Forecasting methods are critical to tackle the wind uncertainty in operating grids with a high wind power penetration. This talk revisits the classical short-term wind power forecasting problem and presents novel methods via adapting recent deep learning to improve the forecasting performance. We will discuss the short-term deterministic wind power forecasting and probabilistic wind speed forecasting.  In wind power forecasting, we propose a K-shape and K-means guided Convolutional Neural Network integrating Gated Recurrent Units method for developing the data-driven model to obtain the deterministic forecast of future wind power. For the probabilistic forecasting of wind speeds, we propose a Mixture Density Attention Network via Wasserstein Distance Based Adversarial Learning. Proposed methods introduce new paradigms of processing wind farm data collected by supervisory control and data acquisition (SCADA) systems as well as engineering data features to facilitate forecasting tasks. We verify the superiority of our proposed method in forecasting tasks by benchmarking against a set of deep and shallow wind power forecasting methods based on SCADA data collected from multiple wind farms. 

Speaker: Dr Zijun ZHANG
Date: 24 February 2021 (Wed)
Time: 13:45pm – 14:45pm
Poster: Click here