用神经网络方法求解Hilbert空间中的优化问题  

Optimization in Hilbert Space by Neural Network Approach

作　　者：杨宗凯[1] 邹理和[1] 

机构地区：[1]西安交通大学

出　　处：《电子学报》1991年第5期55-62,共8页Acta Electronica Sinica

基　　金：国家自然科学基金

摘　　要：本文在线性规划神经网络的基础上,提出了一种求解Hilbert空间中优化问题的神经网络方法。这种方法不是通过利用数字计算机复杂的迭代运算来求优化问题的解,而是建立一个与线性规划网络相应的模拟电路系统,即一个专用的模拟计算机,在稳定状态下,测量电路系统的输出电压即可获得对应问题的解。理论和实验都表明,该方法可在RC时间常数内(纳秒级)对优化问题进行求解,并且具有对电路参数容错性强,结构规正、简单等特点。Based on linear programming neural networks, a neural network approach to optimization in Hilbert space is proposed in this paper. The optimization problems are solved by setting up the associated linear programming circuit rather than by computation in a digital computer. Theoretical analysis and computer simulations show that the approach can solve the optimization problem within RC time constant (order of several hundred ns), and has some advantages such as normal and simple structure and tolerance of inaccuraces in the conductance matrix.

关 键 词：神经网络法 HIBERT空间 线性规划 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]