Friday, October 10, 2014

Very Short, Short, Medium and Long Term Load Forecasting

Load forecasting is so fundamental that it is being used across all sectors in the electric power industry for various business applications. Because of the wide spread of its applications, there are many ways to classify the various load forecasts:
  • based on forecast horizon: very short, short, medium and long term load forecasts;
  • based on resolution of the data or updating frequency (these two concepts are different!): hourly, daily, monthly, seasonal and annual load forecasts;
  • based on business needs: operational, planning and retail load forecasts;
In this post, I will try to touch all of them with the focus on the first group, load forecasting with different forecast horizons. Note that there are other terminologies based on the specific forecasting horizons, such as intraday forecasting and day ahead forecasting.

Anyone can have his/her own definition of very short, short, medium and long term forecasting to separate the detailed tasks under the corresponding jurisdictions. Here are a few examples:
  • The forecasting group in a distribution company that supports rate making and revenue projection may refer short term as one to five years ahead, and long term as 10 to 20 years ahead. 
  • The operations group in an Independent System Operator may use very short term for 5 - 15 minutes ahead, short term for a few hours to one day ahead, medium term for 5 days ahead and long term for two weeks ahead. 
  • The policy makers may treat 30 to 50 years ahead forecasting and long term forecasting, and anything below 30 years as short or medium term. 
  • A retailer may consider short term as one week ahead, medium term as one week to a few months ahead, and long term as up to two years ahead.
If I had done a survey, this list can go as long as the number of people being surveyed. I recognized this as an issue when I was pursing my PhD, so I formally proposed a classification in Section 1.3 of my dissertation that led to the integrated forecasting methodology proposed in Section 1.4. In short, I divided the load forecasting problem into four subproblems, very short, short, medium and long term forecasting with the cutoff points 1 day, 2 weeks and 3 years, respectively. In some other jurisdictions, I have also used another rough classification, which groups short and very short term together into short term load forecasting, and medium and long term together into long term load forecasting.

This classification is primarily based on the information being used to create the forecasts. The longer the forecasting horizon goes, the more information the forecasting process needs:
  • VSTLF only requires past loads; 
  • STLF usually requires past loads and weather information; 
  • MTLF requires weather and economy information; 
  • LTLF needs weather, economy, demographic and sometimes land use information. 
Of course there are overlaps. For instance, STLF can also be generated without weather information, though it is not a best practice for many utilities.

Now that we have defined these terms, how shall we use them? Here are four principles and some examples:

1. Respect the business audience.

When communicating with business users, first figure out what their terms mean, and then use their terms. Do not try to change their terminology, period.

2. Pay attention to the word(s) between "term" and "forecasting".

Long term operational forecasting means one to two weeks ahead load forecasting, which can be referred as "short term load forecasting" in the aforementioned framework. When putting "operational" before forecasting, the horizon of interest is being limited to the lead time for operational purposes right away.

3. Use the narrow definition if the methodology is specifically developed for a small category.

If a forecasting system is based on a univariate technique that does not give a good forecast beyond 6 hours, then name it VSTLF system.

4. Use the broad definition when the narrow definition is not suitable. 

If a regression-based forecasting system is applicable to both hour-ahead and week-ahead forecasting, then name it STLF system.

Recently I wrote a paper on "long term retail energy forecasting", which is essentially "medium term load forecasting for electricity retailers". In retail business, due to the dynamic nature of the business, most companies don't plan for 10 years ahead. As a result, the former one is much more precise and professional than the latter one.


  1. Hi Dr. Tao!

    You mentioned that the resolution of the data and the updating frequency are two different concepts.

    The resulution of the data is the same thing that sampling rate, isn't it?

    So, what is the difference between updating cycle and forecasting horizon?

    Thank you!!

    1. Yes, resolution of the data is based on the sampling rate. If the demand is recorded every hour, then it's hourly data.
      We can also use resolution of the data to name the forecast. For instance, forecasts based on hourly data can be called hourly forecasts.
      The updating frequency is based on how often you update the forecast. If the forecast is being updated every day, then it can be called a daily forecast.
      Here is the confusing part: a daily forecast (updated every day) can be an hourly forecast (based on hourly input data)...


Note that you may link to your LinkedIn profile if you choose Name/URL option.