It is found that the leak location error depends on friction factor based on the research on pressure gradient method. A new method based on BP neural network forecasting friction factor, suitable to the running pipeline, is put forward. This technique has successfully located the leak point in a laboratory pipeline.
In order to minimize damage from leak, it is necessary to study the method for location leak position in long-distance pipeline with a leak, the measuring points just arranged at inlet and outlet of the pipeline. The researchers have been developed many methods for position leak during the past of a few years. These are pressure gradient method [Zhang J], negative wave method[Zhang Dongling], inverse transient method [Pudar, Liggett and VÍtkovský] etc. These methods have their respective advantages and disadvantages for long-distance pipeline. According to the long distance pipeline, only flow rate and pressure at the inlet and outlet used to locate the leak, pressure gradient method is suitable. So, this paper mainly studies pressure gradient method because it agrees better with experiment. It is noticed that the leak location error depends on friction factor based on the research on pressure gradient method. The old determination methods for friction factor, including formula method and least square method, have disadvantages to running pipeline. In order to avoid these shortcomings, a new method for using BP neural network forecasting friction coefficient is put forward. This method uses flow rate as input element and friction factor as output element of BP neural network. Sigmoid function is used as activation function of the network. Firstly, this paper analysis the influence friction factor on leak location error on the pressure gradient method. Secondly, a new method for calculating friction factor is developed. Finally, the new method is used to a laboratory pipeline to locate the leak.
