Indoor CO2 concentration in educational buildings serves as both an indicator of ventilation adequacy and a direct driver of HVAC energy demand. Yet, these two dimensions are rarely examined within a single integrated framework. This study proposes such a framework, combining four coupled components: a whole-building EnergyPlus thermal model of a Mechanical Engineering faculty building (benchmarked against published data rather than calibrated with measured energy data), a physics-based mass balance CO2 model driven by hourly occupancy and ventilation outputs from EnergyPlus, a long short-term memory (LSTM) network for predictive CO2 forecasting, and a systematic six-category multiscenario energy analysis. The mass balance model reproduces peak classroom CO2 concentrations of 1200–1300 ppm during full occupancy, exceeding the ASHRAE 1000-ppm threshold, and recovers to near-outdoor levels within 20–40 min after occupancy ends. The LSTM model delivers 1-h-ahead CO2 forecasts with RMSE = 42 ppm and R2 = 0.94, outperforming persistence (R2 = 0.71), linear regression (R2 = 0.88), and gradient boosting (R2 = 0.92) baselines, with its largest advantage occurring during occupancy transitions. The six-category energy scenario analysis identifies HVAC schedule optimization (17%–22% annual savings), advanced double-pane LowE/LowER glazing (~10%), and reduced window area (~7%) as the most impactful interventions. Under the idealized IdealLoadsAirSystem representation, the fully optimized configuration cuts annual building energy consumption by 34% (30% in winter, 40% in summer), with ±5–7 percentage points of combined uncertainty; real HVAC systems would likely realize smaller savings due to equipment efficiency and control dynamics. Overall, the framework establishes a conceptual and methodological basis for future LSTMdriven demand-controlled ventilation (DCV) strategies in educational buildings. However, no closed-loop control algorithm was implemented or tested here.
