Short and Long-term Temperature and Wind Time Series Prediction using LSTMs

Long Short-Term Memory (LSTM) network based prediction of time-series is getting popular as implementations and computational resources are becoming easily available. We are planning to test the applicability of LSTM approach to predict temperature time series evolution based on single point measurements, as well as coupling with wind speed/direction data to improve prediction. Toward that end, we perform sensitivity experiments with various LSTM configurations, dataset sizes, context windows, and forecast horizons. The experiments will be performed on platforms ranging from a single machine to large clusters to assess the computational performance and extension of datasets in additional dimensions.

 Speaker: Gökhan Sever, Argonne National Laboratory
 Speaker: Julie Bessac, Argonne National Laboratory
 Speaker: Rohit Tripathy, Purdue University