An evolutionary neural network approach to simple prediction of dam deformation

作者：Miao Xinying; Chu Jinkui^*; Zhang Linghan; Qiao Jing

来源：Journal of Information and Computational Science, 2013, 10(5): 1315-1324.

DOI：10.12733/jics20101559

• Back-propagation neural networks
• Dam deformation
• Dam deformation prediction model
• Evolutionary neural network
• Levenberg-Marquardt
• Levenberg-Marquardt algorithm
• Non-linear relationships
• Trial-and-error approach

Due to the nonlinearity between deformation and its influencing factors, it is difficult to establish an effective and practical dam deformation prediction model. As one of the ideal tools to model a nonlinear relationship between inputs and outputs, Back Propagation Neural Network (BPNN) has recently been employed to predict dam deformation. Despite its extensive applications, BPNN has a slow convergence rate and easily falls into the local minimum, and its design and structural optimization are still done via a time-consuming reiterative trial-and-error approach. In this paper, Levenberg-Marquardt with Genetic Algorithm (GA-LM), an evolutionary neural network model combining the Levenberg-Marquardt (LM) algorithm and Genetic Algorithm (GA), has been developed for predicting dam deformation. LM is used to train NN, which shows faster convergence rate than BPNN. The network architecture is optimized by GA. The performance of GA-LM has been compared with that of conventional BPNN and LM algorithm with trial-and-error approach. The comparison indicates that the predicted dam deformation values using GA-LM model are in good agreement with the measured data, and it is proved that the GA-LM model can offer stronger and better performance than conventional neural networks and give superior predictions over the trial-and-error model.

• 出版日期2013