Design of Reliable IoT Systems With Deep Learning to Support Resilient Demand Side Management in Smart Grids Against Adversarial Attacks
Mahmoud Elsisi, Chun‐Lien Su, Mahmoud N. Ali
Abstract
Demand side management (DSM) has become one of the major concerns of the smart grids to cope with the penetration of renewable energy. The availability of new communication technologies can enhance the resilient operation of smart grids for DSM. However, the false injection data and adversarial attacks that are against the operation of signal analysis models represent the biggest challenge against the resiliency of energy management operations by deploying these technologies in the smart grids. In this regard, this article proposes a new reliable industrial Internet of Things (IoT) architecture and deep convolution neural network (CNN) with an image processing strategy based on continuous wavelet transform (CWT) for DSM and resilience operation of energy management. The main contribution involved in this article includes establishing a real-time signal processing model and developing an industrial IoT platform with CWT-based CNN; verifying the IoT architecture with different levels of adversarial attacks; providing a cybersecurity analysis for the smart buildings with DSM considering the device-level attacks, and developing defense strategies from the aspects of detection, mitigation, and prevention. In addition, the proposed deep CNN is designed with proper hyperparameters to counter adversarial attacks. The proposed CWT-based CNN performed the highest accuracy compared with different machine learning and deep learning models. Various testing scenarios are presented and executed to ensure and demonstrate the performance and robustness of the proposed method under different levels of adversarial attacks.