The residential sector accounts for a significant share of the final energy consumption in Germany. Therefore, the demand for accurate electricity consumption forecasting is a strong research topic. This thesis explores the integration of semantic 3D city models with machine learning to simulate residential electricity consumption at the household level. By leveraging CityGML, an international standard for 3D city modeling, and employing deep learning architectures, particularly Long Short-Term Memory (LSTM) networks, the study aims to improve the accuracy of predictions. The structured methodology incorporates geospatial data, statistical data, historical electricity consumption records, and machine learning algorithms to identify patterns in residential energy use. A key innovation of this work is the combination of semantic city models with deep learning approaches, providing adaptability for multiple cases. The thesis extended the use of the official load profile calculation method for households with missing information. Also, the results demonstrate that the LSTM model is promising and can improve prediction accuracy, offering valuable insights for the electricity industry. The findings support the potential of integrating geoinformatics and artificial intelligence for urban simulations.      «

The residential sector accounts for a significant share of the final energy consumption in Germany. Therefore, the demand for accurate electricity consumption forecasting is a strong research topic. This thesis explores the integration of semantic 3D city models with machine learning to simulate residential electricity consumption at the household level. By leveraging CityGML, an international standard for 3D city modeling, and employing deep learning architectures, particularly Long Short-Term...     »