Oil tanker markets modeling, analysis and forecasting using neural networks, fuzzy logic and genetic algorithms

Oil is critical to the world economy and oil shipping is considered of strategic importance by many countries. Oil tanker markets, however, are among the most volatile markets in the world. Traditional econometric and time series modeling methods have had difficulties in modeling these complicated markets due to their highly dynamic and nonlinear nature and the various non-economic factors involved. A new approach to the world oil tanker markets modeling, analysis and forecasting using state-of-the-art techniques of neural networks, fuzzy logic and genetic algorithms is presented.; Neural networks and fuzzy logic systems are nonlinear, adaptive estimators which have been motivated by the human nervous system and human thinking process, respectively. Neural networks and fuzzy systems are developed to model various oil tanker markets and then integrated to form a forecasting system. Heuristic procedures are developed based on extensive experimentation for the identification of appropriate neural networks structures and for the avoidance of under- and over-training of the neural networks. The forecasting performance of the neural networks and time series models are compared. Fuzzy decision modelers are developed to capture aggregate economic behavior of shipowners. Real-coded genetic algorithms are developed to automatically design these fuzzy systems; i.e., simultaneously design both the fuzzy membership functions and fuzzy rules. In addition, a way of incorporating human judgments and non-economic factors (expressed in terms of natural languages) into the fuzzy decision modelers is presented and tested. Results demonstrate that the approach presented is robust in modeling and forecasting complex, nonlinear economic systems and can provide valuable insights into the market behavior that traditional methods can not. It provides a promising, alternative approach to modeling and forecasting large-scale economic systems.