基于改进遗传算法优化自联想神经网络的风机故障诊断  

FAULT DIAGNOSIS OF WIND TURBINE BASED ON AUTO-ASSOCIATIVE NEURAL NETWORK OPTIMIZED BY IMPROVED GENETIC ALGORITHM

作　　者：李政宇 李练兵 芮莹莹 Li Zhengyu;Li Lianbing;Rui Yingying(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300132,China)

机构地区：[1]河北工业大学人工智能与数据科学学院,天津300132

出　　处：《计算机应用与软件》2022年第6期297-302,328,共7页Computer Applications and Software

摘　　要：为了更为准确有效地诊断风机变桨系统故障,提出一种通过改进自适应遗传算法(IAGA)来优化自联想神经网络(AANN)的风机故障诊断模型。在IAGA中对选择算子进行改进,加快自适应遗传算法收敛速率,同时在适应度函数中引入AUC,降低不平衡数据对模型诊断效果的干扰;使用IAGA对AANN初始权值进行优化,通过AANN获得变桨系统正常状态下的残差分布,利用JS散度计算其与故障时刻残差分布的偏移度,判断变桨系统是否故障。利用华北某风电场记录的历史数据进行实验,结果表明,与其他神经网络相比较,IAGA-AANN网络能够有效提高风机故障诊断识别率,缩短模型训练时间。To more accurately and effectively diagnose the fault of wind turbine pitch system,this paper proposes a fault diagnosis model using improved adaptive genetic algorithm(IAGA)to optimize the auto-associative neural network(AANN).The selection operator was improved in the IAGA to speed up the convergence rate of adaptive genetic algorithms.AUC was introduced into the fitness function to reduce the interference of unbalanced data on the model diagnosis effect.This paper used IAGA to optimize the initial weight and threshold of AANN.AANN was adopted to obtain the residual distribution in the normal state of the pitch system,and the JS divergence was applied to calculate the deviation of the residual distribution between normal time and fault time,so as to determine whether the pitch system was faulty.The experiments were performed using historical data of the actual wind field.The experimental results show that,compared with other neural networks,IAGA-AANN network can effectively improve the fault diagnosis recognition rate of wind turbines and shorten the model training time.

关 键 词：风机变桨系统 遗传算法 自联想神经网络 故障诊断 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]