摘要
高拱坝力学性能参数变化规律复杂,使用人工智能算法进行预测已经成为反演参数的重要手段。使用遗传算法对神经网络进行优化来检验优化后算法的性能,并比较不同算法应用于参数反演中预测结果的精度。根据某高拱坝运行期变形监测数据,分别使用RBF神经网络和遗传算法优化的BP(GA-BP)神经网络对不同水位工况下的坝段分区混凝土弹性模量进行反演。基于反演结果进行有限元正分析计算,将所得结果与实测数据进行对比,检验反演精度和效率。结果表明:GA-BP网络的最大预测误差为1.8%,相比于RBF网络预测精度提高了约50%。使用神经网络进行拱坝力学参数反演实用性好,优化后的神经网络比传统BP神经网络在计算精度和效率两方面均有明显改进,且GA-BP神经网络反演比RBF神经网络反演精度更高。
The variation of mechanical parameters of high arch dams is complicated.Prediction using artificial intelligence algorithm has become an important means of inverse analysis.Genetic algorithm is used to optimize the neural network to test the performance of the optimized algorithm and compare the accuracy of prediction results of different algorithms in parameter inverse analysis.Based on the on-site measured deformation data of a high arch dam,Radical Basis Function(RBF)neural network and genetic-algorithm-optimized back propagation(GA-BP)neural network were employed to inversely investigate the elastic modulus of concrete in different dam sections under different hydraulic loadings.The accuracy and efficiency of inverse analysis was testified by comparing the results from finite element calculation with the parameters from inverse analysis with the on-site measured data.The results show that:The maximum prediction error of GA-BP network is 1.8%,which is about 50%higher than that of RBF network.The neural network method is suitable for inverse analysis of mechanical parameters of arch dams.What's more,the optimized neural networks show higher accuracy and efficiency compared with simple BP neural network,and the results from GA-BP neural network are more accurate than those obtained by using RBF neural networke.
作者
关志豪
姜冬菊
黄丹
任青文
GUAN Zhihao;JIANG Dongju;HUANG Dan;REN Qingwen(College of Mechanics and Materials,Hohai University,Nanjing 211100,China)
出处
《粉煤灰综合利用》
CAS
2020年第2期1-6,共6页
Fly Ash Comprehensive Utilization
基金
国家重点研发计划(2018YFC0406703)
国家自然科学基金(51679077)
中央高校基金科研业务费(2017B13014).
关键词
高拱坝
混凝土
反分析
遗传算法
神经网络
high arch dam
concrete
inverse analysis
genetic algorithm
neural network
