TY - CONF
T1 - A novel framework for autoregressive features selection and stacked ensemble learning for aggregated electricity demand prediction of neighborhoods
AU - Khan, Waqas
AU - Walker, Shalika S.W.
AU - Katic, Katarina
AU - Zeiler, Wim
PY - 2020/9/22
Y1 - 2020/9/22
N2 - Demand forecast plays an important role in the power industry, as it sets the basis for decision making in power system operation and planning. Electricity consumption forecasting of individual buildings has been widely used for energy management, planning and energy-saving potential identification in the past decade. Yet, insignificant focus has been put on aggregated demand forecast of neighborhoods. In the context of the future smart grid, short- and long-term demand forecast on a neighborhood level will be an essential task for utility providers to better plan generation and solve congestion problems of the distribution network. Based on a comprehensive literature study, an ensemble learning method is proposed for predicting short- and long-term electricity demand of a campus located in the Netherlands. The ensemble model performed better in demand forecasting of neighborhoods compared to individual models for an hour ahead, day ahead and year ahead with R 2 values of 0.988, 0.951 and 0.943 respectively. Assessing the demand for cluster of buildings with distinct boundaries such as hospitals and campuses at an aggregated level would reduce the amount of data needed to be stored. The proposed technique contributes to short (single step) and long term (multi step) energy self-sufficiency planning and energy balancing systems on a neighborhood scale.
AB - Demand forecast plays an important role in the power industry, as it sets the basis for decision making in power system operation and planning. Electricity consumption forecasting of individual buildings has been widely used for energy management, planning and energy-saving potential identification in the past decade. Yet, insignificant focus has been put on aggregated demand forecast of neighborhoods. In the context of the future smart grid, short- and long-term demand forecast on a neighborhood level will be an essential task for utility providers to better plan generation and solve congestion problems of the distribution network. Based on a comprehensive literature study, an ensemble learning method is proposed for predicting short- and long-term electricity demand of a campus located in the Netherlands. The ensemble model performed better in demand forecasting of neighborhoods compared to individual models for an hour ahead, day ahead and year ahead with R 2 values of 0.988, 0.951 and 0.943 respectively. Assessing the demand for cluster of buildings with distinct boundaries such as hospitals and campuses at an aggregated level would reduce the amount of data needed to be stored. The proposed technique contributes to short (single step) and long term (multi step) energy self-sufficiency planning and energy balancing systems on a neighborhood scale.
KW - Energy consumption
KW - Predictive modeling
KW - Ensemble models
KW - Autoregressive features
KW - Aggregated demand prediction
KW - Autoregressive feature
KW - Energy demand
KW - Energy balancing
KW - Ensemble learning
UR - http://www.scopus.com/inward/record.url?scp=85093690034&partnerID=8YFLogxK
U2 - 10.1109/SEST48500.2020.9203507
DO - 10.1109/SEST48500.2020.9203507
M3 - Paper
ER -