我所殷辰堃老师课题组完成的论文《Interpretable Forecasting of Traction Energy Consumption Based on Nbeats and Temporal Fusion Transformers》于2024-05在《IEEE Data Driven Control and Learning Systems Conference (DDCLS)》正式发表。论文摘要如下：
Abstract: The traction energy consumption accounts for about 50% of the total energy consumption in urban rail transit. Accurately predicting traction energy consumption is an important basis for analyzing energy consumption fluctuations and detecting energy consumption anomalies. However, the high instability of energy consumption caused by complex factors poses challenges to energy consumption prediction. In this paper, a combined approach of Neural basis expansion analysis for interpretable time series (Nbeats) decomposition model and Temporal Fusion Transformers (TFT) prediction model is proposed for energy consumption prediction. Firstly, the Nbeats model decomposes the energy consumption sequence into interpretable trend and seasonality components, which can reveal the impact mechanism of exogenous factors. These components are input as time-varying variables for the TFT prediction model. By integrating the Nbeats model into the TFT prediction model, the accuracy and interpretation of energy consumption prediction are improved. Experiment results show that the Nbeats-TFT model achieves a root mean square error (RMSE) of 0.06. Moreover, the Nbeats-TFT model can generate interpretable outputs and quantify the importance of different external factors, as well as time steps in the prediction. It provides a basis for optimizing energy consumption management in engineering sites.
引用信息：J. Zhihui, Jing Xun, Chenkun Yin, Zixing Guo, Guiying Tian. Interpretable Forecasting of Traction Energy Consumption Based on Nbeats and Temporal Fusion Transformers. IEEE Data Driven Control and Learning Systems Conference (DDCLS), 2024-05. DOI: 10.1109/ddcls61622.2024.10606795