University of Southern California
Research Group

Multiagent Systems Research for Sustainability


Over the decades, energy issues have been getting more important. In the U.S., 48% of energy consumption is from buildings, of which 25% is associated with heating and cooling at an annual cost of $40 billion. Furthermore, on an annual basis, buildings in the United States consume 73% of its electricity.

With rising energy costs, the need to design and integrate scalable energy consumption reduction strategies in buildings calls for novel approaches. There are numerous challenges associated with energy resources such as supply and depletion of energy resources and heavy environmental impacts. The rise in energy consumption in buildings can be attributed to several factors such as enhancement of building services and comfort levels, through heating, cooling and lighting needs and increased time spent indoors.

Motivated by this, we focus on the study of sustainability in order to conserve energy. The goal is not only to construct the correct model for capturing important phenomena in such domains, but to generate optimal strategies to effectively achieve given goals. Specifically, this work includes the following projects:


Sustainability


Current Team

Blake Cignarella

Milind Tambe

Burcin Becerik-Gerber

Pradeep Varakantham

Wendy Wood

David Jason Gerber

Antonia Boadi

Debarun Kar

Timothy Hayes

Farrokh Jazizadeh

 

Collaborators

Nicole Sintov

 

Alumni

Laura Klein

Geoffrey Leyian Kavulya

Jun-young Kwak

 

Collaboration

To effectively address the above challenges, we establish an active collaborative environment with the sustainable energy project under the leadership of Prof. Burcin Becerik-Gerber and her collaborators. Specifically, we are building smart energy systems based on multiagent coordination and collaborating with researchers from different departments including Computer Science (Teamcore Research Group), Civil and Environmental Engineering (Innovation in Integrated Informatics LAB), Psychology Departments and School of Architecture at USC, School of Information Systems at Singapore Management University, and Computer Science Department at California State University Dominguez Hills.

 

Acknowledgments



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.

 

 

TESLA
(Transformative Energy-saving Schedule-Leveraging Agent)

One of Research Testbed Buildings: Leavey Library at USC

 

Energy-oriented Scheduling Agent

Reducing energy consumption is an important goal for sustainability. Thus, conserving energy in commercial buildings is important as it is responsible for significant energy consumption. This energy consumption is significantly affected by a large number of meetings or events in those buildings. Furthermore, a recent study shows that meeting frequency in commercial buildings is significant and continues to grow. In 2001, U.S. Fortune 500 companies are estimated to have held 11 million formal meetings daily and 3 billion meetings annually. Energy-oriented scheduling can assist in reducing such energy consumption. Although conventional scheduling techniques compute the optimal schedule for many meetings or events while satisfying their given requirements (i.e., computing a valid schedule), they have not typically considered energy consumption explicitly. More recently, there have been some trials to conserve energy by consolidating meetings in fewer buildings.

To that end, we present TESLA (Transformative Energy-saving Schedule-Leveraging Agent), an agent for optimizing energy use in commercial buildings. TESLA is a goal-seeking (to save energy), continuously running autonomous agent. TESLA's key insight is that adding flexibility to meeting schedules can lead to significant energy savings. Users in a commercial building continuously submit meeting requests to TESLA while indicating their flexible meeting preferences. TESLA schedules these meetings in the most energy efficient manner while ensuring user comfort; but in cases where shifting meeting times can lead to significant savings, TESLA interacts with users to request such a shift.

This work particularly provides four key contributions. First, it provides online scheduling algorithms, which are at the heart of TESLA, to solve a two-stage stochastic mixed integer linear program (SMILP). This SMILP considers the flexibility of people's preferences for energy-efficient scheduling of incrementally/ dynamically arriving meetings and events. Second, TESLA also includes an algorithm to effectively identify key meetings that could lead to significant energy savings by adjusting their flexibility. Third, this work provides an extensive analysis of the energy saving results achieved by TESLA. Lastly, surveys of real users are provided indicating that TESLA's savings can be realized in practice by effectively leading people to change their schedule flexibility. To validate our work, we used a public domain simulation testbed, which is described below, fitted it with details of our testbed building, and compared the simulation results against real-world energy usage data. Our results show that, in a validated simulation using our testbed building, TESLA is projected to save about 94,000 kWh of energy (roughly $18K) annually. Thus, TESLA can potentially offer energy saving benefits to all commercial buildings where meetings affect energy usage.

TESLA architecture: TESLA is a continuously running agent that supports four key features: (i) energy-efficient scheduling;
(ii) identifying key meetings; (iii) learning user preferences; and (iv) communicating with users.

 

Research Testbed Buildings & Real Data

In this work, we consider two sets of research testbed buildings: i) Leavey library, which is one of main libraries at the University of Southern California and ii) 7 buildings and 1 open space at the Singapore Management University. In these buildings, meetings are requested by users by a centralized online room reservation system. In the current reservation system, no underlying intelligent system is used; instead, users reactively make a request based on the availability of room and time when they access the system.

