Incorporate accurate time series forecasting to solve real-world problems
By Ms Nancy Van Nieuwenhove | 4th May 2020 | *Edits Nov. 2020
2nd Year PhD candidate, Bsc (Hons) in Engineering, BCS
Research interests: machine learning, time series forecasting, optimisation, simulation, modelling
Rakshitha Godahewa is originally from Sri Lanka and has always wanted to become an engineer. Her dream came true when she won the very competitive GCE Advanced Level examination in Sri Lanka. She passed as one of the top 4 students in her district and qualified to enter the Faculty of Engineering, University of Moratuwa. She majored in computer science and engineering, won two awards from University of Moratuwa for academic excellence, completed her degree with a first class honours and was one of the top 4 students in the class of 2016. Meanwhile, she also completed the British Computer Society (BCS) degree. In the final year of her bachelor’s degree, she did a machine learning research project where she used Visitor Location Registry (VLR) data for short-term traffic prediction. The research findings were presented at the Netmob Conference, Italy 2017. There, she found an interest towards machine learning and Artificial Intelligence (AI) research.
After completing her degree, she worked for Codegen International (Pvt) Ltd, as a Senior Software Engineer for 3 years and applied problem-solving principles to help the sector. She dealt with travel-related bookings, built web-based products and gained analytical and programming skills as well as knowledge in algorithms. She realised she could contribute greatly to the engineering field by becoming a researcher. “After browsing through the Monash University website, after seeing the number of researchers and their invaluable contribution to the community, I contacted Dr Christoph Bergmeir (Senior Research Fellow at Data Science and AI Dept., Faculty of Information Technology) and Prof Geoff Webb (Research Director, Monash University Data Futures Institute) with my research and they took me on as a PhD candidate in February 2019.“
During the first few months of her PhD, she developed a new classifier that supports accurate high-dimensional stream classification. Her research paper on the subject was recently accepted to be presented at the International Joint Conference on Neural Networks (IJCNN), UK, 2020.
Rakshitha’s PhD research mainly focuses on improving the accuracy of time series forecasting using global ensemble models; like Recurrent Neural Networks (RNN), Feed-Forward Neural Networks (FFNN) and Pooled Regression models. “We predict the future of time series via different machine learning techniques and increase the forecasting accuracy by optimally aggregating a set of forecasting models. My main goal is to use these models with electricity demand forecasting and energy forecasting. Accurate time series forecasting can be very beneficial for any businesses and industries. Currently, we are using the Ausgrid dataset with our experiments which represents the energy consumption of Australian households. I recently presented my main research findings at the 40th International Symposium on Forecasting (ISF).”.
She has already started using these forecasting models with real-world applications. She worked on a project with Honeywell Ltd where she used temperature forecasts provided by an RNN model to optimise the energy consumption of air conditioning systems especially during the unoccupied periods of lecture theatres. “Often the AC is on in lecture theatres, even if no lectures are happening. We simulated and predicted the future inside temperatures of lecture theatres using machine learning technologies and later used them to optimise the energy consumption of air-conditioning systems. Our simulations have shown that the proposed approach can lead to 20% of energy savings in buildings. We recently submitted a manuscript for a journal based on this work. The conclusions of this research are valid to similar building rooms including hotel conference rooms and libraries, so they have high commercial benefits”.
In 2019, Rakshitha and Dr Arnaud Prouzeau presented their findings at the Honeywell Users Group Conference, in Sydney Australia. “Arnaud focused on the immersive visualisation part of our research and I presented the machine learning part. We received many positive comments from internationally based researchers. I also presented our findings (poster format) at the Monash Energy Conference with Chang Deng, Masters student in IT".
Rakshitha also participated in the AIE poster competition presenting this research project. "The AIE competition was a great experience for me. Our research work will reach for a larger audience including researchers and industry people and that attracted me towards that competition". She has also developed a new forecasting model which is especially suitable to forecast weekly data. They have tested this model on 6 datasets where 2 of them are energy-related.
To help the research community, she built a new repository containing time series benchmark datasets where the researchers can use the datasets to test their newly proposed forecasting algorithms. This repository contains many energy-related datasets.
Her research is mainly funded by co-funded Monash Graduate Research Scholarship and Faculty of Information Technology International Postgraduate Research Scholarship. But, this year, Rakshitha received a top-up scholarship from the Monash University Grid Innovation Hub for her contribution towards energy optimisation research. She notes that all these achievements wouldn't be possible without the endless support and guidance from her supervisors. "This scholarship allowed me to contribute to more energy-related research".
Furthermore, she recently participated in a forecasting competition conducted by Decathlon and was able to obtain the first position. She also participated in the M5 forecasting competition with 3 other Monash students. The method proposed at the M5 forecasting competition, resulted in the lowest error and gave their team a gold medal, along with the 17th position out of 5588 teams globally and 2nd in Australia. Recently she participated with 2 other Monash students in the IEEE CIS Technical challenge 2020 and received the 4th rank. "The goal was predicting the energy consumption of 3248 households. Our submission has been listed as a top submission and we will get an award for that".
In 2021, Rakshitha would like to see more research towards energy optimisation. “The world has a limited amount of energy sources. So, there is a need to increase the research towards renewable energy and energy optimisation. I would also like to see more collaborations between Monash University and industry partners towards energy optimisation research.”
She won the best "Teaching Associate - PhD Student" award from the Department of Econometrics and Business Statistics, Monash University, Australia in January 2021
- Linkedin Profile
- Simulation and optimisation of the air conditioning system
- Short-Term Traffic Prediction Using Visitor Location Registry Data
- Simulation and Optimisation of Air Conditioning Systems using Machine Learning
- New forecasting model suitable to forecast weekly data.
- AIE poster video
- AIE Poster
- Ensembles of Localised Models for Time Series Forecasting: https://arxiv.org/abs/2012.15059
- Non-energy related work. Paper published in IEEE International Joint Conference on Neural Networks: https://ieeexplore.ieee.org/document/9207303
- R. Godahewa, C. Deng, A. Prouzeau and C. Bergmeir. Simulation and Optimization of Air Conditioning Systems using
Machine Learning. June 2020 (https://arxiv.org/abs/2006.15296, submitted to Expert Systems with Applications)
- R. Godahewa, C. Bergmeir, G. I. Webb and P. Montero-Manso. A Strong Baseline for Weekly Time Series Forecasting.
October 2020 (https://arxiv.org/abs/2010.08158, currently preparing a resubmission to International Journal of Forecasting)
- R. Godahewa, K. Bandara, G. I. Webb, S. Smyl and C. Bergmeir. Ensembles of Localised Models for Time
Series Forecasting. January 2021 (https://arxiv.org/abs/2012.15059, submitted to Knowledge-based Systems)