Page - 123 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 3283 3.2.DataPreprocessing 3.2.1. TemperatureAdjustment Generally, the power consumption increases in summer andwinter due to the heavyuse of air conditioning and electric heating appliances, respectively. Since the correlations between the temperatureandelectrical loadin termsofmaximumandminimumtemperaturesarenot thathigh, weneed to adjust the daily temperature formore effective training based on the annual average temperatureof12.5providedbyKMA[34],usingEquation(1)as follows: AdjustedTemp= ∣∣∣∣12.5−MinimumTemp+MaximummTemp2 ∣∣∣∣. (1) Toshowthat theadjustedtemperaturehasahighercorrelation thantheminimumandmaximum temperatures, we calculated the Pearson correlation coefficients between the electrical load and minimum,maximum, average, andadjusted temperatures, as shown inTable 1. In the table, the adjusted temperature shows higher coefficients for all building clusters compared to other types of temperatures. Table1.ComparisonofPearsoncorrelationcoefficients. TemperatureType Cluster# ClusterA ClusterB ClusterC Minimumtemperature −0.018 0.101 0.020 Maximumtemperature −0.068 0.041 −0.06 Average temperature −0.043 0.072 −0.018 Adjustedtemperature 0.551 0.425 0.504 3.2.2. EstimatingtheWeek-AheadConsumption The electrical loaddata from thepast formoneof theperfect clues for forecasting thepower consumptionof the futureandthepowerconsumptionpatternreliesonthedayof theweek,workday, andholiday.Hence, it isnecessary toconsidermanycases toshowtheelectrical loadof thepast in the short-term load forecasting. For instance, if theprediction time is aholidayand the sameday in the previousweekwas aworkday, then their electrical loads can be very different. Therefore, itwouldbebetter tocalculate theweek-ahead loadat theprediction timenotby theelectrical load dataof the comingweek, butbyaveraging the electrical loadsof thedaysof the same type in the previousweek. Thus, if theprediction time is aworkday,weuse theaverage electrical loadof all workdaysofthepreviousweekasanindependentvariable. Likewise, if thepredictiontimeisaholiday, weuse theaverage electrical loadof all holidaysof thepreviousweek. In thisway,we reflect the differentelectrical loadcharacteristicsof theholidayandworkdayin the forecasting. Figure2shows anexampleof estimating theweek-aheadconsumption. If thecurrent time isTuesday,wealready knowtheelectrical loadofyesterday(Monday).Hence, toestimate theweek-aheadconsumptionof thecomingMonday,weuse theaverageof theelectrical loadsofworkdaysof the lastweek. Figure2.Exampleofestimatingweek-aheadconsumption. 123