Our goal is to design new dynamic control strategies for building operations that reduce building energy consumption while increasing human comfort. This goal is achieved through integrating sensing subjective human related data with building operations. This activity aims at developing a method to accurately collect subjective context-dependent and dynamic human data on a continuous and real-time basis.
Design of Novel Agent-based Models and Algorithms:
To achieve the goal of conserving energy in commercial buildings, we worked on two sets of algorithmic contributions. First, we designed a novel BM-MDP (Bounded-parameter Multi-objective Markov Decision Problem) model and robust algorithms including HRMM (Heuristics for Robust Multi- objective optimization under Model uncertainty) for multi-objective optimization under uncertainty for both planning and execution time. The BM-MDP model and its robust algorithms identify key meetings that have significant potential in savings when people’s given constraints are altered. Second, we designed online predictive scheduling algorithms to handle massive numbers of meeting/event scheduling requests considering flexibility, which is a novel concept for capturing generic user constraints while optimizing the desired objective.
These new models have not only advanced the state of the art in multiagent algorithms, but have actually been successfully realized as two real-world applications in the energy domain: SAVES and TESLA. SAVES focuses on the day-to-day energy consumption of individuals and groups in commercial buildings by reactively suggesting energy conserving alternatives. TESLA takes a long-range planning perspective and optimizes overall energy consumption of a large number of group events or meetings together. While SAVES and TESLA differ in their scope and applicability, both demonstrate the utility of agent-based systems in reducing energy consumption in commercial buildings.
We evaluated the proposed algorithms and agents using extensive analysis on data from over 110,000 real meetings/events at multiple educational buildings including the main libraries at the University of Southern California. We also provided results on simulations and real-world experiments, demonstrating the potential of agent technology to assist human users in saving energy in commercial buildings.
Pathway II: Energy Efficient Building Operations
1. SAVES (Sustainable multi-Agent building application for optimizing Various objectives including Energy and Satisfaction):
SAVES focuses on the day-to-day energy- consumption of single individual or single group activity in commercial buildings, to be reactive in suggesting energy conserving alternative to that individual or group. SAVES is deployed at Ralph & Goldy Lewis Hall (RGL) at the University of Southern California. More specifically, SAVES provides the following key novelties: (i) jointly performed with the university facility management team, SAVES works on actual occupant preferences and schedules, and actual energy consumption data, real sensors and hand-held devices, etc.; (ii) it addresses novel scenarios that require negotiations with groups of building occupants to conserve energy; (iii) it focuses on a non-residential building, where human occupants do not have a direct financial incentive in saving energy and thus requiring a different mechanism to effectively motivate occupants; and (iv) SAVES uses a novel algorithm for generating optimal MDP policies that explicitly consider multiple criteria optimization (energy and comfort) as well as uncertainty over occupant preferences when negotiating energy reduction – this combination of challenges has not been considered in previous MDP algorithms. In a validated simulation testbed, we showed that SAVES substantially reduces the overall energy consumption compared to the existing control method while achieving comparable average satisfaction levels for occupants. As a real- world test, we provided results of a trial study where SAVES has led occupants to conserve energy in real buildings.
Results: We provide two sets of evaluations of SAVES. First, we constructed a detailed simulation testbed, with details all the way down to individual electrical outlets in our targeted building and variations in solar gain per day; and then validated this simulation. Within this simulation testbed, we show that SAVES substantially reduces the overall energy consumption compared to the existing control methods while achieving comparable satisfaction level of occupants. Specifically, SAVES reduced the energy consumption in simulation by 31.27 - 42.45% when compared to the manual control strategy. For the average satisfaction level, the manual setting and our novel algorithm showed the best results. This is because the manual setting makes HVACs attempt to reach the desired temperature set point as soon as possible while disregarding the resulting energy consumption, and our method plans ahead of the schedules; thus, these two can achieve the desired comfort level faster than the other control strategies. Second, as a real-world test, we provided results of a human subject study where SAVES is shown to lead human occupants to reduce their energy consumption in real buildings. More specifically, we designed and conducted a validation experiment on a pilot sample of participants under two test conditions: i) feedback without motivation, and ii) feedback with motivation including participant’s own energy use, and environmental motives. From this experiment, we found that w when we provided more informed feedback including environmental motives, people showed statistically significantly higher compliance acceptance rate (68.18%), which strongly supports our claim that more informed feedback was more effective to conserve energy than feedback without motivation.
Publications:
Li N, Kwak J, Becerik-Gerber B, Tambe M (2013) “Predicting HVAC Energy Consumption in Commercial Buildings Using Multiagent Systems”, The 29th International Symposium on Automation and Robotics in Construction, August 11-15, 2013, Montreal, Canada.
