Aiming at the problem that traditional methods are difficult to capture the health status of civil structure comprehensively and accurately, this paper introduces BP neural network to design and optimize the health monitoring and prediction system of civil structure. Static data acquisition uses strain sensors and contact displacement sensors, and dynamic data acquisition uses MEMS acceleration sensors and temperature sensors. The data processing steps include using IForest algorithm for data cleaning and anomaly identification, and using wavelet transform for signal denoising processing, and then using scale energy distribution method to extract signal features. The experimental results show that the initial accuracy of BP neural network, RF, ACA and PSO algorithms are 57%, 42%, 31% and 59%, respectively, and the accuracy after 100 iterations are 97%, 88%, 83% and 92%, respectively. The accuracy of BP neural network is higher than other algorithms.