摘要
为了克服神经网络在基本算法上存在收敛精度低、收敛慢、不收敛和网络结构难以确定等缺陷,采用常用的8种智能算法对故障样本进行诊断性能对比实验,得出网络最优结构的设计步骤和较好的6种智能算法。用这6种智能算法对故障测试样本进行诊断性能对比,得出了train-lm是较佳智能算法,traingdx是收敛稳定性较好的算法;网络最优结构的参数设计过程,为神经网络故障诊断性能的最佳算法和结构提供了系统化设计的实验方法。
In order to overcome the defects of basic algorithm of neural network including low convergence precision,slow convergence,non-convergence and difficulty in determing network structure,with the commonly-used 8 kinds of intelligent algorithm to diagnose fault samples,the optimal network structure and the better 6 kinds of intelligent algorithm were obtained by comparison performance experiments for diagnosis of the fault samples; Then,6 intelligent algorithms were used for performance comparison test for diagnosis,giving a result-that trainlm is the better intelligent algorithm,traingdx is the algorithm with better convergence stability.The optimum fault diagnosis parameters of neural network were obtained,a systematic experimental method of the best algorithm and structure of neural network fault diagnosis performance has established.
出处
《太原理工大学学报》
CAS
北大核心
2010年第2期183-187,共5页
Journal of Taiyuan University of Technology
基金
山西省留学基金资助项目(2004-19)
山西省基础科技平台资助项目(051005)
关键词
神经网络
故障诊断性能
训练与测试
智能算法与结构优化
Neural network
Fault diagnostic performance
Training and testing
Optimization of intelligent algorithms and structures
