Gradient based iterative identification of multivariable Hammerstein-Wiener models with application to a Steam Generator Boiler

Most of the real industrial systems are nonlinear and multivariable which might be correlated with some noises. Therefore, considering a model which can effectively characterize these types of systems are very appealing. In this regard, this paper presents a multivariable Hammerstein- Wiener model for identification of nonlinear systems with moving average noises. For this purpose, this model is first reexpressed as a multivariable pseudo-linear regression problem. Then, a gradient based iterative learning algorithm is invoked which can successfully estimate the matrix of unknown parameters as well as the noises. The efficiency of the proposed identification scheme is investigated through data for a real multivariable nonlinear process as a case study. This process is a Steam Generator Boiler at Abbott Power Plant in Champaign IL which has characteristics of instabilities, nonlinearity, non-minimum phase behaviour, time delays, noise spectrum in the same frequency range of the plant dynamics, and load disturbances. As the results verify, this approach is quite efficient for identification of multivariable nonlinear systems.