From univariate to multivariate – extending the prediction problem

Let’s start by extending a univariate time series prediction problem to a multivariate time series prediction problem. To do that, we can go back to Chapter 4 , Time Series Visualization, where we discussed the visualization of the energy consumption time series dataset.

The raw energy consumption data originates from 6,000 households and businesses in Ireland, and it was collected by smart meters that recorded the energy consumption every half-hour in kilowatts (kW) between July 2009 and August 2010. The original data contains three columns: the timestamp, the ID of the smart meter, and the energy consumption. A few aggregated values have been calculated for every single ID to quantify the amount of energy used at different times of the day and week. Based on such aggregated values, the households have been clustered using a k-means algorithm, where k=30. Finally, the average time series for each...