Using Genetic Artificial Neural Network to Model Dam Monitoring Data

Abstract

A displacement model using the back propagation algorithm of artificial neural networks (BP-ANN) optimized with a genetic algorithm (GA) was presented on the example of an arch-type dam in China. The settlement displacement analysis for a single point located on the dam was performed. The analysis consists of three stages: principal component analysis (PCA), BP-ANN modelling, and deformation forecast. PCA was firstly proposed to select input vectors so as to design the ANN model, and then a GA was adopted to optimize the interconnecting weights and thresholds of ANN, at last the BP learning and forecast were performed. The results demonstrate that the optimized method has better convergence ability than the pure BP-ANN. Compared to stepwise regression, the optimized BP-ANN model is limited in interpreting the displacements contributed by water, temperature, and aging variables, and also it may result in poor predicted ability using low frequency monitoring data.