Transcript Document

Sustainable Engineering MSc Project

Uncertainty in Wind Energy Yield Predictions With Sgurr Energy

• • • •

Robin Odlum Sheikh M. Ali Vijay Dwivedi Antonio Sanchez

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Our Project:

- To study the variation in correlation parameters for the Measure-Correlate-Predict (MCP) method between the wind speed data of pseudo wind farm site and meteorological site.

- A case study: Behaviour of power curve for a wind farm.

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

In predicting wind energy yield from a wind farm, there is uncertainty in:

• • • •

Data Acquisition / Processing Modelling (WAsP, Windfarm etc) Losses in Energy Production Long Term Prediction – MCP Method

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Data Acquisition/Processing:

• • • Uncertainty arises from measuring instrument errors, wind shear, density correction etc.

High probability of human, systematic or random errors reduce the reliability of data Research in this area is not of high interest to our group, so excluded from our project

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Modeling (WAsP, Windfarm etc)

• • Use of assumptions in the modelling software reduces the reliability of energy yield prediction.

Research would require access to the source codes of the software and more resources, so excluded from this project.

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Losses:

• There is uncertainty due to different types of losses in wind energy production on a wind farm like Wake losses, Turbine unavailability, • Blade contamination etc.

Different factors are used to correct the energy output from a wind farm to make better • prediction.

A detailed investigation could become commercially sensitive. However a brief case study is presented to illustrate the main issue.

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Long Term Prediction using MCP Method:

MCP (Measure Correlate Predict) is a statistical technique used for predicting the long term wind resource at a target site. Wind speed and direction measurements from a target and a reference site are correlated and the correlation parameters

Reference Site Wind Speed Wind Direction Target Site Wind Speed Wind Direction

(m,c) are applied to long term historic data of reference site to predict long term wind resource

Correlation slope m: it represents the change in velocity of target site with respect to the reference site intercept c: Correlation Parameters:

at target site.

it gives the velocity of target site when the velocity of reference site is zero

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Long Term Prediction using MCP Method:

Different MCP techniques have been used giving different results. We found it interesting to research in the variation in correlation It was a reasonably parameters with time.

and Dumfries (met site) as: 35 We were interested in 30

Reference Site Wind Speed Wind Direction Target Site

15 Sgurr Energy also 10

Wind Speed Wind Direction

showed its interest 5 0 0 5 10 Y_1979 Y_1983 Y_1984 Y_1985 Y_1986 15 20 25 Wind Speed, m/s (Dumfries, met site)

Correlation slope m: intercept c: Correlation Parameters: it represents the change in velocity of target site with respect to the reference site it gives the velocity of target site when the velocity of reference site is zero

Scope of Project

Uncertainty in Wind Energy Yield Predictions

Sources of Uncertainty Data Acquisition Long Term Prediction Modelling Losses in Energy Production

Scope of Project

Uncertainty in Wind Energy Yield Predictions

Sources of Uncertainty Data Acquisition Long Term Prediction Modelling Losses in Energy Production An Interesting Case Study Analyzing Power Curve Performance Traditional MCP Method Modified MCP Method

. . . . .

- 10 year ref site data, 1 year target site data - Concurrent data gives Correlation parameters (m,c) applied to 10 years ref site data to predict next 10 years velocity - Predict next 10 years energy target site data parameters (m,c) for each concurrent year years ref site data to predict next 10 years velocity

Study of Variation in Linear MCP Correlation Parameters (Linear Regression Parameters)

- 10 year ref site data, 1 year target site data - Concurrent data gives correlation parameters (m,c) applied to 10 years ref site data to predict next 10 years velocity

Predict next 10 years energy yield

Traditional MCP Method Modified MCP Method - 10 years ref site data, multiple years of target site data - Concurrent data gives correlation parameters (m,c) for each concurrent year - Mean value of m and c is applied to 10 years ref site data to predict next 10 years velocity

Predict next 10 years energy yield Steps to carry out the Linear MCP parameters study:

• Select the reference and target sites • Carry out linear regression analysis • Investigate the uncertainty in linear regression coefficients • Compare the energy output between conventional MCP method and modified MCP method Compare Predicted energy to Actual energy for the next 10 years using the predicted and actual wind data of pseudo wind farm for next 10 years

Results

Variation in Linear MCP Parameters Met Station Number

1634-01 1646-01 2083-01 6620-02

D A Name

Turnhouse Edinburgh Lynemouth Dumfries

B C Here we select any two met stations to take one as the reference site and other as the pseudo wind farm site. The purpose is to study the variation in correlation parameters for Measure-Correlate-Predict (MCP) method between the wind speed data of pseudo wind farm site and reference site. We have 20 years (1978~1997) wind data available for these met sites.

Location

Near Edinburgh Airport Blackford Hill Village in Northumberland, England Drungans

Latitude Longitude Terrain category

55.95

55.93

-3.35

-3.19

Complex Complex 55.2

55.05

-1.54

-3.64

Flat Flat

Variation in Linear MCP Parameters We have started evaluating the variation in MCP parameters by making four pairs from the given sites in the following fashion:

We make pairs of site on the basis of their terrain which are classified as ‘complex’ or ‘flat’

Pair No 1 2 3 4

Blackford Hill Lynemouth Turnhouse Dumfries

Pseudo wind farm site

Turnhouse Turnhouse Lynemouth Lynemouth

Terrain comparison

Complex : Complex Flat : Complex Complex : Flat Flat : Flat

Regression Analysis

MCP Parameters Study

Flat-Flat site combination

- Flat Reference Site - Flat Target Site

We can see clearly from this graph that the slope and intercept are varying with the passage of time.

