Multifaceted Clustering of Complex Electricity Time-Series Data to Support Data-Driven Decision-Making in the Energy Sector
Closing Date: 31 July 2020
The primary goal of this project is to create clustering methods and appropriate visualisations especially designed for complex time-series data from the energy sector, in collaboration with the CSIRO. The focus is on the development of highly specialised statistical/machine learning techniques able to properly categorise, visualise, model and interpret relevant electricity consumption patterns, while meeting domain-specific requirements, such as suitable characterisation of anomalies and motifs.
Long time series data is one of the largest sources of residential energy use information in the electricity sector. These time series are obtained from interval meters which typically record electricity for an individual household every half an hour. Since interval meter databases end up capturing data for hundreds of thousands of households, across many years, this time series data becomes large and difficult to understand. Yet there is a real industry need to better understand this data to improve decisions around tariffs, consumer equality, and supply-demand operations. This is where clustering and forming clustered representations of time series become key for the energy sector. There are different approaches to achieve this so-called ‘load clustering’ (more broadly ‘time-series clustering’), where ‘load’ refers to the electrical time series load of a household. However, shaping the choice of approach for this clustering method are industry demands to better understand typical motifs of these clusters and anomalies amongst these clusters. This in part comes to the complicated relationship between humans and electricity usage, especially due to weather and work schedules. The electricity usage of a households is very closely linked to weather. Depending on the season, electricity usage can be wildly different due to space conditioning either heating or cooling large areas. Electricity usage is also very closely linked to work schedules. Usage can vary greatly between workdays and weekends/holidays. Businesses and organisations in charge of operating the electricity network are concerned about knowing the ‘when’ and ‘why’ of high simultaneous usage across residential and commercial sites. This typically happens when it’s very hot, a workday, and in the evening. When this occurs, it can mean outages for anyone on the network. This means a clustering approach should be robust enough to identify these anomalies within groups of households. Regardless of whether a cluster is formed based on an anomaly or a common seasonality present in the households’ time series, there’s also a requirement to have it be sufficiently represented by a motif. This allows industry to apply rules, conditions and considerations to a single representative load for many households, rather than having to consider each of these individually. Knowing these motifs can help improve the planning of electricity supply and policies. Under most time series clustering methods, clusters are formed on the basis of key time series features which are often highly biased by (in this particular case) seasonality. The anomalies despite their importance are often missed out in the process since they play only a small part in the time series and therefore do not bias clustering. As such, these small motifs are often overwhelmed by the dominant seasonalities. Some work in this area has already been completed by CSIRO. However, they have their limitations and would need reworking for this project.
The primary goal of this project is to create a clustering method and appropriate visualisations to support industry decision making.
- What is the best way to cluster together households?
- What is the best way to represent the seasonal variation within a household group?
- Can seasonal variation between household groups – or indeed individual members of those groups – be found and represented?
- Can individual households move between household clusters depending on the time of year?
- How can geographic information be incorporated into the clustering and subsequent visualisations?
- How can the clustering be made sensitive to anomalies between household groups?
- How can the clustering be made sensitive to anomalies between behaviours within a household group?
- When known, how can demographic, building characteristic, appliance uptake and usage, and other properties of households be considered in the clustering and visualisation of clusters?
- What is the best way to visualise the complexities in the above points?
PhD Scholarship details
Living allowance of $28,092 per annum (2020 rate) indexed annually. For a PhD candidate, the living allowance scholarship is for 3.5 years and the tuition fee scholarship is for 4 years. Scholarships also include up to $1,500 relocation allowance (if applicable).
- there may be an opportunity for a living allowance TOP-UP (upon application) to further support the recipient of this scholarship; and
- the selected candidate may be eligible to apply for annual HDR research support from the faculty/school, if available.
Supervisor: Prof. Ricardo J. G. B. Campello
Available to: Domestic students
- Domestic student in Australia
- Honours or Master Degree (Statistics, Computational Statistics, Data Science, Computer Science, Mathematics, Engineering, or related field)
- Verbal and written communication skills
- Programming skills
- Solid statistics/mathematics and/or computational background
The successful applicant must by able to commence by 30 August 2020. Please also refer to the admission eligibility criteria.
Interested applicants should send an email addressing the essential/desirable criteria and expressing their interest along with scanned copies of their academic transcripts, CV, a brief statement of their research interests and a proposal that specifically links them to the research project.
Please send the email expressing interest to firstname.lastname@example.org by 5pm on 31 July 2020.
Applications Close 31 July 2020
|Contact||Prof. Ricardo J. G. B. Campello|
|Phone||(02) 4921 6762|
PhD and Research MastersFind out more