ABSTRACT
An adaptive deep neural network is used in an inverse system identification setting to approximate the inverse of a nonlinear plant with the aim of constituting the plant controller by copying to the latter the weights and architecture of the converging deep neural network. This deep learning (DL) approach to the adaptive inverse control (AIC) problem is shown to outperform the adaptive filtering techniques and algorithms normally used in adaptive control, especially when in nonlinear plants. The deeper the controller, the better the inverse function approximation, provided that the nonlinear plant has an inverse and that this inverse can be approximated. Simulation results prove the feasibility of this DL-based adaptive inverse control scheme. The DL-based AIC system is robust to nonlinear plant parameter changes in that the plant output reassumes the value of the reference signal considerably faster than with the adaptive filter counterpart of the deep neural network. The settling and rise times of the step response are shown to improve in the DL-based AIC system.
Subject(s)
Deep Learning , Nonlinear Dynamics , Algorithms , Computer Simulation , Feedback , Neural Networks, ComputerABSTRACT
Due to prior knowledge being often unavailable in practice, a multi-block strategy totally based on data-driven analytics is an appropriate alternative for plant-wide processes. However, most recent multi-block methods are relatively vague or insufficient for dividing up the process space and lack the comprehensive fault information for quality-related monitoring. This work intends to develop a more reasonable multi-block method and demonstrate the negative impacts of quality-unrelated variables. Both motivations are entirely dependent on the correlation between variables. A major innovation is to determine those independent or related sets of variables, and to provide a more precise indication for those quality-related faults. Sub-blocks with related variables are each modeled by the KPCA, and the rest of the independent variables are treated as an input for a SVDD model. Finally, all of the statistical indicators are aggregated into a single statistic through Bayesian inference. The benefits of the proposed multi-block scheme (MKPCA-SVDD) are elaborated on in detail using numerical simulation, TE benchmark and industrial p-xylene oxidation process.