Bridge Alignment Prediction Based on Combination of Grey Model and BP Neural Network

Abstract

The continuous improvement of bridge construction technology has resulted in an ongoing expansion of bridge spans, which has concomitantly increased the difficulty of controlling the alignment of long-span bridges during construction. In order to address the issue of the grey prediction model exhibiting a significant discrepancy in its alignment predictions for long-span continuous girder bridges, a pre-camber prediction method for bridges based on a combination of the grey model (GM) and BP neural network (GM-BP) is proposed. Firstly, the parameters are identified according to their influence on the pre-camber, and the appropriate parameters are selected as the original data to improve the efficiency of prediction. Subsequently, the original data are preliminarily fitted by the GM(1,1) model, and the predicted values are used as inputs for training the neural network. Finally, the new predicted values are output using the nonlinear fitting ability of the BP neural network. To assess the efficacy of the model, it is applied to the prediction of the pre-camber of the girder segments of a bridge under cantilever construction. The pre-camber prediction for 11#–13# girder sections was based on 10 sets of monitoring data from constructed girder sections. The results demonstrated that the average relative error of the GM-BP combined prediction model was 3.01%, which was 5.68% less than that of the GM(1,1) model, and the overall prediction exhibited a closer alignment with the original data. The GM-BP combined prediction model is an effective method for ensuring the alignment control of bridge construction and is able to achieve high accuracy and stability in its predictions in the case of limited and irregular data.