改进BP算法在模糊神经网络中的应用

引入一种改进的BP算法--动量因子-自适应学习率算法.通过调节动量因子以及在学习过程中的学习率实现自适应,以提高学习速率和增强学习的平稳性.将该学习算法引入到串形结构的多层前向模糊神经网络中,通过学习确定了模糊映射关系,实现了对象的模糊故障诊断.在应用模糊神经网络进行故障诊断时,被监测的故障征兆信号与网络输入层相连,即将输入向量输入到网络中,经过模糊化处理,得到各故障征兆在所定义征兆的模糊子集上的隶属度向量,再利用神经网络的前向计算,得到故障原因的模糊隶属度向量,最后通过对向量的分析确定故障原因的类型.将上述模糊神经网络应用到空气静压轴承中,实现了设备的故障诊断,测试结果验证了该方法的有效性.

Application of improved BP algorithm in fuzzy neural networks

Improved BP algorithm——momentum factor-adapted learning rate algorithm was cited. Based on BP algorithm, the adaptability was realized by momentum factor adjust and the learning rate to improve the learning speed and stabilization. By useing the algorithm in multilayer forward fuzzy neural networks of serial structure, the fuzzy mapping relation and the fuzzy malfunction diagnosis of the object were obtained. When the fuzzy neural network was used in malfunction diagnosis, the inspected malfunction sign connected with the import layer of networks to put the import vector into network. Then the fuzzy subjection limit vectors of the defined sign which belongs to the malfunction sign by dealing with fuzzy were achieved. By counting with the neural networks forward, the fuzzy subjection vectors of the malfunction cause were gained. The type of malfunction cause was analyzed and ensured, and this fuzzy neural network in diagnosing the malfunction of air still press axletree has been applied. The result proves that the method is effective.