Inspection For The Natural Gas Pipeline Leakage Based On EMD And BP Neural Networks

With the role and status of oil and natural gas becoming important, More and more people pay attention to the operation safety of transportation pipeline.Pipes need to be inspected in real time during operation. Pipeline operation state is accurate judgment and the leak is discovered in time can avoid the occurrence of safety accidents.Although the pipeline leakage inspection technology is constantly improving, errors and false reports will still be discovered in the inspection of pipeline transportation processSo, this paper improves the design for the errors and false reports of the natural gas pipeline inspection process.The main work of this paper includes: data acquisition, preprocessing, feature extraction, pattern recognition and parameter optimization. In this paper, the signal is collected by the natural gas pipeline sub acoustic wave sensor, which has the characteristics of low frequency, non line, non periodic, multi signal mixing and so on.The main work is as follows:1. Wavelet threshold function denoising method is introduced. The collected data is contained noise at the construction site. So this paper proposes a method based on the improved wavelet threshold function to deal with the data of the natural gas pipeline.2. Natural gas pipeline data is analysised. In this signal analysis method the detected signal can be decomposed into a sum of finite stationary intrinsic mode functions(IMF) by EMD,among which some IMF components containing the main information of signals are selected to be analyzed on its kurtosis and skewness. Therefore the eigenvectors of the signals led by leakage or other abnormal events can be extracted. These eigenvectors will be used as input to the BP neural network pattern recognition.3. The most widely used BP algorithm in multi layer feedforward network is introduced and studied. Aiming at the deficiency of BP algorithm, the training speed of BP algorithm is put forward. The neural network is trained by the eigenvectors of pipeline data, and the pattern recognition is carried out for the operation of each natural gas pipeline.4. The effect and accuracy of pattern recognition are analyzed, and the parameters of BP neural network are optimized by particle swarm algorithm and genetic algorithm and achieved very good results.