摘要
针对管道漏失诊断多层信息融合决策复杂的问题,提出了一种基于随机森林(RF)融合的管网漏失诊断方法。该方法使用随机森林基于非参数数据驱动的分类特性,利用随机森林的自动抽样技术和节点随机分裂技术,采用独立分量分析(ICA)对负压波和流量信号进行降噪并选择其中的泄漏特征参数作为随机森林的输入,充分利用了不同传感器群组互补和冗余的有效信息进行管道和管网的工况判别。并且随机森林可以衡量传入特征向量中各个特征的重要程度,起到了特征参数数据挖掘的作用,可以进一步简化系统的复杂程度。实验结果表明,该种融合算法相比基于反向传播神经网络(BPNN)及基于支持向量机(SVM)的D-S证据理论算法分别提高了2.2%和6.4%管道漏失诊断的准确率。
To solve the problem that the number of false positives is high and multi-level information fusion calculation is complex, a pipeline network leakage diagnosis based on Random Forest (RF) fusion was proposed. In the method, non-parametric data driven classification characteristic was used and an automatic sampling and random splitting of node was made. Independent Component Analysis (ICA) was used to reduce the negative pressure wave and the flow signal, the leakage characteristic parameter was chosen as the input of the random forest, the complementary and redundant information of different sensor groups was fully used to determine the working condition of pipeline and pipeline network. And the random forest was used to measure the importance of each feature in the feature vector, which further simplified the complexity of the system. The experimental results show that this fusion algorithm can improve the 2.2% and 6.4% accuracy of pipeline leakage diagnosis over Back-Propagation Neural Network (BPNN) and D-S evidence theory algorithm based on Support Vector Machine (SVM).
作者
王学渊
陈志刚
钟新荣
卢宁
WANG Xueyuan;CHEN Zhigang;ZHONG Xinrong;LU Ning(School of Mechanical-Electronic and Vehicle Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Belting Engineering Research Center of Monitoring for Construction Safety,Beijing 100044,China;ChangQing Downhole Technology Company,CNPC ChuanQing Drilling Engineering Company Limited,Shaanxi Xi'an 710000,China)
出处
《计算机应用》
CSCD
北大核心
2018年第A01期20-23,共4页
journal of Computer Applications
基金
国家自然科学基金资助项目(51004005)
住建部资助项目(2016-K4-081)
北京市优秀人才培养资助项目(2013D005017000013)
北京市教育委员会科技计划一般资助项目(KM201610016017)
关键词
漏失诊断
随机森林
管网
信息融合
leakage diagnosis
Random Forest (RF)
pipeline network
information fusion