Wind Power Ramp Event Prediction with Support Vector Machines

Wind energy is playing an important part for ecologically friendly power supply. Important aspects for the integration of wind power into the grid are sudden and large changes known as wind power ramp events. In this work, we treat the wind power ramp event detection problem as classification problem, which we solve with support vector machines. Wind power features from neighbored turbines are employed in a spatio-temporal classification approach. Recursive feature selection illustrates how the number of neighbored turbines affects this approach. The problem of imbalanced training and test sets w.r.t. the number of no-ramp events is analyzed experimentally and the implications on practical ramp detection scenarios are discussed.