基于神经网络的一类大系统动态递阶优化方法

A NEURAL NETWORK BASED METHOD FOR DYNAMICALHIERARCHICAL OPTIMIZATION OF A CLASS OFLARGESCALE SYSTEMS

提出了一种用于求解离散时间大系统动态递阶优化问题的神经网络模型（ＬＨＯＮＮ），该网络以全集成化为特征：１）各子系统的动态方程嵌入相应的局部优化网络中，使得网络结构具有较低的维数，易于硬件实现；２）其上级协调网络和局部优化网络的求解过程同时进行，优化求解速度高，适宜于实时系统优化。

A neural network for dynamical hierarchical optimization of discretetime largescale systems,i.e., LHONN,is presented in the paper. It is a fully integrated network with the following principal characteristics:1)the dynamic equations of the subsystems are imbedded into the local optimization networks, which results in the lower dimension of the neural network, so it is easy for implementation;2) the coordination neural network (CNN) and the local optimization neural networks (LONN) work simultaneously to seek for the optimal solution of the system,which leads to high speed of problem solving. Thus the LHONN is more suitable for realtime optimization problems.