Recurrent neural networks for time series forecasting: Current status and future directions