POSITIVE ENERGY
TOGETHER®
Seeing is Believing: How Data Visualization
Improves the Assessment of Smart Grid
Demand Response Program Impacts
Presented by
Brian Eakin, OG&E and Ken Seiden, Navigant
•
Background and Demand Response Program Overview•
Analytical Infrastructure•
Data Management•
Analytical Tools•
Visualization and Modeling•
JMP Visualizations•
SAS Model Selections•
Lessons Learned•
Questions/CommentsOG&E
• 9 power plants: 6.8 GW • 780 MW - wind
• 789k customers in OK &
AK
• 30k square mile service
area
• 23k miles of overhead
distribution lines
• 500 substations
About Navigant
• Navigant (NYSE: NCI) has 2,700 employees, including 100+ DSM professionals
• We work across all phases of the DSM lifecycle, resulting in a continuous improvement feedback loop
• Demand Response and Pricing impact evaluations are conducted by the
Economics/ Econometrics and Data Management functional teams
• The results being presented today derive from work conducted when Ken Seiden was affiliated with the Cadmus Group
We provide these services for all aspects of DSM including energy efficiency, demand response, renewable energy, distributed generation, and Smart Grid
Strategic Planning Market Studies and Resource Potential Evaluation, Measurement & Verification Program Design Program Implementation Assistance
AMI/Smart Meters Pilot
•
6,600 smart meters•
Installation began October 2007•
Remote reads &connect/disconnects
•
Validated technology performance & operational savings AMI Smart Meter2010-11 Demand Response Study
•
Purpose: Assess
impact on a
customer’s energy
consumption of:
•
different dynamic rate plans•
multiple levels of enabling technology combined withStudy Design: Technology
Energate Pioneer
LSR Rate$aver®
Study Design—Price Plans Off-Peak/Low Standard Medium Residential Commercial High/Critical
VPP
4.5¢/kWh 11.3¢/kWh 23¢/kWh 46¢/kWh 5.0¢/kWh 10¢/kWh 30¢/kWh 60¢/kWhStudy Design—Price Plans Off-Peak Peak Critical Residential Commercial 4.2¢/kWh 23¢/kWh 46¢/kWh 4.7¢/kWh 30¢/kWh 60¢/kWh
TOU
Analytical Infrastructure
• Teradata - Enterprise Data Warehouse (EDW)
– Central repository for all interval meter data
– Massively Parallel Processing – SAS Integration Technologies • SAS – Forecast Server and
Visual Data Discovery (VDD)
– High Performance Forecasting – VDD Software for Visualization
Unique Aspects of OG&E’s Analytics Platform
• To our knowledge OG&E is the first utility to assess a Smart Grid DR program with an in- or on-database analytics system and:
• Utilize industry-specific regression model protocols
• Use with external, independent evaluators as required • Obtain knowledge transfer, including all computer code,
such that OG&E staff can internalize the analytics
• Maintain ultra-conservative data security requirements— not a single byte of data left OG&E’s servers
Visualization and Modeling
• First step in the modeling process is to visually inspect the data and identify any trends or unique behaviors
– Identifying trends and unique behaviors is an
important step in informing the modeling process
• SAS Visual Data Discovery (VDD) is used to interact and explore the interval usage data
– VDD journals were built for commonly used visualizations and data transformations
Visualization and Modeling
• Build models to capture the following: – Weather Effects – Price Effects – Technology Effects – Age/Income Effects – Calendar Effects • Model Types
– Constant Elasticity of Substitution (CES) • Estimate Inter-hour substitution
• On-Peak vs. Off Peak – Average Daily Demand
Visualization and Modeling
• Constant elasticity of substitution model with
customer fixed effects:
– ln(Q
P/Q
OP)
i=
σ
ln(P
P/P
OP)
i+
δ
(CDH
P– CDH
OP)
i+
λ
i+
ε
i–
σ
is the elasticity of substitution between peak
and off peak energy use
• Daily demand model
– ln(Q
i) =
θ
ln(P
i) +
δ
CDH
i+
λ
i+
µ
iHourly Demand Forecast Model
• Several candidate structural econometric model
specifications were developed on a sample of the data • Model specifications included:
– Price terms
– Weather terms (CDH)
– Weekly, Monthly, Hourly, Weekday/Weekend dummies
– Flags for pre/post event hours
Visualization and Modeling
• SAS High Performance Forecasting (HPF) and ETS were used to evaluate candidate forecast models. • For each series in the forecast hierarchy, these
candidate models were evaluated, and the one with the best fit was selected to produce the forecast
• To achieve a consistent set of forecasts across those levels, it was necessary to calibrate the forecasts to a single level of the hierarchy
Key Takeaways
• The analytical infrastructure needed to evaluate a Demand Response program is significant
– SAS and Teradata have been able to meet those needs
• Interactive visualization of interval data is an important part of building the appropriate model
– Extra time spent on data prep and visualizations will save time during modeling
• An efficient data structure and appropriate model selections will streamline the evaluation process
– The data structure for modeling/analysis may vary from the overall Utility Logical Data Model (Teradata table structure)
POSITIVE ENERGY
TOGETHER®
Questions/Comments
Ken Seiden
Navigant DR and EE Economics (303) 728-2479
Brian Eakin
OGE Energy Corp.