针对最差情况Ｈ∞辨识中最常见的一类模型集，给出了一种基于时域数据的两步辨识算法．第一步通过信息一致性原理把辨识问题转化为受限凸规划问题，第二步利用一个多项式逼近定理在第一步结果基础上，对不确定集中的系统进行逼近，得到辨识出的名义模型．最后分析了辨识算法的局部误差、全局误差和算法的收敛性． Aiming at the most common type of model set in worst-case H∞ identification, a two-step identification algorithm based on time-domain data is presented. The first step is to transform the identification problem into a restricted convex programming problem through the principle of information consistency. The second step uses a polynomial approximation theorem to approximate the system with uncertain centralization based on the first-step result and obtain the identified nominal model . Finally, the local error, global error and convergence of the algorithm are analyzed.