Low Earth orbit (LEO) satellites nowadays play a pivotal role in collecting images for the Earth observation. However, the images collected by satellites are possibly tremendous, which causes challenges in dealing with the satellite images. Those challenges include: 1) the unrealistic of transmitting those massive image data to the ground station for centralized analysis because of restricted satellite communication bandwidth and the data privacy issue, and 2) satellite data may be non-independent and identically distributed (non-IID). In this paper, we propose a prototype-guided federated knowledge distillation (Pro-FedKD) approach in an LEO Satellite-high altitude platform (HAP) system, which is designed based on self-knowledge distillation (SKD), federated prototype learning (FedProto) and federated learning (FL). Owing to the adoption of FL, the first challenge can be handled since FL does not require data to leave the local side. To cope with the second challenge, SKD and FedProto are employed. In addition, both model aggregation and prototype aggregation are employed on a pre-defined HAP. To enhance the effectiveness, a top-$N$model aggregation mechanism is proposed, in which among all models,$N$local models that can achieve the top$N$maximum accuracies over the validation dataset of the pre-defined HAP will be selected for aggregation. Experiments demonstrate the error rate gained by the proposed Pro-FedKD method is separately 3.76×, 3.17×, 1.55×, and 1.18× smaller than FedExP, MOON, FedProto, and pFedSD over the EuroSAT dataset, demonstrating a significant reduction. The proposed method also exhibits preeminence in other datasets.

Low Earth orbit (LEO) satellites are capable of gathering abundant Earth observation data (EOD) to enable different Internet of Things (IoT) applications. However, to accomplish an effective EOD processing mechanism, it is imperative to investigate: 1) the challenge of processing the observed data without transmitting those large-size data to the ground because the connection between the satellites and the ground stations is intermittent, and 2) the challenge of processing the non-independent and identically distributed (non-IID) satellite data. In this paper, to cope with those challenges, we propose an orbit-based spectral clustering-assisted clustered federated self-knowledge distillation (OSC-FSKD) approach for each orbit of an LEO satellite constellation, which retains the advantage of FL that the observed data does not need to be sent to the ground. Specifically, we introduce normalized Laplacian-based spectral clustering (NLSC) into federated learning (FL) to create clustered FL in each round to address the challenge resulting from non-IID data. Particularly, NLSC is adopted to dynamically group clients into several clusters based on cosine similarities calculated by model updates. In addition, self-knowledge distillation is utilized to construct each local client, where the most recent updated local model is used to guide current local model training. Experiments demonstrate that the observation accuracy obtained by the proposed method is separately$1. 01\times, 2.15\times, 1.10\times$, and$1.03\times$higher than that of pFedSD, FedProx, FedAU, and FedALA approaches using the SAT4 dataset. The proposed method also shows superiority when using other datasets.

In this paper, cyber-attack prevention for the prosumer-based electric vehicle (EV) charging stations (EVCSs) is investigated, which covers two aspects: 1) cyber-attack detection on prosumers’ network traffic (NT) data, and 2) cyber-attack intervention. To establish an effective prevention mechanism, several challenges need to be tackled, for instance, the NT data per prosumer may be non-independent and identically distributed (non-IID), and the boundary between benign and malicious traffic becomes blurred. To this end, we propose an edge-assisted federated prototype knowledge distillation (E-FPKD) approach, where each client is deployed on a dedicated local edge server (DLES) and can report its availability for joining the federated learning (FL) process. Prior to the E-FPKD approach, to enhance accuracy, the Pearson Correlation Coefficient is adopted for feature selection. Regarding the proposed E-FPKD approach, we integrate the knowledge distillation and prototype aggregation technique into FL to deal with the non-IID challenge. To address the boundary issue, instead of directly calculating the distance between benign and malicious traffic, we consider maximizing the overall detection correctness of all prosumers (ODC), which can mitigate the computational cost compared with the former way. After detection, a rule-based method will be triggered at each DLES for cyber-attack intervention. Experimental analysis demonstrates that the proposed E-FPKD can achieve the largest ODC on NSL-KDD, UNSW-NB15, and IoTID20 datasets in both binary and multi-class classification, compared with baselines. For instance, the ODC for IoTID20 obtained via the proposed method is separately 0.3782% and 4.4471% greater than FedProto and FedAU in multi-class classification.

