This study presents YOLOv11n-GhostLite, an innovative lightweight deep learning architecture optimized for real-time localization of photovoltaic (PV) faults in electroluminescence (EL) images, specifically designed for edge deployment. A Deep Convolutional Generative Adversarial Network (DCGAN)-based synthetic augmentation pipeline is presented to address the issues of class imbalance and limited resource availability, generating high-fidelity, class-conditional EL images that include realistic banding artifacts. This method enhances the representation of minority defect categories by more than 150%, elevating the mean Average Precision (mAP@50) by 4% and decreasing false negatives by 5%. The proposed model incorporates GhostConv for efficient early feature extraction, C3k2 residual blocks for deep representation learning, GhostSPPF for multi-scale context aggregation, C2PSA attention for adaptive feature refinement, and an anchor-free detection head, achieving high performance with only 2.34 million parameters and 6.2 GFLOPs. Detailed experiments on two benchmark datasets PVEL-AD and PV Multi-Defect exhibit the model’s efficacy, attaining 97.2% mAP@50 on PVEL-AD, and 96.4% mAP@50 on PV Multi-Defect, outperforming larger models in both accuracy and speed. The model is further deployed on a Google Coral Edge TPU, demonstrating its real-time functionality with minimal power consumption (~2W) and suitable latency for drone-based solar inspections. YOLOv11n-GhostLite’s integration of efficient architecture and data-driven augmentation renders it an effective solution for scalable, real-time photovoltaic fault detection in resource-limited settings.

The increase in Supervisory Control and Data Acquisition (SCADA) solutions in industrial systems has made these systems more vulnerable to cyberattacks due to insecure communication protocols. This paper proposes a privacy-aware intrusion detection framework that reduces these risks. The proposed framework implements, through machine learning, anomaly detection methods, classification techniques and feature selection methodologies, integrating these approaches to optimize the identification of potential cyber threats, improve model accuracy, and enhance computational efficiency in SCADA systems. Anomaly detection is strengthened through Isolation Forests, time-series analysis, and Fourier-based burst detection while dynamically adapting to concept drift adaptation with Kullback-Leibler divergence. It also utilizes important feature optimization methods, implemented in multiple ways, such as SHAP, Recursive Feature Elimination (RFE), and Principal Component Analysis (PCA). Furthermore, the proposed framework includes a federated learning-based scheme that utilizes differential privacy and homomorphic encryption to ensure the privacy and integrity of the data to enhance model interpretability and efficiency with feature ranking to provide insights into attack patterns and anomaly characteristics. Defences against adversarial attacks use FGSM-based training, feature smoothing, and ensemble-based defences, to reduce susceptibility to evasion tactics. Honeypots and automated IP firewalls enhance attack profiling and proactive mitigation of incidents. The experimental results demonstrate the framework’s operational capability and effectiveness for SCADA network security, achieving 99.29% accuracy, 94.8% recall and 4.1% false positive rates.

The increasing frequency and sophistication of cyberattacks on smart grid infrastructures have raised critical concerns over data integrity, operational resilience, and real-time response capabilities. This study introduces a unified cybersecurity framework for cyber-physical power systems that integrate high-performance anomaly detection with provably secure cryptographic protection. A comprehensive dataset, built upon the IEEE 24-bus test system, includes a diverse set of operational states and five classes of false data injection attacks (FDIAs), including stealthy and replay-based intrusions. To accurately detect both common and sophisticated threats, we implement a suite of supervised learning models—RF, MLP, and Decision Trees—alongside an ensemble strategy termed MVCC, which achieves up to 99.90% accuracy in binary classification and 99.88% in multiclass settings. For defense at the data level, we deploy a two-tier encryption architecture combining AES-GCM (for confidentiality and authenticity) with RSA-OAEP (for secure key management), demonstrating strong resilience against standard attack models (COA, KPA, CPA, CCA) and achieving nearly uniform ciphertext entropy (7.99 bits/byte). The system’s real-time applicability is validated through the deployment of the RF classifier on a PYNQ-Z2 FPGA platform, attaining sub-second inference latency. Further, unsupervised (DBSCAN, K-Means) and temporal (LSTM) models are incorporated for stealthy anomaly localization and early threat prediction. This work presents a scalable, interpretable, and cryptographically secure solution for protecting next-generation smart grids against evolving data integrity threats.