Researchers Jack Umenberger and Anna Osguthorpe Rasmussen from the University of California, Berkeley have developed new algorithms to optimize the use of grid-scale battery energy storage systems. Their work, published in the journal Management Science, addresses the challenges of uncertain electricity prices and battery degradation, aiming to maximize revenue from these storage systems.
The researchers formulated the problem as an online resource allocation task. They proposed an algorithm based on online mirror descent, which performs well in both stochastic and adversarial settings. In the stochastic setting, the algorithm is no-regret, meaning it performs as well as the best fixed strategy in hindsight. In the adversarial setting, it achieves a finite asymptotic competitive ratio, indicating robustness against worst-case scenarios.
One of the key aspects of their work is the incorporation of untrusted advice about the opportunity cost of degradation. They developed a learning-augmented algorithm that leverages accurate advice to perform well, while still maintaining robustness when the advice is poor. This approach allows the algorithm to adapt and learn from the advice, improving its performance over time.
The practical applications of this research for the energy sector are significant. Grid-scale battery energy storage systems are crucial for integrating renewable energy sources and ensuring grid stability. By optimizing the use of these systems, the algorithms can help maximize revenue and extend battery lifetime, making renewable energy more competitive and reliable. The robustness and learning-augmented features of the algorithms ensure their effectiveness in real-world scenarios with uncertain and dynamic conditions.
In summary, Umenberger and Rasmussen’s work provides a robust and adaptive solution for optimizing grid-scale battery energy storage systems. Their algorithms address the challenges of uncertain electricity prices and battery degradation, offering practical benefits for the energy sector. The research was published in the journal Management Science, a leading outlet for research at the intersection of business and economics.
This article is based on research available at arXiv.