Verification of Early Design Stage Machine Learning Model using EnergyPlus Simulation Data
This dataset is used to verify machine learning (ML) energy predictions using the EnergyPlus (EP) simulation. The EP model has been validated for one set of the parameters (SINGH, MANAV MAHAN (2020), “Validation of Early Design Stage EnergyPlus Model for Office Building”, Mendeley Data, V1, doi: 10.17632/x6xwvb2r9r.1). Then, several EP models are developed to generate the data for several combinations of the parameters as required. The pre-trained ML model is used to make prediction for the test dataset. The accuracy of prediction is reported using scatter plot, root mean square error (RMSE) and R2 values. The model shows RMSE of 4.2 MWh/a and R2 value of 0.94. In the scatter, most of the points lie close to the black-dashed line which shows ML predictions are close to the EP simulations. Please note that ML model is developed using a component based approach proposed by Geyer and Singaravel, 2018 and developed further by Singh et al., 2019. Please go through the mentioned paper for more details. P. Geyer, S. Singaravel, Component-based machine learning for performance prediction in building design, Appl. Energy. 228 (2018) 1439–1453. https://doi.org/10.1016/j.apenergy.2018.07.011 M.M. Singh, S. Singaravel, P. Geyer, Improving Prediction Accuracy of Machine Learning Energy Prediction Models, in: B. Kumar, F.P. Rahimian, D. Greenwood, T. Hartmann (Eds.), Proc. 36th CIB W78 2019 Conf., Newcastle, UK, 2019: pp. 102–112
Steps to reproduce
1. Save all the content from Mendeley Dataset on the disk. 2. Install python libraries - Pandas, TensorFlow, Matplotlib and SciKit-Learn 2. Use Programs/AccuracyTest.py to make prediction using the pre-trained ML models and report the accuracy.