Can we improve the freight network solutions of a large retailer by using probability distributions to model demand?

Abstract

The scope of the present thesis was to investigate whether we can improve the quality of our solutions for a retailer's transportation network by employing probability distributions to model freight demand. The quality of the solutions was assessed in the basis of transportation cost and service level. The research work was based on a large scale LP optimization tool, called Stochastic Flow Analyzer. SFA attempts to match transportation capacity to demand by optimally assigning the network flows to either the private fleet or to for-hire carriers. Using SFA we designed a set of annual plans for the retailer's network. Each plan assigned a number of drivers at each of the domiciles. In order to evaluate the performance of each plan we simulated the demand for 26 weeks. We showed that there is a trade-off between transportation cost and service level. Stochastic solutions were performing better in terms of cost but in the same time deterministic solutions were achieving higher service levels. We concluded that the best probabilistic solution was the one where we used the empirical distribution function to model lane demand. However, the deterministic solution with refined fixed volumes offered 5% higher service level, it is easier to implement in real life operations and the increase in transportation cost was less than 4%.