This study addresses the critical issues of rigid energy use and insufficient demand-side responsiveness in office buildings’ Variable Refrigerant Flow (VRF) systems under complex summer conditions. Existing research lacks fine-grained characterisation of short-term load fluctuations and often fails to accurately couple energy efficiency with humidity-adapted thermal comfort. To fill this gap, this paper proposes an integrated Model Predictive Control (MPC) framework driven by load characteristic identification and a novel hybrid prediction model. First, based on actual hourly metered data (683,280 records), K-means clustering was employed to identify three typical load patterns, pinpointing short-term peak loads in core office zones as the primary target for flexible regulation. Second, a high-precision GS-DBO-ELM prediction model—integrating Grid Search and Dung Beetle Optimisation—was developed to capture the nonlinear dynamics of VRF energy consumption and Predicted Mean Vote (PMV). The model achieved an R2 of 0.99 with relative errors constrained within ±5%. Finally, a multi-objective MPC strategy, solved via an improved Artificial Hummingbird Algorithm (HAGSAHA) and weighted by the Analytic Hierarchy Process (AHP), was implemented to dynamically adjust zone-level temperature setpoints. Results demonstrate that the proposed MPC strategy reduces daily cooling energy consumption by 7.95–10.69% and peak loads by 15.3%, while maintaining strict thermal comfort (PMV within ±0.5). Under a time-of-use pricing mechanism, the flexible scheduling strategy achieved a 12.37% total electricity reduction and a 9.54% reduction in operating costs. This work provides a highly replicable, climate-tailored solution for low-carbon, flexible energy management in public buildings.
Han et al. (Wed,) studied this question.