The regional edge cloud data centers support numerous latency-sensitive applications, including autonomous driving, Augmented Reality/Virtual Reality (AR/VR), smart grids. However, dynamic workloads often trigger spurious Virtual Machine (VM) migrations that degrade real-time service guarantees. To address this challenge, we propose a lightweight, proactive VM scheduling framework based on a hierarchical structure (HLFVM). By combining logical region partitioning with low-complexity migration algorithms, it enables rapid localized migration decisions. First, by leveraging the Enhanced Harris Hawk Optimization (EHHO) to optimize the parameters of the Long Short Term Memory (LSTM) model, we propose a Load Forecast method based on the EHHO-LSTM (LFEL) model. This algorithm accurately predicts multiple resource loads on PMs and reduces the lag in migration decision-making. Then, we propose the zone-aware LFEL-based VM Migration (LFVM) algorithm, which includes PM status classification and migration selection mechanism. The migration selection mechanism chooses the VM destinations according to the cost function to expedite the migration decision. Numerous experiments have shown that the execution time of the LFVM algorithm is reduced by at least 70.4% compared to traditional algorithms, while VM migration time is improved by 5.7%. Concurrently, it achieves superior control over energy consumption and enhances resource utilization.

As a cutting-edge technology of low-altitude Artificial Intelligence of Things (AIoT), autonomous aerial vehicle object detection significantly enhances the surveillance services capabilities of low-altitude AIoT. However, the difficulty of object detection is exacerbated by the high proportion of small and obscure objects in the captured images. To address the mentioned challenges, we present an efficient multiband infrared small object detection approach for low-altitude intelligent surveillance services. First, we propose the multiband infrared image fusion algorithm based on cascade-GAN (MIF-CGAN), which produces fused images with high information entropy and high contrast. Then, the Transformer-based multiscale dense small object detection (MsDSOD) algorithm is proposed. The algorithm consists of the global-local object detection (G-LOD) network, the object dense area extraction (O-DAE) module, and the weighted boxes fusion (WBF) module. It extracts small objects features at different scales from infrared images and fuses the global and local detection results to accurately identify small objects in dense scenes. Furthermore, compared to the traditional algorithms, the mean average precision (mAP) of MsDSOD is improved by 0.80% and the average precision in small object detection$({\mathrm { AP}}_{s})$is improved by 0.72%. The proposed algorithm is optimally suited to deal with complex scenes with dense small objects and background occlusion.

Artificial Intelligence of Things (AIoT) utilizes a combination of computing, storage, and networking resources to provide highly reliable and low-latency information services to the industrial production processes. However, with the increasing integration of numerous smart terminals into real-time sensing, autonomous decision-making, and precision manufacturing execution systems, the current task scheduling pattern appears to be insufficient to meet the latency requirements of computationally intensive tasks. To address the above challenge, this paper presents a collaborative edge-terminal task offloading scheme. First, the Task Backlog and Multi-slot Scheduling (TBMS) problem is converted from a long-term offloading problem to a single timeslot scheduling problem by Lyapunov optimization. Then, to simplify the problem, the single timeslot problem is decomposed into three subproblems: the local resource allocation problem, the server resource allocation problem, and the indicator weight selection problem. The two resource allocation problems are proved to be convex, which have been solved by using the Bisection method and the Karush-Kuhn-Tucker (KKT) method, respectively. For the indicator weight selection problem, we proposed the enhanced jumping spider optimization algorithm that integrates the elite opposition-based learning strategy. Extensive experiments show that the proposed algorithm can alleviate the computing pressure of the terminal device. Compared with the traditional methods, the offload system cost is effectively reduced by at least 58.8% and the average execution success rate is increased by at least 6%.

The convergence of transformative technologies, including the Internet of Things (IoT), Big Data, and Artificial Intelligence (AI), has driven private edge cloud systems to the forefront of research efforts. The access to massive terminals and the emergence of personalized services pose serious challenges for efficient resource management in power private edge cloud systems. To address the challenge of inequitable resource allocation in the private edge cloud, this work proposes an intelligent resource allocation strategy with a slicing and auction approach. By formalizing the resource allocation problem as a Mixed Integer Nonlinear Programming (MINLP) puzzle, the method transforms it into a hierarchical allocation challenge for Mobile Network Operators (MNOs), Mobile Virtual Network Operators (MVNOs), and power terminals. The proposed Multi-hop Progressive Auction Algorithm (MPAA) addresses the sliced resource allocation problem between MNOs and MVNOs. Furthermore, a Terminal Resource Allocation Strategy (TRAS) based on improved particle swarm optimization is proposed to solve the spectrum resource allocation problem between MVNOs and power terminals. Extensive simulation results show that the bidding overhead of MPAA is reduced by 6.12% and the average terminal satisfaction of TRAS is improved by about 1.3% compared to conventional methods, thus improving the utilization of wireless resources within the power AIoT.

As an important index of aero-engine, Remaining Useful Life (RUL) is the key content of prediction. Due to the good generation characteristics of Variational Auto-encoder (VAE) and Generation Adversarial Network (GAN) networks, this paper proposes a Health Index (HI) curve generation method based on VAE-GAN. After that, sensor sequence prediction is carried out through Bidirectional Long Short-Term Memory Network (BLSTM). The two networks are parallel, and then RUL prediction is carried out by synthesizing the data of the two networks. As far as the author knows, this is the first use of VAE-GAN in Prognostics Health Management (PHM). It is verified on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset. Finally, the results show that the VAE-GAN network is effective and superior in RUL prediction. At the same time, the proposed parallel network is superior to other RUL prediction methods by generating HI curves.