BP神经网络算法研究及其在燃煤锅炉中应用

A Study on the Algorithm of Neural Networks and its Application in a Coal-burning Boiler

神经网络的种种特性使得它们在动力工程中得到广泛的应用。根据系统的特点研究使用合理的神经网络算法已成为广受关注的课题。BP神经网络是一种最基本的神经网络。国际国内学者围绕如何避免训练陷入局部极小和加速收敛 ,提出了多种措施。但以往方法主要基于如何选择合理的学习率或加入适当的动量项 ,未能充用利用控制系统的特点。通过分析具有不同极点分布的控制系统BP神经网络的误差曲面形状 ,提出在总体误差的下降方向和其相反方向上同时进行一维搜索的方法。算法较多地考虑了有关的细节性问题。此方法被用于某燃煤锅炉 。

Neural networks have found wide application in power engineering because of their special functions.The use of reasonable algorithms of neural networks for the control of a system according to its characteristics has been a subject for study attracting wide attention.A BP neural network is one of the basic type. Researchers at home and abroad have proposed a lot of methods to overcome the difficulty of local minimum points encountered in the training process of a BP neural network and to accelerate the training speed. These methods are based on the choice of a better learning rate or appending of a more appropriate momentum term. They didn′t take advantage of the characteristics of control systems. Analyzing the appearances of error cambers of the BP neural networks of control systems with different limit points distributions, this paper presents an improved algorithm which performs one dimension searching in the descent direction and the opposite direction at the same time. The improved algorithm takes much consideration on related details. The application to a coal-burning boiler shows a great success.