To enable resource-constrained wireless devices (WDs) to process the computation-intensive and latency-sensitive computation tasks, the wireless powered mobile edge computing (WP-MEC) network has been proposed as a promising approach. Incorporating mobile cloud computing (CC) in the WP-MEC network, we investigate the wireless powered mobile edge-CC (WP-MECC) network, where the WDs first harvest energy from a hybrid access point (HAP), and then consume the harvested energy to compute the tasks locally, offload them to the HAP for computation, or offload them to the cloud server (CS) via the relaying of the HAP. To pursue fairness among the WDs, we minimize the maximum task completion time (TCT) of WDs by jointly optimizing the time resources, computing mode selection, and computation resources. We prove the minimization problem is NP-hard. To tackle the problem, we decompose it into the subproblem and top problem, and propose an alternate optimization-based resource allocation and the computing mode selection (ARACM) algorithm with low computational complexity, which achieves a comparable performance with the exhaustive search method in terms of the minimal maximum TCT of WDs. Moreover, we propose a deep reinforcement learning (DRL)-based resource allocation and the computing mode selection (DRACM) algorithm with less execution latency than the ARACM algorithm. Numerical results show that the two proposed algorithms achieve satisfactory performance in terms of the minimal maximum TCT of WDs and execution latency.

Abstract Intelligent computing and social optimization techniques have been used to manage mobile edge (ME) environments for pervasive services. However, the service demand is the prime cause for requiring flawless information management. This article proposes a coherent information management (CIM) process for resolving the convex optimization problems by considering the flaws in raw information management. The information management problem for pervasiveness is defined as a nonlinear problem for user‐to‐information availability. According to the CIM, pervasive convergence is identified for the available information. Furthermore, convolutional neural learning identifies unavailable or utilized information between the service provider and the end‐user layers. This identification improves the training rate for convergence improvement and information coherency. Therefore, the nonlinear pervasiveness is distributed between different service providers, improving efficiency. Therefore, the proposed optimization method achieves a 7.4% high admittance ratio, 8.28% high utilization, 25.25% less allocation time, and 24.4% less waiting time for different availability ratios for different information availability ratios.