As for the MEC system, the combination of D2D and MEC enhances the computing ability of the system and shortens transmission distance where the nearby devices provide close support for simple tasks, while computing-intensive tasks can further obtain sufficient computing and storage resources at the cost of long-distance transmission in the MEC server. There are many types of research on the D2D-assisted MEC system. Ouamri et al. [12] primarily focus on the application of Device-to-Device (D2D) communication in the context of Unmanned Aerial Vehicles (UAVs), aiming to improve energy efficiency. Lingjun Pu et al. [13] put forward a new mobile task-unloading framework based on D2D cooperation, which realizes the dynamic sharing of computing and communication resources through the control assistance of network operators. Dan Wu et al. [14] put forward a dynamic distributed-resource-sharing scheme, which is applied to the unified framework of general D2D communication by jointly optimizing mode selection, resource allocation and power allocation. Yinghui He et al. [15] integrate D2D communications with MEC to improve the computation capacity of the cellular networks, aiming to maximize the number of devices supported by the cellular networks with the constraints of both communication and computation resources. Y Dai et al. [16] develop the interlay mode as a unique D2D model to maximize the system sum rate in the non-orthogonal multiple access (NOMA) cellular networks. However, tasks are offloaded to the MEC server or devices nearby as above. Almost none of them consider the circumstance that mobile devices cannot connect to the MEC server directly in the case of no Internet access or poor signal.

Although the D2D-assisted MEC system solves the problems of limited computing resources and power of mobile devices by reasonable offloading for large-scale and complex tasks, shortening the task-offloading process and reducing the offloading-task data are effective ways to reduce energy consumption further. Therefore, researchers have begun to combine edge caching [17,18] with the MEC system. Reference [19] proposes a D2D-cache strategy using multi-agent reinforcement learning. The D2D-cache problem is formalized as a bandit problem with multiple agents and weapons and Q-learning is used to learn how to coordinate cache decisions. Reference [20] investigates a joint pushing and caching policy in a general mobile-edge-computing (MEC) network with multi-user and multi-cast data based on hierarchical-reinforcement learning. Reference [21] investigates the collaborative-caching problem in the edge-computing environment to minimize the system cost and proposes an online algorithm called CEDC-O to solve this problem. Reference [22] studies edge caching in fog-computing networks, where a capacity-aware edge-caching framework is proposed by considering both the limited fog-cache capacity and the connectivity capacity of base stations. Reference [23] studies the unique problem of caching fairness in edge-computing environments. Reference [24] formulates the Edge Data Caching (EDC) problem as a constrained-optimization problem to minimize the caching-data cost and maximize the reduction in service latency.

Regarding edge caching, most researchers pay more attention to content caching [25,26]. Yu, G et al. [27] propose a content-caching strategy based on mobility prediction and joint-user prefetching. J. Guo et al. [28] propose a novel context-aware object-detection method based on edge-cloud cooperation. Y. Dai et al. [29] integrate DRL and permit blockchain into vehicular networks for intelligent and secure content caching. Wu W et al. [30] propose a content-based D2D cooperative edge-caching strategy. The content can be cached in the user equipment or the surrounding small base stations according to popularity. However, except for content, the task’s size, popularity and complexity are also important factors that deserve consideration.

In addition, the complexity of tasks further accelerates the energy consumption of mobile devices, causing battery consumption to be a critical factor in restricting the development of mobile devices. Meanwhile, executing tasks with low power will prolong the execution time, which cannot meet the quality of service requirements of users. To this end, it is necessary to manage resources reasonably to reduce the energy consumption of mobile devices. Reference [31] focuses on resource allocation in a downlink 5G tri-sectorial cell for non-orthogonal multiple access (NOMA) systems. Reference [32] proposes a proportional fair-based scheduler algorithm that incorporates Signal Interference Noise Ratio (SINR) compensation to address the challenge of resource allocation in cellular networks. Reference [33] studies resource allocation for a multi-user mobile-edge-computation-offloading system under infinite or finite cloud-computation capacity. Reference [34] investigates the latency-minimization problem in a multi-user time-division multiple access mobile-edge-computation-offloading system with joint communication- and computation-resource allocation. Reference [35] studies the problem of joint task offloading and resource allocation to maximize the users’ task-offloading gains. Reference [36] proposes an integrated framework for computation offloading and interference management in wireless cellular networks with MEC, which formulates the computation-offloading decision, physical-resource-block (PRB) allocation. Reference [37] designs an iterative heuristic MEC resource-allocation algorithm to make the offloading decision dynamically. Reference [38] optimizes the joint caching, offloading and time-allocation policy to minimize the weighted-sum energy consumption subject to the caching and deadline constraints. Resource management varies in the MEC system, but few of them jointly optimize computing, communication and caching of the offloading process, which is an all-sided optimization.

With the help of D2D cooperation, devices that cannot access the Internet choose one of the helper devices as a relay to execute and transmit tasks to the MEC server. Combined with edge caching and jointly optimized resource allocation, the model can improve resource utilization, shorten execution time and reduce energy consumption.