TY - GEN
T1 - Forecasting Shor-term Electricity Load of East Java Province using Functional Regression Time-varying Coefficient Model
AU - Sinurat, Paskalis
AU - Akbar, Muhammad Sjahid
AU - Irhamah,
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Electricity is a primary necessity underpinning economic, social, and industrial activities, with East Java as one of Indonesia's major contributors to national electricity demand. The high-frequency and non-storable characteristics of electricity load create significant challenges for short-term forecasting. This study proposes a Functional Regression Time-Varying Coefficient Model (FR-TVCM) within a Functional Data Analysis (FDA) framework, modeling daily load curves as continuous functional observations to capture intraday temporal dynamics more effectively than conventional discrete time-series approaches. Regression coefficients evolve smoothly over intraday time and calendar effects are incorporated through a national holiday dummy variable, which substantially improves forecast accuracy over the lag-only baseline. The proposed FR-TVCM achieves superior forecast accuracy compared to DSARIMA, with particularly pronounced advantages on national holiday dates. Residual diagnostics based on FPCA confirm adequate model fit for the baseline specification, while the holiday-extended model introduces additional complexity warranting further investigation. Beyond prediction, FR-TVCM serves as an interpretable analytical tool for understanding electricity load dynamics as continuously evolving curves, with broad applicability to any power system characterized by high-frequency temporal dynamics and multiple overlapping seasonal structures.
AB - Electricity is a primary necessity underpinning economic, social, and industrial activities, with East Java as one of Indonesia's major contributors to national electricity demand. The high-frequency and non-storable characteristics of electricity load create significant challenges for short-term forecasting. This study proposes a Functional Regression Time-Varying Coefficient Model (FR-TVCM) within a Functional Data Analysis (FDA) framework, modeling daily load curves as continuous functional observations to capture intraday temporal dynamics more effectively than conventional discrete time-series approaches. Regression coefficients evolve smoothly over intraday time and calendar effects are incorporated through a national holiday dummy variable, which substantially improves forecast accuracy over the lag-only baseline. The proposed FR-TVCM achieves superior forecast accuracy compared to DSARIMA, with particularly pronounced advantages on national holiday dates. Residual diagnostics based on FPCA confirm adequate model fit for the baseline specification, while the holiday-extended model introduces additional complexity warranting further investigation. Beyond prediction, FR-TVCM serves as an interpretable analytical tool for understanding electricity load dynamics as continuously evolving curves, with broad applicability to any power system characterized by high-frequency temporal dynamics and multiple overlapping seasonal structures.
KW - Electricity Load
KW - Forecasting
KW - Functional Data Analysis
KW - Functional Regression
KW - Time Varying Coefficient Model
UR - https://www.scopus.com/pages/publications/105047916821
U2 - 10.1109/ICADEIS71120.2026.11644543
DO - 10.1109/ICADEIS71120.2026.11644543
M3 - Conference contribution
AN - SCOPUS:105047916821
T3 - 2026 International Conference on Advancement in Data Science, E-Learning and Information System, ICADEIS 2026
BT - 2026 International Conference on Advancement in Data Science, E-Learning and Information System, ICADEIS 2026
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2026 International Conference on Advancement in Data Science, E-Learning and Information System, ICADEIS 2026
Y2 - 23 June 2026 through 24 June 2026
ER -