Dilated Convolution LSTM for Time Series Forecasting in the Field of Energy Consumption

The forecast models of energy consumption cover a lot of ground, such as energy storage, abnormal detection, energy management system, and control applications. There are still many problems, especially the approaches of extracting global spatial-temporal correlation, to be solved for energy time se...

Full description

Saved in:
Bibliographic Details
Published in2025 5th International Conference on Neural Networks, Information and Communication Engineering (NNICE) pp. 1805 - 1811
Main Authors Xu, Liyang, Lv, Haowen, Wang, Dezheng
Format Conference Proceeding
LanguageEnglish
Published IEEE 10.01.2025
Subjects
Convolution
Correlation
Dilated Convolution
Energy consumption
Energy management systems
Energy storage
Forecasting
Long short term memory
LSTM
Predictive models
Robustness
Spatial-Temporal
Time series analysis
Online AccessGet full text

Cover

More Information
Summary:The forecast models of energy consumption cover a lot of ground, such as energy storage, abnormal detection, energy management system, and control applications. There are still many problems, especially the approaches of extracting global spatial-temporal correlation, to be solved for energy time series forecasting in the subject of numerous research studies. To address this limitation, a novel prediction model named dilated convolution LSTM (DC-LSTM) is proposed based on LSTM in this paper. DC-LSTM contains two main components, including LSTM and dilated convolution (DC). LSTM aims to capture the temporal relationship among input data. Meanwhile, dilated convolution is adopted to discover spatial dependency between variables with emphasis on extracting global spatial-temporal correlation. In evaluating various benchmark datasets, DC-LSTM significantly outperformed baseline methods, surpassing them by an average of 1.08% in RMSE and 8.00% in modified MAPE.
DOI:10.1109/NNICE64954.2025.11063713