Pipeline leakage recognition based on principal component analysis and neural network

Negative pressure waves generated by pipeline leakage may contain multiple intrinsic parameters, which present insignificant differences under different working conditions. Accordingly, first-order differences of negative pressure waves are deployed to extract 8 typical parameters as characteristic parameters of negative pressure wave signals. Principal component analysis is used to reduce dimensions of the 8 characteristic parameters. By using the resulting dimension-reduction features of the typical negative pressure wave signals, training can be made to generate required neural network with self-organized mapping. The network is used to identify samples of negative pressure waves under different working conditions. Relevant results show that the new system can effectively extract major features of negative pressure waves under different working conditions to recognize pipeline leakage. The model is characterized by fast computation and high accuracy.