Benchmarking high performance HVAC Rule-Based controls with advanced intelligent Controllers: A case study in a Multi-Zone system in Modelica

Abstract

The design, commissioning, and retrofit of heating, ventilation, and air-conditioning (HVAC) control systems are crucially important for energy efficiency. However, designers and control contractors adopt ad-hoc control sequences based on diffused and fragmented information; therefore, most of the existing control sequences are diverse and sub-optimal. ASHRAE Guideline 36 (GDL36), High-performance Sequences of Operation (SOO) for HVAC Systems, was thus developed to provide standardized and high-performance rule-based HVAC control sequences with the main focus on maximizing energy efficiency. However, only limited studies verify the performance of these high-performance rules-based control and most of these studies focus on the single-zone systems. In this Modelica-based simulation study, the high-performance rule-based control sequences from GDL36 were compared to the state-of-the-art intelligent controllers (i.e., optimization-based controller (OBC) and deep reinforcement-learning-based controller (DRLC)) in terms of the energy efficiency and thermal comfort. The performance evaluation was conducted in a medium-sized office building with a multi-zone Variable Air Volume (VAV) cooling system. The OBC and DRLC replaced the GDL36 airside supervisory level control loops. In other words, the optimal supply air temperature (SAT) and static differential pressure (DP) setpoints were determined by the optimization problems in OBC and trained control policy by DRLC. The OBC and DRLC were formulated to minimize the HVAC energy consumption and zone air temperature violation. The OBCs with different control intervals and DRLCs with different hyperparameters in the Proximal Policy Optimization (PPO) algorithm were exploited and fine-tuned. The simulation results show that the GDL36 has a comparable energy performance (within a 3% deviation) with DRLC in the cases with high or mild cooling loads. It also has a comparable energy performance (within a 3% deviation) with OBC in the case of a high cooling load. However, the case with GDL36 consumed 7% more energy in the testing shoulder period. For the thermal comfort metric, the GDL36 has slightly more zone air temperature violations in all testing scenarios compared to the two intelligent controllers. From this case study, we can conclude that the GDL36 has demonstrated its comparable performance in terms of energy efficiency and thermal comfort with the two intelligent controllers. The GDL36 airside SOO is good enough considering the complexity and tuning efforts of the intelligent controllers.