基于改进核函数的支持向量机天然气脱硫装置故障诊断方法
Fault diagnosis using Support Vector Machines (SVM) with improved kernel functions in natural gas desulfurization system
浏览(4048) 下载(8)
- 引用格式:
-
何宇琪,张波,王俊超,熊鹏.基于改进核函数的支持向量机天然气脱硫装置故障诊断方法[J].天然气与石油,2024,42(4):94-100.doi:10.3969/j.issn.1006-5539.2024.04.012
HE Yuqi, ZHANG Bo, WANG Junchao, XIONG Peng.Fault diagnosis using Support Vector Machines (SVM) with improved kernel functions in natural gas desulfurization system[J].Natural Gas and Oil,2024,42(4):94-100.doi:10.3969/j.issn.1006-5539.2024.04.012
- DOI:
- 10.3969/j.issn.1006-5539.2024.04.012
- 作者:
- 何宇琪1 张波1 王俊超1 熊鹏2
HE Yuqi1, ZHANG Bo1, WANG Junchao1, XIONG Peng2
- 作者单位:
- 1. 中国石油天然气股份有限公司西南油气田分公司重庆气矿, 重庆 400021; 2. 重庆大学机械与运载工程学院, 重庆 400044
1. Chongqing Gas Mine, PetroChina Southwest Oil and Gasfield Company, Chongqing, 400021, China; 2. College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing, 400044, China
- 关键词:
- 改进核函数;支持向量机;HYSYS;天然气脱硫;故障诊断
Improved kernel functions; Support Vector Machines (SVM); HYSYS; Natural gas desulfurization; Fault diagnosis
- 摘要:
针对传统脱硫故障诊断方法反应慢、诊断准确率低的问题,根据Mercer理论,改进了支持向量机(Support Vector Machine,SVM)的核函数及其参数,建立了一个由多项式核函数、Sigmoid核函数和高斯径向基核函数复合成的改进核函数,在此基础上提出了一种基于改进核函数的SVM天然气脱硫装置故障诊断方法。相对于传统SVM,改进SVM体现了各单一核函数的优点,并具有更好的学习效率及诊断准确率,在小样本数据条件下仍然具有较好的泛化能力。利用HYSYS软件建模并与现场数据进行对比实验,由实验结果可知改进SVM的误差率降低到传统SVM误差率的约30%,验证了新方法能有效提高脱硫装置故障诊断的准确率和效率。研究结果有助于天然气脱硫装置故障诊断系统工作的智能化开展,同时也为故障诊断方法的研究提供了借鉴。
To address the issue of slow response and low diagnostic accuracy inherent in traditional desulfurization system fault diagnosis methods, this paper proposes a Support Vector Machine (SVM) fault diagnosis method for natural gas desulfurization system based on improved kernel function. According to Mercer's theory, this method redesigns the SVM's kernel function and its parameters, by integrating polynomial, Sigmoid, and radial basis kernel functions into a single composite kernel. Compared with the traditional SVM method, this improved approach not only combines the benefits of each single kernel function but also offers higher learning efficiency and diagnostic accuracy, maintaining strong generalization capabilities even with limited sample data. Experimental validation conducted using HYSYS modeling and field data demonstrates that the improved method reduces error rate to approximately 30% of the its original method before improvement, thereby verifying the methods effectiveness in improving the accuracy and efficiency of the fault diagnosis of desulfurization systems. The research results contribute to the intelligent operation of fault diagnosis systems in natural gas desulfurization systems and also provide a reference for the study of fault diagnosis methods.
