10/15, 13:00–13:30 (Asia/Tokyo), pyconjp_3
言語: English
Leadership in Energy and Environment Design (LEED) rating system claims to combat the increasing carbon footprint of buildings and certifies the ones with superior energy consumption performance. This talk presents a python led data-driven analysis to determine whether LEED certified buildings have reduced carbon footprint as claimed and test the validity of the LEED rating system.
Description:
Developed by the US Green Building Council (USGBC), LEED is universally the most celebrated rating system. In recent years engineers and project owners are increasingly inclined towards obtaining this coveted LEED certificate because of the alleged “green” and “eco-conscious” connotations that comes with a high LEED score. However, a high LEED score is not enough to convince many engineers and researchers that the building will be as environment-friendly as LEED claims it to be. In this presentation I will outline how machine learning is used using python scikit-learn, beautiful soup, pandas, follium, matplotlib, seaborn packages to check whether LEED score can be a good predictor of a building’s energy efficiency. This analysis will help engineers to make the better choice regarding the carbon footprint reduction of construction industry which holds the utmost importance right now.
Program:
o Self introduction (1 minute)
o Overview of the session (1 minute)
o Motivation (1 minute): A significant number of professionals and academics are not convinced by the efficacy of LEED score or whether it adds any value to construction industry. To test this hypothesis, a systematic analysis of large number of building data is required.
o Objectives (2 minute)
o Dataset introduction and extraction using python beautiful soup (4 minutes)
o Data processing (6 minutes)
o Use pandas to create and filter LEED building dataframe (2 mins)
o Augmentation with LEED score breakdown using custom web scrapper (2mins)
o Handling missing values using pandas (1 min)
o Feature selection and outlier removal using numpy (1 min)
o Early data analysis and visualization (6 mins):
o Mapping location of LEED certified building based on their zip code and geofile using folium, json and web-browser (2 minute)
o Visualizing and comparing the average points earned by LEED buildings for each LEED scorecard category using matplotlib (1 minute)
o Visualizing the correlation between LEED score and building energy attributes using seaborn. (3 mins)
o Analyze the impact of LEED score as an attribute to predict energy use intensity (EUI) for LEED building dataset (2 mins)
o Algorithm selection (Multilinear Regression, Support Vector Machine, Random Forest, using scikit learn and Artificial Neural Network implemented using Tensorflow) (1 minute)
o Result comparison using mean squared error (MSE), mean absolute error (MAE) and R-squared. Predict LEED score of non-LEED certified buildings (1 minute)
o Train a predictive model to predict EUI when training set is the complete building dataset (LEED + NON-LEED) to analyze the impact of LEED score on the predictive performance (1 minute)
o Result discussion and summary (3 minutes)
o Is LEED score a good predictor for building energy performance (1 minute)
o Which learning algorithm performs well for the prediction model and why? (1 minute)
o Limitations of open data accuracy and recommendations (1 minute)
Tabassum Mushtary Meem is a Masters student at the Department of Construction Engineering at École de technologie supérieure (ÉTS), Montreal, QC, Canada. Her research focus is on efficiency and sustainability in construction process and automation in building information modelling (BIM).
Meem is a strong believer in the ability of sustainable design and research practices combined with critical thinking for strengthening our communities and having a positive effect on the ecological balance of the world. She is from Bangladesh and is working passionately to encourage women from under developed and developing countries to pursue a career in engineering.