QUANTITATIVE PERFORMANCE EVALUATION OF A WIND TURBINE

GENERATOR CLUSTER USING STATISTICAL TECHNIQUES

S. V. Joshi1*; S. Sainis1; M. D’Souza2

1 School of Mechanical and Building Sciences, VIT University, Vellore, India

2 Suzlon Energy Limited, One Earth, Pune, India* Tel: 0763-956-7408, E-mail: saumitra.vjoshi@gmail.com

INTRODUCTION• Power generation from wind has emerged as a major sector in the

energy market.

• Besides technology development, wind power companies focus equally on development of a strong customer base.

• The performance of a Wind Turbine Generator (WTG) is directly linked with monetary profit or loss to the customer.

• The important factors that play a vital role in the performance of a wind site are:

• Wind at site

• Machine availability (MA): This is the factor Wind Energy companies have a control over

• Grid availability (GA)

INTRODUCTION

• Estimation of power generation losses is crucial for evaluating the performance of a WTG Cluster.

• A suitable technique that can effectively calculate the total power generation losses due to lack of MA and GA is imperative to the company.

• Power generated from the WTG cluster was predicted using WAsP.

• Wind Atlas, Analysis and Application Program (WAsP) is a powerful PC program that can predict power generation from Wind Turbines and Wind Farms. The predictions are based on wind data measured at meteorological stations in the same region. The program includes a complex terrain flow model, a roughness change model, a model for sheltering obstacles, and a wake flow model.

RESEARCH PROBLEM

Fig. 2 WAsP-predicted and actual generation at centum MA and GA

• Problem statement: Identify reasons for shortfall in actual and predicted power generation, and quantify these losses using actual data

• Approach: Define different techniques of calculating losses caused due to lack of MA and GA in order to determine the best suited method based on accuracy of results as well as practical utility

Where is the

shortfall?

How much is

the shortfall?

How do we

quantify it?

SITE DESCRIPTION

• For our research, the WTG cluster scrutinized has 10 WTGs, located in Kuchchh region in India and is maintained by Suzlon Energy Limited, India.

• Relatively flat terrain (5 meter rise for every kilometer distance)

• Proximity of mast to area of study

• Generator Model: S82 (Rated Capacity: 1.5 MW)

Fig. 1 Location of cluster WTGs and Mast. X and Y scale is in UTM

DATA COLLECTION AND VALIDATION• Data obtained from Suzlon Energy Group’s Supervisory Control and

Data Acquisition (SCADA) Monitoring Centre.

• Missing data was completed using two approaches:

• Measure-Correlate-Predict (MCP) Method: for data points with lack of MA + GA

• Cluster Nacelle Velocity Average (CNVA) Method: for lack of MA only

• Validation:

Fig. 3 Scatter plots of Wind Vane 1(65m Height) reading vs. Wind Vane 2(50m Height) reading

Fig. 4 Scatter Plot of (V65/V50) vs. Wind Direction

Fig. 5 Wind speed at 65m (Blue) and at 50m(Red) vs. Time

PROPOSED STATISTICAL METHODS• The statistical correlation techniques used in this study for prediction is

based on a simple approach of determining the weightage of points (xi, yi)

on the argand plane to find the best fitting polynomial through the set.

• The basic steps followed in this technique are as follows:

• Identification of an independent variable x. In our case, we have used wind speed which was obtained from completed data sets obtained by MCP and CNVA methods as described in the previous section

• Selection of a suitable dependent variable y such that y = f(x) + c

• Obtaining the spatial distribution of y with x to carry out a weight analysis on the argand plane

• Expressing y as a polynomial function of x.

• Calculation of yj at conditionally selected points xj

PROPOSED STATISTICAL METHODS

Calculation

Determination of

missing x

Identification of x

Velocity

MCP Method

Curtailment Loss

Generation Loss (100 MA, GA)

Turbine Cluster Average

Generation Loss (100

GA)

Fig. 6 Schematic of procedure

CURTAILMENT LOSSES• Data analysis of three quarantined WTG for pitch angle

variations with wind speed show that curtailment exists for T002 only

Fig. 7 Scatter plot of pitch angles versus wind speed for T002. Notice the anomalous trend at higher speeds.

Fig. 8 Actual Power Generation versus wind speed for T002

Fig. 9 Actual Power Generation for T002 versus Wind Speed showing fitted polynomial after filtering the curtailment points

CURTAILMENT LOSSESStatistically correlated, ninth-degree, centered-and-scaled data polynomial is calculated as shown below that relates actual T002 power generation to wind speed

Using above equation, curtailment losses are evaluated as follows:

LACK OF MACHINE AND GRID AVAILABILITY• Main reason for shortfall: Lack of Machine Availability in high wind

season

• Numbering in figure is based on fiscal months (1represents April, and so on)

Fig. 10 Lack of MA for T002, T003 and T005 is in high wind season

CALCULATION OF MA + GA LOSSESI. Extraction Factor Method (EFM)Determination of a relation between the ratio () of maximum theoretical extractable power to actual power generated and MCP wind speed.

Fig. 11 Plot of versus Wind Speed for T002

OTHER METHODS• Power generation loss due to lack of MA+GA

II. Monthly Correlation Method (MCM)

- Correlation of actual monthly power generation to MCP wind speed

• Power generation loss due to lack of MA

I. Turbine Specific Power Curve Method (TSPCM)

- Correlation of actual power generation to CNVA wind speed

II. Rated Power Curve Method (RPCM)

- Correlation of rated power to CNVA wind speed

RESULTS

Initial comparison

Coarse Extrapolation

PFM

MCM

TSPCM

RPCM

Lack of MA + GA Lack of MA

CONCLUSIONS• TSPCM gives best and most consistent results

• TSPCM manages to bring down the error in calculation of ac-tual generation losses well below 10%.

• Calculates generation losses due to lack of MA, which is of more utility to WTG companies.

• EFM calculates results very well for a temporally distributed lack of MA

• MCM is well suited for a temporally concentrated lack of MA

• RPCM provides a good way to compare two machines’ per-formance – not suggested for quantification

SCOPE OF THIS WORK

• Refinement of the TSPA technique by incorporation of more complex concepts of statistics to make calculations more accurate.

• Making all methods computationally less tedious by determining a way of automation of polynomial feeding into prediction codes.

QUESTIONS?