Spotlight on a Student: Fateme Fahiman
Machine Learning techniques for time-series forecasting with application to Smart Grids
What is your research about?
My research focuses on Machine Learning techniques for time-series forecasting with application to Smart Grids. With an ever-growing population, global energy demand is predicted to keep increasing. Furthermore, with the integration of renewable energy sources (to reduce carbon emission and dependency on fossil fuels) and grid modernization, the power system has become more volatile and less predictable than ever before. For example, renewable energy sources are weather dependent, which cannot be predicted exactly; we also need to cope with uncertainty in energy demand, consumer behaviour, energy prices, and transmission constraints. In fact, there are many sources of uncertainty associated with both demand and supply modelling. From a short-term perspective (seconds to hours ahead), maintaining frequency stability at the operational frequency acceptable range requires the demand and supply to match each other as quick as feasible. Frequency instabilities cause major security problems in power grid like grid black/brown out. Thus, maintaining the balance between demand and supply at all times is a challenging task in the operation of electric power grids. Furthermore, trying to produce more accurate load and renewable generation forecasting models has become a major research challenge for energy suppliers, energy marketers, financial markets, and other parties that contribute to electric power generation, distribution and transmission.
To this end, the questions that my research aims to address are:
- How much electricity is going to be consumed for the next 5 minutes to 24 hours ahead from customer level to grid level?
- What will be the temperature, wind speed, humidity, etc. for the next day and how do they affect the generation and consumption of the electricity?
- How much energy will be generated by each wind farm in the next 30 minutes all over the Australia?
- What will be the peak demand in the next day for each region in Australia, and at what time will that peak demand occur?
Answering these questions requires forecasting several future observations from a given sequence of historical observations, called a time series. Historically, time series forecasting has mainly been studied in econometrics and statistics. In the last decade, machine learning, a field that is concerned with the development of algorithms that can automatically learn from data, has become one of the most active areas of predictive modelling research. This success is largely due to the superior performance of machine learning prediction algorithms in many different applications.
Thus, by applying machine learning algorithms, I can extract intuitive patterns and trends from large data sets to produce more accurate forecasting models for electricity demand, as well as renewable energy supply from wind and solar generation.
Who are your supervisors?
My supervisors are Professor Christopher Leckie and Doctor Sarah Erfani from the Department of Computing and Information Systems.
Prof. Christopher Leckie has over 30 years’ experience in developing Artificial Intelligence and Machine Learning algorithms for application in cyber security, energy modelling and telecommunications. He has having led research teams at Telstra Research Laboratories, NICTA and the University of Melbourne. His research on using data mining for anomaly detection, fault diagnosis, cyber-security and the life sciences has led to a range of operational systems used in industry, as well as over 300 articles published in leading international conferences and journals. He is currently Associate Director of the new Oceania Cyber Security Centre.
Doctor Sarah Erfani has strong knowledge of machine learning, data mining, deep learning, distributed systems and data privacy. Her research interest is in understanding the underlying principles of learning and intelligence and applying the state-of-the-art machine learning techniques to real-life application scenarios.
What do you want to do next?
I have just under 4 months remaining before I submit my PhD thesis. I am also working on demand and wind generation forecasting projects at the Australian Energy Market Operator (AEMO) since May 2017 on part time basis. I have developed and deployed cutting-edge machine learning algorithms to improve the forecast accuracy of 5 minutes to day ahead operational power demand forecasting models for five regions (VIC, TAS, NSW, SA, QLD) in Australia. My proposed model improved the accuracy of AEMO’s day ahead peak demand forecasting model by 50%. My model is running in AEMO’s system and helps their control room to make operational decisions for the power system all over these five states in Australia.
Improving the models’ forecasting accuracy is a never-ending process. One of the current challenges is to predict the annual growth of "behind the meter" renewable energy (mostly solar) production. As managing the high penetration of renewable energy resources will become increasingly difficult, market prices for more clients will become flexible and more client groups will be encouraged to either store their own production, shift their demand towards off-peak hours or sell their production through blockchain-based markets. I expect this to open a completely new and very interesting chapter in energy forecasting and I am passionate about using technology to contribute positively to future sustainability practices.
Fateme and her supervisors warmly welcome any enquiries about her research. You can email her at firstname.lastname@example.org, Prof. Christopher Leckie at email@example.com, and Dr. Sarah Erfani at firstname.lastname@example.org .
- Best Student Paper Award in IEEE International Conference on Fuzzy Systems, Italy, 2017 (FUZZ-IEEE 2017): F. Fahiman, J. Bezdek, S. Erfani, M. Palaniswami, C. Leckie, “Fuzzy c-Shape: A new algorithm for clustering finite time series waveforms”. In Proceedings of the 2017 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE 2017), Naples, Italy, 9-12 July 2017, 8 pages.
- First ranked prize for the Melbourne Energy Institute's Energy Hack 2016.