摘要
针对传统油气管道泄漏监测和诊断准确率不高的问题,提出一种基于BP神经网络的多层D-S证据理论融合算法。用压力传感器、流量传感器、温度传感器等传感器监测管道泄漏信号。这些信号经预处理后输入到已训练好的BP神经网络。该网络的输出结果被用作各初始命题的基本概率赋值函数(BPAF)。再用多层D-S证据理论,将不同源、不同测点以及不同时刻的传感器信息依次进行融合。并将此方法并应用于实验室中,模拟管道泄漏的检测和诊断。试验结果表明,用多层D-S证据理论算法,能够有效提高管道泄漏诊断的准确率,降低泄漏识别的不确定性。
Abstract : For the sake of improving the accuracy of pipelines leakage diagnosis, a novel multi-layer date fusion leakage diagnosis algorithm was worked out based on BP network and D-S evidence theory. Leakage signals were got by several kinds of sensors such as pressure sensor, flow sensor, and temperature sensor. The data were preprocessed first, and then put into the well trained BP neural network to obtain the values of BPAF. Pieces of information from sensors were fused according to multi-layer D-S evidence theory. This method was used to simulate pipeline leakage in authors'laboratory. The results show that using this meth- od can reduce the uncertainty of diagnosis recognition and improve the accuracy. Key words: oil and gas pipelines ; leak diagnosis ; Dempster-Shafer (D-S) evidence theory ; BP neural network; basic probability assignment function(BPAF) ; multi-layer date fusion
出处
《中国安全科学学报》
CAS
CSCD
北大核心
2013年第8期171-176,共6页
China Safety Science Journal
基金
国家自然科学基金资助(51005247)
北京市科技新星计划(2010B068)
关键词
油气管道
泄漏诊断
D—S证据理论
BP神经网络
基本概率赋值函数(BPAF)
多层次信息融合
oil and gas pipelines
leak diagnosis
Dempster-Shafer (D-S) evidence theory
BP neural network
basic probability assignment function(BPAF)
multi-layer date fusion