Mikkel Slot Nielsen

Now Big Data Can Be Better Utilised in Probability Calculations

PROBABILITY THEORY

PhD Mikkel Slot Nielsen has solved some important problems regarding the understanding of time series. And then he has helped Vestas assess the risk of extending the service life of their wind turbines.

By Filip Graugaard Esmarch

In his PhD project, mathematician Mikkel Slot Nielsen has worked on the further development of highly specialised methods within probability theory. His thesis contains as much as eight published articles about models for understanding time series data, i.e. data arranged on the basis of measurements over time. With such models you can calculate probable scenarios for the future development of the measurements – within e.g. meteorology, climate studies and economics. 

‘If you are an investor, you may look up the stock price on a daily basis and wish to be able to predict when it will stop falling. Or you may be interested in the probability that it will exceed 100 at a later point’, Mikkel Slot Nielsen explains.

Using approx. 40-year-old time series models you can simulate e.g. 1,000 scenarios and determine how many of them will meet a specific criterion. But the classic models fall short when using a time series with more detailed observations where you have e.g. measured a stock price every minute rather than every day. But Mikkel Slot Nielsen found a solution.

Ready for BIG DATA

‘With the type of models that I have looked at, I have shown how to make a similar setup that allows for a lot of observations and including all the data available. In this way, the method utilises big data and strong computing power better than what has been possible so far’, he says.

In short, he was trying to combine results from old time series models with methods from stochastic differential equations.

‘This requires fairly detailed probability theoretical results. And at the same time, it leads to econometric problems, e.g. concerning co-integration, which is what is evident from Pepsi’s stock price compared to Coca Cola’s – their individual fluctuation may seem extreme, but they share this extremity’, says Mikkel Slot Nielsen.

Helping Vestas

The ninth and final article in Mikkel Slot Nielsen's thesis shows how applicable probability theory-based research can be. This is evident from an industrial collaboration with Vestas, which, based on data from their wind turbine sensors, needed help to calculate how much wear and tear a wind turbine will experience through its service life, and thus how long you can justify to keep it running.

‘The machine learning algorithm we used typically reveals the most likely degree of wear and tear on the wind turbine tower. However, with regard to their risk assessment Vestas was more interested in determining the extent of potential tear and wear in the one percent most extreme cases. Therefore, we had to adjust the algorithm’, says Mikkel Slot Nielsen.

A postdoc scholarship from the Independent Research Fund Denmark has enabled him to continue his research at Columbia University in New York, where he is currently exploring unique properties of the same machine learning algorithm.