Course contents *
The course consists of an introductory reading part covering classification problems and regression problems, statistical methods such as kNN, Artifical Neural Networks and Decision trees. These topics are divided into three separate study blocks, for each study block there are 2-3 lectures, and 1-2 seminars. The study blocks are:
Study Block 1 - Statistical Methods
The topic includes basic statistical tools for data analysis and pre-processing of data. In addition, distance based methods such as Nearest Neighbour and other non supervised learning techniques are covered.
Study Block 2 - Artificial Neural Networks
The topic covers design and optimisation of learning in ANN single and multilayer networks. Methods for optimisation and learning such as Kohonen feature maps are also included. In addition, methods for structured search in problem spaces are covered in this study block.
Study Block 3 - Decision Trees
Analysis of problems and design of Decision Trees for classification and regression is covered in this study block. Methods for optimal splitting and validation are included.
The individual project is consists of 4,5 ECTS credits and involves developing a decision support application that utilizes machine learning techniques. The project shall at least consist of identification of topic, reading of related work on application of machine learning techniques to the problem, gathering and cleaning of data needed for the application, selection of suitable machine learning approach, development of solution and documentation of the solution in the form of a conference paper suitable for an entry level conference focused on applications such as IEEE ISGT or similar.
Intended learning outcomes *
After completing the course the student should be able to:
· Create algorithms for data capture and cleaning from heterogeneous data sources using statistical tools.
· Develop and validate Decision Tree algorithms with application to power system problems
· Develop algorithms for forecasting of production of renewables (wind, PV) using statistical methods and data based methods such as ANN.
· Explain differences between chosen algorithms for forecasting including aspects of data availability and forecasting accuracy.
· Given a power system problems, and appropriate test data to develop a function of machine learning that leaves decision within example one of the following areas.
o Pattern recognition in consumption or generation profiles
o Prediction of production from renewable energy sources like wind power or solar panels
o Analysis of fault scenario using the phase angle measurements.
o Implement appropriate solution in software and visualize the results.
No information inserted