摘要
针对传统BP神经网络实现模拟电路故障诊断时存在的缺陷容易收敛于局部最优值且训练时间过长等,提出了利用遗传算法(GA)优化的BP神经网络来对模拟电路进行故障诊断的方法。实验结果证明,优化后的BP网络可有效地避免收敛于局部最优值,大大地缩短了训练时间。同时为了提高遗传优化的收敛速度和精度,避免"早熟"现象,本文提出了一种引入移民算子的遗传算法,仿真结果表明了该算法的有效性。
Because tradition BP neural network converges on local optimum easily and the training time is too long when it is used in fault diagnosis of analog circuit, a genetic optimization algorithm that can search for the optimum parameters of BP neural networks is proposed. The results show the optimized BP neural network can effectively avoid converging on local optimum and reduce training time greatly. At the same time, in order to improve convergence rate and precision and overcome premature convergence, a new genetic algorithm based on immigration operator is put forward. Simulation results illustrate that the new algorithm is feasible and effective.
出处
《电子测量与仪器学报》
CSCD
2007年第1期20-24,共5页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金资助课题(编号:60372001)
关键词
模拟电路
故障诊断
遗传算法
BP神经网络
移民算子
analog circuit, fault diagnosis, genetic algorithm, BP neural network, immigration operator.
