Hysteresis Compensation by Deep Learning Algorithms

We propose a Deep Learning (DL) algorithm, which connected to a Preisach-memory updating rule, allows to describe rate-independent memory phenomena. Memory rules could be realized through the employment of Play or Stop operators, which, as known, show counter- or clockwise loops. The former are suitable to represent the classical operators of Preisach type, while the latter, showing clockwise cycles, is well behaved to describe the inverse or compensator operator. The paper is aimed to propose a unified approach that, through a specific structure of the hysteresis model (splitting the memory rules from the rest) is able to optimize performances of a Deep Learning (DL) algorithm and describe both hysteresis process and its compensator with good accuracy and computational efficiency. The approach would pander the actual stream that locates the need to extract the basic features of the phenomenon in connection to a neural algorithm in order to improve the modeling performances of the whole model, specifically aimed to implement compensator operators for e.g. control tasks.