On Tuesday, March 24, we held our first brown bag seminar of the year, “Comparison of Peak Forecasting Methods,” which provided an overview of peak forecasting methods and thoughts on ways to structure stronger peak forecasting models. Modeling peak demand is an important topic and we felt it was a strong topic to kick off our annual brown bag series. During the session, we conducted a quick poll and 93 percent of respondents indicated their company develops forecasts of peak demand, highlighting the importance of this topic. We plan to launch a survey in the next month to find out more about industry practices.

In case you missed it, during the seminar Dr. Stuart McMenamin, Itron vice president of forecasting, led a discussion about alternative methods for peak forecasting and summarized the results that we have developed. We looked at modeling daily, monthly and annual peaks using data from 1998 to 2014. In our study, we examined various weather variables (e.g., temperature, humidity, cloud cover, etc.) and economic drivers (e.g., GDP, GMP, number of households, etc.) that are typically included in peak models. We did some exploratory analysis using artificial neural networks to determine the relative importance of these variables and to guide us in construction of explanatory variables for regression models. This included development of optimized THI variables, development of multi-part weather variables and construction of weather indexes that combine average temperature, humidity, maximum temperature, minimum temperature and lagged temperatures. These variables are then interacted with day type variables and seasonal variables to allow for slope shifts. The result of this structuring was a very strong and robust regression model.

We also examined using subsets of the data (e.g., all days, weekdays only, extreme weather days only, etc.) with hourly and daily modeling approaches, as well as using quantile regression for comparison against least squares. Ultimately, in-sample and out-of-sample tests were conducted to compare the performance of these approaches, which yielded some very interesting results.

Needless to say, a lot of information was discussed. Outside the technical discussion of building strong models, an additional takeaway from our analysis is that the relationship between energy use and the economy has changed dramatically since the Great Recession. This needs further study to better understand the causes of this change and what this means for longer-term forecasts.

For those who registered for the brown bag seminar, you should have received an email with a link to the recording of the presentation and the slides, so you can review it at your leisure. Let us know if you did not receive it.

We would like to thank everyone who attended for taking interest in our brown bag seminars. We have a few more coming later this year, including a session on June 9 about modeling the impact of new technologies. Be sure to mark your calendars and register at www.itron.com/forecastingworkshops. Please feel free to leave any comments about this brown bag seminar – we’d love to hear your thoughts! Keep your eyes open for the upcoming survey as all participants will receive a copy of the results. If you aren’t sure that you are on our mailing list or would like to be added, contact us at forecasting@itron.com.