Machine Learning (ML). It’s arguably one of the most cited buzzwords of our time. But what are its practical uses in the energy sector? On the 15th of March 2019, FiRRM hosted an insightful workshop in conjunction with Uniper and SFB 823 to answer this question. In this article, we’ll be looking at some of the highlights of our speakers’ talks.
How does it look on paper?
Before attempting to understand ML within the context of the energy industry, we should first do a quick review of the research. Aiding us in this endeavour was Dr. Nico Piatkowski, a graduate of TU Dortmund and research associate at the Department of Computer Science’s Artificial Intelligence Unit. Dr. Piatkowski laid the groundwork of our discussion in his presentation on the basics of research in ML and its practical applications.
Usually, building and testing models within a laboratory setting will produce expected results. However, applying them to the real world is where it gets complicated. The key question is: Does the proposed model really work? Unseen data is arguably the biggest factor in the effectiveness of an algorithm. In the real world, previously unforeseen factors can significantly skew the results of a given model. This is also known as empirical risk. One crucial step in applying a ML-model to the real world is validation. One way to do this is with Cross-Fold-Validation. Essentially, this involves applying a test data set to the existing training data set in a particular ratio. This is then repeated in as many combinations as possible in order to validate your model. The goal is to improve the ML model’s ability to generalize a pattern.
An important takeaway from Dr. Piatkowski’s presentation is the question of whether developing and implementing a ML model is even necessary in some real-world situations. It is important to consider the resources you need. Do you need the potentially massive computational power involved in ML? Or will a regular statistical model like simple regression do the trick? What tools will you need for the development of a ML model? And when should you use them?
Applying ML in Business Operations and Finance
Now, how would ML algorithms fit into business and trade? This was the central question of Uniper’s own Dr. Pascal Heider’s presentation. With readily available computational power provided by Amazon Web Services or Azure, the resources for data-processing are extremely accessible. Of course, it’s worth remembering that this can incur high costs. Despite the barriers to development such as cost, time constraints, and software expandability, Uniper manages to actively implement algorithms in its operations. For example, in predicting transport vessel (e.g. cargo ships) location and customer behaviour. Additionally, ML plays an important role in the company’s finances. In particular, employing algorithms as aids in automated Swing contract dispatching and in portfolio-allocation.
Broadening the topic within the context of finance is FiRRM’s own Dr. Timotheos Paraskevopoulos. In his talk, Dr. Paraskevopoulos provided insight into ML and its potential application in hedging. Inspired by the original work on Deep Hedging, Dr. Paraskevopoulos assessed the effectiveness of an ML agent at hedging European options in discrete time. In his approach, no assumptions regarding the underlying process are required. And as a result, it is model-free. Furthermore, he included customisable loss functions within the implemented modular framework. The implications of his work could hold a lot of potential for future research in reinforcement learning and its role in hedging strategy. For a more detailed view on this topic, read more in our blog post: Deep Hedge.
The potential of ML within the energy sector and, indeed business as a whole, is promising. As the research continues to develop, ML will play a leading role in the day to day operations in the ways mentioned above. For now, however, the challenge of effectively implementing ML in the real world remains, as before, one of the key barriers to its complete integration in the industry.