Small-scale wind energy: methods for wind resource assessment

Small-scale wind energy is a renewable energy technology with exciting prospects in a low carbon energy future. However, in order for the technology to be fully utilized, techniques capable of predicting the wind energy resource quickly, cheaply and accurately are urgently required. Specifically, the direct measurement approaches used in the large-scale wind industry are often not financially or practically viable in the case of small-scale installations.
The subject of this thesis is the development of low-cost, indirect methods for predicting the wind resource using, (i) analytical models based on boundary layer meteorology and (ii) data-driven techniques based on measure-correlate-predict (MCP). The approaches were developed and tested using long-term (11 years) wind data from meteorological stations, short-term (1 year) data from a field trial of small-scale turbines, and output from an operational forecast model.
As a first step, the performance of an existing boundary layer scaling model was evaluated at 38 UK sites and found to result in large site-specific errors. Based on these findings, a revised model was developed and shown to improve prediction accuracy. However, uncertainty analysis and comparison with onsite measurements revealed average errors in the predicted wind power density of over 60% due to uncertainties in the model input parameters. Hence, it was concluded that such an approach is best applied in a scoping context to identify sites worthy of further study.
To investigate the ability of low-cost, data-driven techniques to reduce these uncertainties, MCP approaches were trialled using onsite measurement periods as short as 3 months at a subset of 22 of the above UK sites. In addition to established linear approaches, non-linear Gaussian process regression and bivariate conditional probability approaches were developed. Using a 3 month measurement period, the best performing MCP approaches resulted in average errors in the predicted wind power density of 14%, compared to 26% when using the boundary layer scaling approach at the same sites. The effect of seasonal variability in the prediction errors was investigated in detail and found to be most significant at coastal sites. This variability was found to be reduced by using output from an operational forecast model in place of long-term reference wind data.
This work provides a means for low-cost and rapid wind resource assessment in cases where traditional approaches are not viable.