Here is my submission to a data science competition to model and forecast a multivariate time series from data harvested on an off shore oil rig. Do to a tight competition time schedule I used a basic stacked LSTM to model the data but believe a state space hidden Markov model would perform best.