粒子群优化神经网络在船舶辅锅炉故障诊断中的应用  

APPLICATION OF PARTICLE SWARM OPTIMIZATION NEURAL NETWORK IN FAULT DIAGNOSIS OF MARINE AUXILIARY BOILER

作　　者：高鹤元 甘辉兵[1] 郑卓 杨远达 Gao Heyuan;Gan Huibing;Zheng Zhuo;Yang Yuanda(College of Machine Engineering,Dalian Maritime University,Dalian 116001,Liaoning,China)

机构地区：[1]大连海事大学轮机工程学院,辽宁大连116001

出　　处：《计算机应用与软件》2020年第8期137-141,148,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(51479017);工信部高技术船舶科研项目“智能船舶综合测试与验证研究”(工信部装函[2018]473号)。

摘　　要：针对自组织特征映射(Self-Organizing Feature Map,SOM)神经网络训练过程依赖权值向量,网络初始化的随机性影响收敛速度和聚类精度的缺陷,提出一种粒子群(Particle Swarm Optimization,PSO)算法优化SOM神经网络原始权值的学习规则。将优化算法应用于船舶辅锅炉燃烧故障诊断的仿真研究中,对使用DMSVLCC模拟器运行的样本数据进行分析,并与单一SOM网络的分类结果进行比较。仿真结果表明,优化后的算法能够对船舶辅锅炉某工况下故障样本数据进行有效的准确聚类,准确度高于传统的SOM网络,具有良好的可训练性和模式识别能力。Aiming at the shortcomings that self-organizing feature map(SOM)neural network training process depends on weight vector,and the randomness of network initialization affects the convergence speed and clustering accuracy,we propose a particle swarm optimization(PSO)to optimize the learning rule of original weight of SOM neural network.The optimized algorithm was applied to the simulation study of combustion fault diagnosis of marine auxiliary boiler.The sample data of DMSVLCC simulator was analyzed,and the classification results of single SOM network were compared.The simulation results show that the optimized algorithm can effectively and accurately cluster the fault sample data of a marine auxiliary boiler under a certain working condition.The accuracy is higher than the traditional SOM network,and it has good trainability and pattern recognition ability.

关 键 词：SOM神经网络 粒子群优化算法 船舶辅锅炉 故障诊断 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]