Agent-Based Phase Space Sampling of Ensembles Using Ripley's K for Homogeneity.

In future energy systems, decentralized control will require delegation of liabilities to small energy resources. Distributed energy scheduling constitutes a complex multi-level optimization task regarding the underlying high-dimensional, multi-modal and nonlinear problem structure. The multi-level issue as well as the requirement for model independent algorithm design are substantially supported by appropriate machine learning flexibility models. Generating training sets by digital twins works well for single energy units. Combining training sets from individually modeled energy units, on the other hand, results in folded distributions with unfavorable properties for training. Nevertheless, this happens to be a quite frequent use case, e.g. when an ensemble of distributed energy resources wants to harness the joint flexibility for some control task. A fully decentralized agent-based algorithm is proposed that samples from distributed twins maximizing coverage of flexibility and simultaneously minimizing the discrepancy of the sample by using Ripley’s K measure. Applicability and effectiveness are demonstrated by several simulations using established models for energy unit simulation.