Prediction of Metro Train Air-conditioning Return Air Temperature Based on Real-time Field-measured Data

Abstract

［Objective］To enable proactive adjustment of the cooling capacity for metro train AC (air-conditioning) system based on ambient temperature, it is essential to study the prediction of return air temperature in metro train AC systems. ［Method］A time series forecasting method is employed to predict the variation trend in return air temperature of metro train AC systems. Real-time operational data from the air-conditioning system of a Guangzhou Metro train are collected through AC sensors. Outliers are removed using a boxplot, and a sliding window approach is applied to handle the time span of input and output data. An LSTM (long short-term memory) neural network model is then constructed to predict the return air temperature of the AC units, and the impact of different sample sizes on the model prediction accuracy is comparatively analyzed. ［Result & Conclusion］The LSTM neural network model can learn the temperature control logic of metro train AC system, with the predicted temperature curve closely matching the actual temperature curve, therefore is suitable for predicting the return air temperature of metro AC units. Through parameter optimization, the model accuracy is improved to 0.84, enabling precise prediction of unit return air temperature. Increasing the sample size for a single training session can shorten model training time but may reduce the model final prediction accuracy to some extent correspondently.

Keywords

• metro
• air-conditioning system
• long short-term memory