The increasing penetration of renewable energy resources, electric vehicles, and intelligent grid infrastructures has introduced significant challenges in energy forecasting, grid regulation, and real-time power management due to renewable intermittency, climatic variability, and fluctuating demand. To address these issues, this research proposes the Variability Energy Regulation Grid Equilibrium (VERGE) framework, which integrates Climate-MambaFormer, Dynamic Spatio-Temporal Graph Transformer, and Safe Reinforcement Learning–Model Predictive Control (RL-MPC) into a unified energy-management architecture. Climate-MambaFormer performs weather-aware forecasting of electricity demand and renewable generation, while the graph transformer learns dynamic interactions among solar units, wind sources, battery storage systems, EV charging stations, load centers, and grid nodes. The Safe RL-MPC controller generates optimal energy-regulation actions while satisfying operational safety constraints. Experimental evaluation demonstrates that VERGE achieves a forecast accuracy of 99.1% and a frequency stability of 98.9%, outperforming existing state-of-the-art energy-management approaches. Statistical analyses including p-value testing, confidence interval analysis, and ANOVA validate the significance and reliability of the obtained results. The findings confirm that VERGE effectively improves renewable-energy utilization, reduces power imbalance, enhances grid stability, and supports adaptive energy regulation in modern smart-grid environments. In conclusion, VERGE provides an efficient, reliable, and scalable solution for intelligent energy forecasting, grid-equilibrium maintenance, and real-time renewable-energy management under dynamic operating conditions.