基于概率神经网络的凝汽器故障诊断研究

Condenser Fault Diagnosis Using Probabilistic Neural Networks

基于概率神经网络(PNN)提出了一种新的汽轮机组凝汽器故障诊断方法。PNN是一种可用于模式分类的神经网络,其实质是基于Bayes分类规则与Parzen窗的概率密度函数方法发展而来的一种并行算法。PNN可以克服反向传播神经网络(BPNN)学习收敛速度慢、易陷入局部极小值等问题,而且优于常见的凝汽器故障诊断方法:PNN学习规则简单,训练速度快,可以满足训练上实时处理的要求;训练不需要太多样本,模式分类能力强,而且具有很高的运算速度;抗干扰能力强,对传感器测量噪声具有较强的诊断鲁棒性;新的训练样本也很容易加入以前训练好的分类器中,很适用于在线检测。将该方法用于某汽轮机组凝汽器故障诊断中,仿真结果表明了该网络在分类应用中的快速性、准确性,而且易于工程实现。

Based on the Probabilistic Neural Networks (PNN), this paper presents a new scheme for condenser fault diagnosis. As one kind of neural networks for pattern classification, the PNN implements the Bayesian classification rule and Parzen density estimation algorithm. PNN can increase the convergence rate and overcome the local optimization of Back Propagation Neural Networks (BPNN). Due to its trivial and instantaneous training and fast diagnosing speed, PNN can be used in the real time diagnosis. As additional patterns are observed, they can be added into the system easily. Consequently, the generalization will be improved and the decision boundary can be obtained more accurately. Thus the on-line and incremental learning can be easily achieved. The result of diagnosis shows that the proposed scheme is fast, accurate, and easy to use.