A class of learning/estimation algorithms using nominal values: Asymptoticanalysis and applications

A class of estimation/learning algorithms using stochastic approximation in conjunction with two kernel functions is developed. This algorithm is recu rsive in form and uses known nominal values and other observed quantities. Its convergence analysis is carried out, the rate of convergence is also ev aluated. Applications to a nonlinear chemical engineering system are examin ed through simulation study. The estimates obtained will be useful in proce ss operation and control, and in on-line monitoring and fault detection.