With the rapid development and convergence of the mobile Internet and the Internet of Things (IoT), computing-intensive and delay-sensitive IoT applications (APPs) are proliferating with an unprecedented speed in recent years. Mobile edge computing (MEC) and energy harvesting (EH) technologies can significantly improve the user experience by offloading computation tasks to edge-cloud servers as well as achieving green and durable operation. Traditional centralized strategies require precise information of system states, which may not be feasible in the era of big data and artificial intelligence. To this end, how to allocate limited edge-cloud computing resource on demand, and how to develop heterogeneous task offloading strategies with EH in a more flexible manner are remaining challenges. In this paper, we investigate an EH-enabled MEC offloading system, and propose an online distributed optimization algorithm based on game theory and perturbed Lyapunov optimization theory. The proposed algorithm works online and jointly determines heterogeneous task offloading, on-demand computing resource allocation, and battery energy management. Furthermore, to reduce the unnecessary communication overhead and improve the processing efficiency, an offloading pre-screening criterion is designed by balancing battery energy level, latency, and revenue. Extensive simulations are carried out to validate the effectiveness and rationality of the proposed approach.

In order to improve life mobile users' service experience, mobile cloud computing (MCC) is promoted, which can offload these compute-intensive applications to cloud. Although MCC can alleviate the burdens of Smart mobile devices (SMDs), it also aggravates the computing and storage overheads in cloud center and bandwidth overhead on wireless links for applications offloading. Therefore, we should carefully design the offloading policy to decrease these overheads while easing the burdens of SMDs. To this end, we investigate the offloading policy in heterogeneous wireless networks. In this paper, a queue is built to formulate the mobile users' workload offloading problem and Lyapunov optimization is used to make trade-off between the system offloading utility and the queue backlog. First, based on the deterministic WiFi connection, a Lagrangian optimization method is proposed to decide the optimal offloading workloads. Furthermore, considering the random WiFi connection durations, a multi-stage stochastic programming is adopted. The experimental results show the effectiveness of the Lagrangian optimization offloading method for deterministic WiFi connection and the multi-stage stochastic programming method for random WiFi connection.

Mobile edge computing (MEC) enables various services to be cached in close proximity to the user equipments (UEs), thereby reducing the service delay of many emerging applications. However, the limitation of storage, computation, and radio resources, the dynamics of the decentralized MEC environment, and the complex spatial relationships of service request types and wireless network states between edge nodes make it difficult to realize efficient edge computing services. To address these challenges, this paper integrates the digital twin (DT) technology with a multi-cell MEC network to study an intelligent cooperative task offloading and service caching scheme, aiming at maximizing a quality of services (QoE)-based system utility. Specifically, we first construct a digital twin edge network (DITEN) to reflect the physical MEC system in real-time and provide data for training. With the help of DT technology, it is easy to access data resources in the DITEN to improve the simulation ability and reduce the communication cost. Then, we propose a graph attention-based multi-agent reinforcement learning (GatMARL) algorithm to learn the optimal task offloading and service caching strategies in the DITEN. The GatMARL employs a graph attention-based value decomposition network to capture the potential spatial relationships between edge nodes to learn better attentive cooperation policy. Simulation results demonstrate that the proposed GatMARL algorithm exhibits an effective performance improvement compared with state-of-the-art benchmarks.

Accurate spatial-temporal traffic modeling and prediction play an important role in intelligent transportation systems (ITS). Recently, various deep learning methods such as graph convolutional networks (GCNs) and recurrent neural networks (RNNs) have been widely adopted in traffic prediction tasks to extract spatial-temporal dependencies based on a large volume of high-quality training data. However, there exist data scarcity problems in some transportation networks, and in these cases, the performance of traditional GCNs and RNNs based approaches will degrade sharply. To address this problem, this paper proposes an adversarial domain adaptation with spatial-temporal graph convolutional network (Ada-STGCN) model to predict traffic indicators for a data-scarce target road network by transferring the knowledge from a data-sufficient source road network. Specifically, Ada-STGCN first develops a spatial-temporal graph convolutional network that combines the GCN and gated recurrent unit (GRU) to extract spatial-temporal dependencies from source and target road networks. Then, the technique of adversarial domain adaptation is integrated with the spatial-temporal graph convolutional network to learn discriminative and domain-invariant features to facilitate knowledge transfer. Experimental results on the real-world traffic datasets in the traffic flow prediction task demonstrate that our model yields the best prediction performance compared to state-of-the-art baseline methods.

With the exponentially increasing number of mobile devices, crowdsensing has been a hot topic to use the available resource of neighbor mobile devices to perform sensing tasks cooperatively. However, there still remain three main obstacles to be solved in the practical system. First, since mobile devices are selfish and rational, it is natural to provide cooperation for sensing with a reasonable payment. Meanwhile, due to the arrival and departure of sensing tasks, resource should be allocated and released dynamically when sensing task comes or leaves. To this end, this paper designs a game theoretic approach based incentive mechanism to encourage the "best" neighbor mobile devices to share their own resource for sensing. Next, in order to adjust resource among mobile devices for the better crowdsensing response, a semi-truthful auction based task migration algorithm is proposed, which can guarantee the truthfulness of announced price of auctioneer. Moreover, taking into account the random movement of mobile devices resulting in the stochastic connection, we also use multi-stage stochastic decision to take posterior resource allocation to compensate for inaccurate prediction. The numerical results show the effectiveness and improvement of the proposed multi-stage stochastic programming based distributed game theoretic methodology (SPG) for crowdsensing.

With the rapid advancement of the Internet of Vehicles and artificial intelligence (AI) technologies, the cooperative intelligent transportation system (C-ITS) has drawn great attention in recent years. To provide an ultra-reliable, low-latency computation experience of C-ITS, computation offloading is deemed indispensable by working with edge-cloud servers. In this paper, we first investigate a distributed dynamic computation offloading model for multi-access edge computing (MEC) enabled C-ITS under a heterogeneous road network, in which the multiple and heterogeneous computing power sources cooperatively provide computation offloading services for vehicles. Considering the autonomous offloading manner of the vehicles, we formulate the task offloading and computing power allocation as a distributed Stackelberg game, where the MEC servers as the leader to allocate computing resources and manage local energy, and the vehicles as the followers to offload local computation task. Since the observable states in the game is incomplete, the problem of resolving the optimal strategies for each game player is modeled as a partially observable Markov decision process (POMDP) to maximize the long-term cumulative reward. Then we develop a computation offloading algorithm using Stackelberg game-based multi-agent deep deterministic policy gradient (SG-MADDPG), which uses a centralized training and decentralized execution method to learn the optimal computing power allocation and computation offloading policies. Finally, extensive simulations are carried out and show the rationality and effectiveness of the proposed algorithm.