|
Usability of contractual information for prediction of household energy consumption volume
Škorvaga, Cyril ; Janda, Karel (advisor) ; Khymych, Olha (referee)
5 Abstract This thesis investigates usability of contractual information, enriched with publicly available sociodemographic and environmental statistics, for predicting household energy consumption volume. The aim is to assess the usability of this type of information to enhance prediction accuracy as well as to uncover relationships between energy consumption and various independent variables derivable from the contractual information, such as appliance groups, location, age, and sex. Regression trees, a machine learning technique, are employed to develop a prediction model. Thesis focuses on households in the Czech Republic. The results demonstrate the efficacy of the prediction model, with low bias and improved accuracy compared to existing estimators for newly set meter points. The inclusion of regional-level variables enhances prediction accuracy only moderately. However, patterns derived from extensive datasets yield statistically significant conclusions regarding the effect of these variables. Challenges in incorporating certain variables and lack of longitudinal data limit the study. Future research directions may include exploring how different customer groups react to time- variant factors to enhance the accuracy and applicability of energy consumption predictions. The findings provide utility...
|