Long Term Probabilistic Load Forecasting at NCEMC

1
Long Term Probabilistic Load
Forecasting at NCEMC
Jason Wilson
Load Research Analyst
North Carolina Electric Membership Corporation
jason.wilson@ncemcs.com
2
Agenda
•
•
•
•
•
Introduction
Multiple Linear Regression
Forecasting Process
Results
Takeaways
3
Many Masters to Serve
4
5
6
Forecasting at NCEMC
• Company Functions
Power Supply
Asset
Management
Resource
Planning
Portfolio
Management
• Types of Forecasts
Transmission
Services
Regulatory
Affairs
Wholesale
Rates
7
Long Term Load Forecasting
• Business needs
–
–
–
–
Power Supply
System Planning
Risk Management
Portfolio Analysis
• Challenges
–
–
–
–
Data Quality
Weather Diversity
Time Constraints
Regulatory Requirement
8
Long Term Forecasting at NCEMC
• Several decades of evolution
– Since formation of cooperative
• Past implementation
– Using monthly data
– Moved to hourly data using ARIMAX model
• Desire for improvements in LTLF
– All forecasts can be improved
– More reasonable load shapes
– Easier to understand and explain
9
Multiple Linear Regression
10
Pay attention to Patterns in Data
8000
7500
7000
6500
6000
5500
Forecast Trend Drivers
GDP ($Mil) and Households
(Thousands)
5000
4500
4000
3500
3000
households
gdp_typ
11
Load-Temperature Relationship
• Nonlinear
• Changes throughout year
• Changes hourly
12
Load-Temperature Relationship
13
14
Forecasting Process
• Started with a short-term forecasting model
– Hong (2010), Short Term Electric Load Forecasting, NCSU
PhD Dissertation
• Augmented the STLF model substituting GDP for
trend variable
• Developed weather and economy scenarios
15
Long Term Load Forecasting
• Low resolution monthly data
• Inaccurate forecasts
• Unexplainable errors
• Point Forecast
16
Long Term Load Forecasting
• 1 to 50 years ahead, typically 10 to 30 years
• Planning, rates, finance, DSM, portfolio and risk
management
• Weather data quality issues
• Unfair to judge LT forecasts via point forecasts
• Probabilistic approach preferred
17
Results
18
Results
19
Results
20
Results
21
Budget to Actuals
Typical
2013 Mild Weather Low Growth
3,500
3,000
2,500
2,000
1,500
1,000
500
0
1
2
3
4
5
6
7
8
9 10 11 12 1
2
3
4
2012
Energy
5
6
7
8
2
3
4
5
Weather Normalized
Typical
7
8
9 10 11 12
2013 Mild Weather Low Growth
Actual
Actual
6
2014
2013
1,500
GWh
9 10 11 12 1
Severe
Forecast
Weather Normalized
Actual
Actual
Forecast
MW
Demand
1,000
500
0
1
2
3
4
5
6
7
8
2012
9 10 11 12 1
2
3
4
5
2013
6
7
8
9 10 11 12 1
2
3
4
5
6
2014
7
8
9 10 11 12
22
Process Improvements
• Smart Grid brings high resolution data
• Hourly data can have quality issues due to human,
meter and communication errors
• Weather and Economy data quality most
uncontrollable components
• Increasing weather stations to 25 from 6 improved
accuracy
23
Takeaways
• Data quality issues and solutions
• Hourly data improves accuracy
• Scenario based forecast provides range of options for
planners and portfolio managers
• Potential to expand hierarchies to individual
customer level using smart meter data
• Never-ending process of improvement
24
Contact Information
Contact: Jason Wilson
Load Research Analyst, NCEMC
Jason.Wilson@ncemcs.com
http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6595138&url=http%3A%2F%2Fieeexplore.iee
e.org%2Fiel7%2F5165411%2F5446437%2F06595138.pdf%3Farnumber%3D6595138