Assessment of BPNN and NARX models for precise piezometric level prediction in earth dams

Dam safety is a critical issue, as numerous failure cases have been reported, leading to severe disasters. In the design of an earth dam, water seepage represents a major concern, as it can cause internal erosion when the seepage capacity exceeds its allowable limit. Due to the crucial role played by the piezometric water level in dam safety analysis, many researchers have focused on developing prediction tools using artificial intelligence (AI) to generate highly accurate forecasts of a dam’s piezometric water level. In this study, the objective is to evaluate the predictive performance and generalization capability of two AI methods. The methods used are the Backpropagation Neural Network (BPNN) and the time series method (NARX). These models are applied to piezometric data to predict the hydraulic behavior of the dam. The correlation between piezometric levels and the reservoir level was tested and compared using the analysis of recorded piezometric measurements combined within the dam profiles. According to the obtained results, the comparative analysis confirms the high performance and reliability of both methods. The BPNN model, an artificial neural network, excels in modeling nonlinear phenomena through its adaptability to varying degrees of nonlinearity between variables, making it ideal for regression tasks.