A new model reduction strategy for nonlinear distributed parameter systems

A new model reduction strategy is proposed for nonlinear distributed parameter systems which the first principle modeling described by partial differential equation have dominant linear terms. Spectral method, combining balanced truncation model reduction and neural networks (NN) are used to construct the low-dimensional approximation of nonlinear distributed parameter systems. The strategy amounts to finding computationally efficient and stable substitute models for the nonlinear distributed parameter systems. The potential of the strategy is illustrated using spatio-temporal temperature evolution of catalytic rod as an example and The simulation shows that the present strategy is superior to spectral method directly to nonlinear DPS.