Comparative Study On Prediction Models Of Four Crude Oil Moisture Contents

Crude oil moisture measurement technology plays an important role in the development,processing,storage,transportation and refining of crude oil.In order to meet the actual needs of oil fields,a variety of parameter information processing techniques have been introduced in the study of crude oil moisture content measurement.It is of great theoretical significance and practical value to develop a crude oil moisture meter with wide application range,high precision and high cost performance.Aiming at the short-wave method crude oil moisture content detection sensor,the effects of temperature and salinity on the measurement accuracy of short-wave crude oil moisture content sensor were studied by using the characteristics of different short-wave absorption capacity of oil and water.The fitting curve was obtained by single factor test data fitting analysis,and the influence of temperature and salinity on the measurement accuracy of crude oil moisture content was obtained.Through the multi-factor test data fitting analysis,the influence of temperature and salinity on the measurement accuracy of crude oil moisture content was obtained.It provides data and theoretical basis for further research on crude oil moisture prediction model.On-line measurement of crude oil moisture content is affected by various factors such as flow regime,temperature,salinity,and oil-water two-phase flow.In this paper,an indoor experimental device based on multi-parameter oil-water two-phase flow is designed.The mathematical relationship between temperature,salinity,calibration water content and actual water content is studied.The temperature,the salinity and the calibrated water content are selected as inputs,and the actual measured results of the sensor are used as outputs.The multiple regression analysis method,the standard BP neural network method,the improved BP neural network method and the genetic algorithm optimization BP neural network method are respectively used.The prediction model of crude oil moisture content was established.Through the comparison experiments of four prediction models,the comparative study of the regression model,stability and practicability of the prediction model was carried out.Experiments show that the BP neural network prediction model optimized by genetic algorithm greatly improves the influence of temperature and salinity on the measurement of crude oil water content,and has strong generalization ability.It is a new high-precision intelligent measurement method.