Abstract-Wiener-typed nonlinear systems with hard input constraints are ubiquitous in industrial processes. However, owing to their complex structures, there are very few achievements on their control algorithm. Aimed at this problem, an improved dual-mode control algorithm is put forward. Firstly, the detailed procedure of this algorithm is proposed. Then, its feasibility, stability and convergence are analyzed by using the invariant set theory combined with LMI(linear matrix inequalities) technique [8] . In contrast to traditional algorithms, this one has the capabilities of maximizing the size of the closed-loop stable region and decreasing the online computational burden. Finally, the proposed algorithm is performed by simulations with promising results. Key words -Dual-Mode control algorithm, Wiener-typed nonlinear systems, Zeroin algorithm, invariant set, LMI I. INTRODUCTION uring a lot of real industrial processes such as distillation [9] , pH neutralization control [2,5] , heat-exchanger [16] , chemical reaction [14] , and biological visual process [16,17] , there widely exists a type of nonlinear systems which can be described by Wiener mode. The background of this model can be even extended to the areas of communication [11] , signal processing [11] , psychology [17] and sociology [17] . It consists of a linear dynamic element followed by a memoryless nonlinear element while Hammerstein model contains the same elements in the reverse order [9] . Wiener-typed systems correspond to processes with linear dynamics associated with general nonlinear operators. In recent years, the control of Wiener-typed systems has become one of the most urgent and difficult tasks in nonlinear control field [2,5,9,11,16,17] . Due to the particular structure of Wiener model, the identification and control algorithms of this system are much more complicated than the counterparts of Hammerstein-typed system. Suny and Lee [18] worked over the method approaching nonlinear element with polynomials based on the suggestion that the output of linear block be feasible. However, in practical industrial control, this assumption can not be always satisfied. In 1997 [9] , Kalafatis and Wang proposed a method identifying the two parts of Wiener model at the same time, but an assumption must be satisfied that the inverse of the nonlinear element can be approximated by P order polynomials with satisfactory precision, which greatly limited its applications. In the same year [14] , Yamanaka et al developed a new kind of dynamic neural network which is composed by a Laguerre function filter and a memoryless nonlinear block. Based on this model they presented a model reference adaptive control scheme for Wiener-typed nonlinear systems. Stability analysis was also given, but these conditions were too rigid and the initial weights were difficult to optimize. In 2001 [7] , Duwaish et al developed an approach to control Wiener models using MPC combined with genetic algorithm, but the computational burden was so heavy that its real-time performances...