Smart Grid Network Templates for Electricity Distribution Analytics

Smart Grid Network Templates for Electricity Distribution Analytics

Loading
Loading Social Plug-ins...
Language: English
Save to myLibrary Download PDF
Go to Page # Page of 17

Description: This electronic presentation for Electricity Distribution Analytics to smart grid LV network templates. Now, more about Western Power Distribution. DNO oprate 4 network East Midlands, West Midlands, South West, South Wales which have 7.8 million customers, 55,300 square km service area, 221,000 km overhead lines and cables, 185,000 substations.

Normalization values expressed relative to the daily peak. This prevents clusters for different sized substations with similar load shape. Low Carbon Technologies increase in peak load.

Generation impact on network voltages. Less predictability season and day variations peak load timing future load scenarios. Smart Meter Data Events & Readings - Significant new opportunities Potentially large data volumes will need new processing to integrate with existing systems.

Existing network headroom, technology impact, customer behavior. Short term planning – quick connections, minimal reinforcement. Long term planning – keeping the lights on and the bills down.

Improve asset management – understanding loads, voltages, asset environment. Future technology uptake modelled by propensity matrix. Load Scenarios built to reflect assumptions as “lever positions”.

.

 
Author: Jenny Woodruff (Fellow) | Visits: 2116 | Page Views: 2641
Domain:  High Tech Category: Consumer Subcategory: Smart Meters 
Upload Date:
Link Back:
Short URL: http://www.wesrch.com/electronics/pdfEL1AB9FFRAPBW
Loading
Loading...



px *        px *

* Default width and height in pixels. Change it to your required dimensions.

 
Contents:
Grid Analytics 2013
Electricity Distribution Analytics in Action
Load Estimation

About Western Power Distribution
Distribution Network Operator (DNO)
Own and Operate 4 networks

East Midlands
West Midlands
South West
South Wales

7.8 million customers
55,300 square km service area
221,000 km overhead lines and cables
185,000 substations

Already have
“quite a bit” of data.

Electricity Distribution – Analytics in Action

Why are DNOs interested in Data Analytics?
Low Carbon Technologies
Increase in peak load.
Generation impact on network voltages.
Less predictability
• season and day variations
• peak load timing
• future load scenarios

Moving to Smarter Grid
Increased monitoring & communications
Data could have more than one use.
e.g. monitoring installed for automatic load
transfer may assist asset management.
Smart Meter Data
Events & Readings - Significant new opportunities
Potentially large data volumes will need new
processing to integrate with existing systems.

Improving our Business - New data & techniques
Planning & Asset Management
existing network headroom, technology impact, customer behavior
• Short term planning – quick connections, minimal reinforcement.
• Long term planning – keeping the lights on and the bills down.
• Improve asset management – understanding loads, voltages, asset environment.

Operations
• Outages, faster notification, speedier more accurate location.
• Better understanding of network voltages.
• More accurate reporting –outage statistics, losses

Network Planning - Project Examples
Method

Data

Timeframe / application

LV Network
Templates

> 1 year of 10 min real power
averages
>800 substations

Short term planning
2-3 years, then may need to
reformulate templates.

Falcon
Elexon profiles

Connectivity, Estimated Annual
Consumptions, Profile Coefficients,
Half Hourly meter data

Short term planning

Falcon
Energy Model

Connectivity, Building Data,
Efficiency Measures applied,
Demographic data, Non domestic
customer data, SAP ratings

Short term planning
Long term planning
Split by end use
(Potential for operational use)

All methods involve significant data handling.

LV Network Templates

Normalisation

Clustering

Mapping

Scaling

Normalisation
Values expressed relative to the daily peak.
This prevents clusters for different sized substations with similar load shapes.
Clustering (Agglomerative Hierarchical Method – Top Down)
Optimised at 10 clusters – trade off between shape variations & mapping accuracy.
Classification (Multinomial logistic regression)
Which cluster best represents this substation? Uses commonly available fixed data.
Scaling (Non linear regression)
What size should the peak be? Uses same fixed data.

LV Network Templates
Example profiles
Calculated for each
Seasons & Day type

Indicative average voltages

LV Network Templates

LV Network Templates
• Once templates developed, then using classification tool is easy.
• Template refresh period not yet known.
• Data required relatively easy to acquire & process. (All DNOs)
• Very good for domestic substations.
• Harder to predict where half hourly metered customer load is
dominant.

Falcon – Elexon Profiles
Recreate the Settlement process at a smaller scale





Collecting and processing data is quite complex
Works better for larger substations.
No profiles for half hourly metered customers - backward looking.

Energy Model
Physics + Statistics

Detailed heat load &
hot water calculation
Other energy usage uses profiled
reflecting type of house & occupant
e.g. lighting are people in/out,
how many rooms occupied,
what type of lighting,

Future technology uptake modelled by propensity matrix.
Load Scenarios built to reflect assumptions as “lever positions”

Energy Model

• Customer data – Data Protection issues need managing.
• End Use Profiles derived from existing EST research.
• Flexible format – should allow for future tweaking.

Other approaches for Load Estimation
•De Montfort University
•Reading University & SSE

•Low Carbon London
•CSE & Bristol University– Data mining

Customer
Profiles

•LV Network
Templates

Clustering &
Regression
models

Physics
Models
•Falcon Energy Model

Agent Based
Modelling

Energy
Modelling
Smart Meter
& Grid Data

Time Series
Analysis
Other
Clever stuff

•Real time summation.
•Calibrating estimates.

•Dublin Energy
Institute ( Fourier
& Gaussian)
•Auto regression ,
Wavelets

•Machine Learning
•Neural Networks
•Fuzzy Logic
•Probabilistic techniques

Comparing the models – Quality Metrics
Metric

Elexon

LVNT v 6.0

Correlation Coefficient

0.8

0.5

R squared

0.7

0.4

Intercept

-0.5

-8.9

Slope

1.1

1.9

Skew

0.7

2.2

Standard Deviation of Percentage Error

0.8

59.3

Average Absolute Percentage Error

1.0

32.4

Minimum Percentage error

-0.2

0.3

Maximum Percentage error

8.3

457.2

Maximum-Weighted error metric

0.3

4.4

Minimum-Weighted error metric

0.5

27.2

Other metrics might include total energy, load factor, time of peak, etc.

Energy Model Results

Reflections
Finding

Recommendation

It’s still early days for monitoring and
new communications technologies.
Data can be missing /wrong.

Ad hoc ETL processes will need to be
standardised. Early / frequent checks to
ensure that opportunities to capture data
are not lost.

Legacy Data Issues
4 DNOs

Need flexibility in data handling
Allow time for data issues in projects

Hidden Data Quality Issues
External matching rates poor

Cleanse addresses, more validation at
point of data entry.

Engineers are not Statisticians.

Time to develop in-house skills.

Lots of activity – sharing helps everyone

Make LV Network Templates data available
as far as possible.

What Next?
Energy Model

Refine and validate, develop scenarios

LV Network Templates

Validate with other DNO data
Implement into Business As Usual

Cross industry cooperation

Continue to support sharing data and learning

Smart Meter Data

Develop our strategy for integration

Big Data

Continue to look for opportunities

Subscribe
x