University campuses represent complex microgrids with highly stochastic energy demands driven by fluctuating occupancy and diverse building functions. Traditional Building Management Systems (BMS) often operate on static, reactive schedules, leading to significant energy waste. This paper proposes a novel Digital Twin (DT) framework integrated with Long Short-Term Memory (LSTM) networks to enable proactive energy optimization. The system leverages an IoT sensor network to collect real-time environmental and electrical data, which is synchronized with a virtual replica of the campus building. The core of the DT is an ensemble of LSTM models tailored for different room types (e.g., lecture halls, laboratories), capable of capturing long-term temporal dependencies in energy consumption patterns. Experimental results conducted at a university facility demonstrate that the proposed LSTM model achieves a high prediction accuracy with a coefficient of determination (R²) of 0.97. By integrating these predictions into an automated HVAC and lighting control strategy, the system achieved a 22% reduction in total energy consumption compared to baseline operations. This research contributes a scalable, data-driven architecture for transitioning university campuses toward Sustainable Development Goals (SDGs) through intelligent energy management.