<p>The rapid integration of Distributed Energy Resources (DERs) including photovoltaic (PV) systems, energy storage units, hybrid electric vehicles (HEVs), and smart home energy management systems (SHEMS) has increased the vulnerability of smart grids, particularly microgrids, to cyber-attacks and data integrity threats. As modern microgrids operate as Cyber-Physical Systems (CPS), ensuring secure and trustworthy data exchange is essential. This paper proposes a security-enhancement framework that combines a Self-Modified Pelican Optimization Algorithm (SM-POA) with a Spatio-TemporalNet (STNet) hybrid deep learning model for multi-level cyber-threat prediction and defense. The system integrates a six-level authentication architecture to safeguard Blockchain-based data transactions and maintain operational integrity. SM-POA optimizes feature selection and decision-making for precise security assessment, while STNet processes spatiotemporal information from microgrid components for high-accuracy forecasting. The system assesses network traffic, physical layer signaling, and protocol level interactions to provide multilevel security prediction. A Blockchain layer enables decentralized and tamper-proof data storage. On a simulated 33-bus radial microgrid test system, the performance analysis of the SM-POA + STNet model shows an accuracy rate of 96.3%, precision of 94.8%, recall of 95.2%, and F1-Score of 95.0%, evaluated under identical preprocessing and training conditions against six baseline models. Thus, it is effective in proactive attack detection and energy consumption reduction. In summary, the combination of the two techniques improves the robustness of smart grids, cybersecurity, and efficiency, although the framework remains untested against advanced persistent threats and zero-day exploits, which we identify as limitations for future work.</p>