Integrated renewable energy and demand-side management for low-carbon commercial buildings in tropical climates: A matched-configuration benchmark with predictive control and a multi-agent DRL architecture

Abstract

Commercial buildings account for a major share of global energy use, particularly in tropical regions where high cooling demand, intermittent renewables, and grid instability complicate effective management. A persistent confound in the renewable-enabled BEMS literature is that proposed advanced controllers are compared against rule-based controllers operating on different physical assets, conflating the contribution of the renewable hardware with that of the control strategy. We address this by introducing a matched-configuration baseline (RBC-Full-RE) operating on the same renewable asset set as the proposed system: a 250 kWp solar PV array, 500 kW h battery, 200 m2 solar thermal, 800 kW heat pump, and 30 kW biogas CHP. Against this matched baseline we evaluate a 12-step receding-horizon model predictive controller (MPC-Full-RE) on a 10-story, 12 500 m2 commercial office building in Chittagong, Bangladesh, using a calibrated 3R2C thermal model and a synthesised weather year matching the local climatology. Full-year simulation gives a clean decomposition of savings: renewable hardware contributes 28.2% reduction in annual grid electricity (Baseline 2,030 MWh/yr → RBC-Full-RE 1,457 MWh/yr) under identical rule-based control, and MPC contributes a further 2.8 percentage points on the same hardware (1,411 MWh/yr), totalling 30.5% relative to the all-electric baseline. Peak demand falls 34.3% (612 kW → 402 kW); thermal comfort improves from 94.8% to 97.8% of occupied hours within the ASHRAE 55 Cat. II band. Simple payback is 12.1 years at 2024 pricing. We additionally specify a Safe Multi-Agent DRL controller with MPC safety filtering (SMA-DRL-MPC); the MPC-Full-RE result establishes a principled lower bound on what the proposed DRL extension must improve upon. Simulation code is released for reproducibility.