Developing an agent-based modeling platform for river basin management using AI and machine learning techniques

Hu, Xinchen

Permalink

https://hdl.handle.net/2142/124659 Copy

Description

Title

Developing an agent-based modeling platform for river basin management using AI and machine learning techniques

Author(s)

Hu, Xinchen

Issue Date

2024-04-17

Director of Research (if dissertation) or Advisor (if thesis)

Cai, Ximing

Doctoral Committee Chair(s)

Cai, Ximing

Committee Member(s)

• Zhang, Zhenxing
• Ju, Tingju
• Tartakovsky, Alexandre

Department of Study

Civil & Environmental Eng

Discipline

Environ Engr in Civil Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Agent-based model
• machine learning
• reinforcement learning
• irrigation scheduling
• reservoir operation

Abstract

A river basin can be understood as a coupled nature-human system (CNHS), which is featured by the dynamics and feedback between environmental processes and human activities. Agent-based models (ABMs), which are recognized as useful tools for studying the relations between individual-level behaviors and system-level emergence, have their growing use for integrated river basin management (IRBM) problems which cover environmental, technical, economic, social, and legal aspects, and the interactions between these aspects. However, it is difficult to setup an ABM, especially in describing appropriate behavior rules for different agents involved in IRBM. The primary goal of this dissertation is to develop a general platform aimed at assisting modelers lacking proficient programming skills in setting up ABMs without starting from scratch. This dissertation includes three parts. The first is on the structure of the platform and the design of the platform. A webapp-based platform is implemented to provide user interfaces. The agent component is coupled with an environment component that simulates the natural processes. A modular design method allows a convenient extension of the platform. The second part addresses how to derive appropriate behavior rules when data availability is limited. A reinforcement learning (RL) framework is built as an example to derive irrigation rules for farmers. It shows the capability of AI models for deriving human behavior rules. The third part of this dissertation discusses and demonstrates the coupling of AI models to derive behavior rules under the condition of with or without sufficient data support. While the RL-based deep learning method is used to deal with the condition of limited data availability, a generic data-driven reservoir operation model (GDROM) is used to show an example to derive reservoir operation rules when sufficient data is available for machine learning applications. Furthermore, the potential of using advanced AI models such as ChatGPT to derive agent behavior rules is demonstrated. Overall, this dissertation presents a general ABM platform for integrated river basin management, which is to help ABM modelers reduce the effort on building their model from scratch . This dissertation also shows the effectiveness and potential of using data techniques (machine learning, data mining) and AI models to derive agent behavior rules to build more realistic ABMs.

Graduation Semester

2024-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2024 Xinchen Hu

Owning Collections