Construction of Cross Energy Type Data Model based on Spatiotemporal Data Mining

Abstract

In order to ensure the accuracy of oilfield development dynamic data, the author starts from analyzing the characteristics of development dynamic data, and conducts in-depth research on the characteristics of development dynamic data, the algorithm set for accuracy detection of development dynamic data, and comprehensive analysis methods. Firstly, in response to the spatiotemporal heterogeneity in developing dynamic data, combined with the design concept of a multi detector combination algorithm based on spatiotemporal mixed patterns, the accuracy detection algorithm is evaluated and selected. Based on this, the author proposes a development dynamic data accuracy detection method that considers the influence of multiple factors (FAGTN); Secondly, ARIMA, MGLN, STGCN, and FAGTN algorithms were selected as the algorithm sets for developing dynamic data accuracy detection, in order to complete the data accuracy detection based on monthly oil well data as the research object; Then, a combined weighting based analysis method was proposed to comprehensively analyze the accuracy detection results of dynamic data development, and the results showed: The dynamic data accuracy detection method based on ARIMA has the worst performance, with detection accuracy below 70% in different detection attributes, which is relatively not high enough; The development of dynamic data accuracy detection method based on MGLN achieved an accuracy rate of 80.53% when detecting sleeve pressure, but the accuracy rate did not reach 80% when detecting oil pressure, dynamic liquid level, monthly oil and water production, and the detection effect was relatively unstable; The accuracy of developing dynamic data accuracy detection methods based on STGCN fluctuates around 80%; Realize comprehensive evaluation of detection results; Finally, the experiment and evaluation of the comprehensive detection method for developing dynamic data accuracy were completed using real sample data.