Joshua New, Ph.D. Oak Ridge National Laboratory newjr@ornl.gov 865-241-8783 Seminar 55 Simulation Calibration Autotune Calibration Learning Objectives • Describe how ASHRAE Guideline 14 defines calibration criteria for energy simulation • Describe how high performance computing (HPC) resources can be used to efficiently distribute simulation runs across multiple servers • Describe how machine learning algorithms can be used to support the development of efficient calibration techniques • Describe the disadvantages of each of the three calibration techniques presented • Describe the advantages of each of the three calibration techniques presented • Describe realistic scenarios for model calibration that can be utilized by practitioners today ASHRAE is a Registered Provider with The American Institute of Architects Continuing Education Systems. Credit earned on completion of this program will be reported to ASHRAE Records for AIA members. Certificates of Completion for non-AIA members are available on request. This program is registered with the AIA/ASHRAE for continuing professional education. As such, it does not include content that may be deemed or construed to be an approval or endorsement by the AIA of any material of construction or any method or manner of handling, using, distributing, or dealing in any material or product. Questions related to specific materials, methods, and services will be addressed at the conclusion of this presentation. Acknowledgements • • • • • • • • • Amir Roth – DOE’s BTO Aaron Garrett – JSU Jibonananda Sanyal – ORNL Richard Edwards – UT Mahabir Bhandari – ORNL Som Shrestha – ORNL Buzz Karpay – Karpay Associates XSEDE OLCF Outline/Agenda • Motivation • What is Autotune? – Calibration as search • How does it work? – Methods for speeding up the search • How good is it? – Calibration process and accuracy • How can I use it? – Deployment as web service Motivation 3,000+ building survey, 23-97% monthly error ASHRAE G14 Requires Using Monthly utility data CV(RMSE) 15% NMBE 5% Using Hourly utility data CV(RMSE) 30% NMBE 10% 5 The Autotune Idea Automatic calibration of software to data E+ Input Model . . . 6 Calibration as Search Problem/Opportunity: ~3000 parameters per input file 2 minutes per simulation = 83 hours 7 Supercomputers for Buildings • • • • EnergyPlus is a desktop app Writes files during a simulation Use RAMdisk Balance simulation memory vs. result storage • Validate simulation output • Bulk write data to disk • Design of Experiments for Uncertainty Quantification • In-Situ data analysis • Scalable Architecture for Big Data Mining • 270TB of simulation data CPU Wall-clock Data EnergyPlus Cores Time (mm:ss) Size Simulations 16 18:14 5 GB 64 32 18:19 11 GB 128 64 18:34 22 GB 256 128 18:22 44 GB 512 256 20:30 88 GB 1,024 512 20:43 176 GB 2,048 1,024 21:03 351 GB 4,096 2,048 21:11 703 GB 8,192 4,096 20:00 1.4 TB 16,384 8,192 26:14 2.8 TB 32,768 16,384 26:11 5.6 TB 65,536 32,768 31:29 11.5 TB 131,072 65,536 44:52 23 TB 262,144 131,072 68:08 45 TB 524,288 8 Suite of Machine Learning • Linear Regression • Non-Linear Regression • Feedforward Neural Network • Self-Organizing Map with Local Models • Regression Tree (using Information Gain) • Support Vector Machine Regression • Time Modeling with Local Models • K-Means with Local Models • Recurrent Neural Networks • Gaussian Mixture Model with Local Models • Ensemble Learning (combinations of multiple algorithms) • Genetic Algorithms Integrated mixture of Commercial, Research, and Open Source software 9 MLSuite Architecture Data Preparation PBS XML Supercomputer #1 MLSuite Linux #1 30x LS-SVMs validation folds 1-10 input orders 1-3 … Linux #218 Supercomputer #2 MLSuite Example • EnergyPlus – 2-10 mins for an annual simulation !- ALL OBJECTS IN CLASS Version, 7.0; !- Version !- SIMULATIONCONTROL === SimulationControl, No, !-Do Zone Sizing Calc No, !-Do System Sizing Calc … • ~E+ - 4 seconds AI agent as surrogate model, 90x speedup, <5% error; “brittle” <156 input changes 11 Getting more for less • EnergyPlus is slow – Full-year schedule – 2 minutes per simulation • Use abbreviated 4-day schedule instead – Jan 1, Apr 1, Aug 1, Nov 1 – 10 – 20 seconds per simulation r = 0.94 Monthly Electrical Usage r = 0.96 Hourly Electrical Usage 12 Evolutionary Computation How are offspring produced? Thickness Conductivity Density Specific Heat Bldg1 0.022 0.031 29.2 1647.3 Bldg2 0.027 0.025 34.3 1402.5 (1+2)1 0.0229 0.029 34.13 1494.7 (1+2)2 0.0262 0.024 26.72 1502.9 • Average each component • Add Gaussian noise • … “AI inside of AI” 13 Evolutionary Process Island Hopping 1. 2. 3. 4. 4 of 19 experiments Surrogate Modeling Sensor-based Energy Modeling (sBEM) Abbreviated Schedule Island-model evolution 14 Final Calibration Accuracy XSEDE and DOE Office of Science DOE-EERE BTO Industry and building owners Commercial Buildings Monthly utility data Hourly utility data Features: Works with “any” software Tunes 100s of variables Customizable distributions Matches 1+ million points Uses commodity hardware ASHRAE G14 Requires Autotune Results 15% 5% 30% 10% 0.32% 0.06% 0.48% 0.07% CVR NMBE CVR NMBE Residential home Within 30¢/day (actual use $4.97/day) Leveraging HPC resources to calibrate models on commodity for optimized building efficiency decisions 15 Working Internal Website 60+ fields (optional) Determine Inputs to Calibrate #Inputs #Groups #Inputs #Groups Restaurant Hospital 49 49 227 146 Large Hotel Large Office 110 71 85 45 Secondary Stand-alone Small Hotel Small Office School Retail 231 282 72 59 128 136 61 56 Medium Office 81 38 Strip Mall 113 89 Midrise Apartment 155 82 Super Market 78 73 Primary School 166 113 Quick Service 54 54 Warehouse TOTAL 47 45 1809 1143 Provide Data Calibrated Results Metric Input error average Input error maximum Input error minimum Input error variance IDF + CSV = XML CV(RMSE) CH4:Facility [kg](Monthly) CO2:Facility [kg](Monthly) CO:Facility [kg](Monthly) Carbon Equivalent:Facility [kg](Monthly) Cooling:Electricity [J](Hourly) Electricity:Facility [J](Hourly) … NMBE CH4:Facility [kg](Monthly) CO2:Facility [kg](Monthly) CO:Facility [kg](Monthly) Carbon Equivalent:Facility [kg](Monthly) Cooling:Electricity [J](Hourly) Electricity:Facility [J](Hourly) Electricity:Facility [J](Monthly) 143+ outputs Value 24.38 66.12 0.09 228.53 9.95 15.42 20.40 14.42 1577.96 10.48 -9.57 -14.78 -19.52 -13.83 592.77 -9.52 -9.52 Performance and Availability ASHRAE Autotune G14 Results Requires Monthly utility data Hourly utility data CVR NMBE CVR NMBE 15% 5% 30% 10% 0.32% 0.06% 0.48% 0.07% ASHRAE Autotune G14 Results Requires Monthly utility data Hourly utility data CVR NMBE CVR NMBE 15% 5% 30% 10% 1.20% 0.35% 3.65% 0.35% Results from 24 Autotune calibrations Results from 20,000+ Autotune calibrations (3 building types - 8, 34, 79 tuned inputs each) (15 types – 47-282 tuned inputs each) FY15 project to begin integration of Autotune web service as OpenStudio application Free to use. Pay for cloud computing. Bibliography • • • • • • • Garrett, Aaron and New, Joshua R. (2014). "A Scientific Study of Automated Calibration applied to DOE Commercial Reference Buildings." ORNL internal report ORNL/TM-2014/709, December 31, 2014, 114 pages Ostrouchov, George, New, Joshua R., Sanyal, Jibonananda, and Patel, Pragnesh (2014). "Uncertainty Analysis of a Heavily Instrumented Building at Different Scales of Simulation." In Proceedings of the 3rd International High Performance Buildings Conference, Purdue, West Lafayette, IN, July 14-17, 2014. Sanyal, Jibonananda, New, Joshua R., Edwards, Richard E., and Parker, Lynne E. (2014). "Calibrating Building Energy Models Using Supercomputer Trained Machine Learning Agents." In Journal on Concurrency and Computation: Practice and Experience, March, 2014. Garret, Aaron and New, Joshua R. (2013). "Trinity Test: Effectiveness of Automatic Tuning for Commercial Building Models." ORNL internal report ORNL/TM-2013/130, March 7, 2013, 24 pages. Edwards, Richard E., New, Joshua R., and Parker, Lynne E. (2012). "Predicting Future Hourly Residential Electrical Consumption: A Machine Learning Case Study." In Journal of Energy and Buildings, volume 49, issue 0, pp. 591-603, June 2012. Bhandari, Mahabir S., Shrestha, Som S., and New, Joshua R. (2012). "Evaluation of Weather Data for Building Energy Simulations." In Journal of Energy and Buildings, volume 49, issue 0, pp. 109-118, June 2012. Garrett, Aaron and New, Joshua R. (2012). "An Evolutionary Approach to Parameter Tuning of Building Models (Experiments 1-17)." ORNL internal report ORNL/TM-2012/418, April 2012, 68 pages. Questions? Joshua New newjr@ornl.gov