Compu&ng Electricity Consump&on Profiles from Household Smart Meter Data
O. Ardakanian, N. Koochakzadeh, R. P. Singh, L. Golab, S. Keshav University of Waterloo
3rd Workshop on Energy Data Management
March 2014 Real-Time Distributed Control for Electric Vehicle Chargers
Omid Ardakanian, S. Keshav, and Catherine Rosenberg Cheriton School of Computer Science, University of Waterloo
Abstract
Problem Definition
Solution Approaches
spatial & temporal uncertainties
EV charging load is significant
~
impacts on the grid
At high penetrations, uncontrolled electric vehicle (EV) charging has the potential to cause line and transformer congestion in the distribution network. Instead of upgrading components to higher nameplate ratings, we investigate the use of real-time control to limit EV load to the available capacity in the network. Inspired by rate control algorithms in computer networks such as TCP, we design a measurement-based, real-time, distributed, stable, efficient, and fair charging algorithm using the dual-decomposition approach. We show through extensive numerical simulations on a test distribution network that this algorithm operates successfully in both static and dynamic settings, despite changes in home loads and the number of connected EVs. We find that our algorithm rapidly converges from large disturbances to a stable operating point. Given an acceptable level of overload, we show in a dynamic setting that only 30 EVs could be fully charged without control, whereas up to around 300 EVs can be fully charged with our control algorithm, which compares well with the ideal maximum of 383 EVs.
Control
Pre-dispatch Real-time
Centralized
Utility-oriented
User-oriented
Distributed
Utility-oriented
User-oriented
Control Architecture
Single Snapshot Optimization Problem
Results
References
max log 𝑟𝑎𝑡𝑒∈𝒮
subject to 0 ≤ 𝑟𝑎𝑡𝑒 ≤ 𝑚𝑎𝑥𝑟𝑎𝑡𝑒 ∀𝑠 ∈ 𝒮 𝐸𝑉 𝑙𝑜𝑎𝑑 + ℎ𝑜𝑚𝑒 𝑙𝑜𝑎𝑑 ≤ 𝑠𝑒𝑡𝑝𝑜𝑖𝑛𝑡 ∀𝑙 ∈ ℒ
MCC nodes
Smart chargers
Every line/transformer is associated with a nameplate rating and a setpoint
line/
trans
form
er lo
adin
g
1 2 3 4
(measured by an MCC node)
(setpoint)
available capacity
Certain distribution branches may be subject to significant overloads, while the whole system has sufficient capacity
this provides proportional fairness
O. Ardakanian, S. Keshav, C. Rosenberg, “Real-time Distributed Control of Electric Vehicle Charging”, submitted to IEEE Transactions on Smart Grid. O. Ardakanian, C. Rosenberg, S. Keshav, “Distributed Control of Electric Vehicle Charging”, Proc. ACM e-Energy, May 2013. O. Ardakanian, C. Rosenberg, S. Keshav, “Fast Distributed Congestion Control for Electrical Vehicle Charging”, ACM SIGMETRICS Performance Evaluation Review, December 2012.
Your Smart Meter is Watching!
From: hSp://www.thestar.com/opinion/2009/11/17/your_smart_meter_is_watching.html 2
Smart Meters are Ubiquitous
Electricity Gas Water
Energy Retail Associa&on Projects
Trials
3
Mo&va&on for Smart Metering
4
Electricity Consump&on Profiles
5
The Need for Electricity Consump&on Profiles
6
Prior Work on Electricity Consump&on Profile Genera&on
• Rely on data that is not easily available
• Use a black box method which is not interpretable
• Are not robust to noise
• Do not remove the effect of temperature and ac&vity – cannot be extended to other regions and ac&vity paSerns
7
Takeaways
• Electricity consump&on profile genera&on has several applica&ons
• A profiling framework must be simple, interpretable, yet prac&cal
• Time series analy&cs can be used to generate such consump&on profiles
8
Key Observa&ons
9
Residen&al Load Varies with Temperature
10
Residen&al Load Varies with Ac&vity
11
Residen&al Load Varies with Ac&vity
Time of Day
0 1 2 3 4 5 6 7 8 9 10 11 12
Power (W
)
12
Ac&vity level must be inferred from data
Our Methodology
13
PARX Model
recent history temperature-‐sensi6ve load
outliers intercept and noise terms
season index
14
PARX Model – cont’d
Cooling
Hea&ng
Overhea&ng
15
Handling Outliers
10% of Observa6ons
10% of Observa6ons
16
Compu&ng Consump&on Profiles
• Parameter Es&ma&on – Number of seasons – Coefficients
• Subtrac&ng the effect of exogenous variables
17
Weekday and Weekend Profiles
18
Comparison – Predic&ve Power • Data set – Residen&al hourly electricity consump&on data of 1000 homes from March 2011 to October 2012
– Hourly air temperature data of that region
• Prior work – 3-‐Line Method
• Fits a tree-‐piece linear regression aker removing outliers – Hourly Mean – Convergent Vector
• The same as ours but does not remove the effect of exogenous variables
19
Results Av
g. RMSE
20
Conclusions • Electrical consump&on profile genera&on is important and has many applica&ons – water and gas consump&on
• Time series auto-‐regression framework enables us to remove the effects of temperature and ac&vity
• We demonstrated a simple, interpretable, and prac&cal profiling model with high predic&ve power
21