Abstract: When using an ARMA model to predict monitoring data of bridges, difference processing causes some data lose. To improve prediction accuracy, this paper makes use of the advantages of wavelet analysis, i.e., no information is lost after wavelet transform. The time series with a clear trend are divided into two parts. The low-frequency part representing strain trend is modeled using GM(1,1), and the high-frequency part representing random disturbance using ARMA. The predicted value is then obtained by combining the two parts. Validation is made with the strain data acquired from an on-line monitoring system on a Yellow River suspension bridge. The results show that prediction accuracy of the combined GM(1,1)-ARMA model is higher than the traditional ARMA. The method is applicable to early warning of similar bridges.

Key words: suspension pipeline, strain prediction, wavelet transform, GM(1,1)-ARMA