A particle swarm optimization and prior knowledge fusion seismic damage prediction of concrete structures

Abstract

A data-driven algorithm is developed to predict local seismic damage distribution of concrete structures based on the measured global structural response. Based on the algorithm, the signal of reaction force and displacement at one location is only necessary for seismic damage prediction. The algorithm is established based on the improved particle swarm optimization with two innovative strategies. One is the probabilistic mutation procedure, which can consider the prior knowledge of the positive correlation between the strain/stress level and damage level in the seismic damage optimization process. Another is the dynamic condition-based mutation and cross procedure, which can increase the diversity of the particle swarm in the optimization process to get rid of the possible local optimum. A representative example of a concrete column under cyclic load is designed and modeled to examine the performance of the algorithm. The prediction results based on the algorithm are compared with the traditional particle swarm optimization and the previous damage inversion algorithm based on ant colony optimization. The comparison results support that the local seismic damage distribution prediction based on the algorithm is closer to the corresponding experimental result. In addition, the error of the predicted macroscopic response in the final seismic stage based on the algorithm is 1.7%. The prediction error of the traditional particle swarm optimization algorithm is 12.6%, and the prediction error of the previous damage inversion algorithm based on ant colony optimization is 8.5%. The ability of the proposed algorithm is supported, which can be capable of seismic damage prediction of concrete structures subjected to earthquakes.