非均匀采样系统多新息随机梯度辨识性能分析

Performance analysis of multi-innovation stochastic gradient identification for non-uniformly sampled systems

针对一类非均匀采样系统,提出了其输入输出表达的多新息随机梯度辨识方法.该方法将随机梯度算法中的新息项扩展为向量,有效利用了历史新息所包含的信息,从而提高辨识精度和算法的收敛速度,同时又保留了随机梯度算法计算量小的优点.仿真例子通过改变新息长度,验证了所提出辨识算法性能的优越性.

A state space model is derived for non-uniformly sampled systems. Based on the obtained input/output representation, a multi-innovation stochastic gradient identification algorithm is presented by expanding the scalar innovation to an innovation vector. The proposed algorithm uses both the current innovation and the historical innovations, which improves the stochastic gradient algorithm for the identification accuracy and convergence rate. Simulation example verifies the superiority of the proposed algorithm by adjusting the innovation length.