基于极化因子神经网络的火电厂制粉系统故障诊断技术

Neural Network with Polarization Factor for Pulverizing System Fault Diagnosis

摘要:

制粉系统是火电厂的主要设备,其安全稳定运行对发电企业的经济生产具有十分重要的意义;针对制粉系统的运行特性和故障分析,提出了基于极化因子神经网络的火电厂制粉系统故障诊断方法,该方法将故障征兆相应的过程变量作为输入,将制粉系统故障类型作为输出,通过训练神经网络建立其系统故障诊断模型,其中训练过程中采用极化因子来自动调整神经网络的收敛速度,从而在满足误差目标的前提下,防止其陷入局部极小;选取实际火电厂制粉系统3个典型故障及其相对应的9个故障征兆参数进行了实验;结果表明,该方法具有良好的收敛性,完全可以满足火电厂制粉系统现场故障诊断的要求。

Abstract:

Pulverizing System is an important part of the power plants and it is crucial to keep the system working safely and stably. According to the operation characteristics and fault analysis knowledge of the system, a fault diagnosis method based on neural network with polarization factor for the pulverizing system of the power plant is proposed. The method builds the diagnosis model by treating a neural network. The neural network uses the process variables that are related to the fault symptoms as the inputs and the fault types as the outputs. Moreover, a polarization factor is used to adjust the convergence speed of neural network automatically. Thus, the method can accomplish the treatment of the neural network and avoid the local minimums. The experiments are performed with three typical faults and their nine corresponding fault symptoms parameters derived from the pulverizing system of a real power plant. The experimental results verify the good convergence of the proposed method. The proposed method can achieve the requirement of on-site fault diagnosis of the pulverizing system of the power plants.