Search

Due to the substantial interdependencies between water and energy infrastructures, the water-energy nexus (WEN) is regarded as an appealing focal point for potential malicious intruders, particularly in the context of smart cities. Driven by the increasing vulnerability of these critical infrastructures to sophisticated cyber-attacks and the lack of comprehensive protection strategies, this study aims to address these challenges. This paper introduces a pioneering deep learning-driven framework for detecting, locating, and mitigating stealthy false data injection attacks (FDIAs) targeting electrical and water networks within the WEN. The proposed framework, which leverages two lightweight deep learning models, is designed to accurately detect and locate cyber-attacks in real-time while effectively mitigating their impact, thereby enhancing the operational resilience and security of the WEN. These models undergo training and testing using a wide range of challenging cyber-attack scenarios, including Pulse, Random, Ramp, and Scaling attacks, to ascertain their efficacy and resilience. The detection model accurately identifies and locates the cyber-attacks, achieving a remarkable accuracy rate of 99.73% for identifying attacks aimed at the electrical network and 96.66% for detecting attacks directed at the water network. After detecting the cyber-attack, the mitigation model reconstructs manipulated measurements to ensure the continued operation of the WEN and prevent the occurrence of disruptions or failures. Remarkably, the model attains an outstanding recovery success rate of 98.98% for electrical measurements and 87.64% for water measurements. Additionally, it effectively reduces the operational cost increase (OCI) due to cyber-attacks from 2.19% to 0.0048% for the electrical network and from 5.17% to 0.34% for the water network. The dataset employed in this study is characterized by its imbalanced nature, with cyber-attack instances being infrequent compared to normal instances. Consequently, the adoption of balanced affinity loss, designed to tackle the dataset’s inherent imbalance, achieves superior performance, with an accuracy of 99.73% and an area under the curve (AUC) of 99.86%. The proposed framework offers a robust and efficient solution for safeguarding the WEN against cyber-attacks, making it a valuable tool for enhancing the resilience and security of smart city infrastructures.

Image-based 3D reconstruction is a challenging task that involves inferring the 3D shape of an object or scene from a set of input images. Learning-based methods have gained attention for their ability to directly estimate 3D shapes. This review paper focuses on state-of-the-art techniques for 3D reconstruction, including the generation of novel, unseen views. An overview of recent developments in the Gaussian Splatting method is provided, covering input types, model structures, output representations, and training strategies. Unresolved challenges and future directions are also discussed. Given the rapid progress in this domain and the numerous opportunities for enhancing 3D reconstruction methods, a comprehensive examination of algorithms appears essential. Consequently, this study offers a thorough overview of the latest advancements in Gaussian Splatting.

The integration of Artificial Intelligence (AI) services within the framework of Software-as-a-Service (SaaS) cloud architecture has significantly permeated our everyday routines. These AI services diverge from traditional applications by offering a more personalized user experience. That is why a predefined instance configuration is not an optimal approach for these applications. The challenge is further compounded by the unpredictable nature of user demand, making optimal resource allocation to these instances a complex task. This paper introduces an innovative algorithm, termed Deep-Hill, designed to enhance cloud resource allocation through precise prediction of SaaS instance configurations. It is a combination of a 5-layer Deep Neural Network (DNN) and a Hill-Climbing algorithm. This unique approach classifies the instance configuration in one of the five classes with 96.33% accuracy, 90.83% precision, 90.96% recall, and 90.86% F1-score. On average, it reduces the number of active hosts by four, contributing to 13.33% less power consumption. The remarkable performance of the Deep-Hill algorithm underscores its potential to set a new benchmark in the optimization of SaaS cloud resources. It paves the way for more cost-effective SaaS applications, marking a significant step forward in the evolution of cloud computing.

Performance modelling of a deep learning application is essential to improve and quantify the efficiency of the model framework. However, existing performance models are mostly case-specific, with limited capability for the new deep learning frameworks/applications. In this paper, we propose a generic performance model of an application in a distributed environment with a generic expression of the application execution time that considers the influence of both intrinsic factors/operations (e.g. algorithmic parameters/internal operations) and extrinsic scaling factors (e.g. the number of processors, data chunks and batch size). We formulate it as a global optimisation problem and solve it using regularisation on a cost function and differential evolution algorithm to find the best-fit values of the constants in the generic expression to match the experimentally determined computation time. We have evaluated the proposed model on three deep learning frameworks (i.e., TensorFlow, MXnet, and Pytorch). The experimental results show that the proposed model can provide accurate performance predictions and interpretability. In addition, the proposed work can be applied to any distributed deep neural network without instrumenting the code and provides insight into the factors affecting performance and scalability.

Electrical equipment is prone to different types of Partial Discharge (PD) failures that are varying between minor and severe level. In this paper, Three developed models for Convolution Neural Network (CNN) are proposed to detect and classify four different partial discharge types which are arcing, corona discharge, tracking, looseness as well as healthy equipment situation. Notably, the resulting models exhibited an impressive overall accuracy of more than 94%, which is particularly significant considering the inherent presence of noise in the real-world samples obtained as representative field failures. These findings underscore the robustness and effectiveness of the CNN models in accurately identifying PDs, despite the intricate challenges associated with real-world data.