Approximate Dynamic Programming Applied to Biofuel Markets in the Presence of Renewable Fuel Standards

Abstract

Due to the increased attention on renewable fuel standards to mitigate the effect of greenhouse gas emissions from automobiles, the US government has started implementing new policies that require gasoline distributors to mix biofuel (such as corn-based ethanol or other similar fuels based on biomass) into gasoline. We focus on the problem of managing a biomass plant in the presence of renewable fuel standards in this thesis. We develop a model where we decide how much biomass to sell, how much biomass to produce and how much capacity to add. We solve this policy using a backward Markov Decision Process (bMDP) and Value Function Approximation (VFA) where we approximate the value functions in the latter policy using the Concave, Adaptive Value Estimator (CAVE). We show that under certain assumptions on the approximated value functions, the VFA, is a must faster algorithm and obtains a policy that achieves a pro t that is close to that of the bMDP. We survey various machinery commonly utilized in approximate dynamic programming throughout this thesis.