Agent-based approach for system identification and optimal control of high performance buildings

Commercial buildings have strong impacts on humans and the environment. They not only affect occupants’ comfort, health, and well-being but also consume more than 19% of the total energy consumption in the US. High performance building designs can achieve significant energy savings, with new building technologies such as advanced building envelopes, thermally activated building systems, on-site power production and thermal storage; dynamic effects related to variability in occupancy and environmental conditions; diversity in occupant thermal preferences; and the integration of these diverse technologies into an overall control system design. Model-based predictive control (MPC) is a promising approach for the realization of high performance buildings as operations can be optimized for the specific building and climate through an estimated process model that predicts the future evolution of the system, while incorporating the most up-to-date information on weather forecast and system dynamics. Despite of the advantages, there are still significant obstacles associated with the realization of MPC implementation in actual buildings. First of all, the process of generating a control-oriented building model, which is referred as system identification, can be complex and not easily reproduced, due to the customized design of buildings and HVAC systems. Also, MPC computation could become intractable due to the large decision dimension for large-scale systems. To date, the formulation, solution, and integration of optimal controls into existing building management systems (BMS), may not be easily scalable to other buildings on account of the design customization and control intractability. It is envisioned that in the future, with new technology for sensing, information processing and communication, distributed intelligence would be embedded into devices and would be widely deployed into actual buildings. Towards the realization of this plug-and-play intelligent building operation, the research objective of this thesis is to develop a multi-agent system approach to optimal control of high performance buildings, based on new algorithms for distributed system identification and distributed model predictive control (DMPC). From the application perspective, the focus is thermal environment control of open-plan office spaces. Radiant floor systems are evaluated as high performance features and used as test-beds to demonstrate the proposed agent-based framework for zone and local environment control. As a first step, a multi-agent systems approach for data-driven grey-box building models is introduced. Each zone is divided into sub-systems (agents), and a parameter set for each subsystem is first estimated individually, and then integrated into an inverse model for the zone using the dual decomposition algorithm. Two case-studies are designed and conducted using the Living Laboratories at Purdue’s Herrick Building as test-beds to validate the estimated control-oriented models under realistic operation conditions. The results show that the model prediction accuracy of the new approach is fairly good for implementation in predictive control while models can be developed and integrated with improved efficiency, flexibility and scalability, compared to centralized approaches. In the next step, a centralized MPC strategy is developed for zone thermal environment control in an occupied office space with radiant comfort delivery along with a chiller and boiler as HVAC sources. The MPC controller deploys an optimization algorithm based on constraint quadratic programming with hard comfort bounds, which yields an exact numerical solution, and it is straight forward and robust for this application. Results from the MPC implementation during the cooling season show that more than 34% cost savings are achieved by load shifting to utilize higher chiller efficiency with lower outdoor air temperature, and lower electricity prices. In the heating application, the energy use reduction from the optimized control is around 16% compared to conventional control. In the final step, a distributed optimization algorithm, inspired by the Proximal Jacobian Alternating Direction Method of Multipliers (PJ-ADMM), is introduced. It includes multiple MPCs run iteratively while exchanging control input information until they converge. With this tractable approach, agents solve individual optimization problems in parallel, through information exchange and broadcasting, with a smaller scale of the input and constraints, facilitating optimal solutions with improved efficiency. The developed algorithm is tested using field data from an occupied open-plan office space with localized comfort delivery along with distributed sensing, control, and data communication capabilities. The radiant comfort delivery system with predictive control is capable of providing localized thermal environments, thereby improving occupant satisfaction, while achieving more than 27% reduction in electricity consumption compared to baseline feedback control. In summary, this thesis introduces a new agent-based approach for system identification and MPC, which is implemented and tested using an actual building as test-bed. The results show significantly improved performance compared to conventional systems and controls. The overall methodology could be packaged into a toolbox integrated into open-source building control platforms, existing building management systems, or embedded into new smart devices. It is a scalable solution that can be extended to other smart and connected environments, e.g., multiple building systems, multi-zone buildings, building clusters integrated with power grids and automobiles