Machine Learning Algorithms for BESS Energy Trading

Background
In a few years there will be a steady supply of used batteries from vehicles. The idea, on which this project proposal is based on, is to assemble such packs into Battery Energy Storage Systems (BESS). When connected to the grid and can assist with services, for example act as backup power or support grid frequency balancing. A BESS supports energy cost management by charging when there is plenty of supply (low prices) and discharge when supply is limited (high prices). The price of energy varies over time. Historically the price has been lower at night, due to lower demand but this is now changing in markets with a high penetration of solar energy.


To maximize the financial and environmental benefits of this type of arbitrage, an intelligent energy trading algorithm is critical. The income for sold energy shall be higher than the cost of bought energy and battery degradation. An enticing, but somewhat naive, approach is to set up some simple rules such as "discharge the power plant with 100 kW if the electricity price exceeds 2 SEK/kWh". Using a set of rules like that is probably sub-optimal leading to low returns. Also, the parameters of the rules will be local with a need to be adapted for a particular market, city, region or country. Expensive engineering work required to make such an adaptation will limit the scalability of the solutions. An optimal and model-based energy management is therefore required.




Suitable background
Computer science, mathematics, energy systems, physics

Description of thesis work
This master thesis addresses the following questions:
How can machine learning, reinforcement learning in specific, be used for energy trading with a BESS?
If there are multiple adequate machine learning algorithms available to answer the above question, what are pros and cons between these candidates?

The student(s) will be provided with know how in the area and required mathematical models.
The thesis work will include various fields such as economics, artificial intelligence, simulation and optimization. Personal interest in machine learning, programming or renewable energy solutions is seen as benefit. The work will be carried out at Volvo Energy in Gothenburg, in collaboration with Research Institutes of Sweden.


Thesis Level: Master

Language: Swedish and/or English


Starting date: Jan 2024


Number of students: 1-2


Tutors
Niklas Thulin, Volvo Energy, +46 739 027 383
Jonas Hellgren, RISE, +46 730 353 761