Synthesis of Intelligent Learning Control Systems by Evolutionary Neural Networks

在本项目中，我们提出了一种基于神经网络的智能控制系统的实现方法。神经网络与传统的信息处理方法相比，具有并行处理、分布式存储、学习等特性。本课题旨在利用这些特性进行智能控制系统的综合以及遗传算法和进化规划等进化计算，并将其应用于实际的对象控制问题，以显示所提方法的有效性。为了完成项目研究，我们采用了以下方法来综合这些智能控制系统：(1)神经网络的学习能力我们利用通信通道的信息度量来描述神经网络的学习能力。(2)遗传算法进化机制的出现。为了提高神经网络的学习能力，考虑了(1)，引入遗传算法，寻找学习曲线的全局极小值。(3)神经网络图像理解的构建为了实现图像理解的智能化，我们不仅使用了数字数据，还将图像信息处理作为一个多媒体数据处理。(4)基于图像信息和动作行为的规则获取系统及其应用在未知环境下，我们实现了利用图像信息和动作行为的智能控制系统。这些方法已被应用于实际控制问题，如倒立摆的稳定和加热炉的温度控制。从这些实验结果来看，本文提出的方法具有良好的控制效果，可以应用于其他控制问题。

In this project, we have proposed a new approach to realize the intelligent control systems based on neural networks. The neural networks have several specific properties over conventional information processing such as parallel processing, distributed memory, learning, etc. The present project is to utilize these properties for synthesis of intelligent control systems as well as evolution computation such as genetic algorithms and evolution programming and then apply them to real plant control problem to show the effectiveness of the proposed methods. To complete the project study, we have adopted the following approach to synthesize these intelligent control systems :(1) Learning Ability of the Neural networkWe have formulated the learning ability of neural networks by using information measure of the communication channel.(2) Emergence of Evolution Mechanism by Genetic AlgorithmsTo improve the learning ability of neural networks considered (1), we have introduced the genetic algorithms and find the global minimum of the learning curves.(3) Construction of Image Understanding by Neural NetworksTo realize the intelligence we have used not only numerical data but also image information processing as a multi-media data processing.(4) Rule Acquisition System Based on Image Information and Action Behaviors and their ApplicationsUnder the unknown environment we have realized the intelligent control systems using the image information and action behaviors. These approaches have been applied to real control problems such as stabilization of inverted pendulum and temperature control of a heating furnace. From these experimental results the proposed approach in this study are good control results and could be applied to the other control problems.