66
Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation Antonio Lorenzo Dissertation Defense April 14, 2017

April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite

Imagery, and Data Assimilation

Antonio LorenzoDissertation Defense

April 14, 2017

Page 2: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

2

Problem & Hypothesis

Problem: imperfect solar power forecasts Hypothesis 1: ground sensors will improve forecastsHypothesis 2: hybrid methods will reduce errors

Pow

er

Time

Page 3: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

3http://www.utilitydive.com/news/caiso-notches-record-serving-567-of-demand-with-renewable-energy-in-one/439085/

Page 4: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

4

Background: Solar Variability

Power from solar plants can be highly variable due to clouds

Page 5: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

5

Background: Solar Variability

Page 6: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

6

Background: Solar Variability

A 20 MW ramp is about equivalent to the demand of 10,000 homes

Page 7: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

7

Coal provides base power that cannot change quickly

Background: Solar Variability

Page 8: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Background: Solar Variability

8

Output from a gas turbine can be ramped quickly. Helps backup solar

Page 9: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

9

Utilities are accustomed to controlling their generators from a control room

Background: Solar Variability

Page 10: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

10

Background: Solar Variability

Utilities are accustomed to controlling their generators from a control room

Clouds/weather control the output of solar power plants

Page 11: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

11

UA Forecasting Partners

TEP

APSPNM

Page 12: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

12

Operational Forecasting for Utilities

● Our work includes a web page with graphics and information meant to help the utilities understand and use the forecasts

● Also have a HTTP API for programmatic access

Page 13: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Irradiance to Power Conversion

1313

PV System Model

Page 14: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Irradiance to Power Conversion

1414

PV System Model

Page 15: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

15

Context & Hypothesis

• TEP, APS, PNM need solar forecasts because variability is an issue

• Hydrology & Atmospheric Sciences provides forecasts from a weather model (WRF)

• Physics department explored cloud camera and sensor network approaches

Hypothesis 1: ground sensors will improve forecastsHypothesis 2: hybrid methods will reduce errors

Page 16: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Outline of my work

• Benchmark forecasts• Irradiance network forecasts• Satellite data assimilation

1616

Page 17: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

17

Clear-Sky Index

Clear-Sky Index = Observations / Clear-Sky Expectation

Page 18: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

18

Terms

Page 19: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

19

Benchmarks

Page 20: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

20

Persistence Forecasts

• Forecast assuming a quantity doesn’t change, e.g.

• “the power output tomorrow will be the same as today”

• “the GHI in 15 minutes will be the same as it is now”

Page 21: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

21

Persistence Forecasts

Page 22: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

22

Persistence Forecasts

Page 23: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

23

Persistence Forecasts

Page 24: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

24

Benchmarks

Page 25: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

25

Benchmarks

Page 26: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

26

Benchmarks

Page 27: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Outline of my work

• Benchmark forecasts• Irradiance network forecasts• Satellite data assimilation

2727

Page 28: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

28

Irradiance Sensor Network

• Custom sensors• Inexpensive ($500)• Solar panel + battery

power• GSM modem to transmit

data in real-time• Built and deployed in

2014• Rooftop PV power data

• 5 minute average power• Proxy for irradiance• Thanks to Technicians for

Sustainability! A. T. Lorenzo, W. F. Holmgren, M. Leuthold, C. K. Kim, A. D. Cronin, and E. A.

Betterton, “Short-term PV power forecasts based on a real-time irradiance monitoring network,” in 2014 IEEE 40th Photovoltaic Specialist Conference (PVSC), 2014, pp. 0075–0079.

Page 29: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

29

Network Forecasts: Basic Premise

• One sensor predicts the output of another

• Map cloud field with enough sensors

Page 30: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

30

Network Forecasts: Implementation

Page 31: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

31

Network Forecasts: Results

• 20% improvement over clear-sky index persistence on average for 3 months of data

• Surprise: this 20% improvement was seen even at 2 hours

• Comparing only RMSE is not enough

Page 32: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

32

Taylor Diagram

Source: Taylor, 2001 doi: 10.1029/2000JD900719

Page 33: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

33

Taylor Diagram

Page 34: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

34

Taylor Diagram

Page 35: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

35

Taylor Diagram

Page 36: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

36

Taylor Diagram

Page 37: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

37

Taylor Diagram

Page 38: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

38

Taylor Diagram

● How the forecast will be used is important to assessing forecast quality

● Analyzing only RMSE may not be enough

Page 39: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

39

Taylor Diagram for Network Forecasts

● Network forecasts may have larger RMSE at some time horizons, but they better capture the observed variability

Page 40: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

40

Published work on irradiance network

A. T. Lorenzo, W. F. Holmgren, and A. D. Cronin, “Irradiance forecasts based on an irradiance monitoring network, cloud motion, and spatial averaging,” Sol. Energy, vol. 122, pp. 1158–1169, 2015.

