Technology

Researchers at the University of Kassel present a machine learning approach that represents specific target topologies (Tts) as actions

The electricity generation landscape has undergone a profound transformation in recent years, driven by the urgent global movement on climate change. This shift has led to a significant increase in renewable energy (RE) generation, resulting in a grid increasingly subject to variable feed-ins. The rise of heat pumps and electric vehicles has further increased consumer demand for electricity, and consumers are also beginning to contribute to the grid by generating their own electricity through photovoltaic panels.

Transmission system operators (TSOs) need to adapt their electricity infrastructure in innovative ways to deal with unpredictability. Switching buses at the substation level to change the network topology is a promising method that is gaining increasing attention in the academic community. To a certain extent, the grid can be stabilized in important parts through intelligent switching, as explained in. Particularly in DRL, which stands for Deep Reinforcement Learning, deep learning technologies could drastically reduce computing costs, which is why scientists are proposing to use them to solve this problem. French transmission system operator RTE was the first to test such methods as part of the L2RPN challenge. Due to its realistic representation of power grids, ongoing development, and difficulties, L2RPN has become the community’s preferred standard for DRL-based grid simulations.

The problem arises when these behaviors are often examined independently of one another. Although they could be useful for the following phase, they could lead to the creation of non-ideal topologies. Contrary to popular belief, autonomous substation activities are not taken into account during network operation. As an alternative, the gradual replacement of several substations is being considered. Nevertheless, DRL studies on network optimization barely touch on these comprehensive topology techniques. This could be due to the expensive calculations required to determine the combinations, or it could be a limitation of the L2RPN Grid2Op environment design, which only allows one substation change per time step.

In their current study, researchers at the University of Kassel are taking a new direction in which the focus is on the topology of the power grid and not on the switching processes of individual substations, but on the arrangement of all buses at all substations. The basic assumption is that some topologies (TTs) are more stable than others. Attempting to achieve close TTs takes precedence when our current topological state is not sufficiently persistent. Because the Target Topology (TT) can be achieved from almost any topology configuration, there is no need to understand specific combinations of substation activities. This is an advantage and is particularly useful in more complex networks as TTs can result in numerous substation actions being executed in sequence.

The study presents a search technique for TTs that meet the criteria. The results show that TTs are stable to instability using this technique when considering a collection of existing substation activities. Furthermore, the researchers integrate a greedy search component with TTs into their previously reported CAgent technique to create a topology agent (TopoAgent85–95%). The team ran the agent on the WCCI 2022 L2RPN Challenge validation grid to verify whether their method was useful for optimizing the grid. A multi-seed evaluation with 500 TTs was used to evaluate the impact of the proposed topology agent on the WCCI 2022 L2RPN environment. The TopoAgent85-95% agent achieved a 10% higher score and 25% longer median survival than the benchmark. Further investigation revealed that TopoAgent85−95% is close to the base topology, demonstrating its performance stability.

Overall, the study shows that using TTs as a greedy iteration hardly increases the running time. They believe that the research community should study TTs more, especially in combination with DRL.


Visit the Paper. All credit for this research goes to the researchers of this project. Also don’t forget to follow us Twitter. Join our… Telegram channel, Discord channelAnd LinkedIn Grupp.

If you like our work, you will love ours Newsletter..

Don’t forget to join our 41k+ ML SubReddit


Dhanshree Shenwai is a Computer Science Engineer with good experience in FinTech companies in Finance, Cards & Payments and Banking with keen interest in applications of AI. She is passionate about exploring new technologies and advancements in today’s evolving world to make everyone’s lives easier.




Source link