基于混合式多类特征选择暖气管道腐蚀程度研究

Research on Corrosion Degree of Heating Pipeline Based on Hybrid Multi Class Feature Selection

近年来,北方工业厂房中存在暖气管道因腐蚀而造成供暖不足导致能源浪费,为实现暖气管道更精准的腐蚀预判,提出了一种使用敲击法采集暖气管道声音信号提取筛选特征并结合模式识别的腐蚀程度预测方法。其中针对经典ReliefF算法未考虑特征之间冗余和SVMRFE算法只能处理二分类以及运行速度慢的问题,提出了一种改进的ReliefFP-NMSVMRFE-SVM算法。前期对ReliefF算法处理后的特征集使用Pearson相关系数去除冗余特征,后在改进的MSVMRFE线性和非线性算法中二次筛选,分别获得一组得分递减排序的特征子集,再结合分类器BP、SVM对特征子集进行分类预测。结果表明,ReliefFP-NSVMRFE-SVM算法模型识别精度最高且用时短,在训练集上预测结果为99.85%,独立测试集上预测结果为97.14%,对暖气管道内部腐蚀程度的检测具有一定的适用性。

In recent years,there has been energy waste caused by insufficient heating due to corrosion of heating pipes in northern industrial plants.In order to achieve more accurate corrosion prediction of heating pipes,a corrosion degree prediction method using tapping method to collect sound signals of heating pipes,extract screening features,and combine pattern recognition has been proposed.An improved ReliefFP-NMSVMRFE-SVM algorithm is proposed to address the issues of classical ReliefF algorithm not considering redundancy between features,SVMRFE algorithm being able to only handle binary classification,and slow running speed.In the early stage,Pearson correlation coefficients were used to remove redundant features from the feature set processed by the ReliefF algorithm.Then,a second screening was performed in the improved MSVMRFE linear and nonlinear algorithms to obtain a set of score decreasing sorted feature subsets,which were then combined with classifiers BP and SVM to classify and predict the feature subsets.The results show that the ReliefFP-NSVMRFE-SVM algorithm has the highest recognition accuracy and short time,with a prediction result of 99.85% on the training set and 97.14%on the independent test set.It has certain applicability for detecting the degree of internal corrosion in heating pipelines.