2. TESLA (Transformative Energy-saving Schedule-Leveraging Agent):
TESLA takes a long- range planning perspective and optimizes overall energy consumption of a large number of group events or meetings. TESLA is a goal-seeking (to save energy), continuously running autonomous agent. TESLA’s key insight is that adding flexibility, which is a novel concept for capturing user scheduling constraints, to meeting schedules can lead to significant energy savings. TESLA provides three key contributions: (i) three online scheduling algorithms that consider flexibility of people’s preferences for energy-efficient dynamical scheduling of meetings and events; (ii) an algorithm to effectively identify key meetings that lead to significant energy savings by adjusting their flexibility; and (iii) surveys of real users that indicate that TESLA’s assumptions exist in practice. TESLA was evaluated on data of over 110,000 meetings held at nine campus buildings during eight months in 2011–2012 at USC and SMU. These results showed that, compared to the current systems, TESLA can substantially reduce overall energy consumption.
Results: We have used a public domain simulation testbed and validated this simulation. Within this testbed building, our results show that, in a validated simulation, TESLA is projected to save about 250 kWh of energy (roughly $17K) annually. If this pilot is successful, TESLA can offer energy saving benefits to other commercial buildings, where meetings affect energy usage. More specifically, we evaluated the performance of our scheduling algorithm, which is at the heart of TESLA, compared to two other heuristics (myopic and full-optimization methods), while varying flexibility. We conclude that (i) our predictive non-myopic method is superior to the myopic method; (ii) the predictive non- myopic method performs almost as well as the full-knowledge optimization (about 98%); and (iii) the full flexibility is not required to start accruing benefits of flexibility. Furthermore, we conducted two surveys on a pilot sample of participants (students on campus): (i) an online survey to understand flexibility of those who are using the testbed building; and (ii) a survey to measure flexibility change due to messaging. The measured average time flexibility was 25.34% and their responses fell in a range of 9.86% and 42.86%. The average location flexibility was 16.05% and its range was 0 to 38.24%. In addition, the survey showed that when we provided more informed feedback including environmental motives, participants tripled their flexibility increase percentage (17.12% compared to 5.15%).
Publications:
Kwak J, Varakantham P, Maheswaran R, Chang Y, Tambe M, Becerik-Gerber B, Wood W. “TESLA: An Extended Study of an Energy Saving Agent that Leverages Schedule Flexibility”, Journal of Autonomous Agents and Multi-Agent Systems, Accepted August 2013
Kwak J, Varakantham P, Maheswaran R, Chang Y, Tambe M, Becerik-Gerber B, Wood W. (2013) “Why TESLA Works: Energy Saving Agent Leveraging Schedule Flexibility”, MASS2013, The 1st International Workshop on Multiagent-based Societal Systems, held in conjunction with the Ninth International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS 2013), May 6-10, 2013, Saint Paul, Minnesota, USA
Kwak J, Varakantham P, Maheswaran R, Chang Y, Tambe M, Becerik-Gerber B, Wood W. (2013) ”TESLA: An Energy-saving Agent that Leverages Schedule Flexibility” Autonomous Agents and Multiagent Systems (AAMAS 2013), May 6-10, 2013, Saint Paul, Minnesota, USA
Kwak J, Varakantham P, Maheswaran R, Chang Y, Tambe M, Becerik-Gerber B, Wood W. (2014) “TESLA: An Extended Study of an Energy Saving Agent that Leverages Schedule Flexibility,” Journal of Autonomous Agents and Multi-AgentSystems, Vol: 28, Issue: 4, pp: 605-636
Design and Evaluation of a Novel Comfort Scale for Adaptive Control:
One of the goals of the project is sensing subjective human related data and integrating the collected human related data to building operations. As a first step towards this goal, this activity aims at developing a method to accurately collect subjective context-dependent and dynamic human data on a continuous, and real-time basis. To do this, we have designed and evaluated a novel perception-preference comfort scale, which establishes a relationship between users’ input and their expectations about indoor environmental conditions. This new comfort scale establishes a relationship between users’ input and their expectations about indoor thermal conditions on a real time and continuous basis.
Results: We have designed and evaluated a novel perception-preference comfort scale with different guiding features (slider with temperature values, slider with snapping options, etc.). The guiding features were added to explore the following research questions: What is the influence of slider design variations on a user’s expression of his/her preferences? What is the influence of a slider’s initial position on a user’s description of his/her preferences? Does the preference scale improve a user’s comfort level expression compared to the perception scale? To answer these questions, we have designed an intermediary (facilitates communication between building’s systems and building’s occupants) with different guiding features for smartphones and tablets.
We have adapted a systematic approach similar to the ones used in the human computer interaction field to evaluate different alternatives and conducted several usability studies using evaluation methods including think aloud studies and task execution surveys with human participants. The results confirmed that the proposed sensation scale outperforms the ASHRAE scale and the addition of the guiding features helps increase the consistency between users’ expectations and their votes (about 10% increase in consistency of user votes). Specifically, the snapping feature found to be the best option.
In addition, field experiments were conducted to explore the driving factors, which effect users thermal comfort votes to determine the HVAC control parameters. The results of the field studies showed that the ambient temperature is the most effective factor on users’ votes in the Southern California climate.
