|
Title:
|
|
Identification of variables for site calibration and power curve assessment in complex terrain JOR-CT98-0257 : task 8, a literature survey on theory and practice of parameter identification, specification and estimation (ISE) techniques
|
|
|
|
Author(s):
|
|
|
|
|
|
Published by:
|
Publication date:
|
|
ECN
Wind Energy
|
1-4-2001
|
|
|
|
ECN report number:
|
Document type:
|
|
ECN-C--01-015
|
ECN publication
|
|
|
|
Number of pages:
|
Full text:
|
|
23
|
Download PDF
|
Abstract:
This document presents the literature survey results on Identification, specification and Estimation ISE) techniques for variables within the SiteParIden project. Besides an overview of the different general techniques also an overview is given on EU funded wind energy projects where some of these techniques have been applied more specifically. The main problem in applications like power performance assessment and site calibration is to establish an appropriate model for predicting the considered dependent variable with the aid of measured independent (explanatory) variables. In these applications detailed knowledge on what the relevant variables are and how their precise appearance in the model would be is typically missing. Therefore, the identification (of variables) and the specification (of the model relation) are important steps in the model building phase. For the determination of the parameters in the model a reliable variable estimation technique is required.In EU funded wind energy projects the linear regression technique is the most commonly applied tool for the estimation step. The linear regression technique may fail in finding reliable parameter estimates when the model variables are strongly correlated, either due to the experimental set-up or because of their particular appearance in the model. This situation of multicollinearity sometimes results in unrealistic parameter values, e.g. with the wrong algebraic sign. It is concluded that different approaches, like multi-binning can provide a better way of identifying the relevant variables. However further research in these applications is needed and it is recommended that alternative methods (neural networks, singular value decomposition etc.) should also be tested on their usefulness in a succeeding project.
Increased interest in complex terrains, as feasible locations for wind farms, has also emphasised the
need for adequate models. A common standard procedure to prescribe the statistical methodology for setting up the required experiments and the analysis of the data would be beneficial for all the partners in the various wind projects. Joint efforts of applied statisticians and the experts in the field should be undertaken to achieve this aim, preferably within the framework of a new common project. For the current SiteParIden project it is concluded that, as long as the proposed alternatives have not been investigated further, the multivariate regression technique seems an appropriate all-round
technique, under the condition that it is applied with care.
Back to List