关键词: 控制依赖自适应评估设计(ADACD), 评估网络, 辨识器, 模型参考自适应控制(MRAC), 非线性系统

Abstract: The action-dependent adaptive critic design (ADACD) is introduced into the identification of nonlinear systems to obtain the best model. A total evaluation function is defined to express the total identification error of a system within the time it has experienced, then a critic network is constructed to approximate to this evaluation function. And an identifier network is further constructed taking its output as the critic network input. In such a way the best model can be obtained by minimizing the critic network output. Here the parameter adaptation rule is based on minimizing the critic network output, i.e., the total identification error of a system within the time it has experienced, instead of minimizing the instant error. As a result, the convergence rate is quickened remarkably with more accurate identifiability. After obtaining the model of the system, the design of the model reference adaptive controller (MRAC) for this model is studied by using neural networks. The simulation results showed that the performance of this novel identifier is excellent, and the MRAC based on it can make sure that a system to be controlled tracks the reference model quickly and stably.