University of Thessaly (UTH)
Exploration and development of methodologies for complex systems modelling and decision-making support using advanced fuzzy logic and cognitive maps techniques
Abstract
dc:descriptionModelling and simulating complex dynamic systems have always been a demanding and challenging task especially when considering the formulation of strategies and the development of certain policies in the context of decision-making, with respect to sustainability planning. In this direction, Fuzzy Cognitive Maps (FCMs) which constitute a powerful Neuro-fuzzy quasi-quantitative modelling methodology, are the right candidate to address both complexity and non-linearity with respect to complex systems’ behavior. Additionally, they can handle the lack of reliable quantitative data along with the presence of uncertain information, in terms of knowledge representation and reasoning during the simulation process. Moreover, they seem to be a promising tool to incorporate human experience and other existing knowledge as well as new aggregation participatory approaches for aggregating numerous individual models designed by experts or stakeholders, that will produce more reliable models for decision-making. Afterall, FCMs have found extreme and extensive applicability in multiple research domains which is proved by the enormous list of publications in the literature, over the last decade.In the context of policy-making, energy demand forecasting is another significant and challenging task for tailoring efficient policies towards energy sustainability management and planning. Thus, the selection of accurate forecasting techniques is important for policy-makers to apply precise forecasts highly essential for supporting them in choosing the right strategies. Most of the forecasting methods until recently, are characterized by structure complexity, while they are slow and difficult to use by inexperienced Artificial Intelligence (AI) users. In this direction, FCMs, characterized by their learning capabilities, are emerged in the recent literature as simple, flexible and highly accurate soft computing methods, adequate to cope with relatively small datasets in data handling, and addressing fuzziness in a certain degree. Combining them with other efficient AI methods makes it a promising research direction in providing accurate predictions in the context of energy policy and demand forecasting.This dissertation contributes to the subjects of FCM modeling, aggregation and scenario analysis of complex systems, through the introduction of a FCM development framework which can be used to perform an analysis of the systems’ dynamics for efficient strategic social, economic and environmental sustainability planning. Two research directions are investigated within the framework of FCM development. They particularly concern the aggregation of FCMs proposed by multiple and numerous experts/stakeholders with a focus on scenario analysis, and a prediction algorithm deploying FCMs for demand forecasting purposes, with a focus on sustainable socio-economic and environmental planning.The first direction focuses on a new aggregation method for FCM modeling based on learning Ordered Weighted Averaging (OWA) operators in the aggregation of weights. It aims at aggregating knowledge from multiple sources, thus improving the reliability of the examined model and eliminating possible erroneous beliefs provided by particular participants. This choice of the OWA aggregation method was motivated by its efficiency to aggregate a sizeable amount of individual FCMs designed by experts and/or participants, considering their confidentiality. In this context, a new and easy to use java-based software tool is implemented for the automatic aggregation of individual FCMs based on the proposed OWA learning aggregation method. Additionally, a powerful and easy to use, web-based tool, namely FCMWizard, is incorporated to efficiently perform automatic FCM modeling and scenario analysis under different configurations in diverse domains. It is proved to have a supporting role for policy makers and governments in the analysis of complex real-life problems, predicting their behavior as well as eliciting accurate outcomes of proposed policies that deal with sustainable social-economic-environmental development strategies.The second direction is oriented in a new and highly accurate demand forecasting method with generalization capabilities for time series prediction. It is comprised of an ensemble of AI-oriented models, based on evolutionary FCMs, artificial neural networks (ANNs), and their hybrid structure (FCM-ANN). Experimental evaluation based on a case study regarding natural gas load forecasting, reveals that the proposed architecture attains better accuracies and an improved overall performance against other individual AI and soft computing forecasting models, for the examined case study. At the same time, and for the same dataset, a promising Adaptive Neuro-Fuzzy Inference System (ANFIS) model architecture is investigated for consumption demand forecasting purposes. The proposed ANFIS approach reveals its remarkable prediction performance, in terms of its ability to tackle fuzziness in data handling, flexibility in large datasets, easiness of use and low execution time requirements.On the whole, the innovations offered in this dissertation are devoted to the introduction of the aforementioned methodologies and newly developed FCM-based tools for modelling, aggregation and prediction tasks. Accordingly, they can help policy-makers and other regulatory authorities to identify suitable strategies towards efficient social, economic and environmental sustainability planning.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papageorgiou, Konstantinos
- Παπαγεωργίου, Κωνσταντίνος
Subjects
dc:subject × 14- Ασαφή γνωστικά δίκτυα
- Λήψη αποφάσεων
- Πολύπλοκα συστήματα
- Γνωστικά δίκτυα
- Fuzzy cognitive maps
- Decision making
- Complex systems
- Cognitive maps
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
dc:rights- Statement dc:rights
-
- BY_NC_ND_4_0
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/49365