Neural Networks And Fuzzy Control for Bulk Terminal Operating Management
From Proceeding of Summer Computer Simulation Conference,SCSC95, Ottawa, July 24-26 1994
Pietro Giribone, Agostino G.Bruzzone

---

INTRODUCTION

The application of ANN in management has always created problems due to the impossibility of understanding the proposed decision process.
Such a complication limits the use of the ANN techniques in this type of application, thus it is inevitable that the simulation must be used to validate the developed logic scheme.
In this paper, the authors propose an integrated FL and ANN system which basically aims at defining operative planning activities for a Bulk Terminal.
Management depends on market and terminal operating conditions as well as the choice of managerial criteria imposed by the manager using special fuzzy variables. Therefore, the neural network is trained to adjust the fuzzy parameters to develop a planning scheme that complies with the pre-defined targets, starting from a clearly defined operating situation.
Once the neural network suggests how to adjust the control system, the output variables are fuzzified to obtain parameters which can provide an intelligible explanation of the decision process implemented by that neural network and that also provide an immediate logical validation of the choices proposed to the manager. Therefore, the key issues discussed in this paper refer to the integration of the two techniques to guarantee correct learning of physical system operation to obtain an efficient and precise planning system and to produce a system that can self- handle the decision flow even when operating with neural networks.
The decision to use neural networks as a decision support tool is based on the lack of knowledge relative to the correlation between the multiple independent input conditions of the bulk terminal system with the yard situation.
The difficulty in building meta-models which can correlate these variables is due to the type of system being examined which in fact cannot be predicted since it is chaotic.
The chaotic behaviour of the process could be identified by utilising methods which are currently implemented to analyse chaotic systems, such as the construction of the phase plan in addition to an analysis of its sections and bifurcation diagrams.
These techniques have shown that plant operating methods are implemented in fact during the chaotic condition. Therefore, to perform a predictive analysis and then build a control and decision support system, it was decided to use neural networks to study the time series produced by the simulation. Thus, the paper focuses on the analysis of the specific application for a plant that is currently in the advanced design stage and which should be built on the island of Sardinia (Italy). However, the method proposed can be applied to similar problems related to bulk port terminals. It also refers to similar problems developed by the authors relative to port terminals for containers and different types of cargo.