Study On Early Warning Model Of Pumping Well Operation Based On Clusterin

With the rapid development of Internet of Things and computer technology,digitalization and intellectualization have become the main trend of oilfield development in China.With the installation of various sensors,the mass production data of pumping units are collected in large quantities.The acquisition of a large number of production data of pumping wells brings new opportunities and challenges to oil and gas production.How to make use of the massive data generated in the production process of oil production is particularly important.Based on the multivariate data of pumping units collected on the spot,this paper studies the algorithm which accords with the characteristics of oilfield data,and applies it to the identification of pumping units' working conditions and the early warning of pumping wells' working conditions.Based on the three main sub-nodes of pumping pump,sucker rod and ground motor,the mathematical model of pumping unit under different working conditions is established.The original production data of pumping units are pretreated and converted into secondary data which can better reflect the production status of pumping units.Aiming at the characteristics of production data of pumping wells,such as large amount of data,large number of unlabeled samples and serious imbalance of data,an algorithm framework based on clustering sampler is proposed.Under this framework,the early warning model of pumping wells and the fault identification model of pumping wells are established.Through field data acquisition model performance experiments,the effects of data imbalance on the performance of fault identification models generated by different algorithms are studied.The performance differences between the models based on indicator diagram feature perspective,electric parameter feature perspective and full-view feature perspective are studied.The performance of the early warning model and fault identification model generated by different samplers is also studied.The performance of condition warning model and fault identification model generated by different basic classifiers.The results show that the working condition early warning model and fault identification model based on cluster sampler have good performance.Light GBM-Weight algorithm is used as the basic identifier to generate condition warning model and fault identification model,which can effectively avoid the performance degradation of the model caused by data imbalance.Compared with the model based on single-view feature,the model based on full-view feature training can better identify the working state of pumping wells.