Inverse Analysis of Thermal Parameters of Concrete Box Girder Based on DE-BP Neural Network

In view of temperature cracks in concrete box girders easily occurring during construction, an inverse analysis method based on uniform design theory and differential evolution back propagation (DE-BP) neural network was proposed to accurately obtain the thermal parameters of concrete box girders and ensure the reliability of temperature analysis of concrete box girders. This method established the nonlinear relationship between the temperature peak of characteristic points and the thermal parameters through the DE-BP neural network. By using the uniform design method and the Abaqus finite element numerical model, 130 sets of sample data were generated. Based on the ratio of 12∶1 for training samples to test samples, the back analysis model was trained. The results show that the mean absolute percentage errors EMAPE of the DE-BP neural network model are all less than 3%, and the relative errors are less than 5%. This indicates that the prediction accuracy of the BP neural network can be effectively improved by the DE algorithm. The maximum error of the temperature peak for the characteristic points based on inversion analysis is 2.05 ℃, and the calculated temperature histories are in good agreement with the actual ones. In a word, the back analysis method of thermal parameters for the concrete box girder based on the DE-BP neural network and uniform design theory demonstrates high accuracy and a stable inversion process with good reliability, which can provide a theoretical basis for temperature control of other similar projects.