In this paper we describe techniques relating to intelligent data processing, on line identification and neural networks predictors for complex nonlinear systems with multiple switching models. The initial model identification is provided based on the realization theory non-iterative algorithm.