Projecting Generation Decisions Induced by a Stochastic Program on a Family of Supply Curve Functions

We propose to post-process the results of a scenario based stochastic program by projecting its decisions on a parameterized space of policies. By doing so the risk of overfitting to the set of scenarios used in the stochastic program is reduced. A proper choice of the structure of the space of policies allows one to exploit them in the context of novel scenarios, be it for Monte-Carlo based value estimation or for use in real-life conditions. These ideas are presented in the context of planning the exploitation of electric energy resources or for evaluating the economic value of a portfolio of assets.