More on Conditioned Sampling in the Simulation of Stochastic Networks

Abstract

The technique of conditioned sampling has been shown to improve simulation efficiency in the estimation of stochastic activity network duration c.d.f.'s. This paper describes a.method for generalizing the conditioned sampling approach from its current use of product-form estimators to the use of product/convolution-form estimators. Estimators of the latter type are constructed and demonstrated to require fewer samples per realization (hence increased estimation accuracy) in almost all networks. An algorithm for estimator construction is presented and proven to apply to any given activity network. It is also shown that the derived product/convolution-form estimators may require a precedence structure within the sampling sequence which creates their corresponding realizations.