A streamlined universal method for VRF unit configuration in the early design stage

Abstract

This study presents a streamlined and practical methodology for optimizing variable refrigerant flow (VRF) system configurations during the early design stage, using only basic weather data and building type as input. Recognizing the substantial impact of HVAC systems on building energy use and emissions, the proposed strategy addresses the limitations of conventional selection approaches that overlook building-specific and climate-specific factors. An artificial neural network (ANN) model, trained on 2,470 simulation cases, encompassing 13 building types across 10 locations and spanning all 19 ASHRAE climate zones, was developed to predict hourly part load ratios (PLRs), achieving high accuracy with R² values of 0.85 (cooling) and 0.78 (heating), and RMSE values of 0.068 and 0.035, respectively. These predictions were used in conjunction with a curve-based approach to estimate annual energy consumption across various configuration scenarios. Results reveal that optimal VRF configurations differ significantly depending on building type and climate zone, with energy use varying by an average of 4.0% between best and worst cases. Compared to conventional selection methods, the proposed approach achieved an average energy savings of 2.9%, with savings reaching up to 7.0% in specific scenarios such as hospitals and climate zone 3C. The methodology offers a widely applicable, data-efficient solution that supports improved energy performance and emissions reduction in diverse building sectors.