Multivariate time series predictions for an IIoT project

Require future predictions based on historical time series data. Require the training of the model and provision of a trained model that can be effectively relearnt with transfer learning. Require the predictions for 1 hour, 3 hrs, 6 hours out. Looking at potentially a LSTM through Keras. Require the confidence intervals / min/max bands for prediction to also be made available.

Require the model to be trained via GPU.

Dataset approx 1M rows and 50 columns wide