Sequential parameter identification method for nonlinear systems

In this paper we propose an accurate online parameter identification algorithm for nonlinear time-varying systems. For such systems, techniques based on cross-validation to achieve regularization or model selection are not possible, and classical least square techniques are not reliable when the dynamics of the system are highly nonlinear. To overcome these problems, an identification algorithm devised from Sutton´s dynamic learning rate techniques and based on a learning window and forgetting factor criterion has been used. In doing so, the proposed algorithm avoids the need for heuristic choices of the initial conditions and noise covariance matrices required by Kalman filtering. The performance of the proposed method is demonstrated on aircraft flight dynamics parameter identification in the horizontal plane.