Multi-step and multi-output forecasting with multivariate time series

In this recipe, we’ll extend the LSTM model to predict multiple steps of several variables of a multivariate time series.

Getting ready

So far, in this chapter, we have built several models to forecast the future of one particular variable, solar radiation. We used the extra variables in the time series to improve the modeling of solar radiation.

Yet, when working with multivariate time series, we’re often interested in forecasting several variables, not just one. A common example occurs when dealing with spatiotemporal data. A spatiotemporal dataset is a particular case of a multivariate time series where a real-world process is observed in different locations. In this type of dataset, the goal is to forecast the future values of all these locations. Again, we can leverage the fact that neural networks are multi-output algorithms to handle multiple target variables in a single model.

In...