Comprehensive evaluation of dam seepage safety combining deep learning with Dempster-Shafer evidence theory

Abstract

Seepage state is one of the most important factors affecting the safety and durability of dams, which calls for accurate observation and evaluation. Traditional dam seepage evaluation mainly focusses on single monitoring points, and ignores the uncertainty and subjectivity with multi-sensor fusion. This paper presents an evaluation method of dam seepage based on Dempster-Shafer (D-S) evidence theory and deep neural network (DNN). In view of the nonlinearity of the monitoring data and the difference of seepage influencing factors, attention mechanism (AM) and long short-term memory (LSTM) model are combined to evaluate dam seepage safety of single monitoring points, which provides the basis for the comprehensive evaluation. The case study demonstrates that the average RMSE, MAE and MAPE values of AM-LSTM are of minimal losses and higher fitting accuracy. The forecast results of the proposed comprehensive evaluation approach are consistent with the actual dam seepage state, demonstrating its reasonability and reliability.