Traditional dam safety monitoring models have limitations in factor selection and nonlinear re ⁃ lationship expression. Although BP neural network possesses excellent nonlinear fitting capabilities
its application is still constrained by issues such as sensitivity to initial weights
slow convergence speed and overfitting. Therefore
this study proposes an IAPSO-BP neural network model optimized by im ⁃ mune particle swarm optimization algorithm for predicting the settlement displacement of earth-rockfill dams. This model optimizes the initial parameters of network through immune particle swarm optimiza ⁃ tion algorithm
and combines a cross-entropy cost function with a weight decay regularization method to effectively improve the global optimization ability and generalization performance of model. Using the settlement monitoring data from a concrete-faced rockfill dam of a pumped storage power station′s lower reservoir as a case study
the results show that compared to the standard BP model and the PSO-BP model
the IAPSO-BP model demonstrates significant improvements in fitting accuracy and prediction stability
with the average relative prediction error reduced to 1.05%
and effectively overcomes the prob⁃ lems of overfitting and local convergence
providing a more reliable and efficient analytical tool for the safety monitoring of earth-rockfill dam settlement.