基于Stacking模型融合的抽油机故障诊断算法

Fault Diagnosis Algorithm for Pumping Unit Based on Stacking Model Fusion

抽油机故障诊断过去基本采用单分类器识别的方法。针对单分类器识别的局限性和抽油机故障诊断的复杂性,提出了一种基于Stacking的多模型融合的抽油机故障诊断算法。首先,对8种不同的抽油机故障图像进行数据清洗,得到二值化图像;然后分别使用AlexNet、VGG16、GoogLeNet、ResNet50作为基学习器,对采集的抽油机故障图像进行分类识别;最后,采用基于Stacking的集成学习方法,将各基学习器的预测结果融合重构后,作为次级元分类器XGBoost的输入,其输出即为最终识别结果。实验结果表明,使用该方法对8种最常见的抽油机故障图像进行实验,平均识别率高达98.16%,基于Stacking的多模型融合的抽油机故障诊断算法显著优于由单一特征组合构建的同类分类器算法,并且具备较好的泛化能力与鲁棒性。

In the past,single classifier recognition method was basically used for fault diagnosis of pumping units.Considering the limitation of single classifier recognition and the complexity of fault diagnosis for pumping units,a multi-model fusion fault diagnosis algorithm for pumping units based on Stacking is proposed in this paper.First,data cleaning is performed on 8 different fault images of pumping units to obtain binary images,and then the collected fault images of pumping units are classified and identified by using AlexNet,VGG16,GoogLeNet,and ResNet50 as base learners,respectively.Finally,an integrated learning method based on Stacking is used to fuse and reconstruct the prediction results of each base learner as the input of the secondary meta-classifier XGBoost,whose output is the final recognition result.The experimental results show that using this method,the average recognition rate of 8 common fault images of pumping units is as high as 98.16%.The multi-model fusion based fault diagnosis algorithm for pumping units based on Stacking is significantly superior to similar classifier algorithms constructed from a single feature combination,and has good generalization ability and robustness.