Ing. Josef Grus

Ing. Josef Grus

Mgr. Petr Novák, Ph.D.

Department of Applied Mathematics

This thesis addresses short-term electricity price forecasting in the Czech day- ahead and intraday energy markets operated by OTE a.s. A modular fore- casting pipeline is developed, combining classical statistical methods, machine learning models, and deep learning architectures, utilizing the AutoGluon li- brary. Market data is enriched with exogenous features such as weather fore- casts and power grid generation predictions to improve forecasting accuracy. The resulting forecasts can serve as a valuable input for energy-aware plan- ning and cost-sensitive decision-making in industrial contexts, particularly in manufacturing environments sensitive to electricity price volatility.

Ing. Josef Grus

doc. Ing. Štěpán Starosta, Ph.D.

Department of Applied Mathematics

This thesis focuses on evaluating and comparing machine learning methods for estimating processing times in real, factory-captured data, spanning a wide spectrum of measured tasks. Because of a lack of features traditionally useful for regression analysis, it explores various data preprocessing approaches as well as regression methods, from simple linear regression to gradient boosting decision trees. The proposed approach utilizes grouping of production tasks by predetermined factors and selects an estimation method for each resulting group, accounting for a wide range of different tasks. Tasks which cannot be classified are estimated using the model of the most semantically similar group. Multiple models that outperform human expert estimations are introduced, with further improvements in the grouping approach when estimated tasks have text features similar to those in training data.