A novel method for prediction of paraffin deposit in sucker rod pumping system based on CNN indicator diagram feature deep learning

Abstract

Paraffin deposit is a major problem associated with water injection in waxy heavy oil reservoirs, therefore accurately and effectively predicting the development of paraffin deposit can avoid some major faults, such as stuck pump, and ensure the stability of the system. Paraffin deposit, as a gradual changing fault, shows a clear trend of development. Based on the convolutional neutral network (CNN) combined with deep learning, the indicator diagram is extracted to calculate the growth rate, and then the timing of serious paraffin problems can be predicted. The paper firstly analyzes and calculates the force of the column in the process of paraffin deposition and obtains the theoretical maximum load in the upper stroke and the theoretical minimum load in the lower stroke, then the load difference is selected as the parameter to characterize the paraffin deposit. In a case study, an experimental well is selected to determine maximum load difference in paraffin deposit process and load difference in normal production, and then the percentage of paraffin deposit is determined based on the growth law as the label of each indicator diagram. The structure of convolutional neural network is designed, and the accuracy rate is taken as the index to optimize the matching scheme of convolutional neural network structure parameters. Finally, the convolution neural network is trained by using training sets of 1.3 million well data. This method is an effective way to predict the paraffin deposit with CNN combined with deep learning and indicator diagram with an accuracy rate of 98.08% and loss less than 0.633. The prediction method in this paper can facilitate the intelligent management of oilfield safety and ensure the efficient production of pumping system.