26
Energy Forecasting to Maximize Use of Renewables Jeff Lerner, PhD Manager of Forecast Operations, Vaisala Eric Grimit, PhD Senior Scientist, Vaisala Thursday, January 29, 2015 Webinar When to consider a probabilistic approach

Energy Forecasting to Maximize Use of Renewables

Embed Size (px)

Citation preview

Page 1: Energy Forecasting to Maximize Use of Renewables

Energy Forecasting to Maximize Use of Renewables

Jeff Lerner, PhDManager of Forecast Operations, Vaisala

Eric Grimit, PhDSenior Scientist, Vaisala

Thursday, January 29, 2015

Webinar

When to consider a probabilistic approach

Page 2: Energy Forecasting to Maximize Use of Renewables

Page 2 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Outline

1) Webinar take home messages

2) Statement of the problem

a. Common practices for scheduling wind power

b. How risk and uncertainty is translated to action

3) Probabilistic forecasts

a. Description

b. Examples

4) Where probabilistic information is contained in a wind power

forecast

5) Use case example: “Avoiding downside risk through the use of

an appropriate prediction quantile”

6) Who benefits from probabilistic forecast use?

7) Advanced Application

Page 3: Energy Forecasting to Maximize Use of Renewables

Page 3 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Webinar Take Home Messages:

Information contained within a wind power forecast is not

fully utilized

Understanding risk and uncertainty can facilitate the use of

probabilistic forecasts

Use of probabilistic forecasts may prove advantageous to

bottom line of operating a wind plant

Probabilistic forecasts aren’t necessarily complicated,

as will be shown in the application of a fixed exceedance

probability

Page 4: Energy Forecasting to Maximize Use of Renewables

Page 4 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Statement of the problem: Common practices for scheduling wind power

These approaches may help reduce penalties, but lose sight of the

“upside potential”

Passing through a deterministic forecast (from ISO or forecast provider)

“Haircutting” (aka scaling) the deterministic

Scheduling only what you can cover during high uncertainty periods (e.g., known reserve capacity)

Page 5: Energy Forecasting to Maximize Use of Renewables

Page 5 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Probabilistic Forecasts:Description and contrast with deterministic forecast

Probabilistic Deterministic

Rain likely, 70% chanceTomorrow’s high temperature

forecast is 48°F

Forecast is for 6–10 inches of snowPJM’s wind forecast is for 1500 MW at

7:00 a.m. tomorrow

There’s a 58% likelihood of an

El-Nino next yearMy tax return will be $528

New England has a 56% probability to

win the Super BowlSeahawks 24, Patriots 20

Probabilistic forecasts assign a likelihood to

each of a number of potential outcomes

Deterministic forecasts are forecasts of a

specific magnitude and time. They contain no

information on the uncertainty.

Page 6: Energy Forecasting to Maximize Use of Renewables

Page 6 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Grocery Store – Choosing a Line:How risk and uncertainty translate to action

Risk: $50

Uncertainty:

3 lines, different

lengths, speeds,

and rules

Possible

solutions…

An action is made based on the perceived seriousness of the risk and

uncertainty associated with the different possible solutions.

Page 7: Energy Forecasting to Maximize Use of Renewables

Page 7 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Outdoor Wedding, Seattle in June:How risk and uncertainty translate to action

Chance of rain: 30%

Risk: Personality,

economic, functional

Uncertainty:

What does 30%

chance mean?

Solutions…

An action is made based on the perceived seriousness of the risk and

uncertainty associated with the different possible solutions.

Page 8: Energy Forecasting to Maximize Use of Renewables

Page 8 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Would you deploy your snow removal equipment and salt reserves?

Page 9: Energy Forecasting to Maximize Use of Renewables

Page 9 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

When would you declare a “state of emergency”?

Page 10: Energy Forecasting to Maximize Use of Renewables

Page 10 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

When would you declare a “state of emergency”?

Page 11: Energy Forecasting to Maximize Use of Renewables

Page 11 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Understanding risk and uncertainty:Wind Power Forecast Prediction Quantiles

Easy to interpret, but no context

Uncertainty is not contained in this forecast

Page 12: Energy Forecasting to Maximize Use of Renewables

Page 12 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

F1

8

F8

0

5 GWP10

P90

Understanding risk and uncertainty:Wind Power Forecast Prediction Quantiles

18-hour: 2800 MW prediction interval; 80-hour: 5100 MW prediction interval.

Page 13: Energy Forecasting to Maximize Use of Renewables

Page 13 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

F1

8

F8

0

5 GWP10

P90

Prediction Interval expresses the probability that the actual production

will be observed in this band.

