Weather Normalization and Load Normalization

In the electric power industry, there are two variables often associated with "weather normalization": reliability and load. If you are interested in reliability, you may refer to the activities of IEEE Working Group on Distribution Reliability. In this post, I'm going to focus on load.

Planning - the business driver of weather normalization

The electric power grid is so important to the daily life of human beings. Electric utilities, the companies running the power grid and delivering electricity to our homes, have to conduct rigorous and comprehensive planning processes to ensure the financial stability of the company and system reliability of the grid. Because a major driving factor of the load is weather, which is quite unpredictable beyond couple of weeks, utilities can hardly predict the load in the medium and long term with high accuracy. Consequently, utilities would like to estimate the typical load first, and then to try to understand the range of extreme loads. In other words, most planning decisions have to rely on the typical load and the margin on top of it. Although there is not a formal definition of weather normalization, the process of estimating the typical load is in fact weather normalization.

Traditional practices of weather normalization 

At a high level, the traditional weather normalization process includes two steps:

1. develop a model based on actual observations of historical load and weather;  
2. apply the model to a normal weather to obtain the load under normal weather. 

Many parameters in this process would affect the results of weather normalization, such as the frequency of the input data to model the relationship between load and temperature, cross-validation in the modeling stage, selection of normal weather and so forth.

Over the past decades, utilities have been using annual or monthly data to develop predictive models for weather normalization. The models have been selected primarily based on the goodness of fit on the historical data. The normal weather published by NOAA is often used as the scenario for typical load.

Load normalization against weather

There are many issues with the traditional practices mentioned above: monthly or annual data does not provide detail load and temperature profiles nor enough observations for comprehensive analysis; goodness of fit does not imply predictive power; load under normal weather may not lead to normal load.

To address these issues together with many other problems around weather normalization, I proposed the concept of load normalization in my recent paper Long Term Probabilistic Load Forecasting and Normalization. Then weather normalization can be interpreted as load normalization against weather. My proposed process include three steps:

1. develop a model based on actual observations of historical load and weather; 
2. apply the model to a set of simulated weather profiles to obtain scenario based load forecasts; 
3. derive median of monthly peak or monthly energy or other load of interest as the normalized load. 

There are many other factors driving the load in the long term, such as economy, demographics, etc. we can apply the similar load normalization concept to come up with load normalization against economy, demographics, land use development and so forth.

Back to Load Forecasting Terminology.