University of West Attica
Ασαφή, στοχαστικά και εξελικτικά μοντέλα πρόβλεψης και βελτιστοποίησης
Abstract
dc:descriptionThis doctoral thesis explores three distinct research directions: a) the development of methodologies based on evolutionary computation, aimed at enhancing the approximation capabilities of fuzzy transforms and fuzzy projections, b) the systematic investigation of uncertainty emerging in the context of occupancy estimation models in smart buildings, and c) the development of integrated methodologies that combine wavelets, neuro-fuzzy approaches, and surrogate techniques, focusing on addressing the problem of time series prediction, and in particular the accurate prediction of environmental variables. Novel techniques and methodologies are proposed, which, when combined, yield solutions to the above-mentioned research problems.In terms of the first direction, the present research emerges as a result of the complete lack of systematic methods for determining optimal fuzzy partitions, and aims to develop hybrid models of fuzzy transforms and fuzzy projections, filling the existing gap. The problem consists of searching for an optimal fuzzy partition, by determining the locations of the fuzzy sets that generate it. Through the development of an appropriate evolutionary methodology based on differential evolution, the problem is approached in a constrained optimisation context. These constraints concern those properties necessary for the models to be well-defined. Hence, an appropriate feasible set consisting of fuzzy partitions is constructed that promotes a possible solution to the problem of fuzzy systems which manifests as a trade-off between accuracy and interpretability. The developed hybrid — evolutionary-fuzzy — models are evaluated in terms of their ability to approximate unknown functions, based on a set of non-uniform observations. The second research direction is related to the inherent uncertainty that emerges when studying stochastic models in smart buildings. With the need for a systematic quantification of the emerging uncertainty, the research is decomposed into two sub-problems: a) the characterisation of the uncertainty resulting due to insufficient knowledge in dynamic mass balance models, and b) the accurate estimation of occupancy sequences, in the face of such uncertainty. Two novel approaches are developed as solutions to the first problem, combining stochastic differential equations with Bayesian inference, as well as fuzzy techniques. The estimation of occupancy sequences is approached via hidden Markov models, introducing the Viterbi algorithm in combination with stochastic differential equation models.Regarding the last research direction, complete methodologies are formulated, which transform intelligent techniques into a tool for understanding the variability of natural phenomena, and determining temporal patterns extending into the future. Specifically, two distinct integrated methodologies are formulated, which provide high predictive accuracy on real time series data of wind speed, as well as maximum wave height. Novel neuro-fuzzy models, as well as new data-driven methods for training these models, are proposed. Finally, algorithms for determining optimal features are studied; when the resulting features are taken into consideration as model inputs, the highest generalisation performance of the predictive methodologies is achieved.
Degree
thesis:*- Grantor dc:publisher
- University of West Attica
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Korkidis, Panagiotis
- Κορκίδης, Παναγιώτης
Subjects
dc:subject × 22- Ασαφή συστήματα
- Στοχαστικά συστήματα
- Εξελικτικοί αλγόριθμοι
- Βελτιστοποίηση
- Τεχνητή νοημοσύνη
- Υπολογιστική νοημοσύνη
- Fuzzy systems
- Stochastic systems
- Evolutionary algorithms
- Optimisation
- Artificial intelligence
- Computational intelligence
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Άλλες Επιστήμες Μηχανικού και Τεχνολογίες
- Επιστήμες Μηχανικού και Τεχνολογία
- Μηχανική και Τεχνολογίες, άλλοι τομείς
- Computer and Information Sciences
- Natural Sciences
- Other Engineering and Technologies
- Engineering and Technology
- Engineering and Technologies, miscellaneous
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62303