Neural sensitivity methods for the optimization of queueing systems

This paper explains the concept of sensitivity and provides basic information on how neural networks can be used to derive sensitivity functions for the purpose of optimizing the performance of queueing systems. Optimization is the process of determining the optimal solution of a given design or control problem, i.e., the best possible solution in the sense of some prespecified decision criterion. The application of neural sensitivity analysis may provide useful information to system designers, and system control and management functions in determining the cause of the deviation of system performance and provide a controlling basis for them in order to operate the system more efficiently. Sensitivity analysis can be applied to a trained neural network to extract useful information which may be difficult to obtain through traditional analytical methods. Thus, sensitivity analysis using a neural network is based on the study of a trained neural network to obtain useful information about a system other than what can be obtained through conventional recall. Some examples on how neural sensitivity functions can be applied to the optimization of queueing systems are presented in this paper.