ABSTRACT
This chapter focuses on online estimation and application of artificial neural networks (ANNs). The step known as parameter estimation aims at choosing the unknown model parameters such that the model behavior accurately represents the process. This usually involves the solution of a so-called inverse problem where model parameters are adapted to optimize a certain measure or cost function. Practical identifiability analysis gives information of the quality with which parameters can be estimated from the available noise corrupted and discrete measurements. The main goal of any parameter estimation procedure is the adaption of the vector of unknown model parameters such that the model accurately represents the real process by solving the so-called inverse problem. If a parameter estimate is characterized by large uncertainty in terms of a large confidence interval, it is said to be practically non-identifiable. In contrast, if an estimated parameter has a small confidence interval, it is said to be practically identifiable.
