A High-Frequency Data-Driven Machine Learning Approach for Demand Forecasting in Smart Cities

Figure 6

A wrongly forecasted signal. In this case, the forecasted day was on Sunday, 19 April 2015, at the industrial area site. For this site, Saturdays are very similar and Sundays differ depending on other factors (e.g., the time of the year). This led to an improper forecasting since the (k) Saturdays closest to 18 April 2015 do not need to be followed by Sundays similar to the forecasted day.