Refinery Supply Chain Optimization Under Uncertainties

Decision making for supply chain under uncertainties as well as estimation and analysis of these uncertainties are significant to supply chain management both in term of theory and practice, since uncertainties are everywhere within a supply chain. This dissertation mainly focuses on two kinds of problems based on the review of supply chain management: first, making decision for refinery supply chain operation under uncertain demand, incomplete information of supply and transportation and variable production yields; second, estimating the distribution of uncertain production yields, finding the important factors causing the uncertainty and solving the subsequent delay time estimation problem. The details are listed as follows:1) A two-stage discriminating framework for refinery supply chain operation optimization under uncertain demand and incomplete information shared with suppliers and transportation companies. A new work flow is suggested by this framework, which rejects to make sacrifice initiatively as part A facing the supplies or transportation companies. Decision maker aims to maximize the profit in the condition of reaching the ideal customer satisfaction level. The designed framework adopts simulation based optimization. The decisions of optimization can be measured and the expectation value of customer satisfaction level can also be reported by the simulation model based on if-then rules. The decisions can be optimized by the validation loop of the first stage and the negotiation loop of the second stage.2) The variation of production yields during production modes changeover is described by Markov chain. The supply chain operation optimization problem under yields uncertainty is modeled by chance constrained programming. A method is proposed to solve the problem that the probability distributions of each period are unknown. This method can be extended to solve alike problems that stochastic variables coupled with some 0-1 decision variables.3) To show how to estimate the distribution type and parameter values of stochastic variables, we describe the methods to estimate and trace the standard production yields based on the data generated from a schedule simulation model.Furthermore, we use x~2 -fit validation to judge the truth of assumption on the type ofprobability distribution.4) In order to reduce the costs completely, we suggest that the uncertainties should be controlled at first. An analysis flow is presented to find the cause of uncertainties, which combines classification and visualization. To the widely existed delay time between measurement variables, the data is transformed and multiple classifiers system is introduced. The results become more credible by selecting algorithms and parameters for free and combining visualization analysis.5) Due to the requirement of delay time estimation during analyzing the cause of uncertainties, the estimation methods are reviewed. Two estimation algorithms are then proposed as the complement of existing methods, which based on a new concept called time series trends similarity search. These two methods are particularly suitable for directly estimating the delay time between two signals from different sources on the condition that any extra stimulus signals are forbidden.At the end of this dissertation, promising future researches on refinery supply chain management are introduced based on the conclusion of this dissertation.