In collaboration with building system managers, we have been collecting data specifying the past usage of group study rooms, which are collected for 8 months (January through August in 2012) at USC and for 3 months (August through October) in 2011 at SMU. The data for each meeting request includes the time of request, starting time, time duration, specified room, and group size. The data set contains over 110,000 unique meetings, which are used for evaluating TESLA.

 

 

SAVES
(Sustainable multiAgent system for optimizing Various objectives
including Energy and Satisfaction)

 

Multiagent Systems to Conserve Energy

Recent developments in multiagent systems are opening up the possibility of deploying multiagent teams to achieve complex goals in such energy domains that inherently have uncertain and dynamic environments with limited resources. To model and optimize buildings' energy consumption, building agents, facility managers and human occupants are demanding robust, intelligent and adaptable ambient planning techniques. To realize both tangible benefits such as energy and operation savings, value property, reduction in occupant complaints as well as the intangible benefits such as occupant comfort and satisfaction, designers must develop energy adaptive capabilities within the building environmental control systems.

This project focuses on a novel application to be deployed at Ralph & Goldy Lewis Hall (RGL) at the University of Southern California as a practical research testbed to optimize multiple competing objectives: i) amount of energy used in the buildings; ii) occupants' comfort level; and iii) practical usage considerations. This work provides three key contributions. First, we explicitly consider uncertainty while reasoning about coordination in a distributed manner. In particular, we uses a novel algorithm for generating optimal MDP (Markov Decision Problem) policies that explicitly consider multiple criteria optimization (energy and personal comfort) as well as uncertainty over occupant preferences when negotiating energy reduction Second, human behaviors and their occupancy preferences are incorporated into planning and modeled as part of the system. As as result, our system is capable of generating an optimal plan not only for building usage but also for occupants. Third, the influence of various control strategies for multiagent teams is evaluated on an existing university building as the practical research testbed with actual energy consumption data in the validated simulation testbed. Since the simulation environment is based on actual data, this result can be easily deployed into the real-world. For future work, we consider opportunities for direct occupant participation and incentivization via handheld devices and deploy our system to the real-world.

      
Research Testbed Building: RGL at USC Validated Simulation Testbed

 

Human Subject Study on Energy Conservation

We also design and conduct a validation experiment on a group of human occupants in commercial buildings via a set of agents in our system: room and proxy agents. There is a dedicated room agent per office and conference room, in charge of reducing energy consumption in that room. It can access sensors to retrieve room information and energy use and impact the operation of actuators. A proxy agent is on an individual occupant's hand-held device and it has the corresponding occupant's models. Proxy agents communicate on behalf of an occupant to the room agent based on their adjustable autonomy - when to interrupt a user and when to act autonomously. Room agents may directly communicate with occupants without proxy agents, and different room agents coordinate among themselves as well as with proxy agents.

We conduct this investigation: i) to verify if our system can lead to changes in occupants' behaviors and to reduce energy consumption in commercial buildings, ii) to validate the parameter values used during the negotiation process such as the acceptance/compliance rate for the suggestion and iii) to understand what types of feedback are most effective to affect occupants' energy-related decisions.

 

Simulation Demo

 

 

 

 

THINC
(Energy Saving Incentivization )


Abating energy consumption is an important goal to secure a sustainable future. Specifically, commercial buildings are a key contributor to United States energy use, representing roughly 25% of total energy expenditures according to the United Stated Department cooling.

Large number of meetings or events significantly impacts such expenditures. U.S. Fortune 500 companies annually hold more than 11 million meetings daily. Using algorithms and optimization technique, known as THINC, energy minimization is possible by scheduling meetings back to back; reducing reheating or re-cooling costs.

In order to gain a greater global maximize savings we ask participants to specify a single time slot and location for their meeting to be held in a campus library which schedules over 300 meetings per day. When indicating a time slot you may offer a wider window around your preferred time slot and location by being flexible; offering different times and locations.

Assume that you are looking to schedule a meeting using a meeting reservation system, where you can specify a single option in terms of time and the location for the meeting. When scheduling this meeting you may offer a wider window around the previous preferred slot by being flexible; offering different times and locations. This will allow energy savings as an energy efficient algorithm generates a schedule; reducing re-cooling costs of vacant rooms.

Shapley allocation divides the combined energy savings of the group amongst its participants. Incentivizing participants through monetary compensation may yield towards future offerings of flexibility. Shapley value is taken for each participant and is an average over all possible scenarios or combinations of the groups flexibility related to the scheduling effect of a single participant. This effect is the difference of going from a state of no flexibility to their offered flexibility related to the current state of the system. Thus the value is the individual's contributes to the whole as it relates to the contribution of all other participant in the system.

Although the shapely value is algorithmically fair in dividing cooperative gains an individual participant's perception of fairness need to be evaluated. It of utmost importance from a policy and behavioral perspective that people believe that they have been treated fairly as it is advantages for them to participate again. By finding ways to positively promote energy saving behavior by offering flexibility better schedules and greater achievement of energy savings are possible.

 

 

Resources

Links:

Publications:

Software: