Muti-layers Wavelet Kohonen Neural Network Model for Underground PipelineCoating Detection with Galvanostatic Transient Technique

摘要

The galvanostatic transient response method was used to detect the underground pipeline coating on the spot without excavation. With wavelet and neural network method a multi-layer model was established to analysis the detection information. The model was made up based on Self-organization of Kohonen neural networks and the advantage of wavelet to pike up the useful information. The model has three parts, the front five layers have the ability to pike up the information, the weights of this part are the adaptive wavelet coefficient (filter). The sixth layer corresponding to reconstruction of wavelet analysis, in this way some useful information can be resumed. The last layer has the ability of self-training. After self-training the weights were remembered like BP neural network. It is confirmed that the method is correct and convenient for on-the-spot detection by the detection result of actual pipeline between Dezhou and Puyang of Zhongyuan Oil Field.