Inverse and Sensitivity Analysis of Thermal Parameters of Massive Concrete Using BP Neural Network

Abstract

The mass concrete is a kind of important component in real-world constructions, such as the dam, bridge, and pavement. Accurate thermal analysis is always a key point in structural design due to its low conduction performance. However, the parameters obtained from tests are not sufficient to fit real situations in many cases. The BP neural network algorithm is adopted to get thermal parameters as accurately as possible according to monitored inner temperatures, and sensitivity analysis of thermal parameters is conducted based on partial derivative formulas. A published case with monitored data is used for verifying the proposed method. Inverse analysis results show that the BP neural network performs well, and sensitivity analysis results indicate that the specific heat capacity and the parameter a in the hydration model have most influential on temperature results.