Peer Group Discovery in District Heating Substations and Heat Pumps for Self-Monitoring
|Title||Peer Group Discovery in District Heating Substations and Heat Pumps for Self-Monitoring|
|Summary||Finding suitable peer groups to represent district heating and heat pump customers|
|References|| Y. Kim and S. Sohn, "Stock fraud detection using peer group analysis", Expert Systems with Applications, vol. 39, no. 10, pp. 8986-8992, 2012.
D. Weston, D. Hand, N. Adams, C. Whitrow and P. Juszczak, "Plastic card fraud detection using peer group analysis", Advances in Data Analysis and Classification, vol. 2, no. 1, pp. 45-62, 2008.
D. Weston, N. Adams, Y. Kim and D. Hand, "Fault Mining Using Peer Group Analysis", Challenges at the Interface of Data Analysis, Computer Science, and Optimization, pp. 453-461, 2012.
|Prerequisites||Artificial Intelligence and Learning Systems courses; good knowledge of data mining; programming skills for implementing machine learning algorithms|
|Supervisor||Ece Calikus, Sławomir Nowaczyk|
Peer group analysis is an unsupervised method for monitoring behaviour over time. The objective of this method is to characterize the expected pattern of behaviour around the target object by monitoring the behaviour of similar objects, and then to detect any differences between the expected pattern and the target. Peer group analysis can basically be divided into two stages:
(1) building peer groups, (2) detecting anomalous behaviour in the constructed peer groups.
In this project, the aim is to study methods to identify and build appropriate peer groups and evaluating peer group membership of the samples in district heating and heat pumps datasets.
Building suitable peer groups for different heat pump and district heating customers is a very challenging task. There are different types of customers which show varying behaviours changing over time. In most of the cases, they are not easily separable into well-defined clusters. Moreover, it is important to discover groups considering more than just "similar" customers based on their overall behaviour. For example; grouping the most commonly occurring patterns in daily energy consumption, grouping buildings according to their sensitivity to outside temperature etc.
- Finding good features to represent data
- Applying different unsupervised clustering methods and similarity measures
- Interpreting the clustering results in order to define peer groups
- Evaluation strategy to measure the purity of peer groups using different metrics such as silhouette coefficient, entropy, homogeneity etc.
- Evaluation strategy to measure how correctly samples are clustered into their peer groups.
We have collaboration with 3 companies within this project i.e. HEM, Öresundskraft and EasyServ.