基于非线性同伦LM算法的故障诊断方法

A Method of Fault Diagnosis Based on Nonlinear Homotopy LM Algorithm

引入同伦理论并定义了一种广义的非线性同伦映射,基于现有BP算法,将同伦方法与Levenberg-Marquardt(简称LM)优化方法结合,提出了一种非线性同伦LM神经网络学习算法用于神经网络训练,解决了现有学习算法收敛速度慢和局部极小值的问题,提高了神经网络的学习效率。将改进算法用于建立神经网络故障诊断模型,研制出实时诊断系统用于电站锅炉送风机在线故障监测与诊断。应用结果表明,该诊断方法在收敛速度、精度和稳定性能等方面较同类方法有较大改善。

A generalized nonlinear homotopic mapping was defined b y introducing the theory of homotopy. Combined the methods of homotopy and LM(Levenberg-Marquardt) optimization into existing standard BP network algorithm, a nonlinear homotopy LM neural network training method, called LMBP algorithm in this work,was proposed,trying to solve the speed of convergence and local minimum problems existed in the standard BP network algorithm. The proposed algorithm was applied to construct a fault diagnosis model, and used to develop a real-time fault diagnosis system to monitor and diagnose the faults of forced draught fan for a power station boiler. Experimental results show that the diagnosis approach developed in this work performs better than any other existing methods in the convergent speed, accuracy and stability.