Yonghong Tan

Neural-network-based d-step-ahead predictors for nonlinear systems with time delay.

Abstract Three neural-network-based d -step-ahead prediction strategies for nonlinear processes with time-delay are presented here. They are, respectively, a recursive d -step-ahead neural predictor, a non-recursive d -step-ahead neural predictor, and a Smith-type neural predictor that can also be used for d -step-ahead prediction. Both the recursive and the non-recursive predictors have been extended to the case of long-range prediction. It is known that both the non-recursive d -step-ahead predictor and the Smith predictor have been applied to linear processes with time-delay. It would be useful to extend these strategies to nonlinear systems. Therefore, an extension of both the non-recursive d -step-ahead predictor and the Smith predictor principle to the nonlinear case is presented here. These nonlinear predictors can be constructed by using neural networks. This is very useful for the time-delay compensation of nonlinear processes. The neural-network-based predictors have been applied to the predictions of some nonlinear processes. A comparison based on simulation is given in this paper. Finally, the proposed neural-network-based predictors are used to predict the manifold pressure process in an automotive engine. The predictive result of the corresponding first-principles model-based nonlinear predictor is also illustrated for comparison. The experimental results show that the neural-network-based predictive methods have obtained better performance.

A two-stage genetic algorithm for automatic clustering

In this paper, a two-stage genetic clustering algorithm (TGCA) is proposed. This algorithm can automatically determine the proper number of clusters and the proper partition from a given data set. The two-stage selection and mutation operations are implemented to exploit the search capability of the algorithm by changing the probabilities of selection and mutation according to the consistence of the number of clusters in the population. First, the TGCA focuses on the search of the best number of clusters, and then gradually transfers towards finding the globally optimal cluster centers. Furthermore, a maximum attribute range partition approach is used in the population initialization so as to overcome the sensitivity of clustering algorithms to initial partitions. Finally, the efficiency of TGCA has been extensively compared with several automatic clustering algorithms, including hierarchical agglomerative k-means, automatic spectral algorithm and a standard genetic k-means clustering algorithm (SGKC). Experimental results on four artificial and seven real-life data sets show that the TGCA has derived better performance on the search of the cluster numbers and higher accuracy on clustering problems.

Neural-networks-based nonlinear dynamic modeling for automotive engines

This paper presents a procedure for using neural networks to identify the nonlinear dynamic model of the intake manifold and the throttle body processes in an automotive engine. A dynamic neural network called external recurrent neural network, is used for dynamic mapping and model construction. Dynamic Levenberg–Marquardt algorithm is then applied to the weight-estimation problem. Modeling results indicate that the neural-network-based models have a rather simple structure. Early results also confirm that the neural-network-based modeling of the manifold dynamics can result in a model that is comparable if not better than the first-principle-based models. In addition, it was verified that the neural model has good generalization capabilities.

Modeling Hysteresis and Its Inverse Model Using Neural Networks Based on Expanded Input Space Method

A neural network-based approach of identification for hysteresis and its inverse model is proposed. In this method, a hysteretic operator is proposed to extract the change tendency of hysteresis. Then, an expanded input space is constructed to transform the multivalued mapping into one-to-one mapping so that the neural networks are capable of implementing identification for hysteresis. Similar to the method of modeling hystereis, an inverse hyteretic operator is proposed to construct an inverse model for hysteresis. Then the experimental results are presented to illustrate the potential of the proposed modeling technique.

Nonlinear one-step-ahead control using neural networks: control strategy and stability design

A nonlinear one-step-ahead control strategy based on a neural network model is proposed for nonlinear SISO processes. The neural network used for controller design is a feedforward network with external recurrent terms. The training of the neural network model is implemented by using a recursive least-squares (RLS)-based algorithm. Considering the case of the nonlinear processes with time delay, the extension of the mentioned neural control scheme to d-step-ahead predictive neural control is proposed to compensate the influence of the time-delay. Then the stability analysis of the neural-network-based one-step-ahead control system is presented based on Lyapunov theory. From the stability investigation, the stability condition for the neural control system is obtained. The method is illustrated with some simulated examples, including the control of a continuous stirred tank reactor (CSTR).

Impulse noise removal by a global-local noise detector and adaptive median filter

Noise detection-based median filters have been widely adopted to reduce impulse noise. However, the number of misclassified pixels is obviously increased at high noise density. In order to overcome this drawback, a global-local noise detector is proposed in this paper. Based on the noise detection, an adaptive median algorithm is presented. Simulation results show that the new filter can effectively reduce impulse noise and preserve more details of original images.

Nonlinear Modeling and Decoupling Control of XY Micropositioning Stages With Piezoelectric Actuators

In this paper, a modeling method of XY micropositioning stage with piezoelectric actuators is proposed. In the modeling scheme, a sandwich model consists of both input and output linear submodels, and an embedded neural-network-based hysteresis submodel is used to describe the motion behavior of each axis of the stage. Moreover, a neural-network-based submodel is constructed to describe the nonlinear interactive dynamics caused by the movement of another axis. Then, a tracking control scheme combined with a nonlinear decoupling control is proposed to compensate for the effect of the interactions between axes and track the reference trajectory. Then, the robust design method for the tracking and decoupling control is discussed. Finally, the experimental results on an XY micropositioning stage are presented.

Recursive Identification of Sandwich Systems With Dead Zone and Application

In this paper, a recursive identification approach for a class of nonlinear systems called sandwich systems with the dead zone is proposed. In order to handle the effect of the dead zone, several switch functions are introduced into the model based on the so-called key term separation principle. Hence, the sandwich systems with the dead zone can be transformed into a special model where all the model parameters are separated. Then, a modified recursive general identification algorithm (MRGIA) is applied to the parameter-estimations of the obtained model. Moreover, the convergence of the algorithm for such systems will be discussed. Finally, a simulation example is presented, and the experimental results on an X-Y moving positioning stage are illustrated.

Recursive identification algorithm for dynamic systems with output backlash and its convergence

Recursive identification algorithm for dynamic systems with output backlash and its convergence This paper proposes a recursive identification method for systems with output backlash that can be described by a pseudo-Wiener model. In this method, a novel description of the nonlinear part of the system, i.e., backlash, is developed. In this case, the nonlinear system is decomposed into a piecewise linearized model. Then, a modified recursive general identification algorithm (MRGIA) is employed to estimate the parameters of the proposed model. Furthermore, the convergence of the MRGIA for the pseudo-Wiener system with backlash is analysed. Finally, a numerical example is presented.

Adaptive output feedback control of systems preceded by the Preisach-type hysteresis

An adaptive output feedback controller is presented for a class of single-input-single-output (SISO) nonlinear systems preceded by an unknown hysteresis nonlinearity represented by the Preisach model. First, a novel model is developed to represent the hysteresis characteristic in order to handle the case where the hysteresis output is not directly measured. The model is motivated by the Preisach model but implemented by the neural networks (NN). Therefore, it is easily used for controller design. Then, a radius-basis-functional-neural-networks (RBF NN) adaptive controller based on the model estimation is presented by combining the high-gain state observer. The updated laws and control laws of the controller are derived from Lyapunov stability theorem, so that the ultimate boundedness of the closed-loop system is guaranteed. At last, an example is used to verify the effectiveness of the controller.