Traditional Problem Definition - Computational Intelligence in Time Series Forecasting - page 15

2

Traditional Problem Definition

2.1 Introduction to Time Series Analysis

The importance of time series analysis and forecasting in science, engineering, and
business has, in the past, increased steadily and it is still of actual interest for
engineers and scientists. In process and production industry, of particular interest is
time series forecasting where, based on some collected data, the future data values
are predicted. This is important in process and production monitoring, in optimal
processes control, etc .
A time series is a time-ordered sequence of observation values of a physical or
financial variable made at equally spaced time intervals ǻ t , represented as a set of
discrete values 123

xx x etc . In engineering practice, the sequence of values is
obtained from sensors by sampling the related continuous signals. Being based on
measured values and usually corrupted by noise, time series values generally
contain a deterministic signal component and a stochastic component representing
the noise interference that causes statistical fluctuations around the deterministic
values.
The analysis of a given time series is primarily aimed at studying it's internal
structure (autocorrelation, trend, seasonality, etc .), to gain a better understanding of
the dynamic process by which the time series data are generated. In process
control, the predicted time series data values help in deciding about the subsequent
control actions to be taken.
The broad term of time series analysis encompasses activities like

,,, ,

x definition, classification, and description of time series
x model building using collected time series data
x forecasting or prediction of future values.

For forecasting the future values of a time series a wide spectrum of methods is
available. From the system-theoretical point of view they can be

x model-free , as used in exponential smoothing and regression analysis