Soft Sensor Of Moisture Content Of Crude Oil Based On Multi-model Gaussian Process Regression With Mixed Kernel

In the process of oil well production,many process variables are difficult to be acquired online due to technology or cost issues.However,these process variables are also related to the oil production efficiency and fluid production of oil wells,and they are important parameters to control and optimize the oil production process.In order to solve this problem in production,the soft sensor technique is adopted to predict the process variables.By establishing the function relationship between the auxiliary variables and the objective variables,the difficult objective variables can be predicted by the easily obtained auxiliary variables.In the current oil production process,the measurement of the moisture content of the output oil is mainly realized by offline measurement and online measurement.However,the offline measurement has many disadvantages,such as high cost,high data delay and easy to be affected by the sampling method.The methods of online measurement mostly adopt the methods of centrifugal and electric dewatering by using equipment.It is difficult to guarantee the accuracy of measurement results because of the poor stability.Aiming at the above problems,this paper proposes a soft sensor model for moisture content of crude oil based on the multi-model mixed kernel Gaussian process regression.An adaptive variable population fruit fly optimization algorithm based on information entropy is proposed to solve the problem.The truncation distance of peak density clustering is optimized by the adaptive variable population fruit fly optimization algorithm based on information entropy to avoid the truncation distance from affecting the clustering effect.A Gaussian process regression model based on mixed kernel is established to improve the generalization ability of Gaussian process regression model.The proposed method is verified with the historical production data of oil wells,and it is proved that the proposed soft measurement model has better prediction accuracy for the crude oil moisture content.The main research contents are as follows:1)Aiming at the problem that drosophila optimization algorithm is easy to fall into local optimality and the optimization accuracy is not high,a variable population fruit fly optimization algorithm based on information entropy is proposed.In order to verify the effectiveness of the optimization algorithm,the effectiveness of the proposed optimization algorithm is proved by the simulation experiments.2)Gaussian process regression based on single kernel function has some limitations in the prediction of complex industrial system parameters.Data with different characteristics often require different kernel functions for processing.In order to optimize the regression performance of Gaussian process regression,a mixed kernel function composed of square exponential covariance kernel function,periodic covariance kernel function and Matérn kernel function is used to replace the single kernel function.The generalization ability of Gaussian process regression model is increased and the prediction accuracy is improved.The optimal selection of the hyper-parameters in the mixed kernel is carried out by the drosophila optimization algorithm instead of the maximum likelihood estimation,which avoids the local optimal solution caused by the sensitivity of the maximum likelihood estimation to the initial value.The superiority of mixed kernel Gaussian process regression is verified by experiments comparing with other soft sensor models.3)In view of the shortcomings of the method of measuring the moisture content of crude oil at present,a multi-model soft sensor model based on the regression of mixed kernel Gaussian process for estimating the moisture content of crude oil is proposed in this paper.A variable population fruit fly optimization algorithm is used to optimize the key parameters of peak density clustering to avoid the influence of the key parameters on the number of clusters.A mixed kernel Gaussian process regression sub model is established for the clustering sub data set and the global output is formed by Bayesian strategy.The production data of a domestic oil field is obtained by the well acquisition equipment and the crude oil moisture content estimation is simulated and verified.Comparing with other soft sensor models,it is proved that the model proposed in this paper can be better applied to the prediction of moisture content of crude oil.