基于BP优化神经网络的桥梁抗震动力可靠度分析

Seismic Dynamic Reliability Analysis of Bridges Based on BP Optimization Neural Network

为提升BP神经网络响应面的训练效率及拟合精度,基于首次超越准则和虚拟激励法对桥梁结构在地震作用下的主梁动力可靠度展开研究。引入粒子群算法对标准BP神经网络的权值和阈值进行参数寻优,与未经优化的BP神经网络、RBF神经网络响应面的计算结果进行对比,分析了(20+25+20)m跨径的预应力钢筋混凝土连续箱梁主梁动力可靠度指标。结果表明:相对于遗传算法,粒子群算法对BP神经网络的参数寻优具有更高的收敛效率和精度;优化后的BP神经网络相较于未优化的BP神经网络和RBF神经网络具有更高的训练效率和拟合精度,主梁动力可靠度指标相对误差仅为1.54%;桥梁结构在地震作用下的主梁动力可靠度指标大于4,具备良好的抗震性能。

In order to improve the training efficiency and fitting accuracy of the response surface of BP neural network,the dynamic reliability of the main beam of the bridge structure under the action of earthquake is studied based on the first surpassing criterion and the virtual excitation method.The particle swarm optimization algorithm is introduced to optimize the weights and thresholds of the standard BP neural network.Compared with the calculation results of the response surface of the unoptimized BP neural network and RBF neural network,the dynamic reliability index of the main beam of a prestressed reinforced concrete continuous box girder is analyzed.The results show that compared with genetic algorithm,particle swarm optimization algorithm has higher convergence efficiency and accuracy in parameter optimization of BP neural network;Compared with the optimized BP neural network and RBF neural network,the optimized BP neural network has higher training efficiency and fitting accuracy,and the relative error of the main beam dynamic reliability index is only 1.54%;The dynamic reliability index of the main beam of the bridge structure under earthquake action is greater than 4,which has good seismic performance.