Digital twin and AI-driven smart adaptive walls for energy, comfort and carbon optimization in intelligent green buildings

Buildings account for nearly 40% of global energy use and carbon emissions, yet conventional static envelopes lack the capacity to respond to shifting weather, occupancy, or grid carbon signals. This paper presents a Digital Twin-Artificial Intelligence-Smart Adaptive Wall (DT-AI-SAW) framework that couples materially adaptive envelopes with safety-aware learning control for joint optimization of energy, comfort, and carbon. The architecture places a DT in the control loop and pairs Model Predictive Control (MPC) with a Soft Actor-Critic (SAC) reinforcement-learning agent operating behind a safety shield. A unified Energy-Comfort-Carbon (ECC) objective drives decisions using time-varying grid carbon intensity. The SAW combines switchable optical transmittance, adjustable louvers, variable thermal resistance, and phase-change material (PCM) storage as coordinated control variables. Across Helsinki, Antalya, and Najran, including multi-day heatwave events, the framework delivered 36.3% energy reduction, 40.8% carbon reduction, 31.8% peak-demand reduction, and 22.3 percentage-point comfort gains over static-envelope baselines. The safety shield resolved 91.7% of constraint conflicts through projection, with zero default fallbacks. Ablation confirmed synergistic contributions from every component; pure RL achieved the lowest energy but incurred 560% more constraint violations than the hybrid design. Life-cycle carbon payback fell between 1.9 and 2.8 years, with 30-year cumulative emission cuts of 39.5-45.6%. These results show that adaptive envelopes paired with safety-aware intelligent control can simultaneously advance objectives that static designs force into trade-off.