Publications:
Jazizadeh F, Marin M F, Becerik-Gerber B. (2013) “A Thermal Preference Scale for Personalized Comfort Profile Identification via Participatory Sensing”, Journal of Building and Environment, Vol 68, pp:140–149
Design and Evaluation of Dynamic Control Strategies for Building Operations:
One of the goals of the project is sensing subjective human related data and integrating the collected human related data to building operations. As a first step towards this goal, this activity aims at developing a method to accurately collect subjective context-dependent and dynamic human data on a continuous, and real-time basis. To do this, we have designed and evaluated a novel perception-preference comfort scale, which establishes a relationship between users’ input and their expectations about indoor environmental conditions. This new comfort scale establishes a relationship between users’ input and their expectations about indoor thermal conditions on a real time and continuous basis.
Results: Coming soon
Publications:
Jazizadeh F, Ghahramani A, Becerik-Gerber B, Kichkaylo T, Orosz M. (2014) “User-Led Decentralized Thermal Comfort Driven HVAC Operations for Improved Efficiency in Office Buildings,” Journal of Energy and Buildings, Vol. 70, pp: 398-410
Disaggregation of Energy Use for Personalized User Feedback:
Unlike residential buildings, where occupants are in charge of their energy bills and, therefore, they are motivated to take actions in reducing energy costs, the occupants of commercial buildings are not as motivated to save energy. Even worse, commercial building occupants are usually not aware of their energy consumption patterns or their share of contribution. Provision of detailed information about spatiotemporal energy consumption enables building occupants to understand the impact of their energyaffecting behavioral patterns. The availability of the detailed energy consumption by appliance and equipment types and by occupants provides the ground for energy management strategies that encourage occupants to contribute in the energy conservation efforts (Pathway I). Examples of these strategies could be applications of different interventions such as creating energy awareness by providing occupants with their detailed energy consumption information, provision of monetary incentives, comparison of energy consumption between peers, and so forth. Provision of detailed electricity consumption information in buildings calls for sensing techniques, which enable consumption node-level disaggregation. Conventional load monitoring approaches, using utility meters, smart meters, or commercially available sensing solutions, such as TED and eMonitor, usually monitor the energy consumption at the building level or at the circuit level; e.g., floor level. However, room level energy information decomposition is needed for personalized information provision. Appliance level metering solutions, such as “Kill-a-watt”, Greenwave Reality PowerNodes, and “Watts up” are not viable solutions for lighting load monitoring since they require an intrusive in-line installation, which is usually not feasible and prohibitively expensive. Accordingly, we explored a non-intrusive sensing approach for lighting load decomposition at the room level by indirect measurements of energy consumption using light intensity signal. This information could also be used in intelligent building systems operations (Pathway II).
Lighting systems in commercial buildings are major contributors to the electricity consumption while occupants of these buildings are usually not aware of the effect of their energy-related behavior on the electricity consumption. We developed and evaluated a non-intrusive load disaggregation approach for lighting systems in office buildings to provide high granularity spatiotemporal load monitoring. In our approach, we introduced a process for artificial light event detection and power consumption estimation using light intensity signals and rooms’ contextual information (i.e., room area and number of possible states for lighting fixtures). Three event detection algorithms: generalized likelihood ratio test, signal- shape driven event detection, and enhanced naïve event detection were proposed and evaluated. The correlation between the illuminance in a room and power consumption associated with that level of illuminance was used for power consumption estimation.
Results:
Evaluation of the algorithms showed that the signal-shape driven event detection algorithm out-performs other two event detection algorithms with high recall, precision, and F-measure values (0.91, 0.89, and 0.9, respectively). An analytical solution was used for power consumption estimation of each segments of the light intensity. The power estimation used the correlation between illuminance (calculated based on the logarithmic output response of the sensor) in the room and the power consumption associated with that level of illuminance (in the form of number of light bulbs). The power consumption estimation evaluation in rooms without windows showed an error of 8.35% in 21 rooms, when the sensors were installed closer to the floor and the estimation was adjusted with possible lighting states in the room. Our exploration of the effect of sensor height showed that by installing the sensors closer to the floor, the error in power consumption estimation is reduced. The power estimation in rooms with windows showed promising results, considering that the sensor was installed in the mid height position for rooms with windows. Assessment of power estimation in rooms with windows also showed temporal stability. We plan to employ this methodology in office buildings to provide accurate and detailed lighting related energy consumption information to occupants.
Publications:
Jazizadeh F, Ahmadi-Karvigh S, Becerik-Gerber B, Soibelman L. (2014) “Spatiotemporal Lighting Load Decomposition Using Light Intensity Signals,” Journal of Energy and Buildings, Vol. 69, February 2014, pp: 572-583
Acknowledgment and Disclaimer: This material is based upon work supported by the National Science Foundation under Grant No. 1231001. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.