MPE 2013+ Workshop on Data-aware Energy Use

September 29 - October 1, 2014
University of California, San Diego

Yuvraj Agarwal, Carnegie Mellon University, Yuvraj.Agarwal at
Mario Berges, Carnegie Mellon University, marioberges at
Midge Cozzens, DIMACS (Education Chair), midgec at
Rajesh Gupta, University of California, San Diego, rgupta at
Zico Kolter, Carnegie Mellon University, zkolter at
Balarkrishnan Narayanaswamy (Murali), University of California, San Diego, muralib at
Fred Roberts, Rutgers University and DIMACS, froberts at
Presented under the auspices of the DIMACS Special Program: Mathematics of Planet Earth 2013+ and the Special Focus on Energy and Algorithms.
With acknowledgement of additional University of California, San Diego support.
Workshop Announcement

Increasing demand and expensive supply make efficient resource allocation and optimal decision making critical in energy systems. We need to make good choices about today's energy investments, because they will be with us for a long time.

Data can help us make better choices if we can surmount concomitant challenges: large scale and distributed data acquisition; massive amounts of data; incomplete, unreliable or distributed data; real time data; fusing data for decision making; interoperating/distributed decision makers; decision making in dynamic environments of high consequence; consumer privacy - all complex, multidisciplinary problems. We will explore harnessing data to address problems in energy, emphasizing three primary areas: sensor networks, data mining, optimization applied to grids, homes, buildings and electric vehicles.

Sensor Networks and Data Acquisition: While the purpose of the workshop is to study the promise of large energy data sets, it is critical to first understand the possibilities and limitations of current and future sensor and data acquisition networks. New sensing infrastructure including occupancy, plug level meters, CO2 meters, cameras and microphones can all be deployed. Each sensing modality has its own advantages and disadvantages and in combination for rich datasets for upstream processing. At the same time they present privacy, data storage and user inconvenience challenges which must be addressed.

Data Mining: New algorithmic methods to understand, process, visualize data and find anomalies in streaming phasor measurements are required. New measurements will allow rapid understanding of customers' electricity, usage, thus leading to privacy research combining statistical and cryptographical approaches to data. We need: new algorithms to predict grid response to disturbances and the process of restoring it to a healthy state; a new mathematics for characterizing uncertainty in information created from large energy data sets; new methods for processing high-bandwidth data to convey only the most useful information.

Optimization and Control: Today's "smart" data sources enable real-time precision in operations and control previously unobtainable: Real-time pricing and customer engagement from smart meters; enhanced state estimation, real-time contingency analysis and real-time monitoring of oscillatory behaviors; advanced control methods to enable rapid diagnosis and precise solutions.

Smart Grid: Our electric power systems have grown up incrementally and haphazardly. They form complex systems that are in constant change loads change, breakers go out, they are influenced by weather, etc. These systems operate under considerable uncertainty, and cascading failures can have dramatic consequences. Research challenges relating to power systems arise from the huge number of customers; uncontrolled demand; changing supply mix; and the vulnerability that stems from operating "close to the edge". Algorithmic methods can provide: better monitoring and change detection for energy security; enhanced ability to identify and overcome vulnerabilities; and a more responsive grid that uses real-time data for more dynamic pricing and control.

Smart Buildings: A goal for the smart grid is to provide real-time usage and pricing information to building occupants. Buildings account for more than 70% of total U.S. electricity consumption and more than 40% of total carbon emissions, so the potential savings is substantial. "Smart" applications have already paid off in commercial buildings through balancing energy used for overhead lighting and HVAC with energy needed for other activities, and balancing control schedules with anticipated weather conditions and building occupancy. Managing energy use in buildings can benefit from creation of systems that manage themselves with only "high level" guidance from humans, exploiting data mining, machine learning, control theory, and recommender systems. We will explore ways to design complex, adaptive systems that are self-configuring, self-optimizing, self-maintaining, self-healing, and self-protecting, while minimizing energy use.

Scaling up from the building to the town-city continuum poses a whole new suite of mathematical challenges. While considerable mathematical urban planning progress has been achieved in recent years much of this work is of a theoretical nature well suited only for newly designed and constructed, sometime utopian, ecocities, but less to existing, suboptimal infrastructure. Consequently, many of the optimization options the literature explores are not realistically available in existing cities. Alternative approaches are needed for aging cities such as those in the US. We plan to focus on optimization of options for existing cities in which durable infrastructure is assumed a given, but superficial (exterior) optimization is permissible. The mathematical challenges include energy-aware scale-dependent optimal array design (with linear programming and Monte Carlo methods) applied to choices such as green roofs, street forestry, highly reflective pavements, hydrologically smart (partially impervious) surfaces, and smart material based seasonally varying reflectivity exteriors.

Electric Vehicles: Hybrid electric vehicles and pure electric vehicles present intriguing options for reducing dependence on fossil fuels and reducing carbon emissions. For wide adoption, such vehic les need to be cost-effective and convenient to operate. A viable electric vehicle infrastructure might involve a system of charging stations and/or battery exchange sites making up for limited vehicle driving range. Developing such infrastructure provides mathematical challenges in locating and sizing the stations to provide good coverage for drivers while not over-burdening the electric grid. The interplay between this infrastructure and the power network is complex: Fast-charging facilities would enable long trips in all-electric vehicles, but draw large amounts of power. When electric vehicles are brought home and charged at night, this could lead to short-term spikes in demand, which could be mitigated by exploiting tolerance to delays since it may only require six hours of total charge time within a ten hour period. We need to develop system-wide incentives and methods for spreading the demand of many delay-tolerant customers over time. Turitsyn, et al. describe control algorithms that provide randomized electric vehicle charging start times using methods of queuing theory and statistical analysis to keep the probability of circuit overload negligible. We will also study the idea that batteries in electric cars could create a two-way exchange of electricity, storing electricity that could be "sold" to the grid during times of peak demand, thus adding robustness to the power infrastructure.

Next: Call for Participation
Workshop Index
DIMACS Homepage
Contacting the Center
Document last modified on September 18, 2014.