Teacher:
Prof. Dr. Florian Steinke
M. Sc. Andrei Eliseev
M. Sc. Benedikt Grüger

Content:

Digitalization makes the analysis and interpretation of data ever more important, also for engineers. Smart Grids are a term to describe a host of novel data-based services in the field of generation, distribution, consumption, and marketing of (renewable) energy. The lecture presents the recent developments and the underlying machine learning methods. 

For a start, we describe the different problem areas that machine learning methods deal with, review recent developments in the field, and evaluate the impact of machine learning on the energy sector. After such an introductory overview we review the basics of linear algebra and numerical optimization. We then introduce supervised learning problems and study different model classes to solve such problems (linear models, trees, random forests, nearest neighbor, kernel methods, deep learning). We then turn to a probabilistic view and study unsupervised learning problems. Finally, we give an introduction to probabilistic graphical models. Throughout the semester we discuss exemplary applications of machine learning in the energy domain (e.g. renewable forecasting, predictive maintenance, state estimation, probabilistic load flow). 

Practical exercises with Python deepen the understanding and support student’s actively useable skills. 

At the end of the course, students know important machine learning problem settings and key methods for each task. Moreover, the students are able to apply those methods independently to new applications (in the energy domain and beyond). 

Type of course: 2 SWS lecture + 3 SWS exercise  

Language: English

Dates: Tue 8:55 - 11:30 and Thu 9:50 - 11:30

Literature: 

• K.P. Murphy: Machine Learning. A Probabilisitic Perspective
• C.M. Bishop: Pattern Recognition and Machine Learning
• J. Friedman, T. Hastie, R Tibshirani: The elements of statistical learning (https://web.stanford.edu/~hastie/Papers/ESLII.pdf)
• D. Koller, N. Friedmann: Probabilistic Graphical Models. Principles and Techniques

Requirements for successful participation:

Good knowledge of linear algebra required. Basic knowledge of statistics and optimization will be helpful. Using Python for programming the practical examples should pose no difficulty. 

TUCAN ID: 18-st-2020