Machine learning-based planning and operating of flow-dependent renewable urban energy resources

• Applicant: Dr. Mario Rüttgers
• Subject Area: Fluid Mechanics ; Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing ; Electrical Energy Systems, Power Management, Power Electronics, Electrical Machines and Drives
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 535640183

The main objective of this project is to develop a tool that can be used for optimizing the utilization of renewable urban energy resources that are influenced by meteorological flow phenomena. These resources include urban wind turbines that are influenced by urban flow fields, and photovoltaic systems that depend on cloud motion in atmospheric flow fields. The resources are further classified into horizontal-axis wind turbines, vertical-axis wind turbines, external photovoltaic systems, and building-integrated photovoltaic systems. The tool is composed of two functionalities. With the first functionality reasonable and energy-efficient combinations and locations of the different energy resources are determined. This is realized with a reinforcement learning algorithm that moves and/or replaces the different energy resources, while receiving feedback in terms of wind and solar power generation. The algorithm is trained for a past time period. The power generation is based on atmospheric and urban flow fields, while considering shadows of immobile surroundings or the power curves of the devices. The atmospheric flow data are extracted from a database, and the urban flow fields are generated with a physics-aware graph neural network (PA-GNN). The PA-GNN uses boundary information from the atmospheric flow data. The second functionality focuses on operating the energy resources, once their locations are fixed. First, an artificial neural network is used to predict future atmospheric flow fields. Second, the previously mentioned PA-GNN generates the future urban flow field. Rotatable horizontal wind turbines can then be aligned with the flow, and future urban wind power generation is calculated based on information from a turbine's power curve. Future urban solar power generation can be derived from the predicted atmospheric flow fields. Finally, the combined future wind and solar power generation helps to determine how much energy is provided by flow-dependent urban energy resources, and what fraction needs to be provided by further resources to meet the total demand. The tool has the potential for major contributions towards energy self-sufficient or energy positive cities. Furthermore, it lays the foundation for a follow-up proposal for the Emmy-Noether-Programme, in which a GNN-based optimization for applications from life, earth, and energy sciences is targeted.

• DFG Programme: WBP Fellowship
• International Connection: South Korea
• Host: Professor Sangseung Lee, Ph.D.