Conference Agenda

Model Predictive Control (MPC) applied to buildings is starting to see some commercial applications with companies taking up the challenge. However, it is hard to estimate if relative cost savings are enough to justify the cost of MPC implementation with few reported demonstrations. In small commercial and residential buildings, a simplified one-sizefits-all solution can help reduce costs, while in large buildings or districts the potential savings magnitude can cover a tailored solution. This estimation becomes harder for medium commercial buildings, since a cookie cutter solution cannot be adopted and potential savings might not be sufficient. Therefore, different value propositions can make MPC technology more attractive through additional cost savings. One such value proposition is load shifting in response to dynamic electricity prices. On this aspect, MPC could be a key technology in unlocking Building thermal mass flexibility potential to be more grid-responsive. This study shows the experimental results of MPC control of an office building in Berkeley, where different dynamic electricity price profiles were used in the MPC objective function to shift the building load and to calculate hypothetical electricity costs. Results show a potential 30% cost savings with respect to the existing controller.

Introduction

The transition to an environmentally friendly and energy-efficient building sector is an essential element of global climate protection. In addition to energetic refurbishment measures, hardware and software measures on the control and regulation systems of the building represent an effective approach, enabling cost-effective significant energy savings and improvement in user comfort within buildings.

Model Predictive Control (MPC) has been identified in recent years as a promising approach for energy-efficient and intelligent control of building systems. In this method, building dynamics are incorporated into the control algorithm through simulation models, along with the prediction of external and internal disturbance factors such as weather conditions or building occupancy. This contribution presents the results of implementing an MPC controller in a large administrative building over a period of nine months, focusing on heating energy consumption and thermal comfort.

Building Description

The demonstration building is a large, energy-efficient administrative building located in Northern Germany with a net floor area of 46,500 m² and approximately 1,250 offices. It was completed in 2013 with high standards for primary and heating energy consumption. The building has been subject to intensive monitoring as part of the Energy-Optimized Building (EnOB) research initiative since the beginning of its operational phase. The heating of the building is primarily achieved through concrete core activation, implemented as thermoactivated ceilings (TAC), in combination with ground-coupled heat pumps.

Results of the Experimental Investigation

The developed MPC control strategy is integrated into the building automation system, and its performance regarding heating energy consumption and thermal comfort is compared with the conventional heating curve-based control strategy over two heating periods covering a total of nine months. The MPC controller takes control of the flow temperature regulation and pump control of the TAC heating circuits.

Compared to the heating curve-based control strategy, a measured average saving of heating energy ranging from 10% to 30% per month is observed during the study periods. It is evident that the controller's performance is dependent on prevailing weather conditions. In transitional months such as April, the MPC control proves superior, achieving heating energy savings of up to 75%. Additionally, an improvement in thermal comfort is demonstrated compared to previous operational years. The proportion of indoor air conditions within Category I according to DIN EN 16798-1 is significantly increased due to the altered control strategy. The determination of thermal comfort is based on a multitude of measurements of indoor air conditions in over 70 offices.

In this paper, a multi-zone Model Predictive Control (MPC) system with coordination between multiple systems including Air Conditioning and Mechanical Ventilation (ACMV), lighting, and shading has been presented. MPC system includes data-driven models tailored for each indoor zone for forward prediction of its thermal conditions, illumination, and artificial lighting power. The thermal prediction model was trained with historical data from humidity, indoor air and globe temperature sensors, and disturbances including weather parameters and occupancy. The disturbance data were from machine-learning-based weather forecasting and a real-time occupancy detection system. Illumination and artificial lighting power prediction models were trained with historical data of lighting power meters, blind positions, and indoor illumination levels collected via sensors. Nonlinear autoregressive exogenous model with external inputs (NARX) neural network was employed to develop these data-driven models. With these data-driven models, MPC system simultaneously optimizes multiple targets including energy consumption, thermal and visual comfort by solving nonlinear optimization problems. In the case of thermal comfort, the predicted mean vote (PMV) was optimized to seek a default reference PMV setpoint of 0 representing thermal neutrality while constrained to within -0.5 to 0.5 . Similarly for optimum visual comfort, daylight glare probability (DGP) should be within 0.35 and the work-plane illuminance was constrained to 500-3000 lux. An algorithm to take in occupant feedback as an additional consideration was also incorporated into the MPC system. As per occupant feedback, their preferences can bias the reference PMV set point. MPC system was implemented in a multi-use test space located in Singapore with area of approximately 850 m². The test space was partitioned into 6 learning zones, 2 office spaces, and 3 open spaces. ACMV system serving the test space consisted of 2 Primary Air-Handling Units (PAHUs) and 16 Fan Coil Units (FCUs). Chilled water is supplied to cooling coils of these units from central chiller plant of the building. Conditioned air is distributed to test space through motorized diffusers. Lighting and shading system consisted of LED lighting fixtures with dimmable control and motorized roller blinds respectively. Control performance of MPC system was compared against the test building’s original thermostat-based (reactive) control. MPC system (without occupant feedback) achieved over 33% energy savings with higher thermal and visual comfort. When occupant feedback was considered, it was found that the occupants preferred a thermal environment cooler than thermal neutrality (i.e., negative PMV) for certain periods of the day (e.g., when the occupants have just arrived at the space in the morning or after lunch). This led to higher energy consumption compared to MPC without occupant feedback. Overall, MPC with occupant feedback still achieved over 23.5% of energy savings when compared to that of original reactive control system. Despite advantageous control performance, MPC system requires additional sensors for occupancy comfort evaluations, leading to high implementation cost. Much effort is also needed to construct accurate building models, which further adds hurdles for MPC adaptation. In future deployments of MPC, these shortcomings can be mitigated by building comfort models only based on existing sensors in the building.

