Accurate prediction of peak outflows from breached embankment dams is a key parameter in dam risk assessment. In this study, efficient models were developed to predict peak breach outflows utilizing artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS). Historical data from 93 embankment dam failures were used to train and evaluate the applicability of these models. Two scenarios were applied with each model by either considering the whole data set without classification or classifying the set into small dams (48 dams) and large dams (45 dams). In this way, nine models were developed and their results were compared to each other and to the results of the best available regression equations and recent gene expression programming. Among the different models, the ANFIS model of the first scenario exhibited better performance based on its higher efficiency (E = 0.98), higher coefficient of determination (R2 = 0.98) and lower mean absolute error (MAE = 840.9). Moreover, models based on classified data enhanced the prediction of peak outflows particularly for small dams. Finally, this study indicated the potential of the developed ANFIS and ANN models to be used as predictive tools of peak outflow rates of embankment dams.

Dam breach width significantly influences peak breach outflow, inundation levels, and flood arrival time, but uncertainties inherent in the prediction of its value for embankment dams make its accurate estimation a challenging task in dam risk assessments. The key focus of this paper is to provide a fuzzy logic (FL) model for estimating the average breach width of embankment dams as an alternative to regression equations (RE). The FL approach is capable of handling nonlinear behavior, imprecision in discrete measurements, and parameter uncertainty. Historical data from 69 embankment dam failures are used in the development and testing of the FL model. Application of the FL model is also presented for estimating average breach widths of two case studies that have adequately documented data. The accuracy of the FL rule-based model is investigated using uncertainty analysis: the mean prediction error between the FL estimates and the observed average breach widths is very small (=0.03) and comparable to that achieved using the best available RE. Moreover, the FL uncertainty band is found to be approximately ±0.51 order of magnitude smaller than the ±0.56 order of magnitude achieved with the best available RE. The simulation results indicate the potential of the FL model to be used as a predictive tool for estimating the average breach width of embankment dams.

No abstract

No abstract