Implementation of the dual permeability model and application of population-based metaheuristics in inverse modeling

Johanna Ruth Blöcher

Master's thesis

Abstract:

Soil often exhibits a variety of heterogeneities such as fractures, fissures, cracks, and macropores, and can show dynamic instabilities of the wetting front during infiltration, which can lead to preferential flow. Preferential flow can accelerate the movement of contaminants in the unsaturated zone and is therefore of concern to environmental science. The dual permeability model after Gerke and Genuchten (1993) can be used to model preferential flow in soil. It was constructed around the assumption that the soil medium can be separated into two distinct pore systems. In this approach the matrix and fracture systems are each treated as homogeneous media with separate hydraulic properties that can both be described with a Richards equation. In this work, the dual permeability model was implemented in the free software Dual Richards Unsaturated Equation Solver (DRUtES). Different scenarios were designed to test under which conditions preferential flow occurs.In order to perform parameter estimation as part of inverse analysis, global metaheuristic optimization algorithms were reviewed and implemented. Adaptations of Particle Swarm Optimization (PSO) and Teaching-Learning-Based Optimization (TLBO) were used, which are population-based metaheuristics with different learning strategies. These are high-level stochastic-based search algorithms that don not require gradient information or a convex search space. Introduced variants include a simple bad neighborhood approach and shuffling complexes. Despite increasing computing power and parallel processing, an overly fine mesh is too computationally intensive to be used for parameter identification. This creates the need to find a mesh that optimizes both accuracy and simulation time. A bi-objective PSO algorithm was designed and implemented. This algorithm is based on a dynamic neighborhood approach and can generate a Pareto front of optimal meshes to account for both objectives. The global optimizers were successfully tested on commonly used benchmark functions. The created inverse set-up involved the linking of the software to the optimization algorithms and was tested on both virtual inverse problems and real data.

Keywords

Dual permeability model, Preferential flow, Population-based metaheuristics, Inverse modeling, Optimization, Particle-Swarm-Optimization, Teaching-Learning-Based Optimization

Language used: English

Date on which the thesis was submitted / produced: 18. 4. 2017

Thesis defence

• Supervisor: doc. Ing. Michal Kuráž, Ph.D.

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Master programme / field:
Landscape Engineering / Environmental Modelling