Understanding risk and uncertainty:Wind Power Forecast Prediction Quantiles

18-hour: 2800 MW prediction interval; 80-hour: 5100 MW prediction interval.

Page 14: Energy Forecasting to Maximize Use of Renewables

Page 14 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Where probabilistic information is contained in a wind power forecast

A Prediction Interval is an estimate of a

probability interval in which future

observations will fall.

It is usually based on previous forecast errors.

A Prediction Quantile, z, is a non-

exceedance probability. A decile (every 10th

percent) or a confidence level are examples

For example, if P30 = 75 MW, then there is 70%

probability that the observation will not exceed

75 MW.

[μ - zσ, μ + zσ]

μ = mean

σ = standard deviation

z = prediction quantile

P30

Page 15: Energy Forecasting to Maximize Use of Renewables

Page 15 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Where probabilistic information is contained in a wind power forecast

A Prediction Interval is an estimate of a

probability interval in which future

observations will fall.

It is usually based on previous forecast errors.

A Prediction Quantile, z, is a non-

exceedance probability. A decile (every 10th

percent) or a confidence level are examples

For example, if P30 = 75 MW, then there is 30%

probability that the observation will exceed

75 MW.

[μ - zσ, μ + zσ]

μ = mean

σ = standard deviation

z = prediction quantile

P30

Page 16: Energy Forecasting to Maximize Use of Renewables

Page 16 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Considerations From Energy Scheduler’s Perspective

Imbalance Charges

Transmission:

hub or node level congestion

Day-Ahead minus Real-Time

(DART) price spread

Transmission Rights

Page 17: Energy Forecasting to Maximize Use of Renewables

Page 17 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

The Problem: Downside Risk Exposure

3TIER Blend minimizes bias and MAE; 50% downside risk

Page 18: Energy Forecasting to Maximize Use of Renewables

Page 18 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

One Strategy: Scaling the Forecast

50% scaled 3TIER Blend aka “haircut technique”; 28% downside risk

Page 19: Energy Forecasting to Maximize Use of Renewables

Page 19 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Vaisala Probabilistic Forecasts

3TIER Blend and prediction quantiles:

P10, P20, P30, P40, P60, P70, P80, P90

Page 20: Energy Forecasting to Maximize Use of Renewables

Page 20 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Another Strategy: Choosing Risk Tolerance

3TIER Blend and 3TIER 30th quantile or P70;

P70 estimates 30% downside risk

Page 21: Energy Forecasting to Maximize Use of Renewables

Page 21 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Check the Risk Exposure

3TIER P70 has 27% downside risk;

“Haircut method” was 28% downside, very close to P70

Page 22: Energy Forecasting to Maximize Use of Renewables

Page 22 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Net 20.5 GWhmore over a 6-month period!

3TIER P70 and 50% scaled forecast similar risk exposure (27% vs. 28%)

3TIER P70 scheduled 20.5 GWh more energy than scaled forecast!

Compare the Strategies Over Time

Page 23: Energy Forecasting to Maximize Use of Renewables

Page 23 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Net 20.5 GWhmore over a 6-month period!

3TIER P70 and 50% scaled forecast similar risk exposure (27% vs. 28%)

3TIER P70 scheduled 20.5 GWh more energy than scaled forecast!

*Reliable risk and more energy scheduled, day-ahead

Compare the Strategies Over Time

Page 24: Energy Forecasting to Maximize Use of Renewables

Page 24 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Advanced Application of Probabilistic Wind Power Forecasts

» Accounting for risk

varying with time

» Energy prices/DART spread

» Load forecast error

» Transmission congestion

» Develop rules/algorithm

customized to risk profile

» Choose optimal forecast

percentile based on

expected risk

P50P60

P70

P80

P90

P40

P30

P20

P10

Time

Pri

ce

Sp

rea

d (

$)

Objective risk assessment takes the emotion out of the forecast.

Page 25: Energy Forecasting to Maximize Use of Renewables

Page 25 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Questions

Page 26: Energy Forecasting to Maximize Use of Renewables

Page 26 / 28 January 2015 / Probability Forecast Webinar/ ©Vaisala

Questions…

Information contained within a wind power forecast is not

fully utilized

Understanding risk and uncertainty can facilitate the use of

probabilistic forecasts

Use of probabilistic forecasts may prove advantageous to

bottom line of operating a wind plant

Probabilistic forecasts aren’t necessarily complicated,

as will be shown in the application of a fixed exceedance

probability

WEBINAR TAKE-HOME MESSAGES