Sumedh Bhattacharya

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2017-45

May 10, 2017

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2017/EECS-2017-45.pdf

This dissertation is two-fold: the first, and the primary focus of this project, is a study of different available techniques to perform forecasting of the power usage by the HVAC system in Sutardja Dai Hall, while the second consists of a guide to using these predictions for understanding the effects of applying energy efficient control to the building. We begin with an extensive study of different available forecasting techniques including Gaussian Mixture Models, K-Nearest Neighbors, and Recurrent Neural Networks, and a comparison of their relative accuracy in forecasting the HVAC system’s power consumption. We also propose a new Hybrid Model which takes into account both time series forecasting results from past building data and weather based predictions from publicly available weather forecasting data. A high degree of accuracy is necessary as these predictions will be used to estimate the building’s consumption in the absence of control. We then apply a data-driven energy efficient control model to the building to experimentally calculate savings, which had only been measured using simulations before. As such, a low error margin is required to avoid bias when calculating the deviation caused by the applied control. To this end, we see that a hybrid model using both weather and time series data achieves the highest accuracy due to its ability to model both daily and seasonal trends. Subsequently, we use these predictions to showcase an example of how to apply the results of these predictive techniques.

Advisors: Claire Tomlin

---

BibTeX citation:

@mastersthesis{Bhattacharya:EECS-2017-45,
    Author= {Bhattacharya, Sumedh},
    Title= {Baseline Building Power Estimation},
    School= {EECS Department, University of California, Berkeley},
    Year= {2017},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2017/EECS-2017-45.html},
    Number= {UCB/EECS-2017-45},
    Abstract= {This dissertation is two-fold: the first, and the primary focus of this project, is a study of different available techniques to perform forecasting of the power usage by the HVAC system in Sutardja Dai Hall, while the second consists of a guide to using these predictions for understanding the effects of applying energy efficient control to the building. We begin with an extensive study of different available forecasting techniques including Gaussian Mixture Models, K-Nearest Neighbors, and Recurrent Neural Networks, and a comparison of their relative accuracy in forecasting the HVAC system’s power consumption. We also propose a new Hybrid Model which takes into account both time series forecasting results from past building data and weather based predictions from publicly available weather forecasting data. A high degree of accuracy is necessary as these predictions will be used to estimate the building’s consumption in the absence of control. We then apply a data-driven energy efficient control model to the building to experimentally calculate savings, which had only been measured using simulations before. As such, a low error margin is required to avoid bias when calculating the deviation caused by the applied control. To this end, we see that a hybrid model using both weather and time series data achieves the highest accuracy due to its ability to model both daily and seasonal trends. Subsequently, we use these predictions to showcase an example of how to apply the results of these predictive techniques.},
}

EndNote citation:

%0 Thesis
%A Bhattacharya, Sumedh 
%T Baseline Building Power Estimation
%I EECS Department, University of California, Berkeley
%D 2017
%8 May 10
%@ UCB/EECS-2017-45
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2017/EECS-2017-45.html
%F Bhattacharya:EECS-2017-45