Data driven approach in analyzing energy consumption data in buildings Office of Environmental Sustainability Ian Tan
Background Real time energy consumption data of buildings in terms of electricity (kwh) and cooling energy (ton.hr) are being collected by the University Campus Infrastructure cluster. Purposes of capturing these information are monitor consumption, internal charging, detect faulty equipment, promote energy conservation etc. Three common ways an organization can save building energy consumption: 1. Architectural design strategy 2. Energy efficient devices used in the building 3. Optimizing the building management and operation mode.
Business Challenges Energy use in the campus is significant, in terms of tens of millions of dollars per annum. Managing this cost is essential. Difficult for a layman to appreciate the energy use in their building if there is no standard energy profile to benchmark. Time consuming and challenging to analyze the energy consumption in the campus as the buildings are heterogeneous in kind.
Business Objective The key business objective of this investigation is to minimize energy waste at the lowest cost. Measures being taken to achieve the objective: A targeted approach to identify buildings that has high energy saving opportunities. Develop a standard energy use profile for a building, that serves as a benchmark for the occupants (layman) of a building to relate to their daily energy use. Enhance current measures in handling missing and erroneous data that slows the analysis and effective communication of the energy use in a building to its occupants.
Data driven analysis adopted Deployment, Audit Evaluate Business Understanding Drill down analysis Predictive modeling Exploratory data analysis Segmentation of the buildings Benchmark
Proposed Approach
Segmentation of buildings in campus
Benchmark of energy use in cluster A1 Residential related buildings Studies should be done to compare the design and operations of the buildings C,E,H and B with building I. If the energy intensity at buildings C,E and H can be lowered to match building I, the monetary savings opportunities per year is estimated to be $84K per year (upper limit) for building C. Correspondingly, the potential energy savings (upper limit) can also be realized for buildings E & H at $41K and $60K respectively. If we are to compare building B and I, a potential energy saving of $22K may be materialized if the energy intensity of building B can be lowered towards energy intensity of building I.
Typical energy profile of buildings in campus 200 150 100 50 Typical energy profile of an administrative building 0 160 140 120 100 80 60 40 20 0 Typical energy profile of a residential building 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 Kw KW Typical energy profile of a research mixed use related building 600 500 400 300 200 100 0 0:00 0:50 1:40 2:30 3:20 4:10 5:00 5:50 6:40 7:30 8:20 9:10 10:00 10:50 11:40 12:30 13:20 14:10 15:00 15:50 16:40 17:30 18:20 19:10 20:00 20:50 21:40 22:30 23:20 Kw
Drill down investigation on the energy use in buildings C,E, H & I from cluster A1
Drill down investigation on the energy use in buildings C,E, H & I from cluster A1 Building C has an abnormally higher weekend load during the evenings compared to the other buildings. Further investigation is required to understand the reasons for this energy use, but certainly there are opportunities to conserve energy.
Benchmark of energy buildings in cluster A2 Similar in terms of energy intensity range. A type is from the same faculties. B type is office related. C type is residences related E library and mixed used F, G & H & I from respective groups. Investigate why building A2 is much higher than its other counterparts. Bldg H1 has rather high energy intensity given that on Sundays, it is not operational. Bldgs C1 and C2 should appear in sub cluster A1. Also bldg. C2 has higher energy intensity compared to its counterpart. Low weekend operation High weekend operation Scatter Plot of EUI in percentage over the ratio Avg Sun demand over Average total demand for the buildings in Cluster A2 Bldg B1 has higher energy intensity as compared to B2 which is a newer building with a more efficient design. Bldgs E1 & E2 operates on Sunday but not 24 hours.
Energy Profile of a Building B2 Dip Dip weekdays weekends 1 Sun 2-Mon 3-Tue 4-Wed 5-Thu 6-Fri 7-Sat 1 Jan 2-Feb 3-Mar 4-Apr 5-May 6-Jun 7-Jul 8-Aug 9-Sep 10-Oct 11-Nov 12-Dec Should energy analysis consider normalization against outside temperature, for Singapore s tropical weather?
Predictive energy model for building B2 A multiple linear regression was used to create a basic model to predict the daily energy use of this building. Due to limited data, the predictors used were type of day, outside temperature and term period. It is not practical to deploy resources to collect more data to develop a more accurate predictive model. Hence, the model is simple, but adequate enough to help building owners to detect abnormal daily energy use. For the weekends, the model is not appropriate to estimate the energy use on Saturdays and Sundays. A simple average energy use for the particular month can be used to benchmark the typical energy use during the weekends.
Multiple Regression Model Model Summary b Coefficients a Adjusted R Std. Error of the Model R R Square Square Estimate 1.734 a.538.529 104.93 a. Predictors: (Constant), Avg Temperature, Tue, Fri, Wed, Thu ANOVA a Model 1 (Constant) Unstandardized Coefficients Standardized Coefficients B Std. Error Beta t Sig. -495.55 161.36-3.07.002 Model Sum of Squares df Mean Square F Sig. 1 Regressi on 3129367 5 625873 56.842.000 b Residual 2686614 244 11011 Total 5815981 249 a. Dependent Variable: kwh b. Predictors: (Constant), Avg Temperature, Tue, Fri, Wed, Thu Tue -79.78 21.00 -.209-3.80.000 Wed -106.53 21.00 -.279-5.07.000 Thu -124.47 21.02 -.326-5.92.000 Fri -164.06 20.99 -.430-7.81.000 Avg Temperature 83.48 5.82.625 14.34.000 a. Dependent Variable: kwh Predicted Energy = -495.55 + 83.5x(Temp) - 79.8x(Tue) 106.5x(Wed) 124.5x(Thu) 164x(Fri) Temp average outdoor daily temperature in celsius Mon is treated as the reference day Tue, Wed, Thu & Fri are binary inputs of 1 being that particular day and 0 when it isn t that day.
Comparison of predicted and actual daily energy consumption at building B2 2500 kwh 2000 1500 1000 The difference between the actual value and predicted value ranges between -7% to 7%, for at least 90% of the dataset. 500 0 1-Jan-14 2-Mar-14 1-May-14 30-Jun-14 29-Aug-14 28-Oct-14 27-Dec-14 Actual Predicted
Validation: Percentile of the % residuals Percentiles Percentiles Weighted Average(D efinition 1) Tukey's Hinges difference difference 5 10 25 50 75 90 95-7.58-5.08-2.38 0.29 3.07 5.15 6.76-2.37 0.29 3.07 From the table, we can formulate a business rule that any future actual value that is more than +7% or -7% of the predicted value, needs to be flagged out for investigation. The predictive model computed values can also be used as a benchmark to compare energy used in the future. The common approach of comparing daily energy use on a year to year basis, does not consider the difference in average outdoor temperature.
Next steps Extend the investigation on the energy building profile to other buildings in the cluster. Investigate and explore the differences in the energy building profiles for heterogeneous buildings. Future application Incorporate predictive model into the existing energy monitoring system for the respective buildings as an alert system, and data imputation for missing values. Targeted audit by benchmarking