Identification model for dam behavior based on wavelet network

Dam behavior is conventionally evaluated with identification models of deformation, seepage, stress, and crack opening. The identification model needs to be described with a complicated and nonlinear function. Wavelet networks based on wavelet frames were used to establish the identification models of dam behavior for the first time. Firstly, time-frequency analysis for training data was implemented to determine the original structure of the wavelet network. Next, a new method was proposed for iterative elimination of the redundant neurons according to the dependency between the network output and the nodes in the hidden layer. In this method, rough sets theory was used to calculate the dependency. Lastly, this article built the identification models for the displacement and cracks of one concrete arch-dam with the trained wavelet network. The models can represent the connection between loads and the behavior of the dam. The numerical example shows that the proposed models are reasonable, and the denoising effect of the signal is remarkable.