24Fa-EnergyInBuildings-Res

Energy in Buildings - Residential

Introduction

This project intends to build and research upon a surrogate model that could help predict the heating and cooling demands of residential homes specifically in Georgia. The model researched here is very similair to the one used by the sister team Energy in Buildings - Commercial. Whilst the models are similair at first, some differences were applied in our research as some variables are more preferable to Residential Buildings rather than their Commercial counterparts.

About the model

The machine learning model is based on this notebook on Optimization of building energy consumption. The version here is pretty light as of now, and it consists of 3 main python files, where at the end it should give out a SAV file that can be used to make predictions given certain inputs for whichever program they are needed for.

Before the model is ran, the proper Dataset to train the model must be created. We sourced our dataset from NREL-ResStock. We chose the variables we thought would be most important specifically for residnetial buildings and also combined information to create our own independent columns as seen below.

Encoder.py

After the dataset is finalized, it is ran through the Encoder.py file. Since the model relies on numerical data, it is imperative to find ways to convert categorical data to be usable. Specific to this, our inputs for BuildingType and Orientation were simply label encoded. This file must be updated if other categorical or non-numerical data values are inputted for a new dataset. The file also creates a mapping csv to show what the numbers means after the data is encoded.

Main.py

This file is similair to the notebook 1B, except updated to account fo two output variables and specific to the dataset. As the groups later merged into one, a lot of updated research was done so the model currently is not updated. As of now, similair to 1B Notebook, the file creates various plots whoch showcase statistics, such as the correlation in the heatmap shown below.

The file then gives relational data on various ML models and then chooses Random Forest to then show predicted input and outputs. The model is then saved to a SAV file.

Predictor.py

This file is a barebones program that takes in the model via the previously saved SAV and asks the user to give the value of the respective inputs. From here it runs the model and gives out the prediction on the potential residential buildings heating and cooling load.

Presentation

Team