Adaptive neuro-fuzzy inference system based neural network and parameter constraints

Juston, Marius Francois Robert

Permalink

https://hdl.handle.net/2142/124327 Copy

Description

Title

Adaptive neuro-fuzzy inference system based neural network and parameter constraints

Author(s)

Juston, Marius Francois Robert

Issue Date

2024-05-01

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

Norris, William

Department of Study

Industrial&Enterprise Sys Eng

Discipline

Systems & Entrepreneurial Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• ANFIS
• optimization
• fuzzy logic
• parameter constraint
• DDPG

Abstract

This thesis presents a comprehensive approach to designing and optimizing an Adaptive-Network-Based Fuzzy Inference System (ANFIS) for robotics applications. This research investigates the implementation of constraints on the input and the output linguistic variables, such as single-sided and symmetry constraints, with theoretical guarantees to maintain the desired constraint for offline and training purposes. Specifically, the linguistic joint membership functions that underlie the ANFIS are defined, focusing on symmetrical inputs/outputs and jointly optimized trapezoid membership functions. A novel way of representing and computing the Mamdani ruleset was generated and explored. These constraints aim to reduce the number of training parameters and thus increase training speed. Further optimizations for the ANFIS were derived based on design assumptions, including training the membership functions on closed or single-sided domains and improving the training of linear-based membership functions using a softplus-based saturation function. The relationship between the ANFIS and radial basis functions, as a special case of the neural network definition, was expanded and derived. This helped demonstrate the ANFIS network as a universal approximator. The ANFIS's use was examined in applying line following for a differential drive model, where its stability was explored. The optimal output membership weights based on mean square error optimization were also symbolically obtained for offline training. Additional training of the ANFIS's input/output membership functions was performed using the DDPG (Deep Deterministic Policy Gradient) algorithm to train the ANFIS system to run online.

Graduation Semester

2024-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2024 Marius Juston

Owning Collections