Gauge Deterioration Prediction of Urban Rail Transit Lines Based on CEEMD and SVR

Abstract

［Objective］ In order to strengthen the status ma-nagement of urban rail transit line sections, it is necessary to predict the overall deterioration trend of the gauge in space. ［Method］ CEEMD (complementary ensemble empirical mode decomposition) theory is introduced to extract the IMF (intrinsic mode function) of the geometric alignment of the track section. The PSO (particle swarm optimization) algorithm is utilized to optimize the SVR (support vector regression machine) to train and test the extracted data after calibrating the optimal parameters of the prediction model. Thus, the CEEMD-PSO-SVR prediction model is constructed. The prediction model is tested with 1,128 sets of track inspection sample data within the upward track section from K12+134 to K15+743 on Shanghai Metro Line 16. ［Result & Conclusion］ Compared with the PSO-SVR model and the ARIMA (autoregressive integrated moving average) model, the CEEMD-PSO-SVR prediction model has advantages in three performance evaluation indicators, namely root mean square error, mean absolute error, and absolute value of mean relative error.

Keywords

• urban rail transit line
• gauge deterioration
• ceemd
• svr