Research On The Pipeline Leak Location Based On Wavelet Packet Analysis And Neural Networks.

Pipeline is the main mode of transport for crude oil and petroleum products, due to years of service beyond the fixed number of years and illegal mining, pipeline leakage often happen which has an immeasurable impact on the environment and the economy. Thus, it is particularly important for us to find a manner to detect the pipeline leak location timely. As long distance pipeline with the construction of long distance, and its environment is variable and with other characteristics, pipeline leak location is facing many difficulties, China's oil and gas storage and transportation sector has focused lots of time on researching the topics. The artificial neural network is information processing system which can simulate the human brain's structure and function, not only with function of learning and fault-tolerant, but also has the feature be able to approach any nonlinear continuous mapping in theory, and be able to simulate the intrinsic relationship between the causal variables which are Under the influence of uncertain factors, and the data can be established with the leak data and location data.In this paper, the pressure signal of oil pipeline leak is made as the departure point. firstly, denoised the signal collected before, then decomposed the signal after denoised with 3-layer wavelet packet, applicated MATLAB software to calculate the energy on the value of each node to get a large number of pipeline leakage signal eigenvectors. pulled in The neural network theory, adopted BP neural network and RBF neural network which is now relatively mature, BP pipeline leak location model and RBF artificial neural network model were established by MATLAB software. then, test the model with the data not involved in network training, obtained that the relative error of BP neural network model is about 3%, and the RBF neural network moedl's error is about 1%. Finally, compared the results of BP neural network model with the RBG neural network model's, concluded that RBF network model has higher precision than BP neural network model, and its error rate is less than 1%. Concluded that we had obtained good model which can meet the production needs.