On wavelet transform based convolutional neural network and twin support vector regression for wind power ramp event prediction.

Power produced from renewable energy sources carbon negative and promises an increased reliability for grid integration. Wind energy sector globally has an installed capacity of over 650 GW and promises to grow substantially. In this work, wind power ramp events that arise from sudden change in wind power are studied. Forecasting wind power ramp events is an important problem statement in the current wind power industry. Wind power integration to utility grid is impacted by the forecast accuracy. To improve the reliability and security, wind speed and power forecasts are extensively studied. Ramp events are predicted in this manuscript using hybrid techniques employing wavelet decomposition transform in tandem with convolutional neural network and twin support vector machines. Wind speed datasets from Spain, Germany and Argentina are considered and error metrics are computed. It is observed that CNN based method is 28.76% and 26.43% superior from wavelet based random forest and TSVR method. • Comparative analysis of deep learning and machine learning models. • Ramp events analyzed for wind speed datasets. • Large datasets are analyzed from CPU time perspective. • Randomness in wind speed datasets is explored. [ABSTRACT FROM AUTHOR]