中文摘要：

针对复杂系统的输出含有大量局部平稳噪声,不能直接应用于在线建模的问题,本文将自适应局部余弦神经网络作为在线滤波器,并在分析Lipschitz a≥1函数局部余弦变换的基础上,提出网络的学习算法.仿真表明自适应局部余弦神经网络相对于维纳滤波,能更有效地去除局部平稳噪声.

英文摘要：

An adaptive local cosine neural network with the corresponding algorithm is introduced as on-line estimator to remove locally stationary noise in the output of complex system, which often leads to the bad properties of models. The new algorithm is based on the analysis of local cosine transform of the function with Lipschitz α ≥ 1. The simulation shows that the local cosine network can remove the locally stationary noise more efficiently than Wiener estimation.