A stochastic neural network for resource constrained scheduling

Interest in the use of neural networks for the modelling and solution of problems of optimization, and particularly problems of combinatorics, continues to grow at a substantial pace. This new methodology has both some unique advantages as well as certain drawbacks. In this paper, the authors address the results obtained thus far in an ongoing research effort devoted to examination of the use of neural networks for the modeling and solution of one subclass of a particularly important category of problems: those of resource constrainted scheduling. To accomplish this, the resource constrained scheduling problem of interest is first decomposed into a series of multidimensional knapsack models and an equivalent neural network model for this particular representation is established. Then, by extending the work of Hopfield and Tank, and others, the authors have been able to develop an approach that ultimately serves to solve the original problem-while avoiding, to a great extent, such common neural network difficulties as instability and local minima. The approach that has now been developed and evaluated is designed specifically for a certain type of resource constrained scheduling problem-where such problems are subject to sudden, dynamic changes.