A Hybrid Measure-Correlate-Predict Method for Wind Resource Assessment

Abstract

This paper develops a hybrid Measure-Correlate-Predict (MCP) strategy to predict the long term wind resource variations at a farm site. The hybrid MCP method uses the recorded data of multiple reference stations to estimate the long term wind condition at the target farm site. The weight of each reference station in the hybrid strategy is determined based on: (i) the distance and (ii) the elevation difference between the target farm site and each reference station. The applicability of the proposed hybrid strategy is investigated using four different MCP methods: (i) linear regression; (ii) variance ratio; (iii) Weibull scale; and (iv) Artificial Neural Networks (ANNs). To implement this method, we use the hourly averaged wind data recorded at six stations in North Dakota between the year 2008 and 2010. The station Pillsbury is selected as the target farm site. The recorded data at the other five stations (Dazey, Galesbury, Hillsboro, Mayville and Prosper) is used as reference station data. Three sets of performance metrics are used to evaluate the hybrid MCP method. The first set of metrics analyze the statistical performance, including the mean wind speed, the wind speed variance, the root mean squared error, and the maximum absolute error. The second set of metrics evaluate the distribution of long term wind speed; to this end, the Weibull distribution and the Multivariate and Multimodal Wind Distribution (MMWD) models are adopted in this paper. The third set of metrics analyze the energy production capacity and the efficiency of the wind farm. The results illustrate that the many-to-one correlation in such a hybrid approach can provide more reliable prediction of the long term onsite wind variations, compared to one-to-one correlations.