A map of the weather normalized EUI for multifamily housing (Fig.1) and office buildings (Fig.2) were given to give a spatial representation of the data. Fig 4. shows the most numerous building types contained in the dataset. Multifamily housing consists of over 88% of the dataset. Fig 9, 10, 11 also show the relationship between building age and EUI. As the age of the building increases the EUI decreases.
After completing some descriptive statistics, regression models were made to analyze the features the variables affecting energy performance. Two models were made to consider energy performance based on building type. One model was made for multifamily housing and another for office buildings. Other building types were not considered due to having a low amount of samples in the overall dataset. Many features were selected based off their significance in work by Kontokosta (2012). Both models contained most of the same variables which consisted of Building Age, number of floors, total floor area, floor aspect ratio, lot coverage, and primary source. The Multifamily housing model also considered the total number of units in the building. Following is a description of the variables.
- Property GFA - The total building area self reported
- Building Age - Age of the building
- number of Floors - Total number of floors in the building
- FAR - floor aspect ratio
- Lot Coverage - Floor area * Floor Depth / Lot area
- Electricity Source -A binary variable representing if electricity made up more than 50% of energy consumption. Otherwise it was natural gas.
- Units Total - number of Units in the building
Results
Both models provided an R2 of over .86 showing that the models were strong measurements of EUI.All features were significant to the model except for lot coverage which had a pvalue of .07. This feature also had a coefficient of .-0004 per 1 unit increase with a median of 452 it had little impact on the model. The coefficients of Number of Floors, Building Age, FAR, and Primary Energy Source had a very high impact on the the models. Though the effect sizes of these values are large we have to be careful of collinearity. Number of Floors and FAR had a collinearity of nearly .6 which could make it hard to determine each feature's actual effect size. This is also true for number of floors and total units. However these values may still be low enough that we can say these values are important features for determining the change in energy performance.
Discussion and Conclusion