Autotune Articles

You found the oldest, and likely least-useful, set of articles. To date, there have been 4 primary-source articles which have spawned 30+ in the news:

• Article #1 - initial software development (September 22, 2014)
• Article #2 - release of Autotune as open source (September 2, 2015)
• Article #3 - NREL-hosted Lab-Corps program involving 86 interviews to identify industry's specific needs, uses, and value related to calibration (September 28, 2016)
• Article #4 - two case studies, including human vs. Autotune for calibration performance, in the Journal of Applied Energy (November 2016)

For a partial listing of Autotune news buzz, see bit.ly/autotune_publicity.

Autotune Science

While any multi-objective search algorithm can provide an answer, the DOE/BTO-funded Autotune project conducted an extensive search of over 300,000 algorithm instances to quantify performance of the best calibration algorithm in terms of:

• Matching utility bills - ASHRAE Guideline 14 for industry-standard compliance
• Accuracy/runtime tradeoffs - calibration typically follows Pareto Optimal patterns, but can take significant compute time for error rates below the requirements of G14.
• Scalability/Robustness - calibration for any type of building and set (or number) of tunable parameters (400+ tunable parameters tested with realistic runtime)
• Future-proof - calibration for any set of available data, including IoT or sensor data beyond utility bills (700,000+ annual data points tested with realistic runtime)
• Recover the actual building - calibrate to actual building parameters as defined in the new ANSI/RESNET Standard Method of Test for the Evaluation of Calibration Procedures. For practical applications, it is interest to note that matching utility bills is less than 25% correlated with matching actual building parameters.

The research, detailed in over 30 peer-reviewed publications (bit.ly/autotune_science), has been completed and Autotune is now open-source on GitHub (bit.ly/autotune_code) for use by the private sector.

Autotune Application

Regarding practical use of Autotune, there are a few resources that might help:

• Autotune overview video [YouTube]
• Overview slides [PDF] (note: Brian Ball presented OpenStudio's capabilities at this ASHRAE seminar)

• Step-by-step instructions for 3 ways to run Autotune (easiest to hardest/most-flexible):

  1) Standalone demo - download the directory, replace myidf.idf, myparams.csv, myuserdata.csv, and myweather.epw with your own files, and double-click autotune.bat.

  2) Virtual machine - download a virtual machine and immediately run demo, website, or web service in a virtual environment.

  3) Server installation - step-by-step instructions for installing on a Windows or Linux machine/server.

Autotune has been used in 3 countries, integrated into 5 companies' energy modeling pipelines, and being considered for application by 4 U.S. government agencies. Autotune does use OpenStudio (slide 32) for initial model creation, but only EnergyPlus during calibration for the reasons of speed and scalability. While created to be simulation-engine agnostic for easy application with new versions or simulation engines, the primary limitation for most users is that the open-source version uses EnergyPlus 7.0. Ultimately, it is up to the user to determine whether Autotune, OpenStudio, or other private-sector calibration services work best for your needs.