Research On Water Content Measurement Of Crude Oil Based On BP Neural Network

With most of the oil fields in our country coming into the period of high water content, water content has risen close to the limit of economic exploitation in some wells. Oil field managers need to know the production condition of oil well timely and accurately to provide authentic data for the each link of production, and this requires the detection of water content of crude oil in order to assess the value of the oil well, the exploitation extent and the development of exploitation programs to improve economic efficiency. Therefore, the research and application of online detection technology of water content of crude oil has been paid more attention, which demands higher detection accuracy and cost performance.The paper bases on the theory of electromagnetic conductivity method for measuring water content of crude oil,and derives the relationship between water content of crude oil and the sensor useful voltage signal, and verifies the measurement feasibility of water content of crude oil according to oil-water mixture conductivity change. Using the software ANSOFT, the 3D model of the three coil system for measuring the water content of crude oil is established, and detailed simulation has been done to find the relationship between different coils position and useful voltage signal of the sensor, which contributes to accurate parameters for the actual sensor structures,meanwhile temperature and salinity influencing on the output voltage of the sensor are simulated. The results show that the increasement of temperature and ion concentration of mixture can increase obviously the output voltage of the sensor. Therefore, it is necessary to revise the temperature and salinity.The water content of crude oil is affected by many factors, and the influence factors are complex and nonlinear. The paper uses BP neural network algorithm to deal with the experimental data of crude oil water content with electromagnetic conductivity method, the model is established to predict the water content of crude oil. Aiming at the slow convergence speed and easily falling into local minimum problems of BP neural network, respectively, the paper uses heuristic and numerical optimization method to improve the BP algorithm to predict the water content of crude oil. The research shows that: the two kinds of improved BP algorithm both can improve convergence speed and prediction precision of the water content of crude oil. Fletcher-Reeves correction algorithm has a better effect than adaptive learning rate and momentum gradient algorithm in the optimization of BP neural network prediction model for measuring water content of crude oil. The study provides a strong theoretical and experimental basis for the design of intelligent online measuring instrument of water content of crude oil outside pipeline.