgment condition, represents the task T generated by the i-th vehicle i Use the mth 1 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 2 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 3 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 4 Uninstall methods: represents the task T generated by the i-th vehicle i Use the mth 5 Uninstallation methods: S46, coupling index cEDP of the task generated by the vehicle i (m i ,m -i ), construct the decentralized unloading optimization problem, namely: Among them, min means to find the minimum value, st means to satisfy it, C1, C2, C3, and C4 represent different constraints respectively. Indicates any symbol, T max represents the maximum total delay of the tasks generated by the vehicle when computing offloading, per max Represents the maximum privacy exposure risk of tasks generated by vehicles when computing is offloaded.

6. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 5 is characterized in that: Step S5 specifically includes: S51. According to the constructed decentralized offloading optimization problem, a decentralized privacy-preserving offloading game model is constructed, namely: Where G represents the decentralized privacy-preserving offloading game model, cEDP i (m) represents the common cost set of all participants, cEDP1(m1,m -1 ) represents the coupling index of the task T1 generated by the first vehicle, cEDP2(m2,m -2 ) represents the coupling index of task T2 generated by the second vehicle, cEDP n (m n ,m -n ) represents the task T generated by the nth vehicle n The coupling index of S52. Introduce the potential function to transform the constructed decentralized privacy protection offloading game model into a potential game model, that is: Among them, Φ(m i ) represents the potential game model, η、n chnl 、n comp Both represent intermediate temporary variables and have no practical significance. represents the task T generated by the i-th vehicle i The delay of using the kth offloading method, represents the task T generated by the i-th vehicle i Energy consumption of the kth unloading method, represents the task T generated by the qth vehicle i Use the kth unloading method; S53. The potential game model is solved by using game theory and self-learning method based on reliable privacy protection offloading to obtain the optimal offloading matrix. The specific process is as follows: First, initialize the record memory M of the i-th vehicle i , where the vehicle’s recording memory is used to record the unloading methods and coupling indicators of other tasks; Secondly, determine whether the safety relevance coefficient of the task generated by the i-th vehicle is SR i =1, if so, the optimal unloading method for the task generated by the i-th vehicle is initialized as Otherwise, the optimal offloading method for the task generated by the i-th vehicle is initialized as Then, according to the unloading matrix m of the task set T generated by the initial vehicle, the new index of the initial vehicle is calculated And enter the self-learning stage to obtain the best unloading matrix, which is: Initialize the number of iterations ite to ite = 1, and iterate the loop. Generate a random number when no update request message RTU is broadcast. For parallel vehicles, an iterative loop is performed, according to the record memory M of the i-th vehicle i Extract the unloading matrix m of the task set T generated by the vehicle and determine whether the safety correlation coefficient of the task generated by the i-th vehicle is SR i =1, if so, then according to as well as calculate as well as And select the new index of the smallest vehicle as Will The corresponding uninstall method is set to Otherwise, according to calculate At the same time, the new index of the smallest vehicle is selected as Will The corresponding uninstall method is set to judge If so, the random back-off competition method, the minimum cEDP decision method, or the hybrid decision method is used to update the opportunity, otherwise the optimal unloading matrix is ​​maintained in the next iteration. constant; The update is performed according to the received update request message RTU until the update request message RTU is not broadcasted; Finally, the unloading matrix of the final vehicle-generated task set obtained in the last iteration is returned and uploaded to the regional chain network after reaching a consensus to obtain the optimal unloading matrix.

7. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 6 is characterized in that: The process of updating opportunities using the random back-off competition method is: First, using the retreat time δ of the i-th vehicle i Initialize the timer; Secondly, when there is δ i ←δ i -1, judge δ i = = 0, if yes, the competition is successful and the update request message RTU is broadcast, otherwise according to the update request message RTU received from another vehicle, and the received update request message RTU is used to update Among them, ← indicates assignment, and == indicates equality comparison.

8. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 7 is characterized in that: The process of updating opportunities using the minimum cEDP decision method is: First, the monitoring timer is initialized using the monitoring time ψ, and the update request message RTU is broadcasted; Secondly, before the countdown of the monitoring time ψ ends, monitor and store the update request messages RTU of other vehicles; Finally, compare the RTU in each update request message And according to The corresponding smallest Update the unloading matrix m of the task set T generated by the vehicle and 9. The reliable unloading optimization method based on self-learning vehicle edge computing and network according to claim 8 is characterized in that: The process of updating opportunities using a hybrid decision-making approach is: First, judge If yes, the random back-off competition method is used to update the chance, otherwise the minimum cEDP decision method is used to update the chance, and Ω represents the judgment threshold.

Images