Buildings account for 35% of total energy-related carbon dioxide emissions in the U.S. and thereby play an important role in achieving the decarbonization goals set by the administration. Renewable energy such as solar and wind can help reduce grid emissions by displacing fossil fuel; however, the uncertainty and intermittency of renewable energy resources can cause both diurnal and seasonal variability of the electricity carbon intensity. This paper first presents an analysis of the carbon variation patterns across eleven locations in the U.S. A carbon savings potential indicator is introduced to afford first-order estimation of carbon reduction potential based on diurnal variability of the carbon intensity. This analysis is followed by a model predictive control strategy to minimize building carbon emissions through optimized thermal load shifting in response to time varying carbon intensity signals. The control strategy was tested using a commercial building case study subject to hourly marginal carbon emissions of two U.S. electricity markets – the California Independent System Operator (CAISO) and the New York Independent System Operator (NYISO). The test results show that for the NYISO market, the carbon responsive strategy could save 3.4% carbon emissions compared to the energy minimizing strategy, while for the CAISO market, up to 33.5% carbon emission reduction could be achieved because of the more aggressive intro-day variation of the electricity carbon intensity.

Developing the demand response control for the air conditioner system of residential buildings has been proven to be a highly effective strategy in assisting grid supply-demand balance and facilitating the integration of renewable energy to decarbonize the energy system. Model predictive control (MPC) has strong capabilities for unlocking the flexibility of residential buildings to realize DR by responding to electricity prices. However, the high computational requirements and complex control system integration processes make the application of MPC a significant challenge. A hierarchical nonlinear MPC (HNLMPC) is developed to realize grid-responsive control for residential inverter ACs by responding to real-time electricity price signals. The controller consists of three parts: the upper-level supervisor MPC, the lower-level optimal PID controller, and the signal converter. The indoor air temperature is selected as the optimized setpoint sequence passed from the upper level to the lower level. It could utilize cloud-based infrastructure or the Internet of Things, which means the operation not be limited by the local computing power. A nonlinear prediction model is developed considering the dynamic performances of the inverter air conditioner and the coupled thermal response of an air-conditioned room, which unlocks the DR potential as much as possible. As a result, HNLMPC enables plug-and-play capability for practical applications, reducing the dependency on local computing power, maintaining the performances, and improving the robustness. Compared to basic rule-based control, HNLMPC reduces peak-hour energy consumption by 31.6% and total electricity costs by 14.3% over the entire cooling season. Compared with central MPC, the HNLMPC has a lower demand for computing power.

With the urgent call for carbon reduction, the realization of grid-interactive efficient buildings (GEBs) that provide load flexibility (e.g., load shifting, peak demand reduction, etc.) has become a major research area. Model predictive control (MPC), which optimizes the operations of building heating, ventilation, and air-conditioning (HVAC) systems and distributed energy resources (DERs) based on grid conditions, is one of the mature and viable solutions for GEBs of small and medium commercial buildings (SMCBs), where the advanced building management system and sensor infrastructure are often not available. Despite the recent success of practical MPC solutions for SMCBs with simple HVAC systems such as rooftop units (RTUs), the design of MPC for a complex system such as a variable refrigerant (VRF) system is still costly. In this paper, we present a practical MPC solution for a multi-zone VRF system for small and medium commercial buildings. Utilizing heating operational data from a multi-zone laboratory office building, we propose a VRF model for heating operation and MPC structures based solely on the available data. The developed MPC solution was applied to the laboratory office building for a heating week with a dynamic pricing signal. The results indicate that the proposed MPC can achieve reductions of approximately 32% in peak demand and 3% in energy costs by shifting 24% of peak-time load to non-peak time hours, respectively.