Multi-step forecasting with multivariate time series

So far, we’ve been working on forecasting the next value of a single variable of a time series. Forecasting the value of the next observation is referred to as one-step-ahead forecasting. In this recipe, we’ll extend the models we developed in the previous chapter for multi-step-ahead forecasting.

Getting ready

Multi-step ahead forecasting is the process of forecasting several observations in advance. This task is important for reducing the long-term uncertainty of time series.

It turns out that much of the work we did before is also applicable to multi-step forecasting settings. The TimeSeriesDataSet class makes it extremely simple to extend the one-step-ahead problem to the multi-step case.

In this recipe, we’ll set the forecasting horizon to 7 and the number of lags to 14:

N_LAGS = 7
HORIZON = 14

In practice, this means the predictive task is to forecast the next 7 days of solar radiation...