- Title
- A fuzzy adaptive metaheuristic framework for optimisation and prediction problems
- Creator
- Keivanian, Farshid
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2023
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Intelligent systems are all around us, and many assist in automated decision-making. These decision-making processes can be formulated as optimisation problems (OPs), which can be efficiently solved using computational intelligence techniques, including evolutionary computation (EC), fuzzy computing and neural networks (NNs). OPs are often associated with a wide range of uncertainties and challenges, including multiple design variables, which generate a high-dimensional decision space (DS) that requires enhanced local search capabilities. When the search space is non-convex and nonlinear, computation time can also significantly increase. Multimodal single-objective OPs (MSOPs) are those with many local optima. In this case, many solutions in the DS correspond to similar objectives in the objective space (OS), requiring a high global search capability. When dealing with multimodal multi-objective OPs (MMOPs) with increasingly conflicting goals, finding a trade-off between solutions with a satisfactory non-dominated Pareto front (PF) distribution is challenging. For computational intelligence techniques to be accurate and efficient, the parameters must be appropriately set. This creates parameterisation and computational cost uncertainties, requiring the correct level of exploration and exploitation. Given their stochastic nature, computational intelligence techniques also tend to produce noisy objective functions. In this research, a novel computational intelligence framework based on fuzzy set theory is proposed to solve benchmark and practical optimisation and prediction problems with varying uncertainties. In this context, fuzzy set theory can be used to transform descriptive convergence into fuzzy rules that may help determine the appropriate level of exploration and exploitation while identifying and managing parameter uncertainties. Three major studies for this thesis were conducted, and the proposed fuzzy set theory framework was adapted to address different benchmark and practical problems exhibiting characteristics such as nonlinearity, non-convexity, high dimensionality, multimodality and multi-objectivity, among others. The first study in this research presents a novel hybrid fuzzy–metaheuristic algorithm for solving benchmark MSOPs and MMOPs of varying dimensions. More specifically, a fuzzy adaptive enhanced imperialist competitive algorithm (FAEICA) was proposed. Fuzzy adaptive systems are used to select operators and adapt control parameters without requiring initial parameter settings. The algorithm provides faster convergence speeds compared with popular evolutionary algorithms such as differential evolution, particle swarm optimisation and the imperialist competitive algorithm (and its state-of-the-art variants). A multi-objective version of FAEICA (MOFAEICA), incorporated with fast-elitism non-dominated sorting for PF and a novel diversity-aware index, was also proposed. The MOFAEICA was found to avoid areas with similar objective values in an M-dimensional OS, resulting in a well-distributed PF approximation. The second study of this research considers single-objective optimisation (SO) and multi-objective optimisation (MO) for reinforced concrete cantilever (RCC) (SORCC and MORCC, respectively) retaining wall problems. In the SORCC retaining wall problem, a 12-dimensional DS was formulated based on 12 geometric variables and 26 constraints for structural strength and geotechnical stability, thus intensifying nonlinearity. The environmental impacts of concrete and reinforcing steel on the seismic properties of SORCC constructions were also considered. By extending the FAEICA’s search mechanisms, fuzzy inference rules and adaptive velocity limit function (AVLF), a new fuzzy adaptive global learning local search universal diversity (FAGLSUD) algorithm was developed that enables designers to make decisions based on their preference for low cost, low weight or low carbon dioxide emissions. The MORCC retaining wall problem considers two contradictory objectives—cost and earthquake sensitivity—and has 21 dimensions in its DS. For this problem, a MOFAEICA with an AVLF was developed. To select the most suitable RCC design, multi-objective bargaining game (MOBG) theory was proposed to negotiate a minimum cost and sensitivity to earthquake hazards. According to the simulation results, MOFAEICA-MOBG provides an excellent balance between low cost and low sensitivity to variations in the horizontal earthquake coefficient as well as other concerns (e.g. carbon dioxide emissions and weight). The third study in this research examines a healthcare problem related to body fat prediction (BFP) from an optimisation perspective. Intricately linked to feature selection (FS), BFP inherits complexities such as NP-hardness, high nonlinearity and multimodality, as it involves finding the most important features within a high-dimensional space and eliminating unnecessary ones. First, a weighted-sum BFP approach was adopted, where contradictory metrics are combined into a single composite goal optimised using a fuzzy adaptive global learning binary imperialist competitive algorithm (FA-GL-BICA)–based FS-NN framework. Next, the weighted-sum approach was extended using a greater number of contradictory metrics and higher-dimensional search spaces. A FAGLSUD–based FS-NN framework was proposed to address the uncertainties associated with this extended BFP problem. This framework may provide medical practitioners and users with a more accurate and stable estimation of body fat compared with other hybrid and state-of-the-art models. Finally, a multi-objective BFP approach to simultaneously analyse the conflicts between accuracy, stability and dimensionality was investigated. In this context, a multi-objective FAGLSUD (MOFAGLSUD) framework extended FAGLSUD by introducing a diversity-aware index called spatial spread deviation associated with ranking. This extension demonstrated a well-distributed global optimal Pareto set of low-computational, accurate and stable trade-off solutions, enabling medical practitioners and users to make informed decisions about fat deposits in vital organs based on their preferences. This exploration emphasises the complexity of BFP rooted in NP-hard FS. Our study showcases innovative methodologies to address this healthcare challenge. In summary, this research involved three major studies aimed at advancing our theoretical understanding of fuzzy computing, EC search strategies based on subpopulations and NNs to solve benchmark, engineering and healthcare problems with varying levels of uncertainty and fill the research gaps in these areas. Future studies can adapt these proposed methods to solve other real-world problems while dealing with new uncertainty levels.
- Subject
- fuzzy adaptive evolutionary computation; multi-objective optimisation; multi-objective bargaining game theory; sustainable, economical, and earthquake-resistant reinforced concrete cantilever retaining wall design problem; body fat prediction problem; thesis by publication
- Identifier
- http://hdl.handle.net/1959.13/1510585
- Identifier
- uon:56431
- Rights
- Copyright 2023 Farshid Keivanian
- Language
- eng
- Full Text
- Hits: 278
- Visitors: 293
- Downloads: 21
| Thumbnail | File | Description | Size | Format | |||
|---|---|---|---|---|---|---|---|
| View Details Download | ATTACHMENT01 | Thesis | 21 MB | Adobe Acrobat PDF | View Details Download | ||
| View Details Download | ATTACHMENT02 | Abstract | 463 KB | Adobe Acrobat PDF | View Details Download |