摘要
结合基本遗传算法和神经网络各自的优点,采用十进制编码,引入改进的自适应变异操作等算法,分析和建立了GA&NN遗传神经网络。这种混合训练的效率和效果都比单独的遗传进化或BP训练方法有明显的改善。其通用性较好,具有较快的收敛性和较强的学习能力。模型应用于盾构隧洞开挖的变形预测,根据施工中所测的影响因素大小对地表沉降可作出适当的预测,效果优于回归等常规模型。
Combining the advantage of genetic algorithm and artificial neural network, the GA&NN modal was established, in which decimal coding was adopted. Improved adaptive variation was adjusted which could improve the speed of network convergence, and had the good ability to learn. This modal could be used to predict the surface settlement of shield tunnels, in which the influenced factors in tunnel construction process were acted as the input numbers, the output results were the deformation values, the results of the application showed that GA&NN modal was better than the normal regression modals.
出处
《测绘科学技术学报》
北大核心
2007年第1期67-69,共3页
Journal of Geomatics Science and Technology
基金
河南省高等学校青年骨干教师资助计划
河南省教育厅自然科学基础研究项目(200642001)
关键词
变形预测
遗传算法
神经网络
地表沉降
deformation forecasting
genetic algorithm
neural network
ground settltment