Sediment transport prediction due to in-stream mining by evolutionary polynomial regression method / Nadiatul Adilah Ahmad Abdul Ghani

• Main Author: Ahmad Abdul Ghani, Nadiatul Adilah
• Format: Thesis
• Language: English
• Published: 2019
• Online Access: https://ir.uitm.edu.my/id/eprint/83282/1/83282.pdf; https://ir.uitm.edu.my/id/eprint/83282/
• Institution: Universiti Teknologi Mara

This study investigates the use of Evolutionary Polynomial Regression (EPR) technique in genetic programming to predict sediment transport due to in-stream mining. EPR is a data-driven hybrid regression technique and has been used successfully in solving numerous problems in civil engineering that include sewer failure, water distribution in pipelines, groundwater management, water quality model, floodplain, coastal and also sewage sludge prediction. However, this technique has not been fully explored in establishing sediment transport model. In this technique symbolic models were first constructed by integrating the best features of numerical regression. Four established models were selected to predict sediment transport on Malaysian rivers using 390 data from twelve selected rivers in Malaysia. This extensive database was obtained from the Department of Irrigation and Drainage (DID), Ministry of Natural Resources & Environment, Malaysia and previous studies. Analysis has shown below average performance for all equations when tested on local river data. Rigorous testing of the independent variables from the selected equations, have found significant predictors for use in the newly derived equation. The identified predictors found to be of significance are U*/V (ratio of shear velocity to velocity of flow), R/d50 (ratio of hydraulic radius to particle mean size), B/y (ratio of width to flow depth and U*/s (ratio of shear velocity to fall velocity of sediment particle). Development of model makes use of 273 data (training sets) from twelve rivers in Malaysia. Out of this number, 64% of the predicted data fall within the discrepancy ratio of 0.5 - 2.0 to the measured data. While 70% of the data used for validation fall within the acceptable limit. Upon establishing the best model, new set of data with a limitation of range were selected. Only 256 data were selected from 390 data used which are 64% from total data. Based from this new set of data range, 666 new models were developed using this technique. The best proposed sediment transport model for the first group range can be defined as a function U*/V (ratio of shear velocity to velocity of flow), R/d50 (ratio of hydraulic radius to particle mean size). Balance 134 data which is fall outside the range was used to develop second model for a different range of data. A robustness study was performed in order to validate the generalization ability of the developed EPR model. The percentage of sensitivity analysis for every four input parameters shows input value d50, particle mean size gives the most significant to EPR Model with 4.08%. The less significant variables to this model is shear velocity, U*, with 0.23%. Results obtained from the EPR model was compared with those obtained from four other available sediment transport prediction models. Analyses have found that EPR model features predictive potential to solve highly nonlinear sediment data. Comparative analysis of the proposed model with the established sediment transport models using regression technique and modified Graf were carried out. In conclusion, the model developed using the Evolutionary Polynomial Regression technique had been validated using 82 data and it yielded the best results compared to the other models with most of the prediction data plotted 100% in difference ratio.