System identification of bilinear state-space models by modified gradient search method

The problem of estimating parameter of bilinear state-space system is considered in this paper. A gradient based search algorithm is proposed to the optimization of output error cost function. The nonuniqueness of the fully parameterized bilinear state-space is considered and the parameter update is restricted to the directions that are perpendicular to the manifold of input-output equivalent systems. In addition, the convergence rate is also discussed. Finally, numerical simulation is implemented and the results show the improved performance of the proposed method.