Neural networks for modeling of dynamic systems with hysteresis

Neural network (NN) modeling of input-output sequences for measurement transformers (MT) is presented in this paper. The neural models representing the dynamics of the objects are suited for simulation purposes and relay protection. Training the neural models is done using simulation results, obtained from classical models of MT. The potential transformer (PT) mathematical description is a nonlinear system of differential-algebraic equations (DAEs). Hysteresis and eddy current losses are considered into one core loss term, represented by voltage dependent resistance load. A new technique is proposed for solving the DAEs, combined with explicit and implicit relations given as data. The current transformer (CT) mathematical model is based on Jiles-Atherton's (J-A) theory (Jiles and Atherton, 1986) of ferromagnetic hysteresis. For solving this discontinuous system with singular points an appropriate numerical procedure is proposed.