Research On Soft Sensor Of Dynamic Fluid Level Of Oil Based On Improved Black-hole Algorithm

As a valuable and non-renewable energy,oil is the main pillar of the sustainable development of our country’s economy. To meet our demand for oil, efficient exploitation of oil has become an important issue. In the process of oil exploitation, measuring the depth of liquid level indirectly to regulate oil production technology is an important way to improve the efficiency and dynamic liquid level data is the key parameter.The traditional fluid level measurements include echo, pontoon, pressure gauge,etc.,among which echo method is commonly used. As hand operation is involved in traditional measurement, the accuracy, real-time performance and safety of fluid level measurement can not be guaranteed. Therefore, the technology of soft sensor is put forwarded to improve automation and measurement accuracy of oil exploitation in this article.There are 4 aspects which had an influence on the feature of soft sensor. Among them,the modeling is the kernel of soft sensor technology, and least squares support vector machine(LSSVM) modeling is selected to predict the dynamic liquid level after the comparison of two mature modeling methods. Prior to the establishment of prediction model, the existence of redundant auxiliary variables will reduce the accuracy of fluid level prediction even cause model failure. Therefore, gray relative analysis is applied to pick out the variables highly related to dynamic fluid level as a auxiliary variable, including daily fluid production, pump efficiency, casing pressure. After analysis of the characteristics of auxiliary variable, empirical mode decomposition method is applied to decompose data, removing high frequency noise and predicting the dynamic fluid level with the least squares support vector machine(LSSVM)model. The comparison of prediction of dynamic fluid level before and after de-noised auxiliary variables was input validates this method. In LSSVM modeling, the optimization of kernel parameter and penalty parameter is crucial to prediction accuracy. In this article, the black-hole algorithm was used for parameter optimization and improved. Then comparison of the improved black-hole algorithm with particle swarm optimization(PSO), differential evolution algorithm(DE), artificial bee colony algorithm(ABC), and black-hole algorithm(BH) shows that the improved black-hole algorithm has obvious advantages and it is effective to predict the dynamic fluid level with the combination of LSSVM. In order to prevent theaging of model, the joint mutual information method was adopted to calculate the relationship between the prediction value of dynamic fluid level and the auxiliary variable. Setting an reasonable fluctuation range for joint mutual information, if the value is within the range, do not update the model, otherwise update it. Through simulation, the result showed that the method improved the prediction accuracy of dynamic fluid level.Considering the effect of noise, parameters, changes in working conditions on the prediction of dynamic fluid level, a sound dynamic fluid level forecasting system was established. By comparing simulation, the results show that the proposed method improve the prediction accuracy of the dynamic fluid level,and increase the automation level and production efficiency of oil exploration as a whole.