Sarcouncil Journal of Engineering and Computer Sciences

Abstract: With the accelerated trend of digitalization, solar power systems are integrated with Internet of Things (IoT) technology and major breakthroughs have been achieved into energy monitoring, automation, grid access and connection. But connectedness also leads to susceptibilities such as FDI (false data injection), DoS (denial-of- services) and spoofing attacks toward power systems due to cyber space. Conventional rule-based IDS are not adaptable and do not respond quickly to dynamic or emerging threat conditions. In this article, we proposed a novel DRL-based autonomous cyber defense framework for Smart Solar IoT Systems to achieve real-time online threat detection and self-healing from the system point of view. The framework uses a DQN with (Long Short-Term Memory) LSTM layers to encode the best defensive strategy and interact continually with the environment. The DRL agent adaptively adapts mitigation policies (such as to isolate compromisable nodes, or re-configure the communication routes) according to threat context and reward feedback. Experimental analysis on the simulated solar IoT datasets indicates that the proposed framework achieves 96.9% of detection accuracy with 42% lower response latency, and adapts well in identifying unknown attack patterns without re-training. These results highlight the promise of autonomous machine learning driven cybersecurity for protecting future generation energy infrastructures for renewable energy, and in particular towards resilient and intelligent solar energy ecosystems.