TY - GEN
T1 - Optimization based data mining approach for forecasting real-time energy demand
AU - Omitaomu, Olu Femi A.
AU - Li, Xueping
AU - Zhou, Shengchao
PY - 2015
Y1 - 2015
N2 - The worldwide concern over environmental degradation, increasing pressure on electric utility companies to meet peak energy demand, and the requirement to avoid purchasing power from the real-time energy market are motivating the utility companies to explore new approaches for forecasting energy demand. Until now, most approaches for forecasting energy demand rely on monthly electrical consumption data. The emergence of smart meters data is changing the data space for electric utility companies, and creating opportunities for utility companies to collect and analyze energy consumption data at a much finer temporal resolution (e.g., 15-minutes interval). While the data granularity provided by smart meters is important, there are still other challenges in forecasting energy demand; these challenges include lack of information about appliances usage and occupants behavior. Consequently, in this paper, we develop an optimization based data mining approach for forecasting real-time energy demand using smart meters data. The objective of our approach is to develop a robust estimation of energy demand without access to these other building and behavior data. Specifically, the forecasting problem is formulated as a quadratic programming problem and solved using the so-called support vector machine (SVM) technique in an online setting. The parameters of the SVM technique are optimized using simulated annealing approach. The proposed approach is applied to hourly smart meters data for several residential customers over several days.
AB - The worldwide concern over environmental degradation, increasing pressure on electric utility companies to meet peak energy demand, and the requirement to avoid purchasing power from the real-time energy market are motivating the utility companies to explore new approaches for forecasting energy demand. Until now, most approaches for forecasting energy demand rely on monthly electrical consumption data. The emergence of smart meters data is changing the data space for electric utility companies, and creating opportunities for utility companies to collect and analyze energy consumption data at a much finer temporal resolution (e.g., 15-minutes interval). While the data granularity provided by smart meters is important, there are still other challenges in forecasting energy demand; these challenges include lack of information about appliances usage and occupants behavior. Consequently, in this paper, we develop an optimization based data mining approach for forecasting real-time energy demand using smart meters data. The objective of our approach is to develop a robust estimation of energy demand without access to these other building and behavior data. Specifically, the forecasting problem is formulated as a quadratic programming problem and solved using the so-called support vector machine (SVM) technique in an online setting. The parameters of the SVM technique are optimized using simulated annealing approach. The proposed approach is applied to hourly smart meters data for several residential customers over several days.
KW - Data mining
KW - Energy management
KW - Optimization
KW - Process improvement
KW - Simulated annealing
UR - http://www.scopus.com/inward/record.url?scp=84970967336&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:84970967336
T3 - IIE Annual Conference and Expo 2015
SP - 536
EP - 545
BT - IIE Annual Conference and Expo 2015
PB - Institute of Industrial Engineers
T2 - IIE Annual Conference and Expo 2015
Y2 - 30 May 2015 through 2 June 2015
ER -