摘要
针对复杂化工生产过程数据多样性、高维性以及耦合性的特点,提出一种基于交叉验证递归特征消除算法(RFECV)、粒子群优化算法(PSO),并结合随机森林(RF)和支持向量机(SVM)的故障诊断方法。首先利用RF-RFECV方法对混合运行数据进行K折交叉验证学习与重要性排序,抽取并重构故障特征信息;将预处理后的数据作为输入样本,利用PSO与序列最小优化算法(SMO)搜索超参数得到最佳SVM分类器,实现故障诊断。应用于田纳西-伊斯曼(Tennessee Eastman, TE)过程的仿真实验结果表明:RF-RFECV与PSO-SVM融合故障诊断方法泛化能力强、诊断准确率高,识别准确率可达到99.5%以上。
Aiming at the characteristics of data diversity, high dimension and coupling in complex chemical production process, a fault diagnosis method is proposed based on cross validation recursive feature elimination algorithm(RFECV) and particle swarm optimization algorithm(PSO), combining with random forest(RF) and support vector machine(SVM). Firstly, use the RF-RFECV method to perform K-fold cross-validation learning and importance ranking on the mixed operating data, extract and reconstruct the fault feature information;use the preprocessed data as the input sample, and use the PSO and sequence minimum optimization algorithm(SMO) to search the hyperparameters get the best SVM classifier to realize fault diagnosis. The simulation results applied to the Tennessee Eastman(TE) process show that the RF-RFECV and PSO-SVM fusion fault diagnosis method has strong generalization ability, high diagnostic accuracy, and recognition accuracy can reach more than 99.5%.
作者
张伟
王连彪
张广帅
ZHANG Wei;WANG Lianbiao;ZHANG Guangshuai(College of Automation and Electronic Engineering,Qingdao University of Science and Technology,Qingdao 266061,China)
出处
《青岛科技大学学报（自然科学版）》
CAS
2022年第5期101-108,共8页
Journal of Qingdao University of Science and Technology:Natural Science Edition
基金
国家自然科学基金项目(61971253)。
