Neural networks in control systems

Abstract

Neural network structures used for system identification and control are reviewed. Due to the complexity and diversity of the properties of biological neurons, the task of compressing their complicated characteristics into a model is extremely difficult. Toward this goal, an artificial neuron, also called a unit, that receives its inputs from a number of other neurons or from the external world was developed. A weighted sum of these inputs constitutes the argument of an activation function. This is a simple, but useful first approximation of a biological neuron. Using this model, many neural structures, usually referred to as feedforward neural networks, have been reported in the literature. Many of these networks use only present values of inputs, and are therefore called instantaneous or static systems. A natural extension of static networks is the dynamic or recurrent neural network which incorporates feedback in its structure. No general theory for dynamic neural networks has yet developed similar to that for static networks. With the parallel growth in the field of fuzzy logic, many neural models encompassing the principles of neural networks and fuzzy set theory are being developed. An attempt is made to provide the basic concepts of static, dynamic, and fuzzy neural structures.<>