20.500.12556/DKUM-99336 Napovedovanje količine naročil v podjetju s področja industrije umetnih mas z uporabo naprednih napovednih metod Forecasting order quantities in the plastics industry using advanced forecasting methods V magistrskem delu smo raziskovali uspešnost napovedovanja mesečnih naročil v podjetju iz plastično-predelovalne industrije, kjer visok delež materiala v lastni ceni in nihanja povpraševanja otežujejo načrtovanje proizvodnje in nabave. Namen naloge je bil ugotoviti, ali napredni modeli za analizo časovnih vrst prinašajo dodano vrednost v primerjavi z enostavnimi pristopi na specifičnih podatkih izbranega podjetja. Teoretični del temelji na analizi časovnih vrst in njihovih komponent (trend, sezonskost, cikel in slučajna komponenta), na konceptu stacionarnosti ter na statističnih testih, ki jo preverjajo (razširjeni Dickey-Fullerjev in KPSS test). Predstavljeni so postopki priprave podatkov (STL dekompozicija, winsorizacija), merila natančnosti napovedi (MAE, RMSE, MAPE) ter diagnostika ostankov s testoma Ljung-Box in Kolmogorov-Smirnov. Obravnavane napovedne metode so naivna in sezonska naivna metoda, modeli eksponentnega glajenja (ETS), modeli ARIMA ter model NNAR. V empiričnem delu smo v programskem okolju R za tri artikle podatke vnaprej procesirali, analizirali in vizualizirali, nato pa izdelali trimesečne napovedi. Rezultati kažejo, da enotnega najboljšega modela ni: pri prvem artiklu se je najbolje odrezal model NNAR (MAPE 16,8 %), pri drugem ETS (25,8 %) in ARIMA (25,9 %), pri tretjem pa naivna metoda (19,9 %). Diagnostika ostankov je izločila več modelov z navidezno nizko napako, kar potrjuje, da vrednost MAPE sama po sebi ni zadosten kriterij za izbiro modela. Rezultati podjetju omogočajo neposredno uporabo napovedi pri načrtovanju kapacitet, nabavi surovine in določanju varnostnih zalog. This master's thesis examines the accuracy of monthly order forecasting in a plastics processing company, where a high share of material costs and volatile demand complicate production and procurement planning. The purpose was to determine whether advanced time series models offer added value over simple approaches on the specific data of the selected company. The theoretical part is based on time series analysis and its components (trend, seasonality, cycle, and the irregular component), on the concept of stationarity and the statistical tests used to assess it (Augmented Dickey-Fuller and KPSS). It further covers data preparation procedures (STL decomposition, winsorization), forecast accuracy measures (MAE, RMSE, MAPE), and residual diagnostics using the Ljung-Box and Kolmogorov-Smirnov tests. The forecasting methods examined are the naive and seasonal naive methods, exponential smoothing (ETS) models, ARIMA models based on the Box-Jenkins methodology, and a neural network autoregression model (NNAR). In the empirical part, data for three products were preprocessed, analyzed, and visualized in the R environment, followed by three-month forecasts. The results show that no single model is universally best: NNAR performed best for the first product (MAPE 16.8%), ETS (25.8%), and ARIMA (25.9%) for the second, and the naive method (19.9%) for the third. Residual diagnostics eliminated several models with seemingly low error, confirming that MAPE alone is not a sufficient criterion for model selection. The results allow the company to apply the forecasts directly in capacity planning, raw material procurement, and safety stock determination. napovedovanje časovne vrste napredne metode ARIMA ETS NNAR forecasting time series advanced methods ARIMA ETS NNAR true true false Slovenski jezik Angleški jezik Magistrsko delo/naloga 2026-08-13 06:03:11 2026-09-03 11:56:48 2026-09-04 03:08:40 0000-00-00 00:00:00 2026 0 Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 MAG_Kuk_Gal_2026.pdf MAG_Kuk_Gal_2026.pdf 1 93125C18472D9E98695EABB3023753AA a5380a791b4a88a4b6108405cf612cb9bca60e166a24d4851a3d94daaf7781d8 d452b675-a710-11f1-aef5-00155d0001be https://dk.um.si/Dokument.php?lang=slv&id=211372 Fakulteta za strojništvo 0 0 0