Sarcouncil Journal of Engineering and Computer Sciences

Abstract: The trend of digitalisation in solar power plants has increased efficiency and operational intelligence but on the other hand exposed these plants to new multi-faceted cyber threats. Traditional intrusion detection approaches are not capable of coping with the high volume and velocity of interconnected device related data across solar networks. To address this gap, we propose a hybrid Artificial Intelligence (AI) architecture which combines trained Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks and edge analytics to improve cybersecurity resilience in the solar plant infrastructures. The proposed approach is based on combining CNN for spatial feature extraction with LSTM for temporal sequence learning, which permits prediction and classification of cyber threats attack e.g., FDI (false data injection), DoS (denial-of-service) and spoofing in early stages. By deploying edge analytics, latency is reduced through processing data locally, which guarantees that anomaly detection and mitigation occur in real time. The experimental results show that the detection accuracy is 97.2% and false positive rate is less than 3%, which is superior to some traditional machine learning methods. The results also demonstrate the promise of hybrid AI-aware, edge-capable architectures to evolve solar plant cybersecurity to a threat-adaptive, anticipatory and self-defending architecture —preserving the resiliency and sustainability of future renewable energy networks.