Don't miss our weekly PhD newsletter | Sign up now Don't miss our weekly PhD newsletter | Sign up now

  Generative AI in Energy Forecasting


   Department of Biomedical Engineering

   Applications accepted all year round  Self-Funded PhD Students Only

About the Project

Project Overview:

Meeting Net Zero targets will require fully utilizing smart solutions at all levels of the electricity network. This includes home energy management systems, optimal operation of battery storage systems, and local area energy trading. However, progress at implementing these solutions is hindered by the lack of data at the low voltage level. Although smart meters are being rolled out across the network their data is often unavailable or only analysed and available in limited ways through innovation projects. This is unlikely to change in the near future due to both privacy concerns and the value and opportunities that smart meters provide for suppliers.

One option is to utilise the limited open data sets that are available. However, these are often small in number with only tens of thousands of meters per trial, and they also only consist of a year or two of data. In this regard, generative AI (GenAI) is a valuable potential alternative. GenAI has become famous of late due to the release of the large language models (LLMs) such as ChatGPT and Bard, however the power in these algorithms is their ability to generate new but realistic content. Generative AI could be used to simulate new realistic realizations of energy data using only openly available data.

However, there are also other approaches which can be used deal with missing data. Transfer learning techniques enable estimates or forecasts of data instances which are similar to the training data. For example, a forecast could be fine-tuned for a household with limited historical data, by using the training on other household data.

The aim of this PhD will be to consider how generative, transfer-learning and meta-learning methods can be used to support energy system applications where open data is limited. In particular, one application of focus will be optimal schedules for local energy storage systems.

School of Biological Sciences, University of Reading:

The University of Reading, located west of London, England, provides world-class research education programs. The University’s main Whiteknights Campus is set in 130 hectares of beautiful parkland, a 30-minute train ride to central London and 40 minutes from London Heathrow airport. 

Our School of Biological Sciences conducts high-impact research, tackling current global challenges faced by society and the planet. Our research ranges from understanding and improving human health and combating disease, through to understanding evolutionary processes and uncovering new ways to protect the natural world. In 2020, we moved into a stunning new ~£60 million Health & Life Sciences building. This state-of-the-art facility is purpose-built for science research and teaching. It houses the Cole Museum of Zoology, a café and social spaces.

In the School of Biological Sciences, you will be joining a vibrant community of ~180 PhD students representing ~40 nationalities. Our students publish in high-impact journals, present at international conferences, and organise a range of exciting outreach and public engagement activities.

During your PhD at the University of Reading, you will expand your research knowledge and skills, receiving supervision in one-to-one and small group sessions. You will have access to cutting-edge technology and learn the latest research techniques. We also provide dedicated training in important transferable skills that will support your career aspirations. If English is not your first language, the University's excellent International Study and Language Institute will help you develop your academic English skills.

The University of Reading is a welcoming community for people of all faiths and cultures. We are committed to a healthy work-life balance and will work to ensure that you are supported personally and academically.

Eligibility:

Applicants should have a good degree (minimum of a UK Upper Second (2:1) undergraduate degree or equivalent) in Statistics, Mathematics and Engineering or a strongly-related discipline. Applicants will also need to meet the University’s English Language requirements. We offer pre-sessional courses that can help with meeting these requirements. With a commitment to improving diversity in science and engineering, we encourage applications from underrepresented groups.

How to apply:

Submit an application for a PhD in Electronics Engineering at http://www.reading.ac.uk/pgapply.

 

Further information:

http://www.reading.ac.uk/biologicalsciences/SchoolofBiologicalSciences/PhD/sbs-phd.aspx

 

Enquiries:

Prof William Holderbaum, email:

Computer Science (8) Engineering (12) Mathematics (25)

Funding Notes

We welcome applications from self-funded students worldwide for this project.
If you are applying to an international funding scheme, we encourage you to get in contact as we may be able to support you in your application.

References

1] Review of Low-Voltage Load Forecasting, Data Repository https://low-voltage-loadforecasting.github.io/ [2] TimeGPT, Nixtla, available here: https://docs.nixtla.io/
[3] Andrei Klubnikin, ITRex Group, What are foundation models, and how can they help implement AI at scale? https://itrexgroup.com/blog/what-are-foundation-models/
[4] Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados , Yoshua Bengio, Meta-Learning Framework with Applications to Zero-Shot Time-Series Forecasting, The Thirty-Fifth AAAI Conference on Artificial Intelligence, 10.1609/aaai.v35i10.17115
[5] Marcel Arpogaus, Marcus Voss, Beate Sick, Mark Nigge-Uricher, and Oliver Dürr, Short-Term Density Forecasting of Low-Voltage Load Using Bernstein-Polynomial Normalizing Flows, IEEE Transactions on Smart Grid, 14, pp 4902-4911, 2023.
[6] Eoin Brophy, Zhengwei Wang, Qi She, and Tomás Ward. 2023. Generative Adversarial Networks in Time Series: A Systematic Literature Review. ACM Comput. Surv. 55, 10, Article 199 (February 2023), 31 pages. https://dl.acm.org/doi/pdf/10.1145/3559540
[7] Jinsung Yoon, Daniel Jarrett, and Michaela Van der Schaar, Time-Series Generative Adversarial Networks, 33rd Conference on Neural Information Processing Systems, 2019. https://papers.nips.cc/paper/2019/file/c9efe5f26cd17ba6216bbe2a7d26d490-Paper.pdf
[8] Manuel Weber, Maximillian Auch, Christoph Doblander, Peter Mandl, and Hans-Arno Jacobsen, Transfer Learning with time Series Data: A Systematic mapping Study, IEEE Access, 2021. https://ieeexplore.ieee.org/abstract/document/9646532
[9] Julien Herzen, Florian Ravasi, Guillaume Raille, Gael Grosch, Transfer Learning for Time Series Forecasting, Unit8, https://unit8.com/resources/transfer-learning-for-time-series-forecasting/

Register your interest for this project



Where will I study?

Search Suggestions
Search suggestions

Based on your current searches we recommend the following search filters.