Ispány, MártonGandhi, Herry Kartika2025-01-072025-01-072025-01-07https://hdl.handle.net/2437/384509This dissertation proposes novel variations of combining several methods into hybrid forecasting models to increase the forecasting accuracy and decrease the error of predictions within the scope of problems in heavy manufacturing industries. In the first study, the dataset is a non-negative integer-valued, which comes from the number of daily paper roll defects from three different paper machines. The results of INAR(1) will be adjusted to the characteristics of PINAR(1) model or NBINAR(1) model according to their corresponding characteristics. Through model fitting analysis, the PINAR(1) and NBINAR(1) models are shown to fit the actual probability dataset. The second study is demand forecasting from daily electrical consumption. I use a hybrid linear and nonlinear model, with SARIMA model as the first (linear method) and SVR as the second (nonlinear) model. According to the line plot, it can be seen that the prediction value is very close to the actual data. The third study is energy price forecasting. The result of decomposition shows that the imfs are formed from low to high frequency. For the forecasting model, I use three popular techniques from deep learning. In three cases, CEEMD decomposition gives good results compared to all decompositions, while in heating oil price, wavelet decomposition is the best among others. The prediction and actual in one line plot show that the prediction is close to the actual value.155enForecastingIndustryTime_SeriesHybrid_ForecastingPhD, doktori értekezésInformatikai tudományokMűszaki tudományok