A Critical Review of Machine Learning Application Models in Dam Settlement Monitoring

Today, settlement prediction models in rockfill and concrete dams are essential for dam safety analysis. Due to the complexity, uncertainty, and time-consuming nature of numerical finite element models, data-based statistical models are considered a suitable alternative for dam safety monitoring. This article comprehensively examines the development and application of single and hybrid artificial intelligence models in the field of dam settlement monitoring. The results of the review of artificial intelligence models in predicting the displacement and settlement of dams show that hybrid models (27.4%) are used more than other models. Support vector machine (SVM) & support vector regression (SVR) 11.6%, artificial neural networks (ANNs) 9.5%, and random forest (RF) 7.4% models are also the most widely used single models. Also, multiple linear regression (MLR) 17.4% and SVM 15.3% models are the most popular models to compare with the studied model. Criteria RMSE and MAE have been the most used evaluation criteria in studies. In recent studies, the issue of missing data and the use of spatiotemporal hybrid models to predict settlements has become very hot. Therefore, by using a panel data set, it is possible to identify and measure effects that cannot be identified in pure cross-sectional data or pure time series. This comprehensive research reviews 94 research papers from 2002 to 2023 on the performance of artificial intelligence for dam settlement modeling and provides detailed insight into the approach and reliability of intelligent models.