Automatic reconstruction of performance indicators from support vector based search space models in distributed real power planning scenarios

Abstract

In order to allow for a transition of the current central market and network structure of todays electricity grid to a decentralized smart grid, an efficient management of numerous distributed energy resources will become more and more indispensable. The integration of controllable but distributed and individually configured small devices entails new challenges for the design of future control structures. Pooling and joint planning are proposed concepts for future control schemes. Each energy unit may offer an individual set of feasible schedules for some future planning horizon. If a coalition of distributed energy units is supposed to plan the provision of a wanted joint real power schedule, a many-objective problem arises that demands different evaluation criteria to weight alternative schedules against each other. Each schedule causes individual costs and may be described by individual indicators like reliability, environmental impact or robustness criteria for planning. Communication and integration into planning algorithms without having to know the individual technical setting (i.e. model, operational state, or individual constraints and their modeling) of any energy unit is the aim of this paper. We propose a combination of two approaches: surrogate modeling and a decoder approach for constraint-handling that integrates indicators that are annotated to schedules. We show the appropriateness of this approach by scrutinizing the error when reconstructing the values from the black-box model.