TY - GEN
T1 - Deep Learning Networks for Vectorized Energy Load Forecasting
AU - Jaskie, Kristen
AU - Smith, Dominique
AU - Spanias, Andreas
N1 - Funding Information:
AKNOWLEDGMENTS This work was supported in part by the NSF Grant No. CNS 1659871 and NSF CPS REU supplement 1646542. The work was also supported in part by the ASU SenSIP Center.
Publisher Copyright:
© 2020 IEEE.
PY - 2020/7/15
Y1 - 2020/7/15
N2 - Smart energy meters allow individual residential, commercial, and industrial energy load usage to be monitored continuously with high granularity. Accurate short-term energy forecasting is essential for improving energy efficiency, reducing blackouts, and enabling smart grid control and analytics. In this paper, we survey commonly used non-linear deep learning timeseries forecasting methods for this task including long short-term memory recurrent neural networks and nonlinear autoregressive models, nonlinear autoregressive exogenous networks that also include weather data, and for completeness, MATLAB's nonlinear input-output model that only uses weather. These models look at every combination of load sequence data and weather information to identify which factors and methods are most effective at predicting short-term residential load. In this paper, the traditional nonlinear autoregressive model predicted short term load values most accurately using only energy load information with a mean square error of 7.53E-5 and a correlation coefficient of 0.995.
AB - Smart energy meters allow individual residential, commercial, and industrial energy load usage to be monitored continuously with high granularity. Accurate short-term energy forecasting is essential for improving energy efficiency, reducing blackouts, and enabling smart grid control and analytics. In this paper, we survey commonly used non-linear deep learning timeseries forecasting methods for this task including long short-term memory recurrent neural networks and nonlinear autoregressive models, nonlinear autoregressive exogenous networks that also include weather data, and for completeness, MATLAB's nonlinear input-output model that only uses weather. These models look at every combination of load sequence data and weather information to identify which factors and methods are most effective at predicting short-term residential load. In this paper, the traditional nonlinear autoregressive model predicted short term load values most accurately using only energy load information with a mean square error of 7.53E-5 and a correlation coefficient of 0.995.
KW - LSTM
KW - Load Forecasting
KW - Machine Learning
KW - NARX
KW - Neural Networks
KW - Smart Grid
UR - http://www.scopus.com/inward/record.url?scp=85096198759&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85096198759&partnerID=8YFLogxK
U2 - 10.1109/IISA50023.2020.9284364
DO - 10.1109/IISA50023.2020.9284364
M3 - Conference contribution
AN - SCOPUS:85096198759
T3 - 11th International Conference on Information, Intelligence, Systems and Applications, IISA 2020
BT - 11th International Conference on Information, Intelligence, Systems and Applications, IISA 2020
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 11th International Conference on Information, Intelligence, Systems and Applications, IISA 2020
Y2 - 15 July 2020 through 17 July 2020
ER -