Skip to Main Content
Because of kinds of regulating commands, Oil and gas pipelines in operation can occasionally generate a series of conditions consisting of stopping transportation, starting transportation, distribution, increment, internal pump regulation of single station or multi-stations, as well as valve opening-adjustment. These normal conditions together with the detected leakage condition can easily cause false alarm. Therefore, it is quite important to analyze the sampling signal real time, rapidly distinguish leakage conditions from normal cases, reduce false alarm and improve the leakage recognition accuracy. Traditional BP algorithm based on gradient descent method has played an important role in condition recognition of pipeline. However, this method exists some disadvantages such as local minimum point, slow convergence speed, difficultly determining the number of hidden layers with hidden nodes et al., which can result in convergence failure of algorithm, signal recognition failure or long time need of recognition processing. In this paper, a new BP algorithm, named gradient descent algorithm with momentum term and self-adaptive learning rate, is proposed to indentify these conditions, which can overcome the lack of traditional gradient descent algorithm, enhance and speed up convergence. Tests show that compared with the traditional method, this novel method can save time 18.7%, and increase the recognition accuracy of 16%, which meets the field management requirements.