Document Type

Thesis - Open Access

Award Date

2016

Degree Name

Master of Science (MS)

Department / School

Electrical Engineering and Computer Science

First Advisor

Reinaldo Tonkoski

Second Advisor

Semhar Michael

Keywords

data center, load forcast, mixture model

Abstract

The dependency on cloud computing is increasing day by day. With the boom of data centers, the cost is also increasing, which forces industries to come up with techniques and methodologies to reduce the data center energy use. Load forecasting plays a vital role in both efficient scheduling and operating a data center as a virtual power plant. In this thesis work a stochastic method, based on dependent mixtures is developed to model the data center load and is used for day-ahead forecast. The method is validated using three data sets from National Renewable Energy Laboratory (NREL) and one other data centers. The proposed method proved better than the classical autoregressive, moving-average, as well as the neural network-based forecasting method, and resulted in a reduction of 7.91% mean absolute percentage error (MAPE) for the forecast. A more accurate forecast can improve power scheduling and resource management reducing the variable cost of power generation as well as the overall data center operating cost, which was quantified as a yearly savings of $13,705 for a typical 100 MW coal fired tier-IV data center.

Library of Congress Subject Headings

Data libraries -- Energy consumption.

Data warehousing -- Forecasting.

Cloud computing.

Power resources -- Management.

Description

Includes bibliographical references (pages 66-69)

Format

application/pdf

Number of Pages

80

Publisher

South Dakota State University

Download

Share

COinS
 

Rights Statement

In Copyright