A method and system for identifying working conditions of pumping wells based on multi-view learning

Abstract

The invention discloses a method and system for identifying the working condition of an rod pumped well on the basis of multi-view learning and belongs to the technical field of working condition diagnosis. A multi-view learning approach based on a large-data production environment is adopted, an actually measured ground indicator diagram and an electric power signal are adopted as two different main viewing angles, and an actually measured wellhead temperature value and a wellhead pressure signal are adopted as two auxiliary viewing angles; the four actually measured viewing angles are effectively combined and utilized, and a working condition identification model is established by using a small amount of rod pumped well working condition data or combining with a large amount of unknown working condition data, wherein the established working condition identification model effectively integrates a Hessian regularization and multi-view angle learning approach, and large data and massivemulti-source real-time information acquired by a rod pump oil extraction and production system in an environment of oil gas production internet of things can be fully utilized. The method and the system break through the limitation to identification of the working condition of the rod pumped well through a single information source and the technical bottleneck of a traditional multi-source information identification method, so that the precision rate of identifying the working condition of the rod pumped well is further increased, and the engineering practicability is improved.

Description

technical field [0001] The invention relates to the technical field of working condition diagnosis, in particular to a method and system for identifying working conditions of pumping wells based on multi-view learning. Background technique [0002] In the existing pumping well working condition identification methods, most of the working condition identification methods are based on the dynamometer identification technology, mainly using the pump dynamometer diagram or the measured ground dynamometer diagram combined with the artificial intelligence method to identify the working condition; The identification method based on electrical parameters mainly uses the electrical power diagram or the measured electrical parameters to identify the working condition; there are few identification methods based on multi-source data, and the identification method based on multi-source data mainly uses the pump power diagram combined with oil well production information (production, swab...

AI Technical Summary

Problems solved by technology

[0002] In the existing pumping well working condition identification methods, most of the working condition identification methods are based on the dynamometer identification technology, mainly using the pump dynamometer diagram or the measured ground dynamometer diagram combined with the artificial intelligence method to identify the working condition; The identification method based on electrical parameters mainly uses the electrical power diagram or the measured electrical parameters to identify the working condition; there are few identification methods based on multi-source data, and the identification method based on multi-source data mainly uses the pump power diagram combined with oil well production information (production, swabbing parameters, well condition data, etc.) for working condition identification

[0003] The working condition identification method in the prior art has achieved good results in the production of pumping wells, but there are still the following limitations: (1) In a complex nonlinear system coupled with electromechanical and hydraulic systems, a single information source is used to judge oil well operation. (2) Affected by the damping coefficient and the "division by zero" problem, the pump power diagram and electric power diagram calculated by the model will affect the calculation of the accuracy of the characteristic parameter values; (3) affected by the traditional multi-feature connection recognition The recognition effect and model robustness of the existing working condition recognition methods based on multi-source data are affected by the technical limitations of the method, the technical limitations of the massive real-time data acquisition and storage of oil wells in the early stage, the complex and changeable well conditions, and the unreliable artificial statistical data. (4) The working condition identification method in the prior art requires a large number of labeled working condition training samples, but it is difficult and costly to obtain labeled working condition samples in actual engineering, while the method of unlabeled sample training often identifies poor precision

[0005] Affected by the above limitations, the actual application effect of the existing working condition identification methods in the production of pumping wells is not ideal; in addition, the existing technologies are seriously lacking in the establishment of effective fusion of multiple measured information sources suitable for big data production environments. Therefore, it is urgent to research and develop a pumping well working condition identification method and system to improve the recognition accuracy and practicability, and to solve the problem that the pumping well working condition identification method in the prior art is not conducive to the construction of intelligent oilfield production and technical issues of development

Images