Abstract:

In order to improve the on-line measuring accuracy and widen the measuring range of water content of crude oil, a simulated multi-sensor measurement system of oil/water two-phase flow is adopted to detect the para- meters influencing the measurement, and an intelligent compound model for measuring the water content is estab- lished by combining the rough set preprocessor, the support vector machine classifier and the genetic neural network predictor. Experimental results show that the proposed model effectively eliminates the cross interference of oil/wa- ter emulsion modal, temperature and salinity content, overcomes the nonlinearity of sensor itself, realizes the modal identification of oil-water mixture via fuzzy reasoning and self-learning, and adjusts the model parameters by changing working conditions adaptively. Thus, the accuracy of on-line intelligent measurement of water content of crude oil is effectively improved in a wide measuring range.

Key words: crude oil, oil/water two-phase flow, rough set theory, modal identification, genetic algorithm, neural network, support vector machine