35 30 25 20 15 10 Plot of Wind Speed for Lynemouth (pseudo wind farm) and Dumfries (met site) Y_1979 Y_1983 Y_1984 Y_1985 Y_1986 5 0 0 5 10 15 20 Wind Speed, m/s (Dumfries, met site) 25

MCP Parameters Study

Plot of Wind Speed for Turnhouse (pseudo wind farm) and BlackfordHill (met site) 16 14 12 10 8 6 4 2 0 0 5 10 15 20 Wind Speed, m/s (Blackford Hill met site) 25 Y_1979 Y_1981 Y_1983 Y_1984 Y_1986

- Complex Reference Site - Complex Target Site

Plot of Wind Speed for Turnhouse (pseudo wind farm) and Lynemouth (met site) 16 6 4 2 0 0 14 12 10 8 5 10 15 20 Wind Speed, m/s (Lynemouth met site) 25 Y_1978 Y_1979 Y_1982 Y_1984 Y_1985

- Flat Reference Site - Complex Target Site

MCP Parameters Study

65% Estimation of Deviation in Energy output for various Terrain Catagories 52% 58% Deviation in energy output based on modified MCP method Deviation in energy output based on traditional MCP method 31% 31% 25% 23% 23% 8% 0% Benchmark data for energy output Flat-Flat Complex-Flat Flat-Complex Complex Complex M odified MCP method predicts well in comparison with traditional one.

MCP Parameters Study

Uncertainty in Estimation of Energy output for Flat Complex Terrain Category by Traditional/Modified MCP Method 60% 58% 50% 53% 40% 39% 30% 31% 20% 1 2 3

Year

4 Significant improvement is there in 4 th year.

5 25%

MCP Parameters Study Complex Reference Site Complex Target Site Flat Reference Site

There is significant impact in assessment of energy yield if the number of years are increased from

Flat Target Site Flat Reference Site Flat Target Site

Scope of Project

Uncertainty in Wind Energy Yield Predictions

Sources of Uncertainty

An Interesting Case Study Analyzing Power Curve Performance

- 10 year ref site data, 1 year target site data - Concurrent data gives Correlation parameters (m,c) applied to 10 years ref site data to predict next 10 years velocity - Predict next 10 years electrical energy Data Acquisition It is important to realise how vital long term wind prediction is today, with the Traditional MCP Method increasing diversification of energy production. Long Term Prediction It has resulted in power companies investing millions of pounds on potential sites, and made the accurate long-term wind prediction of these sites absolutely vital.

Modelling Our Modified MCP Method This case study using the data from a

real

wind farm will illustrate how the energy output from the same site can vary dramatically from year to year and could deviate Losses in Energy Production from expected values.

. . . . .

target site data parameters (m,c) for each concurrent year years ref site data to predict next 10 years velocity

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Case Study:

The Power Curve Performance

Power curve performance is used to analyse the energy production in a wind farm.

Case Study: A real wind Farm in UK

Technical Details of the Wind Farm:

• • • • • • • Location Total Wind Turbines Cut in speed Rated Speed Cut out Speed Hub height Rating : Northern Part of UK : 15 : 4 m/s : 16 m/s : 25 m/s : 40 m : 850kW

Case Study: Performance of Wind Turbines

kW kW m/s

Expected Performance

m/s

Unexpected Performance

Power Curve Performance

Winter

During the first three months of the year, a turbine was showing an unexpected performance

Power Curve Performance

Summer & Spring

Power

900 800 700 600 500

kW

400 300 200 100 0 0 P.Output

P.expect

200 400 600

Hours

800 1000 1200 1400

During summer & spring time, a turbine was showing expected performance

Power Curve Performance

Parameters under study:

• Wind Direction No wind direction data from wind farm • Wind speed • • Predicted & Actual Power Output comparison Alarm codes

Power Curve Performance

Turbine X has excess power in March, 2006 and power loss in March 2007

2006 2007

Power Curve Performance

In March 2006, Turbine Y has excess power and power loss in March 2007

2006 2007

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

Challenges in this Project:

• The

real

wind farm data has close to 1/3rd missing entries.

• Missing and wrong entries in the weather data from the Met office.

• Understanding statistics better to find good results

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

KEY FINDINGS:

• • • • • Variation in slope and intercept with respect to time.

Modified MCP method predicts well in comparison to traditional method.

There is significant impact in assessment of energy yield if the number of years are increased from one to three or more for collection of wind data.

Distance between met site and proposed wind farm site should be as small as possible in order to get better results.

The energy yield from a wind farm can vary dramatically from year to year.

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS

FUTURE WORK:

• • • One can look into more than one met site to assess the wind energy yield of proposed wind farm.

A hybrid method could be thought of to predict the wind speed of proposed wind farm. For example for lower wind speed, linear regression and for higher wind speed a non-linear method such as quadratic regression or neural network method.

One can look into the effect of varying temperatures on electronic devices in a wind farm.

UNCERTAINTY IN WIND ENERGY YIELD PREDICTIONS