基于深度置信网络的管网泄漏故障诊断方法研究

针对当前管网系统数据量大不利于传统模型方法诊断故障的问题，设计了1种基于深度置信网络的管网故障诊断算法。首先，对管网数据结构以及管网系统运行状态进行分析，选取管网主要数据作为故障诊断网络的输入，确定相应运行状态作为诊断网络输出；其次，设计了基于多个受限制玻尔兹曼机与Softmax分类器级联的深度置信网络，并且利用对比散度算法和BP算法对模型进行预训练与调优，使模型参数达到全局最优；最后，通过实验测试确定所设计的深度置信网络的训练迭代次数与网络层数，使算法诊断准确率达到最优。研究结果表明:提出的基于深度置信网络的管网故障诊断算法对管网故障诊断可以达到良好的诊断结果，泄漏预测准确率在验证集样本上可达96.87%，在管网泄漏检测方面，相较于传统基于模型的方法优势明显。

Study on fault diagnosis method of pipelines leakage based on deep belief network

In view of the large amount of data in the current pipelines system,which is unfavorable to the fault diagnosis of the traditional model method,a pipelines fault diagnosis algorithm based on the deep belief network was designed. Firstly,the data structure of the pipelines network and the operating state of the pipelines system were analyzed,and the main data of the pipelines were selected as the input of the fault diagnosis network,at the same time,the corresponding operating state was determined as the output of the diagnostic network. Secondly,a deep belief network based on the cascade of multiple restricted Boltzmann machines and Softmax classifier was designed,and the model was pre-trained and tuned by using the contrast divergence algorithm and BP algorithm,so as to make the model parameters be the global optimum. Finally,the training iteration times and network layer number of the designed deep belief network were determined by the experimental tests,so as to make the diagnosis accuracy of the algorithm be the optimal. The research results showed that the proposed pipelines fault diagnosis algorithm based on the deep belief network could achieve good diagnosis results for the pipelines fault diagnosis. The accuracy of leakage prediction could reach 96.87% on the verification set sample,and the advantages were obvious compared with the traditional method based on the model in the pipelines leakage detection.