The model of feed-forward neural networks trained by differential evolution (DE) algorithm is presented to overcome the shortcoming of traditional regression model widely used in monitoring the safety and deformation of dams. DE algorithm is a population-based one in global optimization

with the merits of being easy to use and fast convergence. The neural networks trained by DE can effectively avoid the problem of being stuck in any local minimum that often happens in classical BP neural networks model. The case study of deformation monitoring of an arch dam shows that the DE neural networks model proposed results in a better precision

comparing with traditional regression model and BP neural networks model.