Prediction of temporal variation of scour hole dimensions due to plane wall jets: Application of new soft computing techniques

Abstract

Scour due to wall jet is one of the important problems in hydraulic engineering field. Prediction of the scour hole dimensions with acceptable accuracy is necessary for the structures around the jets. In this issue, use of soft computing (SC) methods is unavoidable due to the complex process of jet scour. In this research, SC techniques such as random forests (RF), particle swarm optimization based fuzzy inference system (PSOFIS), slime mould algorithm based fuzzy inference system (SMAFIS) and adaptive neuro fuzzy inference system (ANFIS) were implemented for predicting temporal variation of the scour hole dimensions due to plane wall jets for the first time. The performance indices and the data distribution showed that among the above-mentioned methods, the RF could predict the scour hole dimensions with the highest accuracy. This is attributed to this fact that RF which is a substantial modification of Bootstrap averaging, increases the precision of estimated prediction by a collection of de-correlated regression trees. The hybrid methods, PSOFIS, SMAFIS and ANFIS were ranked from the best to the worst based on the performance indices of the test data. The sensitivity analysis showed that among the effective parameters, the scouring time and sediment standard deviation had the most and the least effects on the temporal variation of scour hole dimensions, respectively. Use of channel width ratio and jet Reynolds number along with other effective parameters such as densimetric Froude number, tailwater depth ratio, sediment size ratio in the models are recommended to reach appropriate accuracies. Comparison with the existing literature indicated that the accuracy of the proposed methods is considerably better than the previous researches.