Research On Wax Deposition Prediction Algorithm For Waxy Crude Oil Pipeline Based On Machine Learning

Wax deposition during pipeline transportation of waxy crude oil is an important problem affecting the safe operation of pipelines.In serious cases,it may even block the pipeline and threaten the safe transportation of crude oil.At the same time,when different kinds of crude oil are mixed,their rheological properties and wax evolution characteristics will change,and the wax deposition law of pipeline is more difficult to predict.Therefore,for different types of waxy crude oil and its mixed oil pipeline transportation process,it is of great significance to establish a wax deposition prediction model with high accuracy and good applicability to guide the safe operation of waxy crude oil pipeline.In this paper,based on the traditional wax deposition dynamics model,artificial neural network and support vector machine(SVM)algorithm,which have the advantage of non-linear processing,an intelligent wax deposition algorithm for waxy crude oil pipelines based on machine learning is studied.The specific research work is as follows:Firstly,Daqing crude oil,Russian crude oil and their mixed crude oil of different proportions were used as experimental oil samples to determine the basic physical and rheological parameters,including the relationship between viscosity and temperature,thermal parameters of wax precipitation and solidification point.On this basis,a laboratory small loop wax deposition simulation experiment device was used to study the influence of four factors on the wax deposition rate: oil wall temperature difference,oil temperature,flow rate and mixing ratio.Based on the grey relational theory,the primary and secondary relationships of all factors except mixing ratio were obtained as follows: oil temperature > flow velocity > oil wall temperature difference.Secondly,in the aspect of dynamics model,considering the influence of wax content in sediment during wax deposition and related uncertainties in wax molecular diffusion coefficient on wax deposition rate,in order to make the prediction more accurate,two wax deposition prediction models,RBF neural network and support vector machine,were used to predict wax deposition rate,respectively,with average relative error.Four indicators,mean absolute error,standard deviation and root mean square error,were used to compare and analyze the predicted results of three models including wax deposition kinetics model.The results show that the average relative error of the predicted value of the support vector machine model based on particle swarm optimization is the smallest and the predicted result is the closest to the experimental value.After optimization,the average relative error of the RBF neural network is reduced more greatly and the optimization effect is better.The four error indexes of the wax deposition depolarization coefficient method are larger than those of the other two models,and the dispersion degree of the predicted data is the largest.Finally,according to the comparative analysis results of the three models,particle swarm optimization algorithm is used to optimize the parameters of RBF neural network model and support vector machine model,and the optimized model is used to predict the wax deposition rate in 37 groups of experimental data.Compared with the non-optimized model,the optimized model has higher prediction accuracy and less dispersion of prediction data.? Based on this experimental basis,the wax deposition rate is predicted for the actual operation of QinghaiHarbin Oil Pipeline.Among them,the average relative error of the optimized RBF neural network and the support vector machine model decreased by 8.40% and 2.27%,respectively.The accuracy of the optimized RBF neural network is higher than that of the support vector machine,but the average relative error of the RBF neural network is still higher than that of the support vector machine.