Short-term bus load forecasting considering interval forecasting information and temporal convolutional network

Authors

• Xinming Liu, Faculty of Electrical and Control Engineering ， Liaoning Technical University ， Huludao 125105 ， China
• Haichen Li, Faculty of Electrical and Control Engineering ， Liaoning Technical University ， Huludao 125105 ， ChinaFollow
• Jianguang Ji, Faculty of Electrical and Control Engineering ， Liaoning Technical University ， Huludao 125105 ， China
• Shuo Sun, Faculty of Electrical and Control Engineering ， Liaoning Technical University ， Huludao 125105 ， China
• Junxin Bian, Faculty of Electrical and Control Engineering ， Liaoning Technical University ， Huludao 125105 ， China

Keywords

bus load forecasting, temporal convolutional network, mode decomposition, interval forecasting

Abstract

The bus load forecasting in substations is essential for the stable operation of power systems.With the large-scale integration of renewable energy generation into the grid，low-voltage bus loads become increasingly volatile，and their uncertainty can hardly be quantified by point forecasting alone.Interval forecasting can enhance decision-making in power dispatching.For this reason，the study proposes a hybrid short-term bus load forecasting method that integrates point and interval forecasting.Specifically，complementary ensemble empirical mode decomposition with adaptive noise （CEEMDAN ） is applied for primary decomposition of the original bus load.The features of the highly volatile high-frequency components from primary decomposition are extracted through secondary decomposition of the high-frequency components using variational mode decomposition （VMD ）.Next，the sample entropies （SEs） of the subcomponents are calculated to assess their complexity and reconstruct them into three groups of representative components.These representative components are then input into an improved sparrow search algorithm and temporal convolutional network （ISSA-TCN ） model，which incorporates Tent chaotic mapping，adaptive weights，Levy flights，and variable spiral search strategies，to enhance point forecasting accuracy.Finally，quantile transformation （QT） is applied to smooth the abnormally distributed errors in kernel density estimation （KDE ），and confidence interval forecasting is performed to obtain probabilistic intervals.Comparative experiments with real bus load data show that compared with both single and hybrid models，the proposed point forecasting method reduces the mean absolute error to 3.52% and improves the fitting degree to 94.62%.Compared with parametric and nonparametric methods，the proposed interval forecasting method lowers the mean interval score （MIS） to 4.2 at 95% confidence，improves the prediction interval coverage probability （PICP ） to 98%，and covers the largest number of actual values.The proposed method effectively quantifies the uncertainty of point forecasting，comprehensively reflects the variation patterns of bus loads，and facilitates the operation and management of power systems.

DOI

10.19781/j.issn.1673-9140.2026.04.009

First Page

98

Last Page

113

Recommended Citation

Liu, Xinming; Li, Haichen; Ji, Jianguang; Sun, Shuo; and Bian, Junxin (2026) "Short-term bus load forecasting considering interval forecasting information and temporal convolutional network," Journal of Electric Power Science and Technology: Vol. 41: Iss. 4, Article 9.
DOI: 10.19781/j.issn.1673-9140.2026.04.009
Available at: https://jepst.researchcommons.org/journal/vol41/iss4/9