Servicenavigation

Project Details

Back

Projekt Print View

Multi-Domain Modeling and Optimization of Integrated Renewable Energy and Urban Electric Vehicle Systems

Subject Area Electrical Energy Systems, Power Management, Power Electronics, Electrical Machines and Drives
Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
Traffic and Transport Systems, Intelligent and Automated Traffic
Term from 2019 to 2023
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 410830482
 
Final Report Year 2024

Final Report Abstract

The number of electric vehicles (EVs) in urban areas is expected to increase rapidly in the coming years. The German government has set a target of fifteen million EVs by 2030. Without the development of strategies for the grid-compliant integration of EV charging, this rapid growth can cause serious strains on power quality during peak demand as well as increased costs for grid maintenance. In order to develop such strategies, both the spatial and temporal distribution of EV charging demand for the area under consideration must be determined as an input variable. Charging demand data with high temporal resolution allow the identification of critical load peaks caused by EVs during the course of the day, while high spatial resolution allows the localisation of such load peaks for small neighbourhoods and districts, so that network planners can target network expansion and reinforcement. A method for determining the spatial-temporal EV charging demand is therefore developed in this project. In addition to the identification of congestion, the results can also be used to promote the development of charging infrastructure in a targeted manner. The integration of distributed renewable energy and urban electric vehicle fleets benefits from the development and implementation of a standardized reference network, which can then play a crucial role in conducting thorough power quality analysis. This reference network should include a substantial portion of distributed renewable energy sources to ensure accurate assessments. By monitoring voltage drops and line capacities, the impact of electric vehicle charging stations on the charging load can be evaluated, considering different connection scenarios. Furthermore, electric vehicle fleets have the potential to participate in balancing markets and offer flexibility services to system operators by effectively utilizing battery energy storage systems. Furthermore, in order to establish efficient interfaces with the results derived from multi-domain models, an integration methodology was introduced. This methodology was primarily designed to optimize the day-ahead operation of electric vehicle fleets. The key goals of this methodology encompassed the minimization of charging and grid operation expenses, reduction of battery usage and aging, and improvement of user acceptance. In order to couple distributed renewable energy and urban electric vehicle fleets in practice, a large amount of data (e.g. data on charging status, maximum charging time, parking time) needs to be collected and processed in different learned models. We therefore developed methodologies for efficient parallel processing of machine learning tasks, with a focus on improving parameter servers to enable training models on large amounts of data efficiently. First, we introduced dynamic parameter allocation (DPA) and implemented Lapse, a parameter server that achieves faster access to local parameters and reduces communication overhead compared to classic parameter servers. Second, we addressed non-uniform parameter access by developing NuPS, a non-uniform parameter server architecture that integrates replication and relocation techniques to handle skewed and sampled data. Finally, we propose AdaPS, an adaptive parameter server that utilizes intent signaling to automatically adapt to the underlying ML task, eliminating the need for manual tuning. AdaPS is efficient without any tuning and outperforms other state-of-the-art servers. Overall, these advancements aim to optimize distributed machine learning training and facilitate better performance in real-world scenarios.

Publications

 
 

Additional Information

© 2025 DFG Disclaimer / Copyright / Privacy Policy

Textvergrößerung und Kontrastanpassung