The construction quality of bridges is complex influenced by various factors such as material characteristics, process parameters, and environmental conditions, which have obvious nonlinear relationships. Traditional quality control methods often find it difficult to accurately describe their interactions. Therefore, this study established a construction quality control model based on backpropagation neural network, which integrates raw material information, process parameters, environmental data, and actual quality inspection results throughout the construction process. The model adopts a three-layer feedforward network structure, which can dynamically predict and optimize key quality indicators. Taking an actual cross river continuous beam bridge project as an example, this article compares the effectiveness of three modeling methods: BP neural network, support vector machine, and random forest. The experimental results show that the backpropagation neural network performs better in multiple performance indicators: it has higher prediction accuracy (coefficient of determination reaches 0.92), more stable error distribution (average absolute error is 3.2%), and stronger adaptability in key construction processes such as maintenance. Compared with the support vector machine model, the prediction error of this network has been reduced by 42%.
Zhihui Cai (Thu,) studied this question.