Research On The Analysis Method Of Oil Pumping Machine Indicator Diagram Based On Machine Learning

Oil pumping is the mainstream production method in the oilfield at this stage.In the process of oilfield extraction,the harsh working environment and complex working conditions may cause pumping well failure and affect the oil extraction efficiency.The indicator diagram can directly reflect the working condition of the pumping unit well,and it is a common method to analyze and study it for pumping unit well fault diagnosis.However,the traditional analysis of indicator diagrams mainly relies on manual experience,and the workload is large and the recognition accuracy is easily affected by human factors,which is difficult to meet the needs of intelligent oil fields.Therefore,this dissertation takes the pumping machine indicator diagram as the research object and conduct research on the analysis method of pumping machine indicator diagram based on machine learning.Firstly,six types of indicator diagrams were selected for the construction of the dataset,and the indicator diagrams were pre-processed,including image normalization,grayscale,binarization,and edge detection,and the indicator diagrams are converted into filled images with the largest boundary area.Through the above image preprocessing operations,the quality of the dataset was improved;Then,three convolutional neural networks,Alex Net,Goog Le Net and Res Net-34,were used to construct the analysis models of oil pumping machine indicator diagram,and to conduct training learning and comparison analysis of experimental results;In order to further improve the recognition accuracy,the best performing Res Net-34 network among the three models is trained by transfer learning and optimized by incorporating CBAM attention mechanism.Finally,we designed and developed an intelligent recognition system for oil pumping machine indicator diagram using Java as the development language,and realized the visualization of model training results and recognition of indicator diagrams.The experimental results show that the accuracy of the analysis models based on Alex Net,Goog Le Net,and Res Net-34 is 83.6%,85.4%,and 90.2%,respectively,while the accuracy of the optimized Res Net-34 network reaches 94.7%,which is 4.5 percentage points higher than that of the unoptimized network,and the convergence speed of the model is faster,which proves the effectiveness and feasibility of this optimization This proved the effectiveness and feasibility of this optimization scheme,which can help the staff to quickly identify the type of indicator diagrams,so as to achieve timely diagnosis of pumping machine well faults,and has certain application value.