Page 41: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

41

Network Forecasts: Results

Hypothesis 1: ground sensors will improve forecasts ✔Hypothesis 2: hybrid methods will reduce errors

Page 42: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

42

Network Forecast: Limitation

Page 43: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Irradiance Sensors• Strengths

• High accuracy• High temporal

resolution• Weaknesses

• Low spatial coverage• Expensive to deploy and

maintain

43

Strengths & Weaknesses

Satellite Imagery• Strengths

• Broad coverage• Freely available

• Weaknesses• Errors introduced

because GHI is a derived quantity

• Low temporal resolution

Page 44: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Outline of my work

• Benchmark forecasts• Irradiance network forecasts• Satellite data assimilation

4444

Page 45: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

45

Satellite Derived Irradiance

model

Light reflected from the tops of clouds Light that gets through clouds

Page 46: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

46

Satellite GHI Error

Page 47: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Satellite-derived GHI estimate

• Two conversion models:• SE: A semi-empirical

model that applies a regression to data from visible images

• UASIBS: A physical model that estimates cloud properties and performs radiative transfer

• Nominally 1 km resolution• Using 75 km x 82 km area

over Tucson

47

Page 48: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

48

Optimal Interpolation

• Bayesian technique derived by minimizing the mean squared distance between the field and observations

• Is the best linear unbiased estimator of the field

• Same as the update step in the Kalman filter

OptimalInterpolation

Better satellite-derived estimate of GHI

Satellite image from http://goes.gsfc.nasa.gov/text/goesnew.html

Page 49: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

49

Optimal Interpolation

OptimalInterpolation

Better satellite-derived estimate of GHI

Satellite image from http://goes.gsfc.nasa.gov/text/goesnew.html

Satellite Derived Irradiance:

Observations:

Page 50: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

OI Algorithm

50

xa= xb + W(y - Hxb)Better GHI estimate

Maps points from satellite

image to observations

Satellite image from http://goes.gsfc.nasa.gov/text/goesnew.html

Page 51: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

OI Algorithm

51

xa= xb + W(y - Hxb)

W = PHT(R + HPHT)-1

Better GHI estimate

Need to a way to estimate these error covariances

Maps points from satellite

image to observations

Satellite image from http://goes.gsfc.nasa.gov/text/goesnew.html

Page 52: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Error Covariances: P and R

• Decompose P into diagonal variance matrix and correlation matrix:

P = D1/2 C D1/2

• Prescribe a correlation between image pixels based on the difference in cloudiness to construct C

• Compute D from cloud free training images

• Assume observation errors are uncorrelated and estimate R from data

52

Page 53: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

53

OI Parameters

Distance Metrics that I studiedCorrelation Functionsthat I studied

500 training images * 2 models * 6 fold cross validation * 50 height adj. * 2 corr. methods * 3 corr. fcns. * ~10 corr. lengths * ~10 inflation params = 200 million OI analyses

Page 54: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

54

OI in action

Page 55: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Results (one image)

5555

Page 56: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

56

Page 57: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

57

• No clouds in satellite albedo image

• No clouds in analysis using cloudiness correlation ✔

• Clouds with a smooth gradient in analysis using spatial correlation ✖

Page 58: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Predicted vs Measured Scatterplot

• Dashed 1-to-1 line indicates perfect model

• Background is biased with time of day dependence

• Analysis removes bias and time dependence

58

Page 59: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Optimal Interpolation Results

• 900 verification images analyzed

• Six-fold cross-validation over sensors performed

• The large bias for the empirical model was nearly eliminated

• RMSE reduced by 50%

59

A. T. Lorenzo, M. Morzfeld, W. F. Holmgren, and A. D. Cronin, “Optimal interpolation of satellite and ground data for irradiance nowcasting at city scales,” Sol. Energy, vol. 144, pp. 466–474, 2017.

Page 60: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

60

Improved Satellite GHI Error

Page 61: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

61

Improved Satellite GHI Error

Hypothesis 1: ground sensors will improve forecasts ✔Hypothesis 2: hybrid methods will reduce errors ✔

Page 62: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

62

Future Work: Satellite Forecast

Hypothesis 1: ground sensors will improve forecasts ✔Hypothesis 2: hybrid methods will reduce errors ✔Hypothesis 3: hybrid methods will improve forecasts

Page 63: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

63

Future Work: Fusing together forecasts

Hypothesis 1: ground sensors will improve forecasts ✔Hypothesis 2: hybrid methods will reduce errors ✔Hypothesis 3: hybrid methods will improve forecasts

Page 64: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

64

Hypothesis & Results

Page 65: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

65

Summary

• Designed and deployed irradiance sensor network,• Used ground sensors to make short-term forecasts

that are superior to WRF or persistence,• Combined sensor data with satellite images to

improve irradiance nowcasts.

Page 66: April 14, 2017 Dissertation Defense Antonio Lorenzo · 2020-02-05 · Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

Thank you!

66

• Alex Cronin• Matti Morzfeld• BG Potter• Will Holmgren• Mike Leuthold• Eric Betterton• Mike Eklund• Ardeth Barnhart• Travis Harty• Rey Granillo

• Technicians for Sustainability

• Kevin Koch• David Stevenson• Mike Kanne

• Tucson Electric Power• Ted Burhans• Carlos Arocho• Nicole Bell• Carmine Tilghman• Sam Rugel

• Arizona Public Service• Jim Hamilton• Marc Romito