Pipeline leak detection method based on RBF neural network under different working conditions
GONG Jun;SHUI Aishe;BAO Jianming;LYU Zhibo;Dept of Logistics Information & Logistics Engineering,Logistical Engineering University of PLA;PLA Unit 78479;PLA Unit 68060;
This paper presents a leak detection method combining principal component analysis and RBF neural network in consideration of heavy data processing load and high false alarm rate for leak detection. On the basis of data preprocessing, the time domain features of pipeline pressure sequence is computed to reduce the data processing load, and the principal component analysis of the time-domain characteristics is conducted for dimensionality reduction to extract new integrated features that can better reflect the pressure change characteristics. Then, the recognition model is established for pipeline leak detection, with integrated features as the input and working conditions as the output of RBF neural network. The field experimental results show that this method not only reduces the data processing load and improve the detection speed, but also effectively distinguish working condition adjustment and pipeline leak, thereby ensuring the detection accuracy.
【CateGory Index】： TE973.6