Page - 73 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Table 2 Mean absolute percentage errors (MAPE) in daily forecasting for the entire year Naive Holt Winter Static 1 Static 2 Model 1 Model 2 Model 3 Model 5 Majorca 6.024 4.784 6.297 6.209 Minorca 6.470 5.522 7.303 5.392 Pitiüses 6.619 5.096 9.118 7.334 Using HPDI Using Tourists’ Arrival Dynamic 1 Dynamic 2 Static 2 Dynamic 2 Model 4 Model 6 Model 7 Model 8 Majorca 2.721 2.675 6.615 2.726 Minorca 2.942 2.958 6.719 2.899 Pitiüses 2.319 2.336 7.854 2.309 Results show how using the HPDI variable and considering the static models (Model 2 and 3) the forecasting errors have decreased for all the islands (Majorca; Minorca and Pitiüses), whereas in the case of the dynamic models (model 4 and 6) this feature is only observed for Majorca island and not for Minorca or Pitiüses. Secondly, the use of airport’s arrival as a substitute to HPDI in our case of study does not improve the MAPE in Majorca for neither of the static nor dynamic models. However, in the case of Minorca and I Pitiüses, the MAPEs decrease for static and dynamic models. Particularly, in the dynamic models that include airport’s arrival have a better forecasting performance than their correspondent that include HPDI, though the values of the MAPEs are very close (2.899 and 2.30 versus 2.95 and 2.33 for Minorca and Pitiüses respectively). Moreover, the real time performance of the model 6 and model 8 seems to be satisfactory, in the sense that the errors are within the bounds that guarantee the electricity supply security and quality, reflected by MAPEs below 5%, a constant limit suggested as a benchmark in the literature [21]. Additional results on forecasting performance by season and by using airport arrivals as substitute to HPDI were also considered.3 Results indicate that the dynamic model incorporating HPDI performs better in all seasons than its correspondent that includes airport’s arrival. This is true for the case of Majorca and Pitiüses, while in the case of Minorca, which has a different pattern, the forecasts deteriorate during high season when HPDI variable is used, however the use of airport’s variable in this particular case improves well the forecasting performances in all seasons. 5. Summary and conclusions Many islands around the world are characterized by small and isolated electric systems and a high level of tourism specialization from an economic point of view. Thus, on the one hand, having an accurate electricity load forecast is of crucial importance to electricity planning in short-term and, on the other hand, variability in population stocks 3 These results are available upon request M.BakhatandJ.RosselloNadal / ImprovingDailyElectricityLoadsForecasting 73