Shaohua Tan

The min-max function differentiation and training of fuzzy neural networks

This paper discusses the Delta-rule and training of min-max neural networks by developing a differentiation theory for min-max functions, the functions containing min (wedge) and/or max (V) operations. We first prove that under certain conditions all min-max functions are continuously differentiable almost everywhere in the real number field R and derive the explicit formulas for the differentiation. These results are the basis for developing the Delta-rule for the training of min-max neural networks. The convergence of the new Delta-rule is proved theoretically using the stochastic theory, and is demonstrated with a simulation example.

Artificial-neural-network-based fast valving control in a power-generation system

Abstract This paper presents an artificial-neural-network-based controller to realize fast valving in a power-generation plant. A backpropagation algorithm is used to train the feedforward neural-network controller. The hardware implementation and the test results of the controller on a physical pilot-scale power system set-up are described in detail. Compared with some conventional fast valving methods applied to the same system, test results (both in a computer simulation and on a physical pilot-scale power system set-up) show that the neural-network controller has quite satisfactory generalisation capability, feasibility and reliability, as well as accuracy.

Artificial neural networks controlled fast valving in a power generation plant

This paper presents an artificial neural-network-based controller to realize the fast valving in a power generation plant. The backpropagation algorithm is used to train the feedforward neural networks controller. The hardware implementation and the test results of the controller on a physical pilot-scale power plant setup are described in detail. Compared with the conventional fast valving methods applied to the same system, test results both with the computer simulation and on a physical pilot-scale power plant setup demonstrate that the artificial neural network controller has satisfactory generalization capability, reliability, and accuracy to be feasible for this critical control operation.

Adaptive fuzzy scheme for efficient, fast valving control

Abstract Fast valving has long been regarded as an effective and economical method of performing transient control in a power generation plant. Due to the inherent nonlinearities that exist in this operation, fast valving controllers designed in a conventional way cannot deliver satisfactory control. This paper introduces a new approach to construct a fast valving controller by applying an adaptive fuzzy technique. A rule-generation and adaptation scheme is formulated in a fuzzy-system framework. Then the implementation issue is discussed in detail. From the outcome of on-line testing, it is shown that the controller constructed in this way not only maintains global stability, but also appears to be robust in different fault situations.

Identification of nonlinear discrete-time multivariable dynamical systems by RBF neural networks

Proposes a recursive identification technique for nonlinear discrete-time multivariable dynamical systems. Extending an earlier result of the authors (1993) to multivariable systems, the technique approaches a nonlinear system identification problem in two stages: one is to build up recursively a RBF (radial-basis-function) neural net model structure including the size of the neural net and the parameters in the RBF neurons; the other is to design a stable recursive weight updating algorithm to obtain the weights of the net in an efficient way.<<ETX>>

Efficient algorithm for the design of multilayer feedforward neural networks

A novel design technique that builds a multilayer feedforward net for an arbitrary set of binary associations is presented. The key to the technique is to decompose the arbitrary mapping into a sequence of operations composed of three primitive operations. As each of the primitive operations is easily realizable in a simple layered feedforward net, the mapping can be realized by cascading the simple feedforward nets corresponding to the decomposition. The time spent in the construction of the net is proportional to k*n/sup 2/, compared to k*2/sup n/ typical of the conventional direct approaches. With such a reduction, the technique can be used in building multilayer feedforward nets for binary association problems of extremely large size.<<ETX>>

Defuzzification, structure transparency, and fuzzy system learning

The issue of defuzzification is explored in the context of fuzzy system structure and learning for nonlinear system modeling. It is revealed that the best-known defuzzification methods may not necessarily result in transparent fuzzy system structures that are universally approximate and yet suitable for developing effective learning algorithms for modeling. This paper then presents a simple defuzzification method that leads to transparent fuzzy system structures based on the min-max operations.

Fast valving control using radial-basis function neural network

Fast valving has long been seen as an effective and economic method to perform transient control in a power generation plant. Due to the inherent nonlinearities that exist in this operation, the fast valving controller designed in the conventional way cannot deliver a satisfactory control. This paper introduces a new approach to control fast valving by using a RBF(radial-basis function) neural network. A controller construction scheme is proposed, in which a stable learning algorithm is embedded. Then the implementation issue is discussed. From the outcome of on-line tests, it is seen that the controller constructed is effective and robust in many different fault situations.

On the learning of min-max fuzzy systems

This paper develops a methodology for learning fuzzy systems that contain min-max operations. It is shown that with the defuzzification properly defined, a min-max fuzzy system can have a transparent structure that is both universally approximating and easy for a learning scheme to be developed. A specific learning scheme based on multi-scale residue extraction is then presented.

On-line stable nonlinear modelling by structurally adaptive neural nets

This paper proposes a neural net based on-line scheme for modelling discrete-time multivariable nonlinear dynamical systems. Taking the advantage of structural features of RBF (Radial-Basis-Function) neural nets, the method approaches the modelling problem by setting up a coarse RBF model structure in the light of the spatial Fourier transform and spatial sampling theory, then devising appropriate on-line algorithms to carry out refinements for both the RBF net structure and the associated weights. Main convergence results are established in the paper along with the analysis backing up the structure initialization and adaptation. The effectiveness of the scheme is illustrated with an simulation example.<<ETX>>