Generation of synthetic consumer orders for urban retail delivery simulations: a Bayesian framework to produce well-conditioned samples from order data

This study proposes a robust Bayesian formulation for stochastic reproduction of well-conditioned synthetic orders for urban delivery operations. The method seeks to produce order patterns that have similar characteristics to order data with respect to frequency of repeat orders, while avoiding over-conditioning issues inherent in order resampling. It identifies a framework by which models for order distribution can be updated with incremental receipt of order data. The method is applied to a data set of delivery orders provided by an industrial sponsor for the city of Cambridge, and shows promising results with respect to order spread and repeat order patterns.