Age of information (AoI) is an important metric of information timeliness, which determines digital twin (DT) consistency and energy management precision. However, AoI guarantee in the time-averaged sense is unreliable to avoid the occurrence of extreme event. In this paper, we propose a novel information timeliness metric named ultra-low AoI (ULAoI). Compared with AoI, ULAoI further considers the occurrence of extreme event and higher-order statistical characteristics of excess AoI value. Multi-dimensional resources of power internet of things (PIoT) are jointly allocated to achieve ULAoI guarantee from the perspective of sensing-communication-control integration. ULAoI-DT-Prioritized deep Q network (DQN) is proposed to achieve coordinated resource allocation by approximating unobservable information with the assistance of ULAoI-DT, and preventing DQN training from using samples with large AoI based on ULAoI-induced priority. Simulation results demonstrate the superior performance of the proposed algorithm in global loss function, ULAoI guarantee, and energy management optimality.

Dispatching and control information freshness conducts an important impact on the training accuracy of distributed energy dispatching and control model. Poor information freshness will increase the loss function of the training model, and reduce the reliability and economy of dispatching and control. Simplified power internet of things can provide plug-and-play and multi- mode fusion communication support, but it still faces challenges of the coupling of model training and data transmission as well as the difficulty in guaranteeing dispatching and control information freshness. In this paper, a semi-distributed federated learning- based framework for dispatching and control model training decision-making is proposed, and a dispatChing and control informAtion fReshness-aware batch size Optimization aLgorithm (CAROL) is presented. CAROL leverages deep Q network and dispatching and control information freshness awareness to learn the batch size optimization strategy. CAROL can minimize model loss function while guaranteeing long-term dispatching and control information freshness constraints. Compared with existing feder- ated learning algorithms, CAROL achieves superior performance in global loss function and information freshness.

Multi-mode power internet of things (PIoT) provides spatio-temporal coverage for low-carbon operation in smart park through combining various communication media. Heterogeneous resources are dynamically and intelligently managed to improve resource utilization and achieve anti-eavesdropping. However, resource management in multi-mode power IoT confronts challenges such as the mutual contradiction in joint communication and security quality of service (QoS) guarantee and the inadaptability to low-carbon services. In this paper, we propose an Adaptive learNing-based secure and enerGy-awarE resource management aLgorithm (ANGEL) to optimize multi-mode channel selection and power splitting for artificial noise (AN)-based anti-eavesdropping. Based on deep actor-critic (DAC) and “win or learn fast (WoLF)” mechanism, ANGEL can realize multi-attribute QoS guarantee, adaptive resource management, and security enhancement. Simulation results demonstrate its superior performance in energy consumption, secrecy capacity, and adaptability to differentiated low-carbon services.

The low-carbon operation of smart park requires to deploy massive internet of things (IoT) devices to provide real-time monitoring and control services. Digital twin (DT) provides accurate guidance for communication network resource management in low-carbon smart park by establishing a digital representation of physical entities. Facing the strict requirements of DT on delay and accuracy, as well as the constraints of access priority and energy consumption, we propose a federated learning-based DT framework and a Latency-awarE diGital twIn assisted resOurce maNagement algorithm (LEGION). LEGION can achieve a well tradeoff between delay and accuracy performances under the long-term constraints of access priority and energy consumption. Compared with existing algorithms, LEGION has superior performance in average iteration delay, DT loss function, energy consumption, and access priority deficit.

The combination of air–ground integrated power Internet of Things (AGI-PIoT) and cloud-edge-end collaboration enables flexible coverage and real-time data processing. However, how to achieve intelligent cloud-edge-end collaboration in AGI-PIoT faces several challenges such as dynamics of aerial networks, coupling of resource allocation in multiple layers, timescales, and dimensions, incomplete information, and dimensionality curse. In this article, we propose a FEderated Deep rEinforcement leaRning-based multi-lAyer multi-Timescale multi-dImensional resOurce allocatioN algorithm (FEDERATION). The multilayer multitimescale multidimensional resource allocation problem is decomposed into three subproblems based on Lyapunov optimization. For the subproblem of joint task offloading and power control, a federated deep actor-critic-based semidistributed algorithm is developed. The subproblem of admission control is solved by quadratic programming. The third subproblem is addressed through smooth approximation and Lagrange dual decomposition. Simulation results indicate that FEDERATION outperforms existing algorithms in queuing delay, energy consumption, and convergence.

Digital twin (DT) provides accurate guidance for multidimensional resource scheduling in 5G edge computing-empowered distribution grids by establishing a digital representation of the physical entities. In this article, we address the critical challenges of DT construction and DT-assisted resource scheduling such as low accuracy, large iteration delay, and security threats. We propose a federated learning-based DT framework and present a Secure and lAtency-aware dIgital twin assisted resource scheduliNg algoriThm (SAINT). SAINT achieves low-latency, accurate, and secure DT by jointly optimizing its total iteration delay and loss function, and leveraging abnormal model recognition (AMR). SAINT enables intelligent resource scheduling by using DT to improve the learning performance of deep Q-learning. SAINT supports access priority and energy consumption awareness due to the consideration of long-term constraints. Compared with state-of-the-art algorithms, SAINT has superior performance in cumulative iteration delay, DT loss function, energy consumption, and access priority deficit.

Collaborative vehicular network is a key enabler to meet the stringent communication and computing requirements of user vehicles (UVs). A UV dynamically optimizes task offloading by exploiting its collaborations with edge servers and vehicular fog servers (VFSs). However, the optimization of task offloading in highly dynamic collaborative vehicular networks faces several challenges such as queuing delay guaranteeing, incomplete information, and dimensionality curse. In this paper, a Deep Reinforcement lEarning-based queue-Aware task offloading algorithM named DREAM is proposed to maximize the throughput of the UVs while satisfying the long-term queuing delay constraints in a best-effort way. Compared with existing task offloading algorithms, DREAM achieves superior performance in throughput, convergence, and queuing delay.

In the Internet of Health Things based e-Health paradigm, a large number of computational-intensive tasks have to be offloaded from resource-limited IoHT devices to proximal powerful edge servers to reduce latency and improve energy efficiency. However, the lack of global state information (GSI), the ultra-reliable and low-latency communication (URLLC) constraints, and the adversarial competition among IoHT devices have imposed new challenges for task offloading optimization. In this paper, we formulate the task offloading problem as an adversarial multi-armed bandit (MAB) problem. In addition to the average-based performance metrics, bound violation probability of queuing delays and statistical properties of excess values are employed to characterize URLLC constraints. Then, we propose an energy-aware and URLLC-aware Task Offloading scheme based on the exponential-weight algorithm for exploration and exploitation (EXP3) named UTO-EXP3. Guaranteed performance with a bounded deviation can be achieved by UTO-EXP3 based on only local information. The effectiveness and reliability of UTO-EXP3 are validated through simulation results.