摘要
目前在概率神经网络的机理研究中存在一个瓶颈问题:在有限的模式样本中提炼出能反映整个样本空间的平滑因子σ,目前的σ估计都基于经验或有限样本聚类的方法,并不能将空间的概率特性很完整地表达出来,而遗传算法可以在没有任何先验知识的情况下发现系统潜在知识。采用遗传算法来优化概率神经网络的σ,并把它应用于电力变压器的故障诊断中。通过MATLAB7.0仿真结果得出,经过遗传算法优化平滑因子后的概率神经网络可以大大提高故障诊断的准确性。
There is a bottleneck Problem in the probability neural network mechanism research., the smooth factor which can reflect the entire sample space is refined in the limited pattern sample, the present smooth factor estimated all based on the experience estimation or the very limited sample cluster method, which cannot express very completely the spatial probability characteristic, but genetic algorithm can discover the latent knowledge of the system without any prioried knowledge. The genetic algorithm is used to optimize the probability neural network smooth factor. It is applied to the power transformer fault diagnosis. Simulation result with the MATLAB 7.0 proves that this method is correct.
出处
《现代电力》
2007年第2期44-47,共4页
Modern Electric Power
关键词
概率神经网络
平滑因子
遗传算法
probability neural network
smooth factor
genetic algorithm