Analysis and design of a recurrent neural network for real-time parameter estimation

The authors design and analyze a recurrent neural network for real-time parameter estimation. There are several desirable features in the proposed neural approach to parameter estimation. The estimated parameters generated by the proposed neural network are optimal in the sense that a least squares performance index is minimized. The convergence rate of the recurrent neural network based parameter estimator can be controlled by selecting a design parameter. The proposed recurrent neural network is easy to implement in electronic circuits and easy to interface with analog sensors. The configuration and design principles of the neural network are discussed. The operating characteristics of the neural network are demonstrated via an application example.<
>