Benoit Vinsonneau

Extended global total least square approach to multiple-model identification

Abstract The paper proposes a novel global total least square procedure which has been appropriately extended for multiple-model identification of nonlinear systems. The resulting scheme which is a hybrid iterative procedure, makes repeated use of both the behavioural and classical approaches. Model parameters are optimised in order to minimise the distance between an observed time series and the simulated time series of the resulting optimised behavioural model. The developed procedure is demonstrated when applied to an arbitrary nonlinear system.

Frisch scheme identification for dynamic diagonal bilinear models

The article addresses the problem of dynamic system identification in the errors-in-variables framework for a class of discrete-time time-invariant input–output bilinear models when subjected to a white input signal. The proposed algorithm is based on an extension of the bias-compensated least squares method and utilises the Frisch scheme equations to determine the parameter vector together with the variances of the input and output noise sequences. The appropriateness of the approach is analysed and its performance evaluated when compared to other errors-in-variables identification techniques by means of a Monte Carlo simulation. The results obtained demonstrate the accuracy of the proposed method and the performance in terms of noise robustness is also observed.

Gradient-Based Approaches for Recursive Frisch Scheme Identification

Abstract An algorithm for recursive Frisch scheme system identification of linear single-input single-output errors-in-variables systems is developed. For the update of the estimated model parameters, a recursive bias-compensating least squares algorithm, which is based on the well-known recursive least squares technique, is considered. The estimate of the output measurement noise variance is determined using a conjugate gradient method, which tracks the smallest eigenvalue of a slowly varying matrix. For the update of the input measurement noise estimate, a steepest gradient search is applied. It tracks the minimum of a model selection cost function, which is based on a set of high order Yule-Walker equations.

Regularized Structured Total Least Norm for the Identification of Bilinear Systems in the Errors-in-Variables Framework

The paper addresses the identification of time-invariant bilinear system (BS) models in the errors-in-variables (EIV) framework. The proposed scheme is based on the structured total least norm (STLN) technique extended here to handle BS. The performance of the presented approach, i.e. the bilinear STLN (BSTLN) with its further extension incorporating the Tikhonov regularization is compared to several other EIV identification techniques via an extensive Monte-Carlo simulation study. The results obtained demonstrate a considerable noise robustness and therefore the applicability of the BSTLN algorithm in the EIV framework.

Hierarchical supervisory control and plant monitoring applied to industrial furnaces incorporating a priori knowledge

The paper proposes a novel model-based hierarchical supervisory control and plant monitoring system. Key to the new approach is a model of a furnace which takes into account the physical nonlinear behavior within each zone at the local level, the inter-zone behavior at the intermediate level and overall plant performance monitoring at the higher level. Through the use of a priori knowledge in the development of the model(s) the paper stresses the fact that optimum plant performance, hence profitability at the higher level, can only be achieved through a thorough understanding of the issues at the lower and intermediate level. In particular, the paper proposes a new predictive supervisory control strategy which is able to proactively administer set points at the intermediate level to ensure that the exit temperature at the final stage is achieved.

On the errors-in-variables extended Kalman filter

Motivation for this paper stems from the need to deal with systems in which uncertainty exists on the measured signals and in the parameters of an assumed system model of known structure, and/or where the LTI assumption no longer holds. The potential of the a approach presented here is demonstrated when it is applied to a system in the presence of both an imposed mismatch in the model parameters as well the case of time-varying parameters. The effectiveness of the new EIV-EKF procedure is compared with that of an EIV Kalman filter, and evaluated on both single-input single-output and multiple-input multiple-output systems.

VIRTUAL REFERENCE FEEDBACK TUNING APPROACH TO FUZZY CONTROL SYSTEMS DEVELOPMENT

Abstract The paper proposes Takagi-Sugeno PI-fuzzy controllers (PI-FCs) dedicated to a class of integral plants. Two PI controllers are developed first by Virtual Reference Feedback Tuning to ensure the desired control system (CS) performance with respect to the reference input. Other two PI controllers are developed using the Extended Symmetrical Optimum method to ensure the desired maximum sensitivity equivalent to good CS performance with respect to the disturbance input. Then, accepting the approximate equivalence between the fuzzy controllers and the linear ones, a development method for the PI-FCs is suggested. Experimental results are included to illustrate the effectiveness of the development.