Robust and fast multi-market optimization

Energy is traded in different energy markets to help balancing out the energy in the power system. Power companies need to decide which markets to trade the energy in to ensure its best use. This is what is called "Multi-Market Optimization." The challenge is to predict how much electricity that will produced and what prices will be in these different markets, often planning for the next day or even weeks ahead. For big companies, this is a huge deal—every day, hundreds of millions of Norwegian kroner are on the line. The tricky part is that this problem is really complex, and the solutions need to be reliable, even when things don’t go exactly as planned. There could be unexpected changes in data or errors in the models you use, so you need methods that can still find good solutions despite these hiccups. There are two main ways to tackle Multi-Market Optimization: traditional optimization algorithms and Machine Learning (ML). Traditional algorithms are like using a calculator—they work, but they can be too slow to keep up with the fast-paced energy markets. On the other hand, Machine Learning can be much faster and often find better solutions. However, ML methods have a big downside—they're not always reliable, especially when things change unexpectedly. This project is all about bridging that gap. The goal is to create new methods that are not only fast but also reliable, so energy companies can make smarter, quicker decisions without worrying about the unpredictability of the markets.

Multi-Market Optimization is an optimization problem where actors in the energy domain attempt to maximize their profits by participating in several energy markets. To do Multi-Market Optimization, market actors must model future power production or consumption and prices in all markets. The problem is very complex and for big actors it involves hundreds of MNOK every day, so it is important that methods for solving the problem are robust. In this case, robust refers to the method's ability to find good solutions even when facing uncertainty, variation in input data and modeling errors. Current approaches to Multi-Market Optimization can be divided into traditional optimization algorithms and Machine Learning methods. The traditional optimization algorithms finds solutions to slowly to have any practical value, and the Machine Learning methods are known to lack robustness. Machine Learning methods have been shown to be able to outperform the traditional optimization algorithms in many cases, but the limited robustness means that the value for industrial actors remains low. The project aims to fill this gap by developing methods for Multi-Market Optimization that are both robust and fast.