论文部分内容阅读
提出一种α稳定分布基函数概率神经网络(A lpha-PNN)结构,该网络隐层的神经元激活函数采用了对称α稳定分布的概率密度函数,和常规的高斯分布函数相比,其具有更好的可变性和延展性,从而使隐层的神经元在函数近似上具有更高的适应性,同时也克服了概率神经网络对输入数据的独立同分布假设,提高了神经网络对局部脉冲突变的近似能力。在此基础之上,提出了一种新的根据系统输入输出数据实现的故障诊断算法,并将其应用到轴承的故障诊断中。仿真结果表明,在有色噪声的背景下,该算法仍然能够实现较好的识别效果,故障的误报率低于概率神经网络方法。
A kind of α-stable distributional-basis probabilistic neural network (A l-PNN) is proposed. The neural activation function of the hidden layer of the network adopts the probability density function of symmetric α-stable distribution. Compared with the conventional Gaussian distribution function, Better variability and ductility, so that the neurons in the hidden layer have a higher adaptability in the function approximation, meanwhile, it also overcomes the assumption of independent and identically distributed input data by the probabilistic neural network and improves the effect of the neural network on the local pulse Approximate ability to mutate. On this basis, a new fault diagnosis algorithm based on the system input and output data is proposed and applied to the fault diagnosis of the bearing. The simulation results show that the algorithm can still achieve a good recognition effect in the background of colored noise, and the false alarm rate of the fault is lower than that of the probabilistic neural network.