Evolutional Modeling for Complex Nonlinear Dynamic Process Based on UKF & NN in Subspace Approximation

The modeling of complex chemical process is of great significance for determining the optimal parameters.Artificial neural networks(ANNs)have proved themselves to be very useful in various modeling applications,because they can represent complex mapping functions.However,the ANNs model normally represent a static relation,can't describe the dynamic properties of the evolutional chemical process.This study the static ANNs model was regarded as the approximating model of the chemical process respect to the operational parameters in subspace.To make the static model can accurately describe the dynamic properties in real time,the Unscented Kalman Filtering(UKF)algorithm instead of the Extended Kalman Filter(EKF)algorithm was used to update ANNs weights for dynamic chemical process modeling,because the UKF performance superior to that of the EKF in computational complexity and precision.The proposed method was applied to approximate the nonlinear dynamic Hydrocyanic acid(HCN)process,numerical simulations showed that the proposed method was good at modeling the HCN process in high-precision.Therefore,the proposed method provided a new solution to getting the evolutional model of the complex nonlinear dynamic process.