‘ Heating and Cooling with geothermal energy’
Dr Kathrin Menberg
Institute of Applied Geosciences, Karlsruhe Institute of Technology
Abstract:
Thermal Energy from the subsurface can provide climate-friendly heating and cooling all year round. While deep geothermal applications are limited to specific geologic settings, shallow geothermal systems can be installed almost everywhere and operate at different scales.
This talk will explore different applications of shallow geothermal systems that are commonly used to extract and store thermal energy in the subsurface. Focusing on the particular conditions in urban areas the potential of the shallow subsurface to cover the heating demand of urban quarters will be assessed. Also, economic and environment benefits of using the shallow subsurface as storage medium for thermal energy will briefly be addressed. Finally, potential detrimental effects on the environment by altering the subsurface thermal conditions will be discussed in the context of climate change.
‘Concept of Renewable Energy Network System and Benchmark of Deep Learning Forecasting Models Considering Building Data Specificity’
Dr Wonjun Choi
School of Architecture, Chonnam National University, Korea
Abstract:
Each renewable energy source available in the built environment has clear advantages and disadvantages. Therefore, relying on a single source is unlikely to achieve a high-performance building energy system. To address this issue, a renewable energy network system is proposed. This system involves connecting multiple sources, such as solar, geothermal, and air heat in a network form to increase operational flexibility and reduce system component size and initial costs. However, the operational flexibility of the system requires a sophisticated control like model predictive control.
In this presentation, the first part introduces the concept of the renewable energy network system and showcases the results of the pilot operation. The second part focuses on developing a deep learning-based load and temperature forecasting model for predictive control. The presentation showcases the benchmarking results of deep learning architectures, specifically focusing on the impact of the specificities of building energy data, such as limited data and seasonality, on forecasting performance. Additionally, the presentation explores the effect of the historical data length, also known as the encoding length, on the model's inner workings and its resultant forecasting behavior.