|
City: |
New York, US |
|
Organization: |
Singapore-MIT Alliance
for Research and Technology (SMART) & National University of Singapore
(Department of Industrial and Systems Engineering) |
|
Project Date: |
2015 |
|
Reference: |
Constantine
E. Kontokosta & Christopher Tull (2017). A data-driven predictive model
of city-scale energy use in buildings. Applied Energy Volume 197, 304-317. https://doi.org/10.1016/j.apenergy.2017.04.005 |
|
Problem Description: |
Need for laws to
encourage transparency in energy efficiency markets and support
sustainability and carbon reduction plans. Energy usage is a severe urban
problem due the exacerbating effect of climate change, and it is imperative
to calculate energy usage and in order to make its use more efficient, at building,
district and city level |
|
Technical Solution: |
· Develop an effective data base for buildings based on size and spatial attributes to make an efficient policy, maximizing energy efficiency while reducing carbon ·
Work closely with city workers to collect data
on natural gas usage on zip code and district level |
|
Datasets used: |
·
Local Law 84 data on
actual energy use and building attributes for 20652 buildings ·
NYC Department of
City Planning’s Primary Land Use Tax Output (PLUTO) data on all 1,082,437
properties ·
Zip code energy data
set for each of NYCs 176 zip code areas |
|
Outcome: |
· Using three different statistical learning algorithms and feature selection approaches, the results suggest that the data from the LL84 sample can produce reasonably accurate predictions of energy use across the city at the building scale, validated at the spatial aggregation of the zip code, particularly for electric energy use intensity. · SVM provides the most accurate prediction for estimation of energy use within the sample of LL84 buildings, while OLS results in the lowest MAE when predicting total building energy consumption at the zip code-level for the entire city. · Found that building use, size, and morphology emerge as robust predictors of energy use at the building- and zip code- levels. The nature of the building occupancy, whether office, retail, or other use type, impacts electric and natural gas EUI. Larger buildings are found to be less energy intensive, while taller build- ings, controlling for size, are more intensive. ·
The shape of the building, operationalized as
the surface-to-volume ratio, is also found to influence EUI, although
additional study is warranted. |
|
Issues: |
· The approach is partially constrained by the temporal frequency (annual) of the data, as additional insight into seasonal and diurnal patterns through monthly and daily data could prove valuable. · LL84 data was limited and for only buildings above 50,000 sq. ft · It was difficult to model for natural gas usage ·
Low data privacy and need for data access by
utility providers to actually be able to change the energy landscape |
|
Status of Project: |
COMPLETED |
|
Created by: |
Sanam
Panjwani (sanam.panjwani@mail.utoronto.ca) |