A chaos-optimized neural network model for dam safety monitoring

Dam deformation prediction is important for dam safety monitoring and has become a focus of increasing interest in recent years. In this study, on the basis of nonlinear dynamic property analysis of the observations of dam displacements, a novel methodology is proposed to establish dam deformation prediction model with improved prediction precision. Firstly, the dynamic properties of observations of dam displacements are studied by combined wavelet transform with fractal, and the results reveal that dam displacements possess certain low dimensional chaotic character. This provides theoretical foundation and transcendental knowledge for relational establishment of dam deformation prediction model. Moreover, derived from the low dimensional chaotic character, a chaos-optimized neural network model for dam deformation prediction is constructed, which is not only capable of capturing the dynamic properties of observations of dam displacements but also of implementing the model's structural optimization and dynamic mechanism refreshing. Finally, in the practical application of dam deformation prediction, the model performance is quantificationally assessed by multiple indices. The result demonstrates that chaos-optimized neural network model holds higher prediction precision than the conventional back propagation (BP) neural network and ARMA models; and therefore, it is promising for dam safety monitoring.