The Challenge

1
Data Mining Using Independent Component Analysis
An Application To National Electricity Market
Data
Presented By Ravi Abeydeera Supervisor Dr
Marcus Gallagher
The Challenge In recent times, deregulation of
electricity markets throughout the world has
initiated a growing amount of interest in the
modelling, monitoring and management of the
market. Large data sets have also evolved and the
analysis of this data has become a topic of
increasing amounts of literature. This research
implements data mining concepts and methods to
Australian National Electricity Market (NEM)
data, generating results that will aid market
participants in better understanding the
complexities of the market.

• Methodology
• Assumption observed multivariate time series
  (bid profiles, prices, generation data and demand
  data) reflect the reaction of a system, such as
  the electricity market, to a few statistically
  independent, but not directly observable factors.
  Data is treated as a set of observed signals and
  approach the problem of extracting these
  independent factors as a Blind Source Separation
  problem.

Figure CRISPDM Reference model used for data
mining

• Exploratory Analysis
• Data Cleaning
• Derive variables, Transform, Merge and Aggregate
  Data
• Apply filters to data in context of acquired
  business understanding
• Build Model Apply ICA and PCA to data sets
• Reconstruct data sets utilising the factors
  obtained
• Model Evaluation
• Review Process

Figure Complexities of data Contribution of 4
Units to Swanbank B Power Station's generation
Figure Process of ICA

• ICA Independent Component Analysis Using JADE
• Objective of ICA Find a transformation of the
  estimated source signals such that obtained
  signals are statistically as independent from
  each other as possible. We utilise the JADE
  algorithm.
• Results are compared to those obtained using
  principal component analysis (PCA) and in some
  cases yield more superior results.