DRSense - Automated and Scalable Assessment of Demand Response for Green Building Portfolios
Seed 2011-12 Project
Our original proposal:
(proposal paper) (proposal presentation slides)
Green energy efficient building portfolios need to incorporate significant
participation in Demand Response (DR) programs to reduce their operating
costs and peak energy consumption. Currently, building owners and managers
are unable to evaluate the DR potential of their portfolio, assign the best DR
strategies for each individual building and perform a cost/benefit analysis of
implementing DR. The main limitation is the very manual one time evaluation in
current practice. Instead we propose a DRSense: a tool that combines real-time
data, models for building consumption and data mining with advanced
statistical methods to derive DR potential. If successful, the open-source web
based tool will enable facility operators to continuously assess building
performance and optimize the dynamic strategy for DR that achieves the most
savings. Given the important role of DR in the future grid, it is essential to
develop tools that guide the configuration of DR for a specific building
(2) Motivation - Practical scenario and potential impact of the research
Green energy efficient buildings make significant use of renewable technologies and
strive for lower operating
costs. Costs are higher if more energy is consumed during peak load times. An
important technology that is essential for greening a building is Demand Response (DR). DR is a
tariff or program established to motivate changes in electric use by end-‐use customers in
response to changes in the price of electricity over time, or to give incentive payments designed
to induce lower electricity consumption at times of high market prices or when grid reliability is
jeopardized. DR provides an approach to reduce operation costs of a building by
shifting or limiting consumption during peak periods. At the same time DR decreases the peak
generation requirements of the power
system, reducing emissions and increasing grid reliability establishing it as an essential need.
Building owners and managers considering incorporation of DR into their portfolios have
to answer a number of
- How much DR is available for a given building and a portfolio of buildings?
- What strategies are best suited for DR for a given building?
- What is the cost and benefit for incorporating DR at a building?
- How to maximize the DR potential for the entire portfolio given the cost /
benefit analysis of individual buildings?
Today, answering these questions requires an energy audit by a specialized energy
management consultant who visits each building and spends considerable amount of time determining the DR strategy
and estimating DR potential for a given building. For large commercial buildings where such an
assessment can be done once, facility managers have no easy way to determine if a given building is
meeting its desired DR goals during a DR event, or where a particular building stands with respect
to its ‘peers’ in terms of its DR performance for a given DR event. This type of benchmarking
could be useful in improving the participation rates for buildings that are already
enrolled in DR programs but are not maximizing the peak energy reduction potential of the building.
In fact, on an average only 30-40% of enrolled DR capacity participates in any given
DR event. A continuous assessment of demand response can give a more accurate portrait of DR for each building in a portfolio accounting
for changes in use patterns and operation decisions, contrasting with a single time assessment and can
significantly improve the overall participation in DR events.
Moreover, the time and expense involved in the assessment process precludes most small
and medium sized buildings from this type of manual DR assessment. In the absence of an automated, low cost
assessment, most customers are not able to determine the benefits and effort involved in signing up for a DR program.
The aim of our research is to develop DRSense:
We believe the tool we propose to be necessary to drive a wider adoption and maximize the value of DR.
- a low-cost automated self-assessment tools for commercial building
managers to determine the strategies and DR potential of their building portfolio.
- integrate realtime building data, such as that generated by Cisco’s Building Mediator, and
combine it with advanced data mining and forecasting analytics to provide a real-time assessment
- enhance the current BIM model from a static view to a dynamic view that
incorporates energy considerations.
- integrated into the ‘continuous commissioning’ process so that DR participation of an enrolled building is monitored
and automatically benchmarked against other ‘similar’ buildings in near real-time to determine
if a building is optimally participating in DR and maximizing the DR potential under the
constraints of occupant comfort. The portfolio benchmarking can also be used to compare the
buildings with least DR potential to those with most to evaluate more permanent energy
efficiency interventions. Wider adoption of DR also has clear societal benefits, as recent
studies show that 60% participation in the program leads to 14% decrease in
(3) Proposed approach
Our findings up to date:
(1) Description of data
The Jerry Yang and Akiko Yamazaki Environment and Energy Building (Y2E2)
at Stanford University is a three story building with a subterranean basement. This 166,000 square-foot building is instrumented with 2370 HVAC system measurement points including sensors and control points. Each point collects data every minute using a computer-based data collection system.
Among these measurements, we are most interested in electrical power and heating/cooling energy consumption data. The electrical power consumption data include lighting power for each floor, plug loads for east and west wings of each floor, and power consumption for laboratories, server rooms, mechanical equipments, emergency, standby, and a cafe.
The heating and cooling energy consumptions are measured in terms of steam and chilled water usage.
(2) Forecasting algorithm and preliminary results
We modeled each power consumption time-series as a time-varying autoregressive process and applied adaptive autoregressive model to it. In order to forecast the power consumption, we explored various model order, prediction horizon, and time window for model fitting.
The preliminary results using the lighting power measurements from January to August 2011 show that this model can predict the next 15 minute consumption with less than 5% error,
the next 30 minute with less than 10% error, and the next 45 minute with less than 12% error.
(3) Future plans
- Identify power consumption patterns from similar-looking previous measurements
- Combine different types of information, such as temperature and lighting loads, to improve forecasting
- Conduct sensitivity analysis to identify important factors for power consumption
- Find optimal demand response strategy to minimize energy cost
Publications and presentations:
(1) Original proposal
(Last updated on Mar 20, 2012)