Evaluating performance of meta-heuristic algorithms and decision tree models in simulating water level variations of dams’ piezometers

Abstract

Monitoring the seepage, particularly the piezometric water level in the dams, is of special importance in hydraulic engineering. In the present study, piezometric water levels in three observation piezometers at the left bank of Jiroft Dam structure (located in Kerman province, Iran) were simulated using soft computing techniques and then compared using the measured data. For this purpose, the input data, including inflow, evaporation, reservoir water level, sluice gate outflow, outflow, dam total outflow, and piezometric water level, were used. Modeling was performed using multiple linear regression method as well as soft computing methods including regression decision tree, classification decision tree, and three types of artificial neural networks (with Levenberg-Marquardt, particle swarm optimization, PSO, and harmony search learning algorithms, HS). The results of the present study indicated no absolute superiority for any of the methods over others. For the first piezometer the ANN-PSO indicates better performance (correlation coefficient, R=0.990). For the second piezometer ANN-PSO shows better results with R=0.945. For the third piezometers MLR with R=0.945 and ANN-HS with R=0.949 indicate better performance than other methods. Furthermore, Mann-Whitney statistical analysis at confidence levels of 95% and 99% indicated no significant difference in terms of the performance of the applied models used in this study.

Keywords

• Data driven models
• dam surveillance
• soft computing, heuristic algorithms, dam engineering