Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing this collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number. PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
REPORT DATE (DD-MM-YYYYJ
SUPPLEMENTARY NOTES
ABSTRACTThis research is intended to extend the knowledge base concerning logistical network modeling and design. Basic research techniques were developed to begin to model logistical networks within a hybrid simulation/analytic framework. The first step in this process is to develop robust approximations for portions of large-scale simulation models. This research examined the novel idea of utilizing neural networks as a meta-modeling technique to replace specific aspects of a simulation. This work started with the replacement of queueing stations within a logistics network. Any logistics network can be formulated as a network of material flowing through processes requiring resources. A new methodology was developed for forming approximations and improved approximators for queueing stations within a logistics network. The motivation for developing this approximation is the integration of such approximation in hybrid simulation/analytic methods for evaluating logistic networks. Future work should investigate the performance of these approximations within the larger logistical network context. We demonstrate the use of neural networks to close the gap between the output of Whitt's GI/G/m approximation and the results obtained via simulations of a GI/G/m queue. Once the neural network has been trained, we can feed in the parameters and resulting output of Whitt's GI/G/m approximation, along with additional information describing the arrival and service distributions of the queue, and yield an acceptably accurate approximation for the expected wait time in a GIIG/m queue. These approximations can be embedded in parametric decomposition algorithms for the logistics network as a whole. The motivation for developing this approximation is the integration of such approximation in hybrid simulation/analytic methods for evaluating logistic networks. The neural network approximation developed easily beats Whitt's approximation for correlated arrivals, but further investigation is needed to assure a robust approximation. Future work should investigate the performance of these app...