【摘要】 为克服传统BP神经网络存在着容易陷入局部极小点、对初值要求高的缺点,采用遗传算法对BP神经网络的初值空间进行多点遗传优化,得到最佳初始权值矩阵,在此基础上按误差前向反馈算法沿负梯度搜索进行网络学习;同时提出了一种用于BP神经网络遗传优化的染色体浮点编码方法,并描述了作用于染色体上的遗传操作算法。仿真研究表明:遗传BP神经网络的收敛和诊断能力优于传统BP神经网络,可有效运用到汽轮发电机组振动故障诊断中。更多还原

【Abstract】 In view of that traditional BP neural network has such weaknesses as that the optimal procedure is easily stacked into the minimal value locally and causes high demands of initial value,the integration of genetic algorithm and BP neural network has been put forward.On the analysis of the characteristic of BP neural network and genetic algorithm,the possibility and method of the application of genetic algorithm to neural network are discussed.Genetic algorithm which can process inherited optimization with many spots in resolving space is adopted to optimize the initial weight space of BP neural network structure at first,thus the optimal initial weight matrix of BP neural network can be obtained,and then error back propagation algorithm is adopted to train the network.An approach to construct a kind of floating-point encoding chromosome for evolution of BP neural network is proposed,and the operations of this kind of chromosome are described.The convergence ability and diagnosis ability of BP neural network based on genetic algorithm are analyzed on simulated test.Suggested by simulation,this method has improved the convergence ability and diagnosis ability of traditional BP neural network.The results show that the BP neural network based on genetic algorithm is useful to deal with weaknesses of traditional BP neural network and can be effectively applied to diagnose the vibration fault of turbine generator-set.更多还原