University of Technology Sydney. Faculty of Engineering and Information Technology. Analysing the behaviour of soft soils under embankments is a significant challenging task for geotechnical engineers. By having more insight into long term soil behaviour and understanding the key parameters influencing the results, there will be more chance to strategically plan and utilise the soft ground for construction purposes. The time–dependent behaviour of soft soils, especially the ground settlements under structural and non–structural loading, is considered as a significant issue, which has been studied for many decades. Prediction of creep settlement of soft soils is a challenging task, as a very long period of time counted in years is involved. Many theories have been proposed along with a large number of laboratory and field measurements in order to provide more precise knowledge of the time–dependent viscous behaviour of soft soils. However, there are still some disagreements between theoretical and practical studies, which may keep the accuracy of the predictions questionable. Among the great number of developed models for soft soils, the elastic visco–plastic model with the non–linear creep function is considered as an effective method to describe the long–term stress–strain behaviour of soft soils. However, the difficulties to determine the model parameters limit the application of the model in practice. Since the relationship between the effective stress and strain during the dissipation of the excess pore water pressure cannot be identified easily, in the current practice the creep strain limit ε_lm^vp and the creep coefficient ψo/V to form the creep function are determined based on the curve fitting of the experimental data after the end of the primary consolidation. As a result, the number of data points available for the curve fitting is limited, and the extremely long tests are required. Moreover, in the conventional procedure for the ease of the curve fitting, the time parameter to in the elastic visco–plastic, which is the time value of the reference time line in the space of ε-log(σ’z), has been assumed as the time at the end of primary consolidation process. Hence, based on this assumption of to, the reference time line would include viscous strain, which is contradict to the definition of a viscous free reference time line. Thus, the value of to influences not only the reference time line parameters, but also the parameters of the creep function. Additionally, the conventional determination approach for the model parameters is influenced by the thickness of the soil sample. Hence, the model parameters obtained by the conventional method may not be unique. As a result, the main objective of this research project is to propose a numerical solution to determine the model parameters for the elastic visco–plastic model adopting the trust–region reflective least square algorithm. The trust-region reflective least square algorithm is an advanced optimisation method for the non-linear equation system. A Crank–Nicolson finite difference scheme is applied to solve the coupled partial differential equations in order to simulate one-dimensional stress-strain behaviour of soft soil with different boundary conditions. The proposed method can adopt the experimental data during the dissipation of the excess pore water pressure to determine all the model parameters simultaneously. In this thesis, a series of laboratory experiments were conducted at the UTS soil laboratory using two sizes of hydraulic consolidation Rowe cell setups. A 29.5 mm thick soil sample of a kaolinite mixture was tested and adopted to determine the model parameters, while an experimental result of a thicker soil sample (i.e. 140.5 mm thick) was compared with the predictions using the optimised model parameters. The Rowe cell setups can measure the volume change, the vertical settlement and the excess pore water pressure continuously. Especially, the large Rowe cell setup to conduct the test on the 140.5 mm thick soil sample was modified to measure the excess pore water pressure at different depth and different distances to the centre line at the base. Moreover, other four validation exercises including two laboratory–based case studies and two field–based case studies were included to verify the ability of the proposed method to analyse the time-dependent behaviour of soft soils. The developed method can be considered as a simple, practical and accurate solution for the model parameter determination. The optimised model parameters allow the predictions of settlement to be in good agreement with the measurements, while the predictions of the excess pore water pressure are reasonably close to the measurement. Additionally, the variations of the creep strain limit, the creep coefficient and the creep strain rate during the dissipation of the excess pore water pressure can be observed. Moreover, the unusual increase of the excess pore water pressure in the early stages of loading can be also predicted. The numerical analysis applying the proposed method is able to illustrate the influence of the soil layer thickness on the time–dependent stress-strain behaviour of soft soil. The proposed approach can be adopted to back calculate the elastic visco-plastic model parameters for real case in the field utilising time-dependent settlement and excess pore water pressure measurements.