Research On Pipeline Leak Detection Based On Wavelet Analysis And BP Neural Network

As the safe and economic transport tools for conveying energy resources such as oil and gas, pipeline transportation has become important part of national integrated transportation and been widely applied in recent years. However, pipeline leak caused by wear of long-time running, natural aging of equipment, changes of geography and atmosphere and man-made damage occur frequently,which results in a huge potential threat to people's lives, property and environment, as well as precious resources waste. Therefore, it is of great economic importance and social significance to establish pipeline leak detection system.There are many ways of studying pipeline leak detection so far, among which the method based on knowledge mainly in the use of neural network and expert system is universally concerned by the industry. Wavelet neural network combined with the self-learning of neural network and the local characteristics of wavelet is characterized by great adaptive distinguish and fault tolerance and therefore has good application prospects in the fields of pipeline leak detection .In this paper, on the basis of detailed analysis of current research on fluid pipeline leak detection, wavelet neural network algorithm has been used in pipeline leak detection.The main research work is as follows:First of all, through studying the principle of wavelet transforms and the method of wavelet analysis in the application of fault detection, the reasonable wavelet base function, wavelet decomposition level and threshold function have been selected for signal De-noising.Secondly, the principle, algorithms and characteristic of back propagation (BP) neural network and wavelet neural network are introduced, focusing on the application of BP neural network and wavelet neural network in the pipeline leak diagnosis. The method based on energy-fault identification of wavelet packet is used in the system of pipeline leak diagnosis. The characteristical vectorial indicator reflecting the features of pressure signal is proposed.Finally, based on comparative analysis about error training curve and test results of BP neural network to wavelet neural network in the diagnosis of pipeline leak, two improved points of error back propagation algorithm in the BP neural network are proposed: Firstly, inertial coefficient has been added into correcting process of network weights in order to reduce the vibratory trends in learning process and increase the training speed. secondly, local adaptive learning rate algorithm based on gradient symbol change is used in the learning process, which can adjust learning rate dynamically by itself according to each of adjustable parameters' state information. Final test results show that the improved algorithm of wavelet neural network is of great training speed, high identified accuracy and strong stability and play an important role in the pipeline leak diagnosis. Both of theoretical derivation and simulation results show that the method based on wavelet analysis and BP neural network can make timely and accurate diagnosis in the pipeline leak detection and has a certain research and practical value. This method will play a bigger role in the pipeline leak detection with further improvement in the theory and practice.