Supervisors: Dr. Wolf-Gerrit Früh, Heriot-Watt University and Prof. David Infield, University of Strathclyde
PhD Student: Christina Skittides, Heriot-Watt University
Status: Closed
This project aims to develop a methodology to provide potential developers with a reliable estimate of the energy yield from a wind turbine. This methodology consists of a hierarchy of input and analysis steps and will provide a realistic energy yield estimate together with an analysis of the accuracy of that estimate.
The data analysis will appear similar to current approaches characterised as 'Measure-Correlate-Predict' (MCP) methods, in that it will generate a prediction of wind speed and direction distribution at the field site by combining some local measurements with a long-term and concurrent data from one or more remote sites. MCP methods generate a transfer function based on instantaneous correlations or regressions between the concurrent local and remote data. This leads to a recommended period for local measurements of around 12 months. As it is not a prediction of the instantaneous wind but of the wind statistics which is relevant to resource assessment, we propose to follow a fundamentally different approach by predicting the underlying structure of the wind distribution directly rather than follow the usual MCP approach of predicting the instantaneous wind and then calculating the statistics from the predicted time series. The proposed method might be more appropriately termed a 'Measure and Model-Correlate-Assess' (MMCA) method.
In terms of dynamical systems, the attractor of the system rather than the time series on that attractor describes best the energy resource. For that reason, we propose to develop an attractor reconstruction method based on multi-variate Singular Systems Analysis which is based on the singular value decomposition of the input data into singular vectors, singular values, and principal components. This technique can reconstruct a multi-dimensional and multi-variate attractor, not only for the whole system of local and reference data but also for one system in reference to another, in a way similar to a previous analysis of a global temperature time series against a surrogate reference data set. Since the techniques can be applied to time series as well as collections of other data, it can be developed to describe the wind statistics through the singular vectors weighted by their singular values based on the input from wind time series as well as the results of the computational modelling for a range of representative wind conditions.
Once a reliable attractor has been determined from the training set of contemporary measurements at the reference and target site has been constructed, the available data from the reference site can be used to reconstruct a full phase space of the full reference-target site attractor, from which the wind resource at the target site can be predicted, not only as a mean statistic but also in a time-resolved way to help with the assessment of, for example, monthly variability of the wind resource. By using ensemble prediction it is not only possible to predict the mean (annual or monthly) resource but also to calculate the intrinsic accuracy of that prediction. Hence, the wind developer will not only be able to provide an assessment of the resource but will also be able to quantify their confidence in that prediction.
The use of multi-variate statistical modelling allows to test a number of different measures to be used for the training and prediction steps. They could include a number of reference sites as well as synoptic measures such as the North-Atlantic-Oscillation index, climate measurements, or weather/climate model outputs.