Page - 69 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

An extension of the equations 2 is proposed in this study by analyzing the inclusion of the population pressure (for residents and tourist) as an additional exogenous determinant. This variable can be easily obtained in isolated territories where a higher level of forecasting accuracy is specially appreciated. Thus, analytically, equation 1 can be extended, to: lnCt= pt+st+CSDt+CWEAt+Pt+ut (3) where Pt denotes the population pressure. In a similar way equation 2 can be rewritten as tttt εuqθXb Cp )(')(ln)( (4) where X' includes the variables p, s, CSD, CWEA and Pt. 3. Data and forecasting evaluation strategy 3.1. Data The dataset used in this study concerns daily time series for electricity consumption for Majorca, Minorca and Pitiüses (Ibiza and Formentera). The Balearics are considered jointly from January 1995 until September 2007. This is 4655 daily observations for each one of the four utility systems considered. The dataset was compiled by the Spanish System Operator Red Electrica de España and no missing observations were present. Charts of the time series. in Figure 1. show a clear trend along the whole sample, and depict different seasonal cycles. In this case, it should be highlighted that for many coastal areas in the Mediterranean Sea, high season is characterized with high temperatures and an important presence of tourists. Therefore, it is important to be aware that the seasonal movement of electricity load is not originated exclusively by weather pattern but also from differences in the population that is on the islands during the year. Meteorological variables were provided by the Instituto Nacional de Meteorologia, the Spanish official meteorological bureau, and are referred to the airport stations. From the experience of previous literature, the weather-related factors that can influence the electricity demand are temperature, humidity, wind and precipitation in decreasing order of importance [9]. However, the non-linear influence of temperature on the electricity demand suggests the use of two temperature derived functions: heating degree-days (HDD) and cooling degree-days (CDD)[30]. When dealing with the non- linearity of the temperature effect, the most frequent approach is to segment temperature into HDD and CDD, defined as shown below: HDDt = Max (Tref - Tt, 0) (5) CDDt = Max (Tt-Tref, 0) (6) M.BakhatandJ.RosselloNadal / ImprovingDailyElectricityLoadsForecasting 69