Intelligent monitoring method for working conditions of sucker rod pump based on reinforcement learning of convolutional neural network (CNN)

Abstract

The invention relates to an intelligent monitoring method for working conditions of a sucker rod pump based on reinforcement learning of a convolutional neural network (CNN). A monitoring system mainly comprises three parts: construction of the CNN for monitoring the working conditions of the sucker rod pump, actual application of the CNN for monitoring the working conditions of the sucker rod pump as well as learning and update of the CNN for monitoring the working conditions of the sucker rod pump. According to the monitoring method disclosed by the invention, by using the CNN, loss of a great deal of useful information in the numeralization process of an indicator diagram obtained by feature extraction is avoided; the CNN can directly identify inputted indicator diagram pictures so as to reduce errors; meanwhile, on the basis of reinforcement learning, the growth opportunity of a CNN intelligent monitoring system can be effectively improved; accuracy of the working conditions is improved in the repeated working condition diagnosis and reinforcement learning process; possible errors of comparison of a small amount of sample data with a large amount of actual production data is avoided, and the effects that the more the CNN is used, the more intelligent the CNN is, and the more the CNN is used, the easier the CNN is are realized.

Description

technical field [0001] The invention relates to the technical field of petroleum engineering, in particular to intelligent diagnosis of working conditions of rod pumps in pumping wells, in particular to an intelligent monitoring method for working conditions of rod pumps based on convolutional neural network reinforcement learning. Background technique [0002] In the process of oilfield exploitation, the rod pumping method is the most widely used method, accounting for 75% of the total number of oil wells in the world. The analysis and diagnosis of the working condition of the rod pump based on the indicator diagram is a very important part, because the oil pumping The machine dynamometer diagram reflects the working state of the oil system in a concentrated manner, and contains rich information on pumping units, pumps, and rods. At present, there are many diagnostic methods based on the dynamometer diagram, such as: expert system, fuzzy mathematics, gray theory and neural n...

AI Technical Summary

Problems solved by technology

[0002] In the process of oilfield exploitation, the rod pumping method is the most widely used method, accounting for 75% of the total number of oil wells in the world. It is a very important part to diagnose the working condition of the rod pump based on the indicator diagram analysis, because the pumping The machine dynamometer diagram reflects the working state of the oil system in a concentrated manner, and contains rich information on pumping units, pumps, and rods. At present, there are many diagnostic methods based on the dynamometer diagram, such as: expert system, fuzzy mathematics, gray theory and neural network. In essence, these methods are a problem of pattern recognition and classification, and there are several common defects: for the working conditions with similar characteristics of the dynamometer diagram, it is difficult to accurately diagnose the specific fault type only by the dynamometer diagram; The process of extraction and numericalization of the indicator diagram loses a lot of effective information; it is necessary to compare the extracted feature value with the feature knowledge working condition database to obtain the diagnosis result. Due to the lack of growth of the technology, there are errors in this process, and the accuracy rate is difficult. improve

Images