Time Series Analysis Part 1 (intro)

A time series is a series of data points indexed (or listed or graphed) in time order. Most commonly, a time series is a sequence taken at successive equally spaced points in time. Thus it is a sequence of discrete-time data.

Time Series Data obviously has to do with time and the first thing that comes to mind is finance. The world of finance is the world of time series, so stocks, and currency exchange rates and interest rates, they’re all Time series.

Another place is in click through rates. We can try to model how people click on certain places over time.

Sales forecasts are time series because they change over time.

Demand forecasts, when we think of the utility sector. One of the major problems that power companies face every day is to predict the demand for the following day so that they could generate enough energy.

Also, there’s a lot of natural phenomena that exhibit time series, like water flow or temperature, which changes every day.

There’s a lot of biological time series as well, like our heart rate or blood pressure.

Thus, there’s a lot of things with a temporal nature to them that we might want to think about.

In order to understand better the importance of Time series analysis, let’s see some common Time series sources:

  • Medical and Biological
    • Nervous and muscular systems activity: Electrical activity from neurons (in brain or muscles)
      • electroencephalogram (EEG),
      • electrocardiogram (ESG)
      • electromyography (EMG)
      • polysomnography (PSG)
    • Genomics: ADN sequences can be analyzed as time series
  • Language: Natural Language, Speech, and Music
    • Speech processing: speech understanding (speech to text), speech synthesis, translation, compression
    • Natural Language processing: understanding, translation, knowledge extraction, synthesis, summarizing
    • Music: classification, synthesis, intelligent composition
  • Nature and Environment
    • Meteorological variables: temperature, humidity, pressure, humidity, rainfall. Mapping, forecast
    • Pollutants: source detection, modeling, forecasting
    • Earthquakes: detection ground motion and waves, modeling, prediction(?). Water waves (tsunami)
    • Astronomy and Astrophysics data
  • Energy
    • Electric power consumption from large grid, micro grid or single consumer: modeling, forecast, control
  • Industrial, Control, Machinery
    • Signal detection and measurement (smart/soft sensors)
    • System control: predictive, non-linear, multivariable, etc.
    • Distributed sensors (internet of things)
  • Financial and Economic
    • Financial markets values (stock, index, commodities, currencies):
    • Econometric and macroeconomic series (PGB, CPI, rates, etc.)
    • Consumer finance: credit risk, consumption, payments, fraud.
  • Communication and Networks
    • Baseline signals: production, detection, errors,…
    • Internet traffic, routing, information
    • Local networks: optimization, detection, assignments
    • Transportation networks: intelligent transportation

Given that we have temporal nature in our data, what are the sorts of things we can do with that?

The first simple thing is just to try to smooth it, because time series can be very noisy if they’re, for example, coming from a natural process like the weather.

There’s a lot of day-to-day variation that isn’t maybe part of a long-term climate trend that we are trying to extract. So we want a smooth our time series.

The other we want to do is to build a model where we decompose the time series into its parts so that we can explain what’s going on. For example, electricity production has a strong seasonal component because, in hot times of the year, consumers use more air-conditioning, or there may be some seasonal industrial production that goes on. So we want to be able to identify that seasonal part, versus what’s a random component, versus what’s just the correlation from one time to the next.

This is the modeling aspect and finally, the real goal is forecasting, and with forecasting, we want to know what the next step is, like if the power company needs to know how many generators should be running, which is quite expensive as you can imagine.

We don’t want to come up short, but we do not want to waste a huge amount of money burning fuel that we never needed to burn. So we want to know over the next few days, maybe even few months, what’s the expected demand for electricity so that we can effectively with minimum cost, actually hit that demand and produce that amount of electricity.

So, forecasting is widely used in all sorts of disciplines and for the obvious reason that having some insight into what is going to happen next in the future, can be hugely valuable and … variable.


This Blog is created and maintained by Iraklis Mardiris


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s