关键词: 轨道工程；跨座式单轨；空调系统；故障数据；时间序列分析模型；故障预测

Abstract: As an important part of urban rail transit, air conditioning system directly affects the comfort of passengers. The prediction of the time series of air conditioning system faults is helpful to reasonably formulate maintenance strategy and parts procurement scheme, so as to control the cost effectively. Aiming at the prediction problem of time series of air conditioning system faults, the seasonal autoregressive integral moving average model (X12-ARIMA model) was constructed by combining with Census X12 seasonal adjustment method, which was based on the analysis of the periodic fluctuation law and change trend of the number of faults. And the BP neural network model was established based on residual series. The predicted value of the improved X12-ARIMA-BP model was obtained by adding the predicted value of the two models, and compared with the predicted values of X12-ARIMA model, BP neural network model and ARIMA-BP variable weight combination. Taking Chongqing Rail Transit Line 3 as an example, four kinds of models were used to simulate and predict the fault number, based on the monthly failure data of air conditioning system for 7 years. The research results show that compared with the actual value, the mean absolute percentage error of the improved X12-ARIMA-BP model is 18.54%, which is 6.38% lower than that of X12-ARIMA model, 11.01% lower than that of BP neural network model and 4.75% lower than that of ARIMA-BP variable weight combination model. Compared with the other three kinds of prediction models, the improved X12-ARIMA-BP model has the best prediction effect.

Key words: track engineering; straddle type monorail; air conditioning system; fault data; time series analysis model; fault prediction