| Title: |
Simulation-based optimization and reinforcement learning methods to improve decision making in agriculture |
| Authors: |
Moeinizade, Saba |
| Contributors: |
Hu, Guiping; Wang, Lizhi; Olafsson, Sigurdur; Schnable, Patrick; Genschel, Ulrike; Industrial and Manufacturing Systems Engineering |
| Publication Year: |
2022 |
| Collection: |
Digital Repository @ Iowa State University |
| Subject Terms: |
Bioinformatics; Agriculture; Engineering; Genetic Gain; Genomic Selection; Operations Research; Optimization; Reinforcement learning; Simulation |
| Description: |
To address the global challenges of growing population, changing climates, and changing diets agriculture products should be more productive, sustainable, and resilient. To permit more informed decisions in agriculture systems, interdisciplinary efforts are needed. In this dissertation, we use operations research and data analytic techniques to enhance decision making in agriculture systems by tackling specific problems in plant breeding. We design algorithms for improving efficiency in genomic selection, which is a special type of nonlinear, non-convex, high-dimensional, and dynamic optimization problem constrained by resource availability and laws of reproductive biology. Furthermore, we develop an algorithm for selecting cultivars in trait introgression where the goal is to transfer the desirable traits from a donor to an elite line which lacks those desirable traits. The common attribute of the proposed algorithms in this dissertation is taking advantage of simulation to look-ahead and make informed decisions based on the estimated future outcome. We present a family of look-ahead algorithms for optimizing selection and mating decisions in breeding programs and use more advanced optimization techniques such as reinforcement learning to optimally allocate resources during a breeding program. What makes these problems more challenging is the uncertainty due to the recombination events. To capture the uncertainties and characterize the behavior of these complex systems, we develop a stochastic simulation framework. This framework enables testing the proposed algorithms and comparing them with conventional methods by designing case studies using realistic data sets. More specifically, in chapter 2, we introduce the look-ahead selection algorithm to optimize selection and mating decisions by evaluating the probability of achieving high genetic gains within a specific time. In chapter 3, we propose multi-trait look-ahead selection algorithm, which maximizes certain traits while keeping others within desirable ... |
| Document Type: |
doctoral or postdoctoral thesis |
| File Description: |
PDF; application/pdf |
| Language: |
English |
| Relation: |
https://dr.lib.iastate.edu/handle/20.500.12876/arY48V4v |
| DOI: |
10.31274/td-20240329-673 |
| Availability: |
https://dr.lib.iastate.edu/handle/20.500.12876/arY48V4v; https://hdl.handle.net/20.500.12876/arY48V4v; https://doi.org/10.31274/td-20240329-673 |
| Accession Number: |
edsbas.95B3E660 |
| Database: |
BASE |