Contact: Jagdish Achara
Project Description:
When we use or analyze real data sets (e.g., a time series of power consumption) in a smart grid environment, the need of using different time scales often emerging. For example, a data time series over a slow time scale maybe suitable for energy management optimization purposes but for testing real-time control schemes data time series over faster time scales should be available. In this project, we aim at analyzing time series of power consumption data obtain via PMU measurements on two different grids, specifically the EPFL grid (let’s call it the dataset B) and a second typical city-wide grid (let’s call it dataset A). Our data analysis has, among others, also the purpose of studying when it is possible to create synthetic but realistic data over a fast time-scale when we are provided with data over a slower time scale.
More specifically, we are provided with dataset A i.e., the time series of power consumption at a bus of a grid for the city-wide grid which is at minutes or hours scale (slower time scale). However, we need these power traces at a finer time-scale (e.g., seconds) scale for various purposes, e.g., for testing real time control schemes under realistic conditions. The main goal of this project will be to generate the time series for this dataset A at sub-second scale by studying a different dataset, i.e., dataset B, of power traces which is given in sub-second scale. During the course of the project, the student will work on studying the dataset B to find features or trends that can be applied to dataset A for generating the power traces at sub-second level. Multi-resolution analysis (MRA) using wavelets [1, 2] is the main technique the student will explore to do the aforementioned task. Wavelets are suitable for analyzing time series datasets in different components each one reflecting the time evolution of the signal at a particular frequency. Thus, they analyze the time series datasets both in time and frequency, compared e.g., with the Fourier transform that analyzes the data only in the frequency domain suppressing the information over time. Using Wavelets, we can study the importance of diverse time scales in the data time series. Finally, if needed, the student may as well need to study and apply other spectral analysis or machines learning approaches to achieve the goal of the project.
[1] http://davis.wpi.edu/~matt/courses/wavelets/
[2] https://www.atmos.umd.edu/~ekalnay/syllabi/AOSC630/Wavelets_2010.pdf
Project Goals:
- Big Data Analytics over real data sets for power systems using Wavelets/spectral analysis/machine learning.
- Development of methods for creating synthetic but realistic datasets over faster time scales when we are provided with time series datasets over slower time scales.
Required Skills:
- Strong interest in the topic as well as in the learning process
- Knowledge on time series analysis techniques, e.g., wavelets, or willing to learn.
- Proficient in either R or Python or Matlab
- Knowledge of machine learning techniques in general.
Supervisors: Eleni Stai, Jagdish Achara