Unmanned aerial vehicles (UAVs) have been conceived as an available solution to substitute terrestrial base stations (TBSs) to provide downloading services for user equipment (i.e. mobile devices) that have difficulty communicating directly with TBSs. However, the mobility of user equipment (UE) and the random nature of the number of UE will cause several challenges including 1) the hardness of determining optimal user association and UAV resource (i.e., bandwidth and transmit power) allocation decision, 2) the burden of network function maintenance owing to the necessity of shutting down the entire system. Therefore, in this article, joint user association and resource allocation are designed for a software-defined network (SDN)-adopted leaderless softwarized UAV network, where each UAV is regarded as a flying SDN controller to enhance the control ability of the considered network. The purpose is to maximize energy efficiency (EE) with satisfying the quality of service (QoS). To this end, a joint method based on hierarchical agglomerative clustering (HAGC) and multi-agent deep deterministic policy gradient (MADDPG) is proposed. Specifically, the HAGC approach is utilized to determine the optimal MDs association with UAVs. Afterward, MADDPG approach is leveraged to obtain the best policy for resource allocation, aiming to achieve the maximum EE. Finally, the effectiveness of the proposed method is confirmed by the evaluation results.

Unmanned aerial vehicles (UAVs) nowadays are developing rapidly for various applications such as UAV taxis and delivery drones. However, the limited battery energy restricts the flight distance of the UAVs. Thus, urban prosumers equipped with drone recharge stations are introduced to provide charging services for the UAVs. In this article, first, a day-ahead energy scheduling problem for UAV charging-enabled urban prosumers is studied, where the objective is to maximize the overall energy satisfaction of the prosumers with ensuring the Quality of Service (QoS) of the charged UAVs. Specifically, to deal with the considered problem, we decompose it into two stages: 1) the day-ahead energy requirement data prediction stage and 2) energy scheduling stage per prosumer. Thus, second, a joint method based on hierarchical federated learning (HFL) on long short-term memory (LSTM) architecture (HFL-LSTM) and stochastic game-based multi-agent double deep$Q$-learning (MADDQN) with community agent-independent approach is proposed. In particular, the HFL-LSTM approach is leveraged to forecast each prosumer’s energy requirement data without centralized collecting local prosumers’ data such that to protect data privacy. Then, the stochastic game is adopted to analyze the formulated problem, aiming to find the Nash equilibrium (NE) strategy. Afterward, MADDQN with a community agent-independent method is utilized to achieve the best energy scheduling strategy per prosumer. Finally, the experimental results demonstrate the superiority of the proposed joint method that can achieve the lowest mean squared error with the value of 0.0152 and the highest energy satisfaction$(36388)$achieved by the NE policy compared with the benchmarks.

Federated learning is a distributed learning system that addresses the distributed difficulty such as communication overhead and private information in machine learning while maintaining high performance. However, the distributed learners have to dedicate their resources to improving the global model, which is not likely to happen voluntarily. This motivated us to design an incentive mechanism for users (data owners) to actively participate in the FL processes. In this paper, we consider multiple co-existing FL service providers (FLSPs) with the need to train their models and multiple data owners (DOs) that can offer that service. In the system, DO, and FLSP will submit their cost and valuation values to the cloud platform. Based on this information, we formulate an optimization problem that aims to maximize the social welfare under the nonnegative utility constraint and maximum gain of FLSPs. Then, we propose a heuristic algorithm, Binary Whale Optimization Algorithm (B-WOA), that can solve our formulated NP-hard problem in polynomial time. Finally, numerical results are shown to demonstrate the effectiveness of our proposed algorithm. Moreover, we also compare the performance of our proposed algorithm with Hungarian and greedy algorithms.

Prosumers nowadays are capable of consuming and generating renewable energy along with providing charging services for public electric vehicles (EVs) through EV support equipment (EVSE). However, the energy demand of prosumers and EVs as well as the renewable energy generation of prosumers have uncertain nature, which causes difficulty for each prosumer to purchase the proper energy at a lower price in advance. Thus, it is paramount important to do energy procurement prediction (EPP) for each prosumer. Nevertheless, submitting data from each prosumer to a centralized server for EPP will result in communication delay and need to consume a huge amount of network bandwidth and energy. Therefore, in this paper, a clustering-based serverless edge computing-assisted federated learning (FL) approach is proposed for EPP, where the objective is to minimize the Huber loss between the predicted and the real value per prosumer. In particular, firstly, normalized Laplacian-based spectral clustering is leveraged to group the prosumers with a similar energy procurement pattern to solve the problem of biased energy procurement forecast caused by updating the model among all the clients. Secondly, long short-term memory (LSTM) in the federated learning setting is utilized to train the global model of each clustered group, where the model aggregation occurs in the serverless edge computing ability-enhanced local edge server with the best performance. The evaluation results demonstrate the proposed method can achieve the lowest Huber loss compared with the baseline methods.

Recently, the deployment of electric vehicles supply equipment (EVSE) and its market is expanding rapidly to support the massive penetration of electric vehicles (EVs). However, to accomplish an effective EV charging mechanism for urban prosumer communities, it is imperative to tackle the challenges of distinct energy generation among the communities, dependency of the total purchasable energy price of each EV based on the distance between EV and EVSE, and extreme uncertainty among the energy demand and generation. Therefore, in this paper, the problem of EV charging of urban prosumer communities is studied. In particular, a joint optimization problem is proposed to maximize both the social welfare and EV charging achieved rate of the considered urban prosumer communities. Consequently, the formulated problem is decomposed into 1) truthful double auction problem for determining the unit price and winners by maximizing social welfare, and 2) EV auction losers charging problem for improving EVs charging achieved rate by purchasing energy from the power grid. Then the breakeven-based double auction (BDA) mechanism is proposed to find the unit price and EV winners’ for charging. Sequentially, a multi-agent deep reinforcement learning-based asynchronous advantage actor-critic algorithm with a long short-term memory layer (A3C-LSTM) is adopted to achieve the optimal grid energy buying decision for ensuring the charging of the losers. Finally, the experimental results demonstrate the efficacy of the proposed model that can increase the number of EV charging up to 57.31%, and prosumer communities have gained 86.04% of their income compared to baseline methods.

The recent flourish of diversified distributed energy resources (DERs) such as generators, consumers, and prosumers brings indispensable cybersecurity challenges for the smart grid controller. Therefore, to assure a resilient smart grid operation, in this paper, we study the problem of continuous-time consensus policy-based DERs control mechanism for the smart grid controller. In particular, we propose an intelligent grid shepherd for the smart grid controller in the power grid framework. That can autonomously detect the abnormal behavior of the received status message from each DER and apply control decisions into the smart grid controller. To do this, first, we propose a continuous-time Markov decision process problem by formulating a resilient control system for the intelligent grid shepherd. Second, we design a data-informed policy-based model-free reinforcement learning framework to find the optimal consensus policy for each DER control decision (i.e., remain connected with the main grid or disconnected). Thus, we devise a distributed energy resources control algorithm for the intelligent grid shepherd. Particularly, we design an advantage actor-critic scheme under the continuous-time domain with the shared neural network mechanism. Finally, experimental results show the efficiency of the proposed intelligent grid shepherd in terms of accuracy and robustness towards a resilient DERs control.