Identification of oil-water-gas flow patterns by super-sparse near-infrared wavelengths sensor

This paper puts forward a method of using two wavelengths of near-infrared light to identify the oil-gas-water mixed flow patterns. Propose a new method of multiple correlation algorithm (MCA) to solve the sensor super -sparse representation problem. Use MCA results as input layer parameters, use BP neural network to train the voltage signal and identify different flow patterns. The experimental results show that the recognition degrees of water, oil bubble, and oil flow patterns have reached 96.77%, 98.39%, and 100%, respectively, and the recognition rate of the water-gas phase has 78.38%. Using a correction model can further improve the accuracy. Test the samples with water cut from 90% to 70%, and the average RMSE= 1.59e-2. This method can be applied to the detection of the water cut of crude oil in oil fields and has a certain reference value for the wavelength selection of near-infrared light sensors.