Electrical grids generate, transport, distribute and deliver electrical power to consumers through a complex Critical Infrastructure which progressively shifted from an air-gaped to a connected architecture. Specifically, Smart Substations are important parts of Smart Grids, providing switching, transforming, monitoring, metering and protection functions to offer a safe, efficient and reliable distribution of electrical power to consumers. The evolution of electrical power grids was closely followed by the digitization of all its parts and improvements in communication and computing infrastructures, leading to an evolution towards digital smart substations with improved connectivity. However, connected smart substations are exposed to cyber threats which can result in blackouts and faults which may propagate in a chain reaction and damage electrical appliances connected across the electrical grid. This work organizes and offers a comprehensive review of architectural, communications and cybersecurity standards for smart substations, complemented by a threat landscape analysis and the presentation of a Defense-in-Depth strategy blueprint. Furthermore, this work examines several defense mechanisms documented in the literature, existing datasets, testbeds and evaluation methodologies, identifying the most relevant open issues which may guide and inspire future research work.

As AI models expand in size, it has become increasingly challenging to deploy federated learning (FL) on resource-constrained edge devices. To tackle this issue, split federated learning (SFL) has emerged as an FL framework with reduced workload on edge devices via model splitting; it has received extensive attention from the research community in recent years. Nevertheless, most prior works on SFL focus only on a two-tier architecture without harnessing multi-tier cloud-edge computing resources. In this paper, we intend to analyze and optimize the learning performance of SFL under multi-tier systems. Specifically, we propose the hierarchical SFL (HSFL) framework and derive its convergence bound. Based on the theoretical results, we formulate a joint optimization problem for model splitting (MS) and model aggregation (MA). To solve this rather hard problem, we then decompose it into MS and MA sub-problems that can be solved via an iterative descending algorithm. Simulation results demonstrate that the tailored algorithm can effectively optimize MS and MA in multi-tier systems and significantly outperform existing schemes.

Network slicing is a critical technology for fifth-generation (5G) networks, owing to its merits in meeting the diversified requirements of users. Effective resource allocation for network slicing in Radio Access Networks (RAN) is still challenging owing to dynamic service requirements. Therein, automatic resource allocation based on environmental changes is of significant importance for network slicing. In this study, we used deep reinforcement learning (DRL) to allocate resources for network slicing in a RAN with the aid of massive multiple-input multiple-output (MIMO). The DRL agent interacts with the environment to execute autonomous resource allocation. We considered a two-level scheduling framework that aims to maximize the quality of experience (QoE) and spectrum efficiency (SE) of slices. The proposed algorithm can find a near-optimal solution. We used the standard DRL advantage actor-critic (A2C) algorithm to implement upper-level inter-slice bandwidth resource allocation that considers service traffic dynamics in a large timescale. Lower-level scheduling is a mixed-integer stochastic optimization problem with several constraints. We combined the proportional fair scheduling algorithm and the water filling algorithm to perform resource block (RB) and power allocation in a small timescale. The results show that the QoE and SE of all slices using the A2C algorithm achieved a significant performance improvement over the other algorithms. The efficiency of the proposed method was supported by the simulation results.

In this paper, we consider the joint power and modulation optimization in two-user non-orthogonal multiple access (NOMA) channels based on error-rate performance criterion and finite-alphabet inputs. Specifically, with joint maximum-likelihood detection being employed at each user's receiver, the users' transmission power levels are jointly determined with the 4-QAM constellation of one of the users being rotated such that the probability of dominant pairwise error event in the NOMA channel is minimized. Towards solving the joint power and modulation problem, our results are twofold. Firstly, with M antennas at the receiver, closed-form expressions for the optimal angles as a function of M, the channel coefficients and the users' power levels are derived. Secondly, closed-form expressions are derived for the joint power allocation with optimal rotation by casting the joint optimization as a piecewise convex optimization problem over its sub-domains. Using the proposed joint optimization, the error-rate performance of NOMA improves significantly and it outperforms traditional NOMA power allocation methods for the selected transmission rates. The proposed approach thus represents a rather distinct approach in NOMA optimization, and it is applicable in the uplink NOMA where users share the same time/frequency channel or in the downlink NOMA where users are multiplexed in the power domain.

Many countries across the globe face the serious issue of traffic congestion. This paper presents a low-cost graph-based traffic forecasting and congestion detection framework using online images from multiple cameras. The advantage of using a graph neural network (GNN) for traffic forecasting and detection is that it represents the traffic network in a natural way. This framework requires only images from surveillance cameras without any other sensors. It converts the online images into two types of data: traffic volume and image-based traffic occupancy. A clustering-based graph construction method is proposed to build a graph based on the traffic network. For traffic forecasting, multiple models, including statistical models and deep graph convolutional neural networks (GCNs), are used and compared using the extracted data. The framework uses logistic regression to determine the threshold of traffic congestion. In the experiment, we found that the Decoupled Dynamic Spatial-Temporal Graph Neural Network (D2STGNN) model achieved the best performance on the collected dataset. We also propose a threshold-based method for detecting traffic congestion using traffic volume and image-based traffic occupancy. This framework provides a low-cost solution for traffic forecasting and congestion detection when only surveillance images are available.