Forecasting Upper Air Wind Speed using a Hybrid SVR-LSTM Model

Upper air wind data is collected by a process known as Radiosounding. Since the experiments are conducted as per requirement, the interval between data points is not regular, resulting in an irregular time series which are not much explored in the existing literature for forecasting. Traditional approaches like autoregression (AR), moving average (MA) and their derivatives donot work on irregular time series. In the first phase of this study, we use simple linear regression (SLR), polynomial regressions (PR) and support vector regressions (SVR) to forecast upper air wind speed. In the second phase, a hybrid SVR-LSTM model is proposed for forecasting. Upper wind data at Chandipur, Odisha from 2007 to 2010 has been used for training and testing the models. The results are compared in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Experimental results show that the proposed hybrid model performes better than traditional machine learning regression techniques.