Speaker
Description
Reliable infrastructure networks like the electricity grid are crucial for modern society.
The demand and supply of electricity need to be always balanced to ensure grid stability.
Various markets have formed to achieve power grid and electricity price stability These different markets enable trading over different time spans and ensure that supply and demand are always met.
Furthermore, parallel markets are connected to achieve this goal, as is the case for different day-ahead markets in Europe via the EUPHEMIA algorithm.
This results in a complex system of interacting parties such as electricity producers and consumers, external factors like regulations and weather conditions, and more.
An increasing amount of openly accessible data regarding the electricity net, power system, and external factors makes research in many different areas like grid frequency research, electricity network robustness, market analysis, and much more possible.
In my bachelor thesis, I took a look at freely available electricity price data from the ENTSOE transparency platform and various fuel price data.
I then identified extreme price events such as very high (> 50€/MWh) and negative prices.
Building on that I showed that a common linear model with dependence on the residual load can not explain said outliers.
Two logistic regression models were used due to their inherent interpretability to classify the outliers.
To classify these sparse events I used different training approaches such as over- and undersampling to compensate for the disbalance in data points (0.5% of the data points were negative, 1.9% over 50€/MWh). The aim of the thesis was mainly to get used to machine learning techniques and data preparation rather than to use big black box neural network machinery.
Despite its simplicity, the model showed expected as well as surprising phenomena that I will explain further in my poster.
