Volume 66

Study on Health Index Evaluation and Prediction Model of Oil and Water Well Casings Laize Liu, Chaodong Tan, Ao Li

Abstract

Aiming at the problems of untimely monitoring, difficult quantitative assessment, and inaccurate prediction of oil and water well casings, this study develops two models: a casing Health Index (HI) calculation model and an HI time-series prediction model. First, the main influencing factors of casing damage were screened and identified. Then, a casing HI model based on the weighted fusion of Principal Component Analysis (PCA) and Mean Decrease Impurity (MDI) importance values was constructed. Furthermore, research on the time-series prediction model of oil and water well casing HI based on Long Short-Term Memory (LSTM) network was conducted. Using 87 production intervals (18 damaged) from 13 casings in an oilfield of PetroChina, 14 key parameters were identified to calculate and predict HI. Results show: the HI model realizes quantitative health risk evaluation; the LSTM model captures the temporal trend of HI with high accuracy. This study realizes real-time, quantitative, and accurate assessment and prediction of the service operation health status of oil and water well casings, and provides good guidance for the early warning of oil and water well casing damage.

Keywords oil and water well casings, analysis of key control parameters, health index, long short-term memory network, prediction model