Algorithms for minimal model structure detection in nonlinear dynamic system identification (Q4364041)

The paper is dedicated to the Minimal Model Structure Detection problem in nonlinear dynamic system identification. The essence of the proposed solution is a new algorithm for Minimal Model Structure Detection (MMSD). It is derived based on the standard orthogonal algorithm and genetic search procedures. It allows to complete the practical search for the optimal orthogonalization path in nonlinear dynamic system identification and provides a suboptimal solution to the combined problem of model structure detection and parameter estimation. The presented solution can not garantee that the minimal model structure is obtained. Its computational efficiency is better then the known MMSD algorithms, but the necessary computations are still quite extensive. To overcome this problem, a Refined Forward Regression Orthogonal algorithm is developed. Simulated results were used to demonstrate the performance of the two new algorithms that can be used in Nonlinear Autoregressive Moving Average Model with Exogenous Input modeling, in the configuration and training of radial basis function neural networks and in fuzzy model building.