POWER FLOW PREDICTION WITH SOFTWARE SPSS

Ko znamo napovedati porabo energije, se lahko tudi optimalno odločimo, koliko jo kupiti ali kako dimenzionirati prenosne naprave. Z uporabo statističnih metod in metod napovedi na osnovi analize časovnih vrst, smo želeli napovedati porabo energije za industrijski kompleks. Za napovedovanje obstaja mnogo programskih orodij, kjer so implementirane danes standardne metode napovedi. Zaradi enostavne uporabe, smo se odločili za program IBM SPSS. Za testni primer napovedi porabe smo uporabili merilne podatke porabe električne energije industrijskega kompleksa Železarne Ravne za leto 2007. Preučili smo problematiko vpliva velikega porabnika, elektroobločna peč, (velik napovednik) na skupni model napovedi porabe. Določili smo dva modela napovedi, ko veliki napovednik obratuje in ko ne obratuje. V diplomi smo se posebej posvetili vprašanju, kako dolgo časovno obdobje meritev uporabiti za učenje modela napovedi. Delno smo se posvetili tudi problematiki ubežnikov med merilnimi podatki. Izvedli smo kratkoročno in srednjeročno napoved. Optimalna modela napovedi sta ARIMA in aditivni sezonski model.Whit the prediction of the energy consumption we can optimally decide how much electric energy to buy or how to adequate construct electric transmission devices. We wanted to predict the electric energy consumption for the industrial facility by using statistical and prediction methods based on the analysis of time series. Today are many software tools for predictions that are implemented as standard methods. Due to ease of use we have chosen IBM software SPSS. As a test example we used the measurement data of electric energy consumption in facility Železarne Ravne for the year 2007. We have studied the issue of the impact of large consumer, electric oven (big predictor), on the common prediction model of consumption. We determined two prediction models, when the big predictor is operational and when he is not. In the diploma thesis, we paid particular attention to how long a time period of measurement data should we use for learning the prediction model. Partly, we dedicated ourselves to the problem of outliers between the measurement data. We carried out short and medium term predictions. Optimal prediction models are ARIMA and additive season model