Research on temperature prediction model for key components in highspeed rail electrical cabinets based on LSTM.

Abstract

The temperature status of key components in high-speed rail electrical cabinets directly impacts the safety and reliability of train operations. To achieve accurate temperature prediction and early warning, this paper proposes a temperature prediction model based on Long Short-Term Memory (LSTM) networks. Addressing the current research gap and the limitations of traditional models in capturing long-term temporal dependencies in temperature data, a multivariate dataset was first constructed by collecting time-series temperature data of key components along with relevant environmental and operational parameters. Subsequently, comparative analyses were conducted by establishing RBF neural network, ARIMA, Prophet, and LSTM prediction models, systematically investigating the impact of LSTM network architecture parameters and time window settings on prediction accuracy. Experimental results demonstrate that the LSTM model significantly outperforms the comparison models in both prediction accuracy and stability, achieving superior performance in key metrics such as RMSE and MAE. The model effectively captures long-term trends and short-term fluctuations in temperature variations, providing reliable technical support for condition monitoring and fault warning of key components.