Research On Soft Measurement Method Of Dynamic Liquid Level Of Sucker Rod Pumping Well

At present,sucker rod pumping is widely used in oil fields at home and abroad.Dynamic liquid level is an important parameter.On the one hand,it is important to master the production status of oil wells,maintain the balance of supply and discharge relationship of oil wells,and ensure high production and high pump efficiency of oil wells.On the other hand,it can be changed according to dynamic liquid level,timely adjust the oil well parameters to ensure stable and safe operation of the oil well.In view of the problems of low security,poor real-time performance and low measurement accuracy of the current dynamic liquid level detection methods,this paper introduces soft-sensing technology into the dynamic liquid level detection of pumping wells,and establishes a dynamic liquid level soft-sensing model based on Least Squares Support Vector Regression to realize the real-time detection of dynamic liquid level.The main work of this paper is summarized as follows:First of all,combined with the production parameters obtained directly or indirectly during the oil recovery process of the sucker rod pump,this paper uses the grey correlation analysis method to determine the auxiliary variables and preprocesses the data,aiming at the problem of low generalization performance of the model established by single kernel function,the radial basis function and polynomial kernel function are selected to construct linear mixed kernel function,and the dynamic liquid level soft measurement model based on Least Squares Support Vector Regression of single kernel function and mixed kernel function is established respectively.The simulation experiment proves that the model based on mixed kernel function and Least Squares Support Vector Regression is feasible.The Least Squares Support Vector Regression model of the kernel function has better prediction performance.Secondly,in the regression model built by the mixed kernel function,the penalty parameters,radial basis kernel function parameters and mixed weight coefficients have an important influence on the learning ability and generalization ability of the model.Therefore,this paper uses the Fruit fly optimization algorithm to optimize the parameters in the model.Aiming at the problem that the fruit fly algorithm has a slow convergence speed and is easy to fall into the local optimum,the algorithm is improved by introducing adaptive step size and mutation operation.The improved fruit fly optimization algorithm was applied to the dynamic liquid level soft-measurement model,and compared with genetic algorithm,particle swarm algorithm and basic Fruit fly algorithm.The simulation verified the effectiveness of the improved fruit fly algorithm.Finally,due to the changing conditions,new data must be added continuously.Therefore,the model should be updated to establish a new model that reflects the current situation to ensure the validity of the model.This paper is based on the Least Squares Support Vector Machine algorithm of online learning,and is improved on the basis of it.It is applied to the dynamic liquid level soft measurement model to realize the online update of the model,which further improves the prediction accuracy and robustness of the model.