Goran Pavković

The fuzzy set theory approach to the vehicle routing problem when demand at nodes is uncertain

Abstract The problem of vehicle routing when demand at the nodes is uncertain is considered. The quantities to be picked up at the nodes are assumed to be only approximately known. A network with one depot from which vehicles depart and to which they return after completing their service is considered. The paper develops a model to design vehicle routing when demand at the nodes is uncertain. The model is based on the heuristic “sweeping” algorithm, the rules of fuzzy arithmetic and fuzzy logic.

A simulated annealing technique approach to the vehicle routing problem in the case of stochastic demand

Research work dealing with the vehicle routing problem has not paid adequate attention to the cases where the demand for services at certain nodes is a random variable. This paper develops an approach to the vehicle routing problem for the case of stochastic demand. The approach is based on the simulated annealing technique. It is illustrated with a numerical example.

The potential for using fuzzy set theory in airline network design

This paper develops a model to design an airline network and determine flight frequencies. The basic input data are the estimated numbers of passengers between pairs of cities. It is often impossible to estimate these numbers with enough precision (i.e. there is a degree of uncertainty surrounding the numbers of passengers between pairs of cities). Uncertainty is also frequent in the estimation of future airline carrier costs. The fuzzy set theory is an extremely convenient mathematical device with which to solve problems containing uncertainty, sujectivity, ambiguity and indetermination. Without aspiring to revolutionize airline netqork design, the basic goal of this paper is to present the fuzzy set tools and their potential application to a generic airline network design problem. The model developed is based on the fuzzy set theory. The first part of the model uses fuzzy logic to determine the route candidates to make the flights, and the second part of the model is based on fuzzy linear programming to determine flight frequencies on the route candidates. The model is supported by numerical examples.

A neural network approach to the vessel dispatching problem

The paper discusses the process of loading, transport and unloading of gravel by inland water transportation. At the loading port, the problem that needs to be solved is the assignment of load barges to pusher tugs for the planned period of one day. However, disturbances of planned schedules are very common. Whenever a disturbance in a daily schedule appears, the dispatcher urgently attempts to mitigate negative effects resulting from the disturbance. Real-time operations limit the amount of time that dispatchers in charge of traffic control have to make decisions and increase the level of stress associated with quick and adequate response. This paper aims to demonstrate the feasibility of a dispatch decision support system that could decrease the work load for the dispatcher and improve the quality of decisions. The proposed neural network with the ability to adapt or learn from examples of decisions can simulate the dispatchers decision process.

A FUZZY APPROACH TO THE VEHICLE ASSIGNMENT PROBLEM.

Transportation companies receive a great number of requests every day from clients wanting to send goods to different destinations. Their fleets often consist of several different types of vehicles. Dispatchers assign vehicles to transportation requests relying primarily on their experience and intuition. When assigning a certain type of vehicle to a transportation request, the dispatcher must bear in mind a large number of different factors. This paper considers the vehicle assignment problem when vehicle dispatching is done “today for tomorrow”. The paper develops a model based on fuzzy logic that assigns different types of vehicles to planned transportation tasks. The model is tested on a real numerical example. The results of testing the model indicate that it outperforms the work of an experienced dispatcher.

An application of neurofuzzy modeling: The vehicle assignment problem

When assigning vehicles to transportation requests, dispatchers usually have built-in fuzzy rules which they use to assign a given amount of freight to be sent to a given distance a given vehicle. Fuzzy systems equipped with learning capabilities can be trained to control complex processes like the dispatcher. They usually begin with a few very crude rules obtained from the dispatcher. Or they may work out the rules from the observed dispatchers behavior. In this paper, a neural network is used to refine and adapt the fuzzy system to achieve better performance. As a result of the study, on a real set of numerical data, it was shown that the proposed feedforward adaptive neural networks with supervised learning capabilities can be used to tune the initial fuzzy systems.

THE MIXED FLEET STOCHASTIC VEHICLE ROUTING PROBLEM.

The mixed fleet stochastic vehicle routing problem is considered in the paper. It is assumed that operations are performed by vehicles of different types. The model developed is based on the “route first — cluster second” approach. A heuristic algorithm based on space-filling curves is used to produce a giant Travelling Salesman tour. The giant tour is divided into smaller parts using the generalized Floyd algorithm. The final set of routes may be chosen after making a suitable multi-attribute decision making analysis.

A NEURAL NETWORK APPROACH TO MITIGATION OF VEHICLE SCHEDULE DISTURBANCES.

In all transportation industries, when planned schedules are modified by delays, breakdowns or peaks of demand, ‘real‐time’ human decision making takes place. Dispatchers in charge of traffic control usually are allowed a very short time to make decisions that should mitigate the negative effects resulting from the disturbance. This paper aims to demonstrate the feasibility of a dispatch decision support system that could decrease the work load for the dispatcher and improve the quality of decisions. The proposed neural network, with the ability to adapt or learn from examples of decisions, can simulate the dispatchers decision process.