A neural network approach to the vehicle dispatching problem

The problem of the dynamic dispatching of vehicles and crews is examined. The authors describe a neural network model that uses examples of expert dispatcher decisions for training. Given a set of attributes that describe typical dispatching situations and the decision that has been taken (i.e. the vehicle chosen for servicing the new request), a neural network is trained to assess the relative quality of each candidate vehicle and select the vehicle with the highest quality. Various ways of encoding dispatching situations for a neural network are described, and preliminary results with systematic dispatching rules are reported.<<ETX>>