基于凸组合算法优化的前向神经网络观测器设计

The Observer Design of Feedforward Neural Networks Optimized by Convex Combination Algorithm

本文利用凸组合算法对单隐藏层前向神经网络进行了优化,通过迭代来更新权值以调整隐藏层的信息.同时引入了一个新的误差函数来评价误差性能,该函数通过对权值进行解耦来求解优化参数,提高了参数的计算速度.在此基础上,提出了一种非线性系统的自适应神经网络状态观测器设计方法.最后通过仿真验证了该神经网络观测器能准确并快速地观测出系统的状态值.

The convex combination algorithm （CCA） is proposed to optimize single hidden layer feedforward neural net-works. This method updates the weights by iterating to massage the information in the hidden layer. And a new error function is set up to measure the performance of the neural networks. The optimized parameters can be obtained by decoupling the weights, which improves the calculating speed of the parameters. On the basis, a design method of an adaptive neural networks state observer for nonlinear systems is proposed.At last,the simulation is used and illustrates that the observer can observe the state values of the system accurately and quickly.