Industrial Multi-step Time Series Forecasting with Machine Learning Methods
Joseph Knutson · Duo Research Archive (University of Oslo) · 2019
Time series forecasting is a powerful tool for predicting the future values of variables involved in industrial processes, identifying their patterns through autoregression.Making forecasting models of industrial processes requires large amounts of industrial data, which is becoming more plentiful due to industrial digitalization.However, companies are aware of their datasets' value and do not necessarily make their data publicly available.Luckily, we have had the fortune of analyzing a utility company's industrial dataset which consists of measurements from thousands of sensors inside of Oslo's district heating system.The goal of our analysis has been to predict local differential pressure, inside a fixed location in the system, up to 40 minutes into the future.Using Artificial Neural Networks, such as LSTMs, it has been shown that they can better predict the differential pressure than models commonly used (as e.g.persistence or linear AR models).To whom it may concern I would like to offer my gratitude to my supervisor Morten and my co-supervisor Håkon.Morten, your lectures have always been my favourites.Your speech and personality have a warmth to them, as well as an entertaining charm.Håkon, I still remember meeting you when I was a bachelor's student, and you a master's student.Like Morten, you had a welcoming attitude and offered me a place to study among the other master's students.As a result, this office quickly became my home.I want to thank you and Morten, for your hospitality, advise, suggestions and proof reading.Secondly, I would like to thank Harald, Are, Espen, Bertil, Leif and the rest of Intelecy.Letting me cooperate with you gave me the opportunity to study subjects that I am highly passionate about.Extra thanks to Harald and Are, whose competence exceeds mine on the subjects surrounding this thesis.Your advice and proof reading have been crucial for the completion of