An Agent-based Control Implementation for the Optimal Coordination of Multiple Rooftop Units

Conference Year

July 2018

Authors

Xiaodong Hou, School of Electrical and Computer Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow
Yingying Xiao, School of Electrical and Computer Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow
Jaewan Joe, Lyles School of Civil Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow
Jie Cai, Aerospace and Mechanical Engineering, The University of OklahomaFollow
Panagiota Karava, Lyles School of Civil Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow
Jianghai Hu, School of Electrical and Computer Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow
James Braun, School of Mechenical Engineering, Purdue University; Center for High Performance Buildings, Ray W. Herrick LaboratoriesFollow

Keywords

Agent-based control, DMPC, RTU coordination, implementation

Abstract

Model predictive control (MPC) has been a popular advanced supervisory control approach for optimizing the operations of building heating, ventilation and air-conditioning (HVAC) systems, with the objectives of reducing energy consumption and delivering better comfort. However, centralized MPC designs are often 1) not scalable to the increasing sizes of the building systems, 2) not adaptive to subsystem addition/attrition, i.e., ‘Plug-and-Play’ implementation. Agent-based approaches, such as distributed model predictive control (DMPC), are attractive alternative. In this paper, taking a multiple rooftop units (RTU) coordination problem as a case study, we experimentally investigate the energy saving potential by implementing an agent-based DMPC strategy to coordinate the operations of multiple ‘virtual’ variable-speed RTUs with diverse unit efficiencies (COP) in an open space with multiple sub-zones. The operations of three RTUs are emulated by three groups of variable air volume (VAV) diffusers that can be individually controlled to provide continuously changing sensible cooling loads into respective zones. A multi-zone model that accurately captures the thermal interactions of different zones for control purposes is developed. This model takes the sensible cooling loads provided by the three ‘virtual’ RTUs as controllable inputs, and ambient temperature, solar radiation, internal heat gains (occupancy, plug, lighting and equipment loads) as exogenous inputs. Three laptop computers are dispatched into the three thermal zones as local agents. A server computer connected to both the Building Automation System (BAS) and the outside internet is responsible for predicting various exogenous inputs and exchanging information with the local agents. Experimental results show that the proposed agent-based DMPC design and implementation are able to achieve over 20% cost savings, in terms of electricity consumption charge with Time-of-Use pricing schedules, while at the same time maintaining local occupancy comfort. The savings can be further broken down into two parts: 1) utilizing the RTUs with higher overall unit efficiency 2) shifting the aggregate cooling load of the room into periods with lower electricity price or higher RTUs’ COP.