In this paper, a neurodynamic optimization approach is proposed for robust eigenstructure assignment problem of second-order descriptor systems via state feedback control. With a novel robustness measure serving as the objective function, the robust eigenstructure assignment problem is formulated as a pseudoconvex optimization problem. Two coupled recurrent neural networks are applied for solving the optimization problem with guaranteed optimality and exact pole assignment. Simulation results are included to substantiate the effectiveness of the proposed approach.
I. INTRODUCTIONEigenstructure assignment is a vitally important problem in linear control systems design. Since poles (eigenvalues) and their associated eigenvectors of a closed-loop system greatly impact on the control performance such as the stability condition and the convergence speed, pole assignment is an effective approach to place poles of the close-loop system at any desired locations on the complex plane via a state feedback law with appropriate gains. In practice, robust control is more desirable as the systems cannot be precisely modeled or the systems are subject to parameter uncertainties. The robust pole assignment problem is to find the feedback gains such that the robustness of the eigensystem is optimized. Kautsky et al. [20] first formulated the robust pole (eigenstructure) assignment by means of minimizing the spectral condition number of the eigenvector matrix. Alternative robustness measures and various optimization approaches in linear control systems design were widely investigated [22], [24]-[26], [32], [35], [41]. Second-order linear systems constitute an important class of systems, as they can capture the dynamic behaviors of many natural phenomena. There exist numerous applications in various fields, such as vibration and structural analysis, spacecraft control and robotics control [1], [2], [7]. Furthermore, as second-order systems can be viewed as special cases of highorder systems, synthesis approach to second-order systems may be applied for higher-order systems. A few results on robust pole assignment in second-order linear systems are available in the literature [3], [5], [8], [9], [23], [30]. In specific, [30]proposed a robustness measure for second-order control by solving a generalized linear eigenvalue assignment problem subject to structured perturbations. However, most existing algorithms cannot guarantee the achievement of global optimality due to the complexity and nonconvexity of the applied measures. In addition, most proposed methods are not applicable for on-line computing.Neurodynamic optimization based on recurrent neural networks is competent for solving optimization problems in