Stacked LSTM Recurrent Neural Network: A Deep Learning Approach for Short Term Wind Speed Forecasting

Renewable energy has now become a key to the future. Renewable energy can be harnessed from many sources, one such source is wind. Wind energy is constantly prone to changes hence making it an intermittent or abrupt source of energy. Thus wind forecasting finds multiple applications in various fields such as selecting the sites to construct the wind farm, forecasting the future wind speed, future wind power prediction for deciding the electricity tariffs, for penalty-free bidding process, and for enhancing the power system reliability thus making it an extensive area of research. Forecasting the wind speed will support these applications in having superior outcomes. However, the prediction of wind energy at any given time is still a major challenge. There are many techniques for predicting future wind speed, but considering the accuracy, training pattern, and testing ability, applying Machine Learning is considered as the finest solution. There are various approaches in Machine Learning for forecasting the wind speed, among which Long Short-Term Memory (LSTM) based forecasting is the contemporary method for time series forecasting. In this paper, LSTM is further layered to obtain better accuracy. This paper explores a novel Stacked LSTM based architectures, which can accomplish a better wind forecasting model that can be administered for Maximum Power Point Tracking (MPPT) for finding the optimal wind power output. Comparing with various existing algorithms, a three-layered stacked LSTM is found to have better performance indices.