A clustering-based approach for fast modeling of memristive devices for energy storage

We propose in this paper a new characterization of two-terminal nonlinear devices, whose complex dynamic cannot be modeled effectively by quantities related to voltages and currents only. The new model is determined considering the general scheme of a nonlinear device based on a memristor, then constituted by linear components and ideal diodes yielding to a decomposition in locally linear models. In the general case, the scheme of a device is not known, so we propose a novel procedure based on clustering for determining the linear regions present in the device as well as the order of its dynamic model. The proposed approach is particularly suited in applications where time-sensitive decisions are essential, especially in real-time smart grid management and battery modeling.