Data Enrichment and Increment for Deep Learning Component-based Energy Prediction Model

Published: 14-06-2020| Version 3 | DOI: 10.17632/9jvh8ckjbw.3
Sundaravelapandian Singaravel,
Philipp Geyer


This dataset is used for training of deep learning (DL) component based machine learning models described in the linked article. The article examines the effect of enriching training data with several building shapes on the prediction accuracy of machine learning models. There are nine building shapes used to collect the training data using EnergyPlus. Please read the full article for the relevant details of component structure and training of DL components. There are seven training dataset BaseCase, E-1, E-2, E-3, I-1, I-2, and I-3 and one test dataset TestData. The trained DL component are saved under Models folder in each dataset. The performance.csv file inside each dataset folder describes the performance of DL components trained on the corresponding dataset.


Steps to reproduce

1. Use Programs/IDFWrite.exe to generate IDF files for EnergyPlus simulation. It uses samples.csv file to generate IDF files. Samples are generated using Sobol sequences based on the range of design parameters as described in the linked article. 2. Run EnergyPlus to simulate all the IDF files generated in the previous step. Programs/SimulateIDFFiles.exe may be used for this purpose. 3. Use Programs/IDFRead.exe to read .csv files from EnergyPlus output and write data files for training deep learning components. 4. Train deep learning component using Programs/ (will take around a day to re-train models). Each dataset contains the trained model, if you skip this step. It will also make prediction on the test dataset using This step will add predictions in the test dataset files and will update performance.csv in the corresponding folders.