Cluster based approach for mining patterns to predict wind speed

One of the challenges with the wind power renewable energy source lies in integrating this intermittent power source into the electricity grid. Prediction of wind speeds is essential for forecasting the periods when intermittent wind power is generated. Several meteorological parameters like wind direction, air temperature, atmospheric pressure, relative humidity can be used to predict wind speeds. We propose a cluster based approach for extracting patterns in meteorological data that aid in predicting wind speed. Clustering is done by using compactness measure and distinctness measure as discussed in the paper. This approach automatically determines number of clusters present in meteorological data. Patterns are mined from the extracted clusters with the aid of ID3 algorithm. The mined patterns are used to predict the wind speed patterns in a usable and comprehensive manner. Extensive experimentation using 3100 data points indicate that the proposed approach produces good results.