Page - 71 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

[22], and analyzed by various authors [14,1]. The HPDI is an estimation of the daily number of people on each of the Balearic islands, and it is based on resident population registers and daily arrivals and departures from the airports and ports. Results from the HPDI estimation for our period of analysis can be seen in Figure 2, where a positive trend and a clear hard seasonal variation can clearly be observed. Figure 2 Population stock (HPDI) for the Balearic Islands The second reference measure for the population pressure was the daily airport arrivals. Although from a theoretical point of view this second measure exhibits different defects, since different tourists’ length of stay is not captured by this measure. However, in the Balearics islands more than 90% of arrivals are via airports (less than 10% are via ports) and tourists’ length of stay remains relatively similar during the year, hence the variability between HPDI and the airport arrival is significantly reduced. In contrast, taking into consideration the forecasting objective of this work, it should be noted that in practice, forecasts should be implemented to get either daily tourist arrivals or HPDI variable. Hence, prediction for airport daily arrivals are easy to undertake through the slots that airport authorities have engaged, while HPDIs' prediction is comparably complicated due to the presence of other historical data incorporated in HPDI measure. 3.2. Forecasting evaluation strategy 84 months of daily data from 1 January 1999 to 31 December 2005 are used to estimate model parameters, and 12 months of daily data from 1 January 2006 to 31 December 2006 to evaluate the different forecasting methods. This period of 12 months gave 365 days for evaluation period of 1 to 10 days ahead. Analysis and forecasts of Balearic’s electricity consumption has been undertaken for every single island separately and jointly, thus considering Majorca, Minorca and Pitiüses. Models were estimated using the multivariate regression and ARMAX models described above. Their forecasting performances were compared to the set of benchmark models also mentioned in the previous section. Table 1 summarizes the benchmark and the rest of models used in the forecasting exercise. M.BakhatandJ.RosselloNadal / ImprovingDailyElectricityLoadsForecasting 71