Improved CNN architecture for multistep univariate/multivariate timeseries forecasting.
Figure above is the 60hr temperature forecast using the model. The model was trained with 2 years of hourly temperature data. Compared with LSTMs Temporal CNNs proved superior for this particular timeseries data in forecasting.
This project has also been used to give a special lecture to computer engineering students about the Keras framework.