Metamodeling of input–output relationships for complex power market models

TLDR: The proposed statistical and bounding methods approximate model behavior, and can help analysts focus Integrated Planning Model (IPM) runs on input assumptions whose results are potentially interesting but uncertain, to reduce computation time and enable incorporation of such models into multimodel decision support systems.

Abstract: We present a general methodology for approximating the input–output behavior of complex energy market models Outputs, such as costs, prices, and emissions, depend on policy, economic, and technological assumptions The proposed statistical and bounding methods approximate model behavior, and can help analysts focus Integrated Planning Model (IPM) runs on input assumptions whose results are potentially interesting but uncertain The statistical methods (Multivariate adaptive regression splines) use past run data to make predictions of IPM outputs given new input data sets Those methods are illustrated using results from the IPM, a large-scale linear program that is widely used by the US Environmental Protection Agency and industry to simulate the behavior of the US power market Meanwhile, bounding methods use mathematical properties of linear programs, in addition to past run data, to bound outputs for new inputs These methods can be used to approximate the outputs of any convex optimization model of power systems to reduce computation time and enable incorporation of such models into multimodel decision support systems The bounding approach is demonstrated by an application to the COMPETES model of the northwest European power market