基于神经网络的过程工业故障诊断方法研究

发布时间：2018-07-09 23:13

本文选题：故障诊断 + 田纳西-伊斯曼过程　；参考：《西南科技大学》2017年硕士论文

【摘要】：过程工业的故障诊断对保证生产安全、提高产品质量有着重要作用。论文以神经网络研究为基础,针对过程工业存在数据分布复杂、高维等特性,对其故障诊断方法进行研究。具体工作包括:(1)过程工业数据的高维及相关特性,会影响诊断结果的准确性。论文先采用主元分析法对数据降维处理以去除相关性,处理后的数据输入学习矢量量化神经网络进行故障诊断,田纳西-伊斯曼(TE)过程的仿真表明该方法(PCA-LVQ)具有更高的识别率。(2)论文采用粒子群优化算法来确定概率神经网络的平滑参数,针对标准PSO算法易陷入局部极值的缺点,采用非线性变化的惯性权重代替线性变化的惯性权重来改进粒子群优化算法(IIWPSO),最后应用于概率神经网络中进行故障诊断。数值仿真和TE过程的仿真表明了该方法(IIWPSO-PNN)的有效性。(3)在实际应用中,操作者通常是缺乏神经网络计算经验的普通技术人员,很难保证实际效果,因此,论文将神经网络集成算法应用于故障诊断中。首先采用Bagging算法生成多个个体网络,再用IIWPSO-PNN算法训练网络并生成多个IIWPSO-PNN分类器,最后采用相对多数投票法生成最终结果。UCI标准机器学习库和TE过程的仿真表明了该方法(Bagging-IIWPSO-PNN)的有效性。
[Abstract]:Fault diagnosis in process industry plays an important role in ensuring production safety and improving product quality. Based on the research of neural network, the fault diagnosis method of process industry is studied according to the characteristics of complex data distribution and high dimension. The main works are as follows: (1) the accuracy of diagnostic results is affected by the high dimension and related characteristics of process industrial data. In this paper, the principal component analysis method is used to reduce the dimension of the data to remove the correlation, and the processed data input learning vector quantization neural network is used to diagnose the fault. The simulation of Tennessee Eastman (te) process shows that this method (PCA-LVQ) has a higher recognition rate. (2) Particle Swarm Optimization (PSO) algorithm is used to determine the smooth parameters of probabilistic neural networks. The particle swarm optimization (IIWPSO) algorithm is improved by using nonlinear inertia weights instead of linearly varying inertial weights. Finally, it is applied to fault diagnosis in probabilistic neural networks. Numerical simulation and te process simulation show the effectiveness of this method (IIWPSO-PNN). (3) in practical application, the operator is usually a general technician who lacks the experience of neural network calculation, so it is difficult to guarantee the actual effect. In this paper, neural network ensemble algorithm is applied to fault diagnosis. Firstly, bagging algorithm is used to generate multiple individual networks, then IIWPSO-PNN algorithm is used to train the networks and generate multiple IIWPSO-PNN classifiers. Finally, the simulation of the final result. UCI standard machine learning library and te process using the relative majority voting method shows the effectiveness of this method (Bagging-IIWPSO-PNN).
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TP277

【相似文献】