Granular Fuzzy Rule-Based Models: Design and Analysis

• Author / Creator

  Hu, Xingchen

• Fuzzy rule-based models have been studied for decades and emerged in a diversity of architectures and design approaches. They play a vital and unique role by forming a human-centric computing framework. In general, fuzzy rule-based models are regarded as numeric constructs; as such they are optimized and evaluated at the numeric level. However, no ideal fuzzy models that fully capture (coincide with) all numeric experimental data. Information granularity is a perspective to represent and recognize the abstraction of information as the observation method of humans. In this perspective, the numeric data can be presented at various levels of resolution or scales. By bringing a concept of information granularity into fuzzy rule-based models, we give up on obtaining precise numeric models, whereas we make them into granular form and produce granular results. Subsequently, the outputs provided by granular fuzzy rule-based models are aligned well with the experimental data and deliver better insight into credibility. The fundamental objective of this thesis is to establish a comprehensive, systematic method for developing granular fuzzy rule-based models, so that the granular outputs of the models can embrace (cover) the target experimental data as much as possible, meanwhile the granular outputs are as specific as possible. To accomplish these objectives, we study several fundamental design issues that emerge in the realm of Granular Computing. First, we propose an advanced scheme of granulation and degranulation to abstract information granules from numeric data. Second, we investigate several commonly known logic operators that are used in fuzzy modeling and granular fuzzy modeling. Afterwards, we design a series of development strategies for granular fuzzy rule-based models by admitting and allocating a certain level of information granularity around numeric values. Our proposed granular rule-based models could be classified into three groups: granular input space of the models, granular processing modules of the models, and granular output space of the models. Unlike the standard numeric-performance measure of fuzzy models that come in the form of the root-mean-square error (RMSE), two pertinent performance measures are introduced and implemented to evaluate the performance of granular fuzzy rule-based models: namely, coverage and specificity. We develop several protocols of forming and allocating information granules to cope with different strategies of granular modeling and analyze how different protocols lead to improve the performance of granular models. Some commonly used population-based optimization algorithms—for instance, particle-swarm optimization (PSO) and differential evolution (DE)—are used to optimize the allocation of information granularity, and coverage and specificity criteria are used to guide the optimization. A series of experimental studies is reported which offers a comprehensive overview of the underlying realization and performance of the granular fuzzy rule-based models.

• Subjects / Keywords

  • Fuzzy rule-based model
  • Allocation of information granularity
  • Coverage and specificity
  • Granular computing

• Graduation date

  Fall 2017

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/R3C24R23X

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.