Every day large volumes of data are collected in the form of time series. Time series are
collections of events or observations, predominantly numeric in nature, sequentially recorded
on a regular or irregular time basis. Time series are becoming increasingly important in
nearly every organisation and industry, including banking, finance, telecommunication, and
transportation. Banking institutions, for instance, rely on the analysis of time series for
forecasting economic indices, elaborating financial market models, and registering
international trade operations. More and more time series are being used in this type of
investigation and becoming a valuable resource in today�s organisations.
This thesis investigates and proposes solutions to some current and important issues in time
series data management (TSDM), using Design Science Research Methodology. The thesis
presents new models for mapping time series data to relational databases which optimise the
use of disk space, can handle different time granularities, status attributes, and facilitate time
series data manipulation in a commercial Relational Database Management System
(RDBMS). These new models provide a good solution for current time series database
applications with RDBMS and are tested with a case study and prototype with financial time
series information. Also included is a temporal data model for illustrating time series data
lifetime behaviour based on a new set of time dimensions (confidentiality, definitiveness,
validity, and maturity times) specially targeted to manage time series data which are
introduced to correctly represent the different status of time series data in a timeline. The
proposed temporal data model gives a clear and accurate picture of the time series data
lifecycle. Formal definitions of these time series dimensions are also presented. In addition,
a time series grouping mechanism in an extensible commercial relational database system is
defined, illustrated, and justified. The extension consists of a new data type and its
corresponding rich set of routines that support modelling and operating time series
information within a higher level of abstraction. It extends the capability of the database
server to organise and manipulate time series into groups. Thus, this thesis presents a new
data type that is referred to as GroupTimeSeries, and its corresponding architecture and
support functions and operations. Implementation options for the GroupTimeSeries data type
in relational based technologies are also presented.
Finally, a framework for TSDM with enough expressiveness of the main requirements of time
series application and the management of that data is defined. The framework aims at
providing initial domain know-how and requirements of time series data management,
avoiding the impracticability of designing a TSDM system on paper from scratch. Many
aspects of time series applications including the way time series data are organised at the
conceptual level are addressed. The central abstraction for the proposed domain specific
framework is the notions of business sections, group of time series, and time series itself. The
framework integrates comprehensive specification regarding structural and functional aspects
for time series data management. A formal framework specification using conceptual graphs
is also explored.
| Identifer | oai:union.ndltd.org:ADTP/219566 |
| Date | January 2006 |
| Creators | Matus Castillejos, Abel, n/a |
| Publisher | University of Canberra. Information Sciences & Engineering |
| Source Sets | Australiasian Digital Theses Program |
| Language | English |
| Detected Language | English |
| Rights | ), Copyright Abel Matus Castillejos |
Page generated in 0.0022 seconds