Other Research Projects
Research Title: Baseline Model Development for Commercial Office Buildings in the Tropics
The baseline model is an essential prediction tool for determining energy savings by comparing pre-retrofit and after retrofit energy use of building facilities. The main objectives of this study are to develop a holistic baseline model for whole building energy consumption and new methodologies for baselining landlord energy consumption in the tropics.
Twelve commercial buildings in Singapore are involved in this research. The widely used single variant and multiple linear regression analysis methods are evaluated on the whole building energy consumption level. The results show that weather conditions may account for more than 85% of the changes in whole building energy consumption. In addition, the Neural Networks method is applied to the prediction of landlord energy consumption and is proved to have a good prediction on the annual basis. However, its tracking performance on the monthly basis is not as good. A new method, called Support Vector Machines (SVMs), is developed for better baseline models of landlord energy consumption. The results show that the prediction coefficients of variance (CV) from SVMs are all below 3% and the mean absolute errors (MAE) are below 4%. The results play a significant role in improving energy use prediction of energy retrofit projects, and in particular, the promotion of performance contracting services in the industry.
Output
Figure 1 Results of Simple Linear Regression Model
Building |
Annual Energy Use (kWh/m 2 /month) |
MAE (%) |
CV |
VAR(Model)
(kWh/m 2 /month) |
90%PI (%) |
|
Measured |
Modeled |
|
|
|
|
A |
20.26 |
20.59 |
1.64 |
3.80 |
1.35 |
10.06 |
B |
32.87 |
32.09 |
2.42 |
2.78 |
2.60 |
2.9 |
C |
28.59 |
24.22 |
15.33 |
3.49 |
29.79 |
48.15 |
D |
25.55 |
25.76 |
0.8 |
3.76 |
2.05 |
2.59 |
E |
18.81 |
18.34 |
2.5 |
3.32 |
2.83 |
3.04 |
Figure 2 Results of Multivariate Regression Model
Building Name |
Baseline Year |
Prediction Year |
MAE(%) |
CV(%) |
VAR |
PI |
|
|
SR |
MLR |
SR |
MLR |
SR |
MLR |
SR |
MLR |
SR |
MLR |
F |
23.52 |
24.11 |
24.14 |
2.51 |
2.63 |
7.11 |
5.18 |
3.96 |
2.05 |
4.15 |
3.30 |
G |
30.31 |
28.63 |
30.35 |
-5.53 |
0.14 |
8.26 |
3.97 |
8.90 |
2.05 |
5.72 |
3.72 |
H |
22.86 |
23.66 |
23.66 |
7.44 |
3.49 |
12.63 |
11.37 |
13.44 |
9.71 |
6.73 |
5.90 |
Figure 3 Results of Neural Networks and Support Vector Machines
Building
Ref. No. |
Actual Value
(kWh/month/m 2) |
Neuron Numbers in The Hidden Layer/Predicted Consumption |
8 |
9 |
10 |
MSE |
Prediction
CV (%)
|
MAE (%) |
G |
10.55 |
10.39 |
|
|
0.78 |
9.67 |
-1.5 |
H |
11.05 |
|
|
11.12 |
3.83 |
15.5 |
0.64 |
I |
9.54 |
|
|
10.23 |
1.58 |
15.12 |
7.33 |
J |
12.59 |
|
11.82 |
|
2.38 |
14.17 |
-6.2 |
Figure 4 Results of Support Vector Machines
Benefits of the Research Works for Commercial and Industrial Use
- Develop a new method for baselining landlord energy consumption
- Provide a standard utility bill analysis method for the commercial buildings in Singapore
- Provoke the blooming of the energy services companies in the tropical region
- Provoke the development of energy performance contracting in Singapore
^Back to top
|
