Robust Cyberattack Detection in Smart Grids Using Chaotic Feature Selection and Belief-Guided Deep Learning Models

• A state-of-the-art intrusion detection system (IDS) designed specifically for Smart Grid Internet of Things networks is presented in this work. • For better cyberattack detection, it employs a hybrid approach that combines an ideal machine learning classification method with a Chaotic Emperor Penguin (ChaEP) feature selector • The technology is designed to manage the difficulties of detecting cyberattacks in large-scale and dynamic Smart Grid IoT networks. • The focus is on enhancing the accuracy, efficacy, and responsiveness of intrusion detection by selecting the most relevant features from large volumes of data. • The ChaEP algorithm uses emperor penguin foraging behavior and chaotic dynamics to select the most relevant features, addressing high-dimensional data challenges. • The system achieves computational economy while preserving IDS performance by lowering the dimensionality of the data, which results in efficient cyberattack categorization and system optimization. Ensuring the security of Internet of Things (IoT) integrated innovative grid systems is more essential in present times, due to its digital infrastructure and rising energy system defenses against cyberattacks. Smart grids, which establish numerous connections, open security vulnerabilities for attackers who threaten power grid reliability, safety, and stability. Current studies in this domain encounter significant difficulties because of sensor data’s high dimensions, multiple types of attacks, and the requirements for precise and efficient detection systems. Standard detection methods face challenges in achieving accurate results efficiently, which creates insufficient performance during operational usage. Numerous detection models lack sufficient capabilities to handle the changing nature of cyberattacks because their traditional detection mechanisms become less effective. This work develops two innovative solutions to improve cyberattack detection capabilities in smart grid systems. The chaos-driven emperorpenguin (ChaEP) feature selector uses chaotic systems to control emperor penguin optimization to perform efficient feature selection. The feature selector method decreases data set dimensions and keeps only the most essential attributes, leading to better model efficiency and performance metrics. The proposed belief-GANNet (BGNet) combines deep learning components of generative adversarial networks (GANs) with belief propagation to create a hybrid model that strengthens attack detection performance. The novelty of this research work is the invention of the ChaEP feature selector and BGNet detection model, two new models to surmount the challenges of conventional methods. The design uses a flexible system that detects existing and completely new cyberattack forms. Experimental testing over benchmark datasets showed the superiority of proposed algorithms: BGNet scored 99.2% detection accuracy, an imperceptible false alarm ratio of merely 0.38%, and 99.2% and 98.9% precision and recall rates, respectively, far superior to state-of-the-art benchmarks. These tests confirm that the proposed ChaEP and BGNet algorithms are a strong and scalable approach toward improving the cyber-resilience of contemporary smart